University of Groningen Missed opportunities? Germany and the transatlantic labor-productivity gap, 1900-1940 Veenstra, Joost IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below. Document Version Publisher's PDF, also known as Version of record Publication date: 2014 Link to publication in University of Groningen/UMCG research database Citation for published version (APA): Veenstra, J. (2014). Missed opportunities? Germany and the transatlantic labor-productivity gap, 1900- 1940. Groningen: University of Groningen, SOM research school. Copyright Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons). Take-down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum. Download date: 10-09-2020
245
Embed
University of Groningen Missed opportunities? Germany and the … · 2016-03-08 · Missed Opportunities? Germany and the Transatlantic Labor-Productivity Gap, 1900 1940 PhD thesis
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
University of Groningen
Missed opportunities? Germany and the transatlantic labor-productivity gap, 1900-1940Veenstra, Joost
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite fromit. Please check the document version below.
Document VersionPublisher's PDF, also known as Version of record
Publication date:2014
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):Veenstra, J. (2014). Missed opportunities? Germany and the transatlantic labor-productivity gap, 1900-1940. Groningen: University of Groningen, SOM research school.
CopyrightOther than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of theauthor(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons thenumber of authors shown on this cover page is limited to 10 maximum.
lasted up to the late 1940s. It was not before 1947 that the dynamics reversed and Ger-
many managed to close in on American levels of labor productivity. Other European
countries shared Germany’s relative backwardness before WW2 and across the Atlantic
a large labor-productivity gap persisted from the late nineteenth century up until the
post-WW2 period.
Figure 1.2 shows that the transatlantic labor-productivity gap manifested also at the
country level. As such, it is a main feature of economic development in the early twenti-
eth century. The persistence, widening even, of the gap is striking and points at the pres-
ence of systematic growth determinants that long-lastingly influenced economic devel-
opment. Of particular interest in this respect is the timing of the labor-productivity gap;
the emergence of the gap coincided with a period of rapid technological development,
a time also referred to as the second industrial revolution.6 If the “Great Inventions”
of the second industrial revolution determined the growth dynamics of the post-1870
period, differences between countries in the adoption of new technologies may explain
the pattern of diverging development.7 Manufacturing industries, employing about 30%
6. R. Lipsey, K. Carlaw, and C. Bekar, Economic Transformations: General Purpose Technologiesand Long-Term Economic Growth (Oxford University Press, 2005).
7. R. Gordon, “Is U.S. Economic Growth Over? Faltering Innovation Confronts the Six Headwinds,”
�
�
“prnt˙thesis” — 2014/1/22 — 12:20 — page 3 — #17�
�
�
�
�
�
Chapter 1. Introduction 3
of the labor force in developed countries during the first half of the twentieth century,
proved particularly receptive to technological change.8 Because many new technologies
were embodied in capital, manufacturing industries could reap the labor-productivity
benefits associated with innovation by adopting modern machinery.9
Figure 1.2: GDP per capita levels (1,000 $1990)
0
2
4
6
8
10
12
14
1870 1913 1929 1938 1950
Europe US
Sources: Bolt and van Zanden, “The First Update of the Maddison Project.” Europe is calculated on
the basis of British, French and German data.
The comparatively high pace of growth in America suggests that the US success-
fully caught the winds of change, while Europe spilled them. This notion of missed
opportunities implies a latent growth potential that Germany failed to fully explore.
In its 2007 Global Economic Prospects report, the World Bank hints at such a latent
and partially unused capacity for growth in Europe.10 The World Bank examined the
historical growth record of G-5 countries and on the basis of GDP data for the thirty
years running up to 1900 ‘predicted’ economic growth for the 50 following years. The
predicted rate of GDP growth turned out significantly higher than the G-5 countries
Centre for Economic Policy Research Policy Insight No. 63 (2012): 5.8. For employment shares, see B. Mitchell, International Historical Statistics. Europe 1850–2005.
Sixth Edition (London: Macmillan, 1951), 153–164 and M. O’Mahony, Britain’s Productivity Perfor-mance, 1950–1996; An International Perspective (National Institute of Economic / Social Research,1999), 12.
9. H. Jerome, Mechanization in Industry (National Bureau Economic Research, 1934); S. Schurr etal., Electricity in the American Economy. Agent of Technological Progress (Greenwood Press, 1990);W. Devine, “From Shafts to Wires: Historical Perspective on Electrification,” Journal of EconomicHistory Vol. 43, No. 2 (1983): 347–372.10. The World Bank, “World Bank Report: Challenge of Geopolitical Shifts for Long-Term Economic
Forecasts: Lessons of History,” Global Economic Perspectives. Managing the Next Wave of Globaliza-tion (2007): 55.
�
�
“prnt˙thesis” — 2014/1/22 — 12:20 — page 4 — #18�
�
�
�
�
�
4 Missed Opportunities?
actually realized during the first half of the twentieth century, a failure fully attributable
to a slowdown in Europe. As the prediction essentially projects the nature of annual
shocks between 1870–1899 on the years afterward, the discrepancy between forecasted
and realized growth implies that after 1900 Europe deviated from its long-run devel-
opment path. Because the forecasted growth trajectory captures the latent production
capacity, the deviation from it suggests that part of that potential remained unrealized.
Had Europe managed to continue after 1900 as before, the gap to the US would have
turned out considerably smaller in the early twentieth century.
For the case of Germany, the notion of missed opportunities does not correspond
well with the conventional outlook on historical development. The long phase of labor-
productivity divergence before WW2 defies the traditional, and largely qualitative, lit-
erature that attributed special features to the German growth experience. Adam Tooze
articulates this tension between the two strands of literature as follows:
“Was there anything peculiar about Germany’s experience of economic
growth? This seems to me to be a question that though obvious and once a
classic topic for student essays is in fact in need of reassessment. Certainly
in many accounts of Germany’s uneven modernization there was a strong
assumption that the modernity of its economy at least was not in ques-
tion. Indeed, in some interpretations of Europe’s economic development,
claims were made for a peculiar sophistication of the German economy. And
yet from a vantage point at the end of the twentieth century Germany’s
long-run economic trajectory surely looks less distinctive than previously
thought. During the era of steel, chemicals and heavy electrical engineering
German industrialism was no doubt surrounded by a formidable aura. And
it certainly was a considerable industrial competitor. However, even then
these dramatic elements of industrialism formed only a part of economic life
in Germany. And their status as defining elements of economic modernity
was not set to last.”11
The question, then, is how the paradoxical lack of fast labor-productivity growth
at a time of fast technological change ought to be perceived. Did the widening labor-
productivity gap result from a German failure to successfully ride the waves of techno-
logical change? With regard to German manufacturing in the early twentieth century,
this question is addressed in the present study.
11. A. Tooze, “Do We Need a New Economic History of Germany?,” H-Net Online, June 2007,www.h-net.msu.edu.
�
�
“prnt˙thesis” — 2014/1/22 — 12:20 — page 5 — #19�
�
�
�
�
�
Chapter 1. Introduction 5
Two issues receive particular attention. First, given the ‘plague of low-quality data’,
attention goes out to an accurate measurement of German labor-productivity levels
in manufacturing on the level of industries. This involves a critical review of both
the data and the measurement techniques traditionally used by researchers studying
historical labor-productivity patterns. Secondly, the contribution of technological
change to Germany’s labor-productivity performance is studied and its contribution to
the widening German/US labor-productivity gap quantified.
Chapter 2 sets the stage for this study by measuring German/US comparative
labor productivity in manufacturing industries for the years 1909 and 1936/35. As
the quality of the time series provided by the German Historical National Accounts
has been called into question, new estimates of labor-productivity growth in Ger-
man manufacturing are necessary.12 For this purpose state-of-the-art techniques are
employed to construct benchmarks of comparative labor productivity. To allow for
effects of composition and to capture inter-industry variation in performance, the
labor-productivity comparisons constructed in this study apply the industry-of-origin
approach, which breaks down the manufacturing sector in manufacturing industries.
Moreover, following the literature, German and US output values are converted to
a common currency using industry-specific purchasing power parities to enable an
international comparison.13 These measurement techniques have previously been
applied only to bilateral comparisons between the US/UK and Germany/UK for years
prior to WW1 and WW2, but never for a study of German/US labor-productivity
differences in periods before 1950.14
12. W.G. Hoffmann, Das Wachstum der Deutschen Wirtschaft Seit der Mitte des 19. Jahrhunderts(Berlin: Springer-Verlag, 1965); R. Fremdling, “German National Accounts for the 19th and Early 20thCentury: A Critical Assessment,” Vierteljahrschrift fur Sozial- und Wirtschaftsgeschichte Vol. 75, no. 3(1988): 339–357; A. Ritschl, “Spurious Growth in German Output Data, 1913–1938,” European Reviewof Economic History Vol. 8 (2004): 201–223.13. A. Maddison and B. van Ark, “Comparison of Real Output in Manufacturing,” Policy, Planning
and Research Working Papers Vol. 5 (1988): 1–33; B. van Ark, International Comparisons of Out-put and Productivity: Manufacturing Productivity Performance of Ten Countries from 1950 to 1990(Groningen: Groningen Growth / Development Centre, 1993), 1–233; B. van Ark and M.P. Timmer,“The ICOP Manufacturing Database: International Comparisons of Productivity Levels,” Interna-tional Productivity Monitor No. 3 (2001): 44–51; R. Inklaar and M. Timmer, “GGDC ProductivityLevel Database: International Comparisons of Output, Input and Productivity at the Industry Level.,”GGDC Research Memorandum No. 104 (2008): 1–81.14. Broadberry, The Productivity Race; S.N. Broadberry and D. Irwin, “Labor Productivity in the
United States and the United Kingdom During the Nineteenth Century,” Explorations in EconomicHistory Vol. 43 (2006): 257–279; S.N. Broadberry and C. Burhop, “Comparative Productivity in Britishand German Manufacturing Before World War II: Reconciling Direct Benchmark Estimates and TimeSeries Projections,” The Journal of Economic History Vol. 67 (2007): 315–349; R. Fremdling, H.J.de Jong, and M.P. Timmer, “British and German Manufacturing Productivity Compared: A NewBenchmark for 1935/36 Based on Double Deflated Value Added,” The Journal of Economic HistoryVol. 67, no. 2 (2007): 350–378; H.J. de Jong and P.J. Woltjer, “Depression Dynamics: a New Estimate
�
�
“prnt˙thesis” — 2014/1/22 — 12:20 — page 6 — #20�
�
�
�
�
�
6 Missed Opportunities?
The benchmarks uncover a large labor-productivity gap in both periods at the level
of total manufacturing, confirming earlier studies by, among others, Broadberry.15 The
variation of comparative performance on the industry level is substantial, however, and
shows that the diverging trend for total manufacturing described in figure 1.1 fails to
do justice to the dynamics in several underlying industries. This applies particularly to
German chemicals, textiles and primary metals, which displayed a labor-productivity
performance close to the level of their American counterpart. Not surprisingly, general
theories as regards the German-American productivity gap have difficulty accounting
for the cross-industry variation. There are nonetheless some patterns recognizable. For
one, production in the strong performing industries involves mainly goods used as in-
termediates in other industries. For instance, the pig iron obtained in primary metals is
further processed by fabricated-metals and transportation-equipment industries, while
the yarn and thread produced in spinning industries function as inputs for weaving in-
dustries. European markets have been associated with heterogeneous demand patterns,
which discouraged the adoption of standardized production processes, but industries
involved in the production of mainly basic goods may not have suffered from this.16
The comparatively strong performing German manufacturing industries share an-
other characteristic as well. There appears to be an overlap between German industries
with relatively high labor-productivity levels and those associated in the literature with
relatively large establishment size and a high degree of vertical integration. Both phe-
nomena are associated with economies of scale, which possibly endowed industries with
relatively high labor-productivity levels.17 Although in general the scale of production
and the degree of vertical integration was much smaller in Germany compared to the
US, the relatively strong-performing German industries lagged only little behind.18 Nev-
ertheless, these are necessary but not sufficient conditions for catch-up and conceivably
of the Anglo-American Manufacturing Productivity Gap in the Interwar Period,” Economic HistoryReview Vol. 64 (2011): 472–492.15. Broadberry, The Productivity Race.16. L. Rostas, “Industrial Production, Productivity and Distribution in Britain, Germany and the
United States,” The Economic Journal Vol. 53, no. 1 (1943): 39–54, 58-59; A. Chandler, Scale andScope: the dynamics of industrial capitalism (Harvard: Belknap Press of Harvard University Press,1990), 1–780, 47; D.S. Landes, The Unbound Prometheus: Technological Change and Industrial Devel-opment in Western Europe From 1750 to the Present (Cambridge University Press, 1969), 247; S.N.Broadberry, “Technological Leadership and Productivity Leadership in Manufacturing Since the Indus-trial Revolution: Implications for the Convergence Debate,” The Economic Journal Vol. 104 (1994):291–302, 291.17. L. Hannah, “The American Mircale, 1875–1950, and After: A View in the Europan Mirror,”
Business and Economic History Vol. 24, no. 2 (1995): 197–220; L. Hannah, “Logistics, Market Size,and Giant Plants in the Early Twentieth Century: A Global View,” Journal of Economic History Vol.68, no. 1 (2008): 46–78.18. J. Kinghorn and J. Nye, “The Scale of Production in Western Economic Development: A Com-
parison of Official Industry Statistics in the United States, Britain, France, and Germany, 1905-193,”Journal of Economic History Vol. 56, no. 1 (1996): 90–112.
�
�
“prnt˙thesis” — 2014/1/22 — 12:20 — page 7 — #21�
�
�
�
�
�
Chapter 1. Introduction 7
explain some, but certainly not all the observed variation in comparative labor produc-
tivity. For instance, the paper and cement industries, both of which can be argued to
produce standard goods, failed to match the high labor-productivity level vis-a-vis the
US displayed by textiles.
Other factors must have been at work also. Perhaps most importantly in this respect
are differences in the mix of factor inputs employed in production. A possible and
frequently-used explanation for observed labor-productivity differences between the US
and the UK in the nineteenth century has been put forward in the Rothbarth-Habakkuk
thesis, which emphasizes the importance of factor endowments for the capital-labor ratio
at which countries choose to operate.19 In the US a scarcity of skilled labor and an
abundance of natural resources provided an incentive to substitute machinery for labor.
This minimized costs and led to a capital-intensive production process. The supply
of factor inputs faced by European producers differed, which induced the adoption of
less capital-intensive technology. As some determinants of relative factor costs, such
as the availability of natural resources or the size and density of the population, are
exogenous to the production process, a country’s initial conditions influence the choice
of technology. In the extreme, if one assumes that these fixed initial conditions fully
determine the relative factor costs and the choice of factor-input mix, the existence of
different technological paths across the Atlantic was foreordained.20
If technological progress is directed toward the technology currently used by coun-
tries, differences in relative factor costs lead to technological lock-in.21 This begs the
question whether such path dependencies effectively blocked the traditional channels
of labor productivity catch-up described by standard neo-classical growth theories?
Provided that the necessary capabilities and resources are available (Gerschenkron’s
idea of ‘appropriate’ economic institutions and Abramovitz’ ‘social capabilities’)
countries distanced far away from the technological frontier can catch-up quickly by
importing or imitating technologies that are already in use in developed countries.22
19. E. Rothbarth, “Causes of the Superior Efficiency of U.S.A. Industry as Compared with BritishIndustry,” The Economic Journal Vol. 56, no. 223 (1946): 383–390; M Abramovitz, “Resource andOutput Trend in the United States Since 1870,” American Economic Review Vol. 63, No. 2 (1956):5–23; H.J. Habakkuk, American and British Technology in the Nineteenth Century. The Search forLabour-saving Inventions (Cambridge: Cambridge University Press, 1962), 1–222.20. N. Rosenberg, “Why in America?,” in Exploring the Black Box. Technology, Economics, and
History (Cambridge University Press, 1994), 109–120.21. P. David, Technical Choice, Innovation and Economic Growth. Essays on American and British
Experience in the Nineteenth Century (Cambridge: Cambridge University Press, 1975), 1–334. Theclassic reference for path dependency is P. David, “Clio and the Economics of QWERTY,” AmericanEconomic Review Vol. 75, no. 2 (1985): 332–327.22. A. Gerschenkron, Economic Backwardness in Historical Perspective; A Book of Essays (Cam-
bridge University Press, 1962); M. Abramovitz, “Catching-up, Forging Ahead and Falling Behind,”Journal of Economic History Vol. 46 (1986): 385–406.
�
�
“prnt˙thesis” — 2014/1/22 — 12:20 — page 8 — #22�
�
�
�
�
�
8 Missed Opportunities?
Also the recent economic-growth literature has often emphasized the importance of
investment-based strategies for follower countries.23 Yet if follower countries refrain
from adopting advanced technology because of different relative factor costs, Europe
may have been trapped on a labor-intensive and low-productive technological path.
For the case of Germany, recent studies have concluded otherwise, though. Re-
search on the machine-tool industry during the interwar years revealed a process of
technology adoption on the part of Germany and finds that at the outbreak of WW2
capital-intensity levels were as high as in the US.24 The tradition of copying and
adopting American machinery contests the notion of technological lock-in. But how
can large labor-productivity differences coexist with a diminishing machine-intensity
gap? Chapter 3 addresses this paradox by specifically accounting for the contribution
of machine-intensity differences to the German/US labor-productivity gap in 1936/39.
For this purpose I use data envelopment analysis techniques, which offer several
advantages over traditional Solow-based level accounting exercises. First, the analysis
allows for localized innovation, a main feature of technological progress ever since
the first industrial revolution.25 Second, the data envelopment analysis involves a
non-parametric approach and, as such, does not require information on capital and
labor prices to proxy the marginal factor returns.26 Third, the analysis uses horse
power per hour worked, which offers a more accurate indicator of machine intensity
than the total capital stock per employee statistics conventionally employed.27
The data envelopment analysis applied here estimates a global best-practice frontier
and the change thereof between 1899–1939. This best-practice frontier indicates for each
point in the range of operated capital-labor ratios the highest labor-productivity level
contemporaneously or previously attained. Positioning German and American manu-
facturing industries in relation to the global best-practice frontier permits a decompo-
sition of the labor-productivity gap in components of capital intensity and efficiency.
23. P. Aghion, “Higher Education and Innovation,” Perspektiven der Wirtschaftspolitik Vol. 9 (2008):28–45; Daron Acemoglu, “Directed Technical Change,” The Review of Economic Studies Vol. 68, no. 4(2002): 781–809; J. Vandenbussche, P. Aghion, and C. Meghir, “Growth, Distance to the Frontier andComposition of Human Capital,” Journal of Economic Growth Vol. 11 (2006): 97–127.24. C. Ristuccia and A. Tooze, “Machine Tool and Mass Production in the Armaments Boom: Ger-
many and the United States, 1929–44,” Economic History Review Vol. 66, no. 4 (2013): 953–974.25. R.C. Allen, “Technology and the Great Divergence: Global Economic Development Since 1820,”
Explorations in Economic History 49 (2012): 1–16.26. S. Kumar and R. Russell, “Technological Change, Technological Catch-up, and Capital Deepening:
Relative Contributions to Growth and Convergence,” The American Economic Review Vol. 92, no. 3(2002): 527–548; M.P. Timmer and B. Los, “Localized Innovation and Productivity Growth in Asia:An Intertemporal DEA Approach,” Journal of Productivity Analysis Vol. 23 (2005): 47–64.27. A.J. Field, “On the Unimportance of Machinery,” Explorations in Economic History Vol. 22
(1985): 378–401.
�
�
“prnt˙thesis” — 2014/1/22 — 12:20 — page 9 — #23�
�
�
�
�
�
Chapter 1. Introduction 9
The former element captures the difference between the labor-productivity level at
the frontier for German and American industries. As such, it measures the difference in
labor-productivity potential between machine-intensity levels operated in Germany and
the US. This difference in labor-productivity potential does not fully account for the
observed variation in labor-productivity levels for the reason that industries exploited
their labor-productivity potential only partly. The extent to which an industry manages
to exhaust its labor-productivity potential is expressed by the second component, i.e.
efficiency. The difference between the efficiency level in Germany and the US accounts
for the remainder of the labor-productivity gap that is left unexplained by variation
in the labor-productivity potential at the level of machine intensity explored in both
countries.
The labor-productivity gap decomposition shows that it was not a low machine-
intensity level that refrained German industries from matching the labor-productivity
performance of their American counterparts. Rather, a relatively low level of efficiency
accounts for more than two-thirds of the labor-productivity gap. The limited impor-
tance of machine-intensity differences can be attributed to a process of machine in-
tensification in Germany during the 1920s and 1930s.28 The rapid move toward high
capital-labor ratios in German manufacturing aligns well with theoretical models of
appropriate technology in which new production knowledge is appropriate only for one
capital-labor ratio; if innovation takes place exclusively at high capital-labor ratios, “fol-
lower countries” must adopt capital-labor ratios already explored by leader countries in
the past to prevent a further widening of the labor-productivity gap.29
However, as much of what one needs to know to employ new production knowledge
is implicit and not available from handbooks, it takes time to assimilate and operate
machinery at the level displayed by countries exploring that capital-labor ratio before,
an effect that possibly explains part of the initially low efficiency levels in German
manufacturing.30 Other factors came into play as well. The findings of chapter 2
hinted at the positive influence that economies of scale exerted on labor-productivity
levels. A relatively large establishment size, for instance, could have made possible
a labor-productivity performance that was otherwise unattainable at a particular
level of machine intensity. Especially interesting in this respect is the below average
28. R. Richter, “Technology and Knowledge Transfer in the Machine Tool Industry. The UnitedStates and Germany, 1870–1930,” Essays in Economic & Business History Vol. 26 (2008): 173–188;R. Richter and J. Streb, “Catching-Up and Falling Behind: Knowledge Spillover from American toGerman Machine Toolmakers,” Journal of Economic History Vol. 71, no. 4 (2011): 1006–1031.29. S. Basu and D. Weil, “Appropriate Technology and Growth,” The Quarterly Journal of Economics
Vol. 113 (1998): 1025–1054.30. B. Los and M. P. Timmer, “The ‘Appropriate Technology’ Explanation of Productivity Growth
Differentials: An Emperical Approach,” Journal of Development Economics Vol. 77 (2005): 517–531.
scope for labor-productivity catch-up through enhanced efficiency levels in textiles and
primary metals, which were both identified in chapter 2 as relatively strong performing
German manufacturing industries. This suggests that the on average smaller scale of
production in the other German manufacturing industries restrained the full efficient
use of machinery, a notion also advanced by Cristiano Ristuccia and Adam Tooze.31
Chapter 4 moves on to critically assess the measurement techniques conven-
tionally employed in long-run economic history by addressing the debate on German
output growth over WW1. Studies on long-run economic growth are often plagued by
limited data availability, particularly for early periods. In the absence of, for instance,
production data, the unobserved output change can be proxied by the behavior of
correlates. However, the correlation between the proxy and target series is never
perfect, which introduces inaccuracy to the estimates and may spark off a debate
concerning the appropriateness of different proxies. This scenario unfolded in the debate
on industrial output growth in pre-WW2 Germany, leading to different time-series
estimates.32 The choice between output proxies carries important implications for the
assessment of Germany’s growth experience; when used to calculate labor productivity,
the different output estimates indicate a German performance prior to WW1, i.e. 1907,
either equal to or well above the British level. Because the exact fit between proxy
and target variables cannot be determined when the latter is unobservable, choosing
between proxies proceeds on the basis of circumstantial evidence. Given the historical
questions at stake, the debate would benefit from a less conjectural approach.
In chapter 4 I apply a new approach to this old debate. Instead of choosing between
the different estimates, I acknowledge that all series are based on correlates of output.
Consequently, the dynamic properties of each observed series must be captured by the
same component, i.e. output change, while the deviation between the series reflects the
different accuracy of the correlates in capturing the unobserved change in output. Us-
ing state space time series analysis, I filter this common component from the output
series.33 This way, I do not discard any data and thus make full and efficient use of all
information. In a second step, the filtered output series is combined with employment
data to derive an index of German labor-productivity change, which, expressed relative
31. Ristuccia and Tooze, “Machine Tool and Mass Production,” 9.32. R. Wagenfuhr, “Die Industriewirtschaft. Entwicklungstendenzen der deutschen und interna-
tionalen Industrieproduktion 1860 bis 1932,” in Vierteljahrsheftte zur Konjunkturforschung, vol. (Son-derheft) 31 (Berlin: Verlag von Reimar Hobbing, 1933); Hoffmann, Das Wachstum; Ritschl, “SpuriousGrowth in German Output Data.”33. J. Commandeur and S.J. Koopman, An Introduction to State Space Time Series Analysis (Ox-
ford University Press Inc., 2007); J. Durbin and S.J. Koopman, Time Series Analysis by State SpaceMethods, vol. 24, Oxford Statistical Science Series (Oxford University Press Inc., 2001).
to its British counterpart, is extrapolated backward from a known German/UK com-
parative level of labor productivity in 1936/35.34 This exercise is repeated twice, using
the point and interval estimates of the filtered common component, respectively. The
former attributes a 15% lead to Germany in 1907, while the latter indicates a range of
about 11% around the point estimate that contains the estimated parameter with 95%
certainty.
This finding takes on significance for the reconciliation between time-series projec-
tions and benchmark estimates. Faced with the different time series of output presented
in the literature, scholars have previously employed 1907 labor-productivity benchmarks
to test the accuracy of the time series estimates.35 As with the latter, however, vari-
ous benchmark estimates are presented, which ascribe a lead to Germany of either 5%
or 25%. Previously, criteria for the fit between benchmark estimates and time series
projections were loosely defined and the procedure applied in this chapter moves the
debate forward by providing a statistical framework to quantify the margin for error.36
Although the benchmarks presented in the literature deviate markedly, they all fall
within the tails of the confidence interval around my time-series projections. All esti-
mates can therefore be reconciled. This suggests that when benchmarks are used as a
check upon time series, taking into account the measurement error leads to a different
assessment of the fit between both measures as compared to the exclusive use of point
estimates.
Of course, this raises the question if such a broad range of German labor-
productivity levels obtained by the methodology advanced in this chapter renders
impossible a concise assessment of Germany’s comparative performance? Paradoxically,
my answer to this question is that working with confidence intervals actually increases
the reliability of the conclusions regarding historical economic development. Any
conclusion drawn from the filtered time-series estimates are explicitly founded on
a solid statistical basis, which provides an increased certainty compared to studies
employing point estimates only. I can confidently infer that, first, Germany had
overtaken Britain in terms of labor productivity already before WW1, yet by a small
margin only. Second, over WW1 there was a statistically significant change in labor
productivity leadership with Germany dropping below the UK. And, third, given
Fremdling, de Jong and Timmer’s 1936/35 German/UK benchmark comparison,
34. Fremdling, de Jong and Timmer, 2007.35. Broadberry and Burhop, “Comparative Productivity in British and German Manufacturing”;
A. Ritschl, “The Anglo-German Industrial Productivity Puzzle, 1895-1935: A Restatement and APossible Resolution,” Journal of Economic History Vol. 68, no. 2 (2008): 535–565; S.N. Broadberryand C. Burhop, “Resolving the Anglo-German Industrial Productivity Puzzle, 1895–1935: A Responseto Professor Ritschl,” Journal of Economic History Vol. 68, Nr. 3 (2008): 930–934.36. Broadberry and Burhop, “Comparative Productivity in British and German Manufacturing,” 326.
Britain’s lead evaporated again in the 1930s and both countries performed roughly on
par shortly before WW2.
In chapter 5 I explore the possibility of a European convergence club in manu-
facturing before WW1. Several of the potential constraints to productivity growth
in pre-WW1 Britain and Germany, e.g. small domestic markets and relative factor
costs less favorable to capital-intensive production in comparison to the US, are easily
extended to other European countries, too. This invites the question whether, or to
what extent, the condition of being ‘European’ determined the growth experience
of countries.37 The notion of convergence in manufacturing labor-productivity levels
is particularly relevant for the pre-WW1 era, as the period in between 1870–1913 is
characterized by openness to trade and globalization. The relative openness to trade
potentially promoted the convergence between European manufacturing industries
toward a common level of performance, as trade theory suggests that differences in
relative factor prices and thus in the mix of factor inputs used in production iron out
under conditions of free trade.38 Chapter 5 explores this possibility by constructing
benchmarks of comparative labor productivity for the US, UK, Germany, France, the
Netherlands and Sweden around the year 1910.
Despite the openness to trade, the benchmarks show that the level of labor produc-
tivity had not converged between European countries before WW1 and marked differ-
ences persisted, both for total manufacturing and manufacturing branches. Moreover,
backward extrapolation of comparative labor productivity to 1870 points out that the
dispersion of performance hovered around a constant level throughout the period and
showed no signs of convergence. These findings are in sharp contrast to total-economy
developments; GDP per capita levels converged steadily between 1870–1913.39 This
finding aligns well with Broadberry’s notion that convergence at the country level was
fueled mainly by changes in the structure of the economy rather than labor-productivity
developments in manufacturing.40 At the same time, it may also mean that the pre-
37. For conditional convergence, see R. Barro, “Economic Growth in a Cross Section of Countries,”The Quarterly Journal of Economics Vol. 106, No. 2 (1991): 407–443; R. Barro and X. Sala-I-Martin,“Convergence,” Journal of Political Economy Vol. 100 (1992): 223–258; J. Fagerberg, “Technologyand International Differences in Growth Rates,” Journal of Economic Literature Vol. 32, no. 3 (1994):1147–1175.38. E. Heckscher, “The Effect of Foreign Trade on the Distribution of Income,” Ekonomisk Tidskrift
(1919): 497–512; B. Ohlin, Interregional and International Trade (Cambridge, Mass.: Harvard Univer-sity Press, 1983); P. Samuelson, “International Trade and the Equalization of Factor Prices,” EconomicJournal (1948): 165–184; P. Samuelson, “International Factor-Price Equalization Once Again,” Eco-nomic Journal (1949): 181–197.39. J. Williamson, “Globalization, Convergence, and History,” Journal of Economic History Vol. 56,
no. 2 (1996): 277–306.40. S.N. Broadberry, “Manufacturing and the Convergence Hypothesis: What the Long-Run Data
WW1 period did not witness convergence of capital-labor ratios in manufacturing. For
instance, Abramovitz argued that the social competence necessary to exploit the new
technology was limited before WW1.41 Indeed, within Europe the level of machine
intensity differed considerably.42
As chapter 3 revealed that Germany operated before WW1 at a relative machine-
intensity level vis-a-vis the US much lower than during the 1930s, I am inclined to
attach more importance to differences in capital-labor ratios in an explanation of the
transatlantic labor-productivity gap before WW1 than at the end of the interwar pe-
riod. This makes the first half of the twentieth century a period of transition in which
German manufacturing gradually moved toward capital-labor ratios that promised a
considerable scope for labor-productivity growth. The stationarity, even deterioration,
of Germany’s comparative labor-productivity performance between 1909–1936 relative
to the US did not reflect the lack of technological progress, but an incomplete adop-
tion of new technology hampered by learning effects. This transition phase is enclosed
on both ends by periods that arguably display very different dynamics. The relatively
low levels of machine intensity in pre-WW1 German manufacturing suggests that the
labor-productivity gap to the US in the period before 1900 was driven largely by the
use of different technology, while the post-WW2 era witnessed a rapid decrease of both
the labor-productivity and capital-intensity gap to the US.43 This process of capital-
intensity convergence, however, had already set in during the interwar years. While it
failed to bring German labor-productivity levels closer to the US in the short run, it
formed the necessary first step on the road to catch-up and may partly explain the
German growth miracle in the post-WW2 period.
Show,” Journal of Economic History Vol. 53, no. 4 (1993): 772–795.41. Abramovitz, “Catching-up,” 395.42. Hannah, “Logistics, Market Size, and Giant Plants,” 71.43. van Ark, International Comparisons of Output and Productivity.
this exercise are confronted with two strands of literature, each highlighting a different
aspect of the German growth experience. First, the preconditions for growth may have
been less favorable in Germany than in the US. It has been argued that relative-factor
costs in Europe discouraged the substitution of machinery for skilled labor, which in
turn constrained the adoption of labor-productivity enhancing technology.5 Moreover,
the literature has suggested that the small scale of European production negatively
affected labor productivity. Large-scale production and standardization was only limited
applicable, because producers faced a small domestic market characterized by a demand
for customized goods.6 Although these conditions have been ascribed to Europe as a
whole, evidence in support of such theories are based mainly on the case of Britain. The
question, then, is whether the British constraints to labor-productivity growth applied
also to Germany?
Arthur Shadwell, a British contemporary, who traveled the UK, Germany and the
US shortly after the turn of the twentieth century in order to compare the qualities of
industrial life in these countries, took note of Germany’s remarkable success in the face
of circumstances potentially detrimental to development:
“Not a rich country, possessing no exceptional resources or facilities, no
extensive and convenient seaboard, with no tide of skilled immigrant labour
to make things easy, and with enemies in arms on both sides of her, she
has yet within the space of thirty years, and while bearing the burden of
an enormous system of military defense, built up from comparatively small
beginnings a great edifice of manufacturing industry which for variety and
quality of output can compete in any market with most of the finest products
of Great Britain. That is no exaggeration but a plain statement of facts, and
it can be said of no other country.”7
Having matched British performance, did Germany subsequently encounter the same
barriers for further growth that prevented the UK from catching-up to the US? This
was not necessarily the case, given that Chandler has likened several elements of the
German system of manufacturing to the US, rather than to the UK.8 Also, the unique
institutional setting in which German producers operated, in particular the cartel-tariff
5. Habakkuk, American and British Technology; David, Technical Choice, 66; P. Temin, “LabourScarcity in America,” Journal of Interdisciplinary History Vol. 1 (1971): 251–264, 162;Field, “On theUnimportance of Machinery,” 379.
6. Rostas, “Industrial Production, Productivity and Distribution,” 58-59; Chandler, Scale and Scope,47; Landes, The Unbound Prometheus, 247; Broadberry, “Technological Leadership,” 291.
7. A. Shadwell, Industrial Efficiency (Longmans, Green, / Co., 1906), 14-15.8. A. Chandler, “Organizational Capabilities and the Economic History of the Industrial Entreprise,”
Journal of Economic Perspectives Vol. 6, no. 3 (1992): 79–100.
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 17
system, has been associated with labor-productivity benefits, which may have counter-
acted some of the ails that European countries suffered from.9
Second, in addition to questions of relative standing and catch-up, this chapter
addresses the issue of labor-productivity in Germany, too. The US is known for its
unprecedented growth spurt during the interwar period and by placing Germany’s labor-
productivity performance in relation to its American counterpart the growth record of
German industries between 1909 and 1936 may easily be underrated.10 In the presence
of rapid growth in the US, a stagnant level of comparative labor productivity reflects
fast growth in Germany, too. Conversely, a relative standing close to the US does not
by necessity imply a rapid development on the part of Germany.
A focus solely on German growth is called for also because doubt has been cast
upon the reliability of the output and employment indices of the German Historical
National Accounts (HNA) constructed during the 1960s under supervision of Walther
Hoffmann. The critique on Hoffmann’s series is directed toward his use of income data
to estimate output growth. Because of the increased bargaining power of labor unions
over WW1, the productivity to wage ratio changed and using the latter as a proxy for
output leads to spurious growth.11 As the quality of the time series has been called
into question, extrapolating backwards from known labor-productivity levels in the
interwar period could potentially lead to inaccurate estimates. New estimates of labor-
productivity growth in German manufacturing are therefore necessary.12
In response to the first issue, i.e. relative standing, this chapter presents a comparison
of labor-productivity levels between Germany and the US in 1909 and 1936/35. The
second question, i.e. growth in Germany, is subsequently addressed by an exclusively
German inter-temporal comparison of labor-productivity levels between 1909 and 1936.
In both cases the latest methodological developments for constructing productivity
comparisons are taken on board to allow for the most accurate analysis possible. This
involves the application of an industry-of-origin approach to the benchmark estimates,
which, among other things, entails a break down of manufacturing in industries to
provide the detail needed to map out an economy’s productivity profile, i.e. its strong
and weak elements. Having set out the methodology and data in sections 2.2 and 2.3,
the results, which are presented in sections 2.4 and 2.5, are finally positioned in the
literature in section 2.6.
9. Hannah, “The American Mircale,” 207–208; Kinghorn and Nye, “The Scale of Production,” 109;M.Levenstein and V. Suslow, “What Determines Cartel Success,” Journal of Economic Literature Vol.44, no. 1 (2006): 43–95, 85.10. A.J. Field, “The Most Technologically Progressive Decade of the Century,” The American Eco-
nomic Review Vol. 93, no. 4 (2003): 1399–1413.11. Ritschl, “Spurious Growth in German Output Data.”12. Hoffmann, Das Wachstum; Fremdling, “German National Accounts.”
Most research focusing on productivity comparisons is conducted on the total-economy
level, as, for instance, the well-known long-run series provided by Maddison.13 As a
result, Maddison’s time series outline major trends of economic development, but the
dynamics that play out within the economy go by unnoticed. As such, the drivers be-
hind the observed total-economy growth patterns remain hidden. For instance, studies
focusing on technological development conducted on the basis of total-economy data
may miss important historical developments, as effects of efficiency-increasing innova-
tions tend to be underestimated when their impact is measured by their contribution to
GDP. In fact, Paul David used this argument to criticize the conclusions of Robert Fo-
gel’s seminal work on the impact of railroads on American economic growth.14 Another
example that stresses the importance of research on the disaggregated level concerns
Broadberry’s analysis of Anglo-American GDP-per-capita difference, which he ascribes
to compositional effects. When employment shifts from low productivity to high produc-
tivity sectors, output per worker on the total-economy level increases and such a change
in employment structure was crucial for America’s growth spurt during the nineteenth
century.15
To allow for effects of composition and to capture inter-industry variance in per-
formance, the labor-productivity comparisons constructed in this study employ the
industry-of-origin approach, which dissects the manufacturing sector in its underlying
components, i.e. manufacturing industries. By accurately measuring the state of man-
ufacturing in a particular year, the industry-of-origin benchmark provides the starting
point for further research that aims to explain a country’s growth experience. In addi-
tion, a benchmark for the pre-WW1 period supplies a check upon time-series projections
extrapolated backward from more recent benchmark estimates. As it is difficult for back-
ward extrapolations to accurately allow for changes in the structure of an economy (or
manufacturing), especially when the projection covers periods characterized by turbu-
lence and rapid change, such as the World Wars or the Great Depression, problems
occur when time series are projected into the distant past.16 A large deviation between
13. A. Maddison, Phases of Capitalist Development (Oxford: Oxford University Press, 1982); A.Maddison, Dynamic Forces in Capitalist Development: A Long-Run Comparative View (Oxford: Ox-ford University Press, 1991), 1–333; A. Maddison, Monitoring the World Economy 1820–1992 (Paris:Organisation for Economic Cooperation / Development, 1995), 1–255.14. P. David, “Transport Innovation and Economic Growth: Professor Fogel On and Off the Rails,”
Economic History Review Vol. 22, no. 3 (1969): 506–525.15. Broadberry, The Productivity Race.16. A. Gerschenkron, “Soviet Heavy Industry: A Dollar Index of Soviet Machinery Output, 1927–28
to 1937,” The Review of Economics and Statistics Vol. 37, no. 2 (1955): 120–130.
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 19
time-series projections and direct level estimates may indicate a degree of inaccuracy
on the part of the former.17 An additional advantage of the benchmark over time-series
projections is that because the level estimates of productivity produced for the former
refer to one year, data on output and employment can be obtained from a single primary
source, which guarantees a consistency between the output and input measures.18
The importance and appeals of the industry-of-origin approach were recognized by
the late 1940s. The first industry-of-origin benchmark was constructed by Rostas in
1948.19 In an attempt to assess the state of the British and American economies, he
broke down the manufacturing sector and pinpointed the comparative performance of
UK industries in relation to their US counterparts. Since then, the industry-of-origin
approach has been adopted by other scholars, most notably among which Stephen
Broadberry, to address the debate on (historic) patterns of convergence and divergence.
The research conducted here follows in this tradition and builds upon the work of these
early pioneers.
Although a productivity comparison on an industry level is a simple and straight-
forward concept, such a comparison can be made in a variety of ways. The methods
used here are refinements of the basic methodologies of comparison set out by Rostas,
Paige & Bombach and Broadberry. The first industry-of-origin benchmarks obtained
productivity figures by taking the ratio between output volume and employment on
the industry level. When output is expressed in volumes an international comparison is
straightforward, since the unit of measurement is the same for all countries, for instance
produced tons of coke per employee. However, as noted by Inklaar and Timmer, the
direct comparison of physical units of output for the measurement of productivity is
only possible for a specified product or a closely related group of products.20 Conse-
quently, this limits the ability to estimate productivity for industries producing a wide
array of heterogeneous goods, which is always the case when comparing productivity at
the industry or total-economy level. In view of these limitations, it is more practical to
compare output values, rather than output volumes.
Unfortunately, when the value approach is applied, the advantage of directly com-
paring productivity levels between countries is lost. While hectolitres and kilograms are
the same in the US and Germany, US$ and German Goldmark cannot be compared
directly. Therefore, a conversion factor is necessary to express the output value of dif-
ferent countries in a common currency. The exchange rate is not an optimal conversion
17. Chapter 4 provides a detailed discussion of this issue.18. van Ark and Timmer, “The ICOP Manufacturing Database.”19. L. Rostas, Comparative Productivity in British and American Industry (Cambridge: Cambridge
University Press, 1948), 1–263; Rostas, “Industrial Production, Productivity and Distribution.”20. Inklaar and Timmer, “GGDC Productivity Level Database,” 6-8.
factor for this purpose, since it only signifies the price relation between internationally
traded goods. Moreover, the exchange rate is a particularly inconvenient instrument
for this research, as the countries under comparison operated under different monetary
regimes, at times. Both countries were on the gold standard at the start of the twen-
tieth century. Exchange rates were effectively fixed and domestic price movement was
determined by a country’s gold supply. However, by the late 1930s the US had abolished
payments stipulated in gold, while Germany still adhered to the gold standard. Finally,
the exchange rate is a total-economy measure of the price ratio that does not allow for
variation thereof between different sectors of the economy.
A more appropriate alternative to the exchange rate is an industry-specific conver-
sion rate based on producer prices. This technique has been set out by van Ark and is
referred to as the International Comparisons of Output and Productivity (henceforth,
ICOP) methodology.21 The building blocks of the conversion rates are formed by prod-
uct prices. As these prices are seldom available in the statistical records, they have to
be derived from data on the produced value and quantity of products. In a bilateral
country comparison, these product prices – referred to as unit values – are computed
for both countries as in equation (1).
pij =vijqij
(2.1)
Where pij is the unit value of product i in country j, vij the output value of that
product and qij the corresponding produced volume. Subsequently, identical products
are selected and matched between the two countries involved in the comparison. The
ratio between the unit value of the same commodity in both countries captures the
product-specific relative price expressed in terms of country n’s currency per unit of
the base country o’s currency, as in equation (2).
uvrio =pinpio
(2.2)
With uvrio as the unit value ratio (henceforth, UVR) of product i, which represents
the relative unit value in country n (pin) compared to the unit value in country o
(pio). In order to derive an industry-level conversion factor, a weighted average is taken
of the product-specific price ratios classified in the same industry group. The weights
allotted to the UVRs for the purpose of aggregation reflect the product’s share in
total industrial output (vi/∑
vi). The aggregated UVRs are traditionally referred to
as purchasing power parities (henceforth, PPPs). The process of aggregation proceeds
21. Maddison and van Ark, “Comparison of Real Output in Manufacturing”; van Ark, InternationalComparisons of Output and Productivity; van Ark and Timmer, “The ICOP Manufacturing Database.”
Germany and the US. The pre-WW1 analysis for the US is based on the Thirteenth Cen-
sus of the United States published by the US Bureau of Commerce.23 For 1935 US I rely
primarily on the Biennial Census of Manufactures 1935 and the Sixteenth Decennial
Census of the United States.24 The US censuses provide an extensive and consistent cov-
erage of the American manufacturing sector in both years. For interwar Germany I use
the comprehensive archival records of the German production census published in Die
deutsche Industrie: Gesamtergebnisse der amtlichen Produktionsstatistik (henceforth,
production census of 1936). This is the first official German census of manufactures
and is available in two forms; a published edition and the original archival records. The
former has been to set up to hide particular manufacturing activities that were related
to the war effort. The archival records contain considerably more detailed and accurate
information and is used in this study.25
Collecting data to calculate labor productivity for pre-WW1 Germany was less
straightforward. The statistical offices of the US and the UK published a census of
manufactures already before WW1. For manufacturing industries these censuses re-
port data on output, employment, installed capital, etc. and as such are ideally suited
for constructing benchmarks. Because the first German census of manufacturing was
not published until 1936, I have to rely on other sources for the prewar period. The
Kaiserlichen Statistischen Amte (henceforth, Imperial Statistical Office) monitored the
economy in a variety of ways from the turn of the twentieth century onwards. Using
a combination of official statistical publications the industry-level data needed for the
construction of benchmarks is obtained. Because it forms the weakest link in the chain
of benchmarks presented here, the computation of German labor-productivity levels
before WW1 requires further elaboration.
Labor productivity in pre-WW1 Germany
To calculate German labor-productivity levels for the prewar period, I mainly rely
on information obtained from the Vierteljahrshefte zur Statistik des deutschen Reichs
(henceforth, statistical quarterlies). In the statistical quarterlies of 1913 the results of
23. United States Department of Commerce: Bureau of the Census, Thirteenth Census of the UnitedStates Taken in the Year 1910, vol. VIII: Manufactures (Washington D.C.: United States GovernmentPrinting Office, 1913). For mining, United States Department of the Interior, United States GeologicalSurvey 1910.24. United States Department of Commerce: Bureau of the Census, US Census of Manufactures 1935 ;
United States Department of Commerce: Bureau of the Census, US Census of Manufactures 1940 (I);United States Department of Commerce: Bureau of the Census, US Census of Manufactures 1940 (II).25. Reichsamt fur Wehrwirtschaftliche Planung, Die Deutsche Industrie 1936 ; for a detailed discus-
sion of this source see: R. Fremdling, H.J. de Jong, and M.P. Timmer, “Censuses Compared: A NewBenchmark for British and German Manufacturing 1935/1936,” Groningen Growth and DevelopmentCentre Memorandum no. 90 (2007): 1–36.
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 23
industrial surveys for the years between 1907 and 1911 are published.26 The surveys
report output and employment data for a number of industries. For those industries
that are included, the surveys do not provide full coverage. Instead, the production of
a sample of firms is reported. Partly this is due to the fact that the surveys are only
sent to firms affiliated with the national health-insurance scheme for workers (Gewerbe-
Unfallversicherungsgesetze). The smallest workplaces are in effect not covered and the
scope of the surveys may be limited to the larger firms in German industries. This
could lead to compatibility problems when comparing Germany with the US. The US
census of manufactures provides almost full coverage as only household industries and
establishments with an annual output lower than $500 are excluded.27
If due to scale advantages labor productivity is higher in large establishments as
compared to small establishments, the benchmark results based on data obtained from
the statistical quarterlies could potentially overestimate the productivity performance
of German industries. Table 2.1 captures the employment coverage of the industrial
surveys. On the left side of the table the average number of employees working in estab-
lishments included in the industrial surveys is reported. On the right side, I included the
same statistic for comparable industries obtained from the Berufs- und Betriebszahlung
published in 1907 (henceforth, occupational census), which has full-employment cover-
age.28 The nomenclature does not match perfectly between both sources. Nevertheless,
the fit is close enough to be reasonably sure that the classification of the occupational
census refers to the same manufacturing activities. For all industries the comparison of
average establishment size shows that the statistical quarterlies report data on relatively
large establishments. This creates a potential bias in favor of Germany.
In order to quantify the potential bias, I need to know which establishment-size
classes are represented by the establishments included in the industrial surveys. If,
for instance, it turns out that the surveys exclude establishments with less than 10
employees, the part of employment covered by those establishments is not represented
by the surveys, which introduces an upward bias in my estimates. Note that I estimate
the representativeness of the surveys and not their coverage. I use the representativeness
of the surveys for two reasons. First, as for some industries the nomenclature between the
industrial surveys and the occupational census differs, a comparison between the number
26. Kaiserlichen Statistischen Amte, “Ergebnisse der deutschen Produktionserhebungen 1913”;Kaiserlichen Statistischen Amte, “Ergebnisse der deutschen Produktionserhebungen 1914.”27. United States Department of Commerce: Bureau of the Census, US Census of Manufactures 1910
(VIII), 19. In the case of the British census of manufactures (1907) household industries, one-personestablishments, and establishments with less than 10 employees are excluded. As a consequence, about25% of employment is not covered. See: Board of Trade, UK Census of Production 1907, 8.28. Kaiserlichen Statistischen Amte, “Gewerbliche Betriebsstatistik,” Abteilung II, Heft 1, Tabelle 8,
Teer Destill. und Petroleum raff. 43 Kohlenteerschwelerei, Petroleumraff. 30
Kokereien 143 Verkokungsanstalten 131
Zementindustrie 166 Zement- und Trassfabrikation 81
Sources: Kaiserlichen Statistischen Amte, “Gewerbliche Betriebsstatistik,” in Berufs– undBetriebszahlung, Statistik des deutschen Reichs (Berlin, 1907); Kaiserlichen Statistischen Amte,“Erganzungsheft zu die Ergebnisse der deutschen Produktionserhebungen,” in Vierteljahrsheftezur Statistik des deutschen Reichs: Erganzungsheft, vol. Vol. 22, no. 3 (Berlin, 1913).
of employees reported by the surveys and the census, which covers total employment,
introduce a degree of inaccuracy. The compared statistics may not refer to exactly the
same unit of production. Second, even in cases where this is not a problem, the coverage
of the surveys does not provide information on the size of the establishments included
in the surveys. For instance, if 70% of an industry’s employment is covered by the
surveys, it is not clear whether the excluded 30% are employed in relatively small or
large establishments. Therefore, the sign of the bias associated with the surveys remains
unclear, too. Instead, the representativeness of the surveys, i.e. the establishment-size
classes represented by the surveyed establishments, does provide a tool to assess the
bias in the survey’s results.
Using a combination of the average firm size reported by the industrial surveys
and information obtained from the occupational census, I have estimated, first, which
establishment-size classes are represented by the surveys and, second, the share of total-
industry employment that is covered by these establishment-size classes.29 The results
indicate that in most industries the surveys represent all but the smallest establishment-
size classes. As, in general, between 95% and 100% of the manufacturing labor force is
employed in establishments-size classes represented in the surveys, there is no reason
to think that the surveys introduce a structural upward bias in the German labor-
productivity estimates.
29. See appendix 2.A for more detail and the results of this exercise.
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 25
Additional sources for the pre-WW1 period
Another potential drawback of the statistical quarterlies is that several manufacturing
industries are not included in the surveys. Data on, for instance, the food industry,
electrical and mechanical engineering, or the instrument industry are not reported.
For some industries important activities are omitted as well. The chemical industry
is poorly represented by coal-tar distillations, potash and sulfuric acid: information
on inorganic chemicals is unavailable. Furthermore, the industrial surveys only report
output in the textile industry, but no employment, making it impossible to calculate
labor-productivity levels. Lastly, due to their incomplete coverage, the industrial surveys
do not provide a complete output structure of the manufacturing sector. Additional
sources are needed to provide the weights necessary to aggregate the UVRs and PPPs
for an analysis on the level of total manufacturing.
To fill the gaps in the data of the statistical quarterlies, three other publications of
the Imperial Statistical Office are used here. First, the Statistisches Jahrbuch fur das
Deutschen Reich (henceforth, statistical yearbook) provides annual data on a limited
number of industries (mostly the production of mines and blast furnaces).30 To a large
extent these industries are in more detail covered by the industrial surveys. However, the
statistical yearbook includes data on the production of taxable goods, such as sugar,
tobacco, and alcoholic beverages and thereby provides information on output in the
food & kindred industry, which remained outside the scope of the industrial surveys.
Unfortunately, the yearbook reports physical quantities only. Hence, labor-productivity
levels for sugar and tobacco are expressed in produced tons per employee.31 For alcoholic
beverages, output is measured in hectoliters. In contrast to output data, the number
of employees working in the food & kindred industry is not reported by the statistical
yearbook. As with the textile industry, for which only output is reported in the industrial
surveys, an additional source is needed to find employment data necessary to calculate
labor productivity. For this purpose the occupational census is used, both in the case of
textiles and food & kindred. The number of workers in the textile industry derived from
30. Kaiserlichen Statistischen Amte, Statistisches Jahrbuch fur das deutschen Reich (Berlin, 1909–1912), 52–133.31. The obtained comparative labor-productivity level is subsequently projected on the US nominal
level of labor productivity (in US$), to obtain the German level of labor productivity expressed in US$.Because the statistical quarterlies report information on output value and volume for some products,starch mainly, it was possible to construct a PPP for this industry, which is then used to convert laborproductivity from US$ to German Goldmark. Of course, the comparative level of German/US laborproductivity has not changed in any sense, but expressing the comparison in output value enables anaggregation scheme along the lines of equation (2.6), which would not have been possible when outputis expressed in volumes. For tobacco, the same procedure has been followed with the difference that noindustry-specific PPP could be obtained and I relied on the total-manufacturing PPP, instead.
the occupation census is adjusted in line with the coverage of the industrial surveys.32
Secondly, the labor-productivity estimates for paper and glass production are based
on reports of trade unions, which were collected during the 1920s and used to recon-
struct developments on the industry level. These reports were subsequently published in
the Ausschuss zur Untersuchung der Erzeugungs- und Absatzbedingungen der deutschen
Wirtschaft.33 Finally, the estimates for tire production (rubber) are obtained from the
Industrielle Produktionsstatistik, a special edition of the Wirtschaft und Statsitik pub-
lished by the Statischen Reichsamt.34 Although these publications report predominantly
production figures for the period since 1925, the interwar data is sometimes comple-
mented with information on years before WW1 for purpose of comparison.
At this point, I am able to calculate labor-productivity levels for many manufac-
turing industries in pre-WW1 Germany. In most cases, the data employed to calculate
labor productivity does not cover an industry’s total output and employment. For the
purpose of aggregation, however, it is recommendable to allot total-industry weights
to the industry-level labor-productivity estimates. This way, the composition of manu-
facturing is properly taken into account. Because total-industry output is not reported
for Germany, but total employment is (by the occupational census), I have estimated
total output by multiplying the German nominal labor-productivity level by total em-
ployment. Essentially, the labor-productivity estimates are thus reweighted according
to industry-employment shares derived from the occupational census. Earlier research
already pointed out that this procedure underestimates the share of high-productivity
industries, but to a small extent only and is unlikely to affect the results substantially.35
This is confirmed by the 1936/35 German/US benchmark. Using the product of nominal
labor productivity and total-industry employment as a proxy for total-industry output
produces the same result as obtained by the use of actual total-industry output.
A potential problem is that most labor-productivity levels calculated on the basis of
the industrial surveys do not refer to the same year as the weighting scheme, i.e. 1907. In
fact, except for the textile and food & kindred industries, all productivity data refer to
either 1908, 1909 or 1910. If the results are to be interpreted as representative for 1907,
labor-productivity levels must have remained constant over this period, which seems
unlikely. In this study I pursue a less stringent approach by choosing the year for which
32. See appendix 2.B for more detail.33. Verhandlungen und Berichte des Unterausschusses fur allgemeine Wirtschaftsstruktur, “Die
deutsche Zellstof-, Holzschliff-, Papier- und Pappenindustries”; Verhandlungen und Berichte des Un-terausschusses fur allgemeine Wirtschaftsstruktur, “Die deutsche Glasindustrie.”34. Kaiserlichen Statistischen Amte, “Industrielle Produktionsstatistik”; Kaiserlichen Statistischen
Amte, “Industrielle Produktionsstatistik”; Kaiserlichen Statistischen Amte, “Industrielle Produktion-sstatistik.”35. Broadberry and Burhop, “Comparative Productivity in British and German Manufacturing,” 320.
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 27
Figure 2.1: Peak and census years, 1900–1913
United States
0
2
4
6
8
10
12
14
16
realG
DP
(1913
=100)
1900 1902 1904 1906 1908 1910 1912 191450
60
70
80
90
100
110
0
2
4
6
8
10
12
unem
plo
ym
ent
rate
(%)
Germany
1900 1902 1904 1906 1908 1910 1912 191450
60
70
80
90
100
Real GDPa
Trend growth GDPc
Unemployment rateb
Census years
a Sources: Angus Maddison, “Historical Statistics of the World Economy: 1–2008 AD,”Groningen Growth and Development Centre, 2008, http://www.ggdc.net/maddison/, Table 2:GDP Levels, retrieved: 23 March 2011.
b Sources: [US] D.R. Weir, “A Century of U.S. Unemployment, 1890–1990,” Research inEconomic History Vol. 14 (1992): 301–346, 341–343; [GER] T. Pierenkemper, “The Standardof Living and Employment in Germany, 1850-1960: An Overview,” Journal of EuropeanEconomic History Vol. 16 (1987): 51–73, 58–59.
c The basic long-run trend growth is fitted as a least-squares polynomial of degree 2, for theperiod 1870–1913.
the most output data is available, i.e. 1908 or 1909, as the basis for the benchmark. This
setting assumes that the composition of the manufacturing labor force has remained
unaltered between 1907 and 1909. As the employment structure is much less volatile
than the movement of productivity levels, projecting the 1907 structure on either 1908
or 1909 does not give cause for concerns.36 As the prewar benchmark is used for a
comparison with America and the latter’s census of manufactures refers to 1909, I
designated 1909 as base for the German benchmark.
The choice of 1909 as the prewar benchmark-year was further strengthened by move-
ments of the business cycle. Whenever possible, I took care to avoid years which are
at a peak or in a through of the cycle. Figure 2.1 shows that the level of real GDP
at the selected census years for both countries was above the long-run trend, and that
36. On the basis of the industrial surveys I am able to calculate the annual change in labor productivitybetween 1907 and 1911 for several industries, see appendix A.1. In almost all of these industries laborproductivity increased (rapidly) over the years 1908–1911 (Δ LP). Assuming that labor productivitydid not change, even in this short period, is therefore problematic. Instead, the employment share ofthese industries changed little.
the unemployment rate at that point in time was relatively low or stable. This is an
essential requirement for my analysis, as I strive to determine the level of potential pro-
ductivity differentials between the countries under comparison. I thus want to exclude
the effects of business cycles and capacity under-utilization as much as possible; which,
I am convinced, is the case for the selected benchmark year.37 Consequently, all German
labor-productivity estimates originally based on data from other years are adjusted to
a 1909-basis using Hoffmann’s industry-level time series of output and employment.
2.4 The transatlantic labor-productivity gap
The methodology and data described in the previous chapters enables me to compare
labor-productivity levels between German and US manufacturing industries. This is
necessary because the extent to which Germany lagged behind the global productivity
leader is not immediately evident from other studies. In the literature German/UK and
US/UK comparisons are presented for prewar and interwar years. Until now, direct Ger-
man/US comparisons were not available and the productivity gap between Germany
and the US could only be obtained indirectly using the German/UK and US/UK com-
parisons. The quality of an indirect German/US estimate depends on the consistency in
the applied methodology and the coverage of industries between the German/UK and
US/UK comparisons, which is never perfect.
Moreover, such a procedure is in particular problematic for the pre-WW1 period,
because the size of the gap between Germany and the UK before WW1 is not undis-
puted. Both Steven Broadberry & Carsten Burhop (henceforth, B&B) and Albrecht
Ritschl have presented German/UK benchmarks for 1907, reporting a productivity ra-
tio in manufacturing of 1.08 and 1.28, respectively.38 Contingent on the choice between
these benchmarks, Broadberry & Irwin’s estimate of a 2:1 American lead over Britain
in 1909/10 implies a German/US productivity ratio of either 0.54 (via B&B) or 0.63
(via Ritschl); a difference of about 15%, which is sizable for this type of research.39
As described in section 2.2, the industry-of-origin approach compares the gross out-
put by industries between countries using an industry-specific conversion factor or
PPP. The inter-industry variation illustrated by table 2.2 highlights the importance
of industry-specific conversion factors. The listed Laspeyres, Paasche, and Fischer gross
37. See Jong and Woltjer, “Depression Dynamics” for an elaborate discussion of the business cycleand capacity utilization effects and a sensitivity analysis for the interwar period.38. Ritschl, “The Anglo-German Industrial Productivity Puzzle,” 549; S.N. Broadberry, R. Fremdling,
and P. Solar, “European Industry, 1700–1870,” Jahrbuch fur Wirtschaftsgeschichte Vol. 2 (2008): 141–171, 93239. Broadberry and Irwin, “Labor Productivity in the United States and the United Kingdom,” 261.
studied for the year 1909. The employment coverage of the prewar comparison amounts
to 34% for Germany and 47% for the US. In 1936/35 the coverage of the prewar industry-
sample was 32% and 33%, respectively.40
The lack of full-coverage data for the pre-WW1 period may introduce a bias in
the estimates. Indeed, on the aggregate level, a comparison between the sample and
full-coverage results for 1936/35, presented in table 2.5 below, shows that the former
overstate Germany’s performance by about 10%. The difference is accounted for by
two effects. First, the performance in some German industries is overestimated by the
sample data. For textiles, chemicals and, to a lesser extent, primary metals the total-
industry results show a much poorer performance on the part of Germany. This indicates
that the production activities covered by the 1909 sample displayed an a-typically high
performance level, a finding that helps explain the strong performance in some parts of
German manufacturing, an issue to which I later return. Second, several industries are
excluded by the sample data and these tended to perform relatively weak. Including
these industries thus drags down the overall level of German labor productivity. Then
again, even after downward adjusting the performance in some industries, the main
characteristics of German manufacturing remain unaltered, as do the conclusions drawn
on the basis of the comparison.
With regard to these conclusions, the top row of tables 2.4 and 2.5 reports the com-
parative performance on the aggregate level and shows that Germany tracked America
at considerable distance, both in 1909 and 1936/35. If I accept the idea that countries
lagging behind look to the universal productivity frontier for catch-up growth, as is
often suggested in the literature, Germany had yet a long way to go by 1909.41 De-
spite this large potential for catch-up growth, at the end of the interwar period German
and American levels of productivity had not converged. Instead, the US extended its
lead and the German/US productivity ratio dropped from 57% to 52%, which might
not come as surprise given the many calamities since 1914. Still, the distance to the
US before WW1 was larger for the UK than it was for Germany. Given Britain’s edge
over Continental Europe all through the nineteenth century, this change in European
productivity leadership signifies a success on the part of Germany in modernizing the
manufacturing sector since the second industrial revolution.
Up till this point I have looked at gross output per employee in Germany and the US.
40. For Germany 1909 95% of mining employment is covered. All employment is taken into accountfor US 1909, US 1935 and Germany 1936.41. The idea of catch-up is old and can be found in the works of, for instance, Gerschenkron, Economic
backwardness, 113, 116 and Abramovitz, “Catching-up,” 387. For more recent frontier analysis see,for example, Acemoglu, “Directed Technical Change,” 39 and Vandenbussche, Aghion, and Meghir,“Growth, Distance to the Frontier and Composition of Human Capital,” 98.
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 33
Table 2.5: German/US comparative labor productivity(US = 100%), sample and full-coverage data
Description Per employee Per hour
Sample All Sample All
Mining 29 29 26 26
Manufacturing 52 46 47 41
Food and kindred products 43 45 41 44
Tobacco manufactures 35 35 34 34
Textile mill products 111 74 103 69
Apparel products . . . 49 . . . 39
Lumber and wood products . . . 49 . . . 46
Paper and allied products 52 52 46 46
Chemical and allied products 105 72 96 66
Petroleum and coal products 55 56 50 51
Rubber products 46 41 43 39
Leather and leather products 57 50 55 48
Stone, clay, and glass products 54 48 50 43
Primary metal products 103 93 88 79
Fabricated metal products . . . 48 . . . 42
Machinery (excl. electrical) . . . 49 . . . 40
Electrical machinery . . . 49 . . . 43
Transportation equipment 24 25 23 22
Sources: see text, section 2.3. The coverage and number of the UVRs: see(for sample) table 2.3 and (for full coverage) appendix A.6 and A.7
There are, however, good reasons to adjust for working hours. Over the interwar period
both the US and European countries saw a rapid drop of hours worked per year. The
increased bargaining power of labor unions, but also the effects of the Great Depression
led to a shortening of the working week and an increasing number of holidays.42 Since
the change in hours worked was larger for the US, adjusting for hours will affect the
labor-productivity comparisons. For the US, total annual hours worked plummeted from
2,718 in 1909 to 1,817 in 1935; a drop of 33%. The corresponding figures for Germany
42. The correction for hours worked is based on data from M. Huberman, “Working Hours of theWorld Unite? New International Evidence of Worktime, 1870–1913,” Journal of Economic HistoryVol. 64, no. 4 (2004): 964–1000 and M. Huberman and C. Minns, “The Times They are not Changin’:Days and Hours of Work in Old and New Worlds, 1870-2000,” Explorations in Economic HistoryVol. 44, no. 4 (2007): 538–567. In addition, several primary sources have been used, i.e. KaiserlichenStatistischen Amte, Statistisches Jahrbuch fur das deutschen Reich, Hoffmann, Das Wachstum, UnitedStates Department of Commerce: Bureau of the Census, US Census of Manufactures 1910 (VIII),International Labour Office, Year Book of Labour Statistics 1939 (Geneva: International Labour Office,1939) and R. Matthews, C.H. Feinstein, and J.C. Odling-Smee, British Economic Growth, 1856–1973(Oxford: Clarendon Press, 1982), 1–712.
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 35
aged to keep up with the US. This was a small group constituted by textiles, chemicals
and primary metals.
When we probe deeper and dissect the manufacturing sector on the SIC 3-digit
level, the pronounced variation between comparative levels of performance persists.46
For instance, a break down of primary metals in 1909 points out that the iron & steel
industries in Germany were not at all inferior to the US. A low efficiency in nonfer-
rous metals, however, depresses the productivity level for German primary metals as
a whole. Similarly, the large gap between Germany and the US in petroleum & coal
production was caused by a low level of productivity in German petroleum refining. In
coke production Germany was no less efficient than America. The comparative perfor-
mance of chemical industries in 1936/35 varied, too; while the German paint production
performed at a third of the US level, Germany enjoyed an advantage over America of
about 2:1 in the fertilizer industry. In short, the range of German industrial performance
relative to the American frontier was large.
In spite of the observed variation in comparative labor productivity between in-
dustries, the pattern of strong versus weak performers persisted over time. Industries
already performing distinctively weak or strong before 1909 did likewise in 1936/35. Es-
pecially industries at the lower end of the performance scale were predominantly station-
ary. Leather manufacturing, paper production, petroleum refining, and transportation-
equipment industries all persistently trailed the US at a large distance. At the opposite
end of the spectrum textiles, primary metals and chemicals, all of which already did
well in 1909, improved their comparative performance. Still, table 2.5 suggests that
the industries included in the sample performed a-typically strong compared to total-
industry comparative labor productivity. In particular, the spinning activities studied
in the sample does much better in relation to the US than the textile industry on the
whole. The same goes for chemicals and, to a lesser extent, primary metals. Even so,
between the sample- and full-coverage comparison the top-three German industries dis-
playing the strongest comparative performance is the same, only their relative levels
drop from parity to about two-thirds the level of the US. The recurrence in 1936/35 of
a productivity profile similar to the case of 1909 suggests that the level of comparative
performance was dictated by long-run growth determinants, rather than the turbulence
of the period.
With respect to this persistent productivity profile, an identifying trade mark for all
strong or weak performing German industries, such as the aforementioned distinction
between light versus heavy industries in the case of the UK, is not directly evident. Yet
46. Compare the data underlying the 2-digit labor-productivity levels in appendices A.2 and A.4. Forthe UVRs needed to express both output values in a common currency, see appendix A.8.
production and printing activities; industries for which the data required by the ICOP
methodology is not available.
Third and final, the intertemporal benchmark presented here corrects for differences
in working hours between the prewar and interwar period, which Hoffmann’s series do
not. As noted above, in Germany the total number of hours worked on an annual basis
decreased by 25%. Taking account of the reduction in labor input leads to an upward
adjustment of labor-productivity growth. As a result, the food, textile and leather in-
dustries no longer display a negative rate of labor-productivity growth, as implied by
Hoffmann. Rather, these industries stagnated or experienced little growth only. The
correction for hours increases the average annual growth rate in industries by about 1
percentage point across the board, with the exception of mining for which the adjust-
ment makes a small difference only. The reduction of hours worked was considerably
smaller in mining, mostly because the time spent below ground was already relatively
low in 1909.
Table 2.6 shows that on the aggregate level German industry realized a moderate
rate of labor-productivity growth, with mining doing better than manufacturing. On
the level of industries, the growth experience varied considerable. Some industries dis-
played rapid growth while others appeared to stagnate or even decline. With respect
to the former, the common denominator of fast-growing industries appears to have
been maturity. In particular ‘young’ industries developed rapidly, the transportation
equipment industry being a prime example. The latter’s fast-paced growth is reflected
by the price ratio between 1909 and 1935. The price level dropped sharply between
1909 and 1936, a characteristic feature of rapidly developing industries. Rubber, which
through the production of tires was closely related to the motor-vehicles industry, and
chemicals & allied belong to this category, too. Industries born (chemicals, motor ve-
hicles, tires, petroleum)48 or extensively modified (primary metals, tobacco)49 during
the late nineteenth century succeeded in raising productivity levels. In contrast, none
of the stagnated industries (e.g. food & kindred, textiles and leather) can be plausibly
typecast as young.50
The results presented in table 2.6 shed new light on the comparative German/US
productivity levels. Germany’s outstanding comparative performance in textiles over
the interwar period suddenly looks less impressive knowing that the relatively small
productivity gap resulted from a lack of any significant progress in both countries. To
48. Landes, The Unbound Prometheus, 234.49. ibid., 235.50. Although these industries modernized, too. In textiles, for instance, the ring spindle gradually
replaced the mule and the food industries witnessed the introduction of new techniques that conservedproducts for a longer time. Broadberry, Fremdling, and Solar, “European Industry,” 158, 161.
the German textile, lumber, leather, and food industries lagged substantially behind; in
these industries the share of workers employed in establishments with over 50 workers
was 3, 4, or – in the case of the food industry – even 5 times smaller than in the US.
Table 2.8: Share of workers employed in establishmentsemploying >50 workers (%)
Description US 1909 Germany 1907
Textiles 93 38
Paper and printing 67 51
Lumber 81 22
Leather 90 25
Iron and steel 99 98
Food 67 13
Ceramics 85 55
Chemicals 85 70
Sources: J. Kinghorn and J. Nye, “The Scale of Production inWestern Economic Development: A Comparison of OfficialIndustry Statistics in the United States, Britain, France, andGermany, 1905-193,” Journal of Economic History Vol. 56, no.1 (1996): 90–112, 99.
For iron & steel and chemicals the small difference in average establishment size coin-
cides with a correspondingly small difference in labor-productivity levels. I am inclined
to relate Germany’s emphasis on large-scale production in these industries to Hannah’s
observations regarding giant plants (>1,000 workers) in Germany and the US. Hannah
states that giant production units were particularly representative for “modern” in-
dustries and in chemicals, shipbuilding, and electrical manufacturing Germany counted
more giant plants than the US.56 The opposite conclusion applies to tobacco and auto-
mobiles.57 With the exception of electrical engineering, the presence of giant plants or
the lack thereof corresponds well to the comparative productivity levels presented in this
research; chemicals performed on par with the US, while the transportation-equipment
industry and tobacco manufacturing trailed the American frontier at considerable dis-
tance. Moreover, following Kinghorn and Nye, Hannah underlines Germany’s overall
smaller average establishment size in manufacturing and suggests it might have been
the bulk of small workshops that drove Germany’s low overall labor productivity.58
So far as the data of Kinghorn and Nye go (table 2.8), the small share of employment
working in large-scale establishments reported for the German textile industry is difficult
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 43
Table 2.9: Distribution of employment over establishment-size classes (%) inGerman manufacturing industries, 1907
SIC Industry ≤ 50 51–1,000 ≥ 1, 001
Textiles Totala 28 72 0
Cotton spinning 31 69 0
Linen spinning 18 82 0
Jute spinning 7 93 0
Silk spinning 51 49 0
Chemicals General chemicals 29 58 13
Petroleum and coal Totala 21 79 0
Petroleum 66 34 0
Coke 18 82 0
Stone, clay and glass Cement 12 84 4
Primary metals Totala 15 59 27
Iron & steel 5 55 40
Cast iron 30 60 11
Nonferrous metals 9 82 9
Transportation equipment Motor vehicles 21 50 29a Industry totals are a weighted average calculated using employment weights. Foremployment data, see appendix A.2.May not sum to total due to rounding. Sources: Kaiserlichen Statistischen Amte,“Gewerbliche Betriebsstatistik,” in Berufs– und Betriebszahlung, Statistik des deutschenReichs (Berlin, 1907).
to reconcile with the strong labor-productivity performance it delivered in both 1909 and
1936/35. This puzzle can be explained by table 2.9, which reports for several industries
covered by the labor-productivity comparisons the distribution of employment over
establishment-size classes. For 1909 the textile industries included in the comparison
concern spinning activities and table 2.9 shows that in these industries the employment
share working in establishments with over 50 employees was much higher than the 38%
reported by Kinghorn and Nye for the whole of textiles. In cotton spinning, which was
the largest spinning industry in terms of employment, this share amounted to 69%. The
other textile industries, i.e. jute, linen and silk spinning, employed 93, 82 and 49% of
total labor in large-scale establishments, respectively. Clearly, the spinning industries
not only displayed above average labor-productivity levels, as the 1936/35 comparison
testifies, they were also characterized by relatively large establishments. For the other
two strong performers, i.e. iron & steel and chemicals, the employment share working
in large establishments differs not between the industry sample of the comparisons and
Kinghorn and Nye’s data. So each of the three German strong-performing industries
produced on a scale not much smaller than their American counterparts.
The establishment-size data provides some empirical support for the notion that in
textile spinning, iron & steel, and chemicals Germany faced little problems when it
comes to production scale. If these industries were indeed involved in the production of
predominantly basic goods and did not face a demand for customized consumer goods,
the comparable establishment size in both countries suggests that technical constraints
to standardized production should be no less in Germany than in the US. However,
the efficiency effect of establishment size may be offset by differences between German
and US firm size. Optimal firm size is partly determined by the relative costs of market
transactions and if markets function poorly these transaction costs can be lowered by
integrating several (or all) stages of the production chain within a single firm. As the
size of firms was typically larger in the US than in Europe, this potentially endowed
American producers with an advantage over their German competitors.59
But there is a problem with this argument. In the case of firm size, being larger is not
always better. More specifically, as optimal firm size is determined by transaction costs,
the smaller firm in Germany might simply reflect a well-integrated market that reduced
the incentive for firms to extent their control over more stages of the production chain.
Country-specific conditions conducive to low transaction costs can thus limit the size
of firms. Cartels – a much favored model of industrial organization in Germany – could
have provided such conditions and the smaller firm size in Germany need not have been
a sign of backwardness.60 Cartels offered an alternative way to attain a reduction of
transaction costs. Through the control exerted by the cartel over different stages of the
production chain, coordination problems could be addressed efficiently without having
to integrate these production stages in one firm. Related to this, the stability offered
by cartels potentially induced higher rates of investment, leading to capital deepening
and productivity growth.61
Although cartels are associated with a reduced intensity of competition, moving Ger-
many away from competitive capitalism, the literature on Germany is strikingly positive
about the effect of cartels on economic development.62 If German cartels tended toward
a monopoly control of the market, they could have closed the door on technological
development, yet Burhop and Lubbers conclude that in the case of German coal-mining
corporations productivity was not significantly affected by cartel membership.63 Over
59. Chandler, “Organizational Capabilities,” 83.60. Kinghorn and Nye, “The Scale of Production,” 109; Hannah, “The American Mircale,” 207–208.61. Levenstein and Suslow, “What Determines Cartel Success,” 85.62. J. Kocka, “Entrepreneurs and Managers in German Industrialization,” The Cambridge Economic
History of Europe Vol. 7 (1978): 492–589, 564.63. C. Burhop and L. Lubers, “Cartels, Magerial Incentives, and Productive Efficiency in German
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 45
the period 1881–1913 there is no evidence that the Rheinisch-Westphalien Coal Syndi-
cate, one of the longest-lasting cartels, adversely influenced levels of technical efficiency.
In similar vein, Kinghorn argues that German coal and iron & steel cartels around
the turn of the century did not lead to true monopoly power, yet they did allow firm
members to use more efficient production technologies.64 Strikingly, the top three of
industries with the largest number of cartels included iron & steel, chemicals and tex-
tiles, i.e. precisely those industries that the benchmark comparisons showed to deliver
a strong performance relative to the US.65
Vertical integration: protectionist policy
Vertical integration was encouraged not only by the cartel system, but also by the pro-
tectionist policy that Germany maintained to restrict foreign competition. The case-
studies in the work of Webb are particularly useful in this respect. Being the leading
advocates of protective tariffs, the iron foundries, cotton-spinning mills, and large-scale
agriculturalists are centrally placed in Webb’s research. To interpret the impact of tariffs
on domestic production accurately, he measures the effective rate of protection, which
captures the tariff-instigated percentage increase in value added and thereby takes into
account the price change in both intermediate inputs and finished products.66 Webb
concludes that, together with cartels, protection encouraged vertical integration; as tar-
iffs raised domestic market prices above the world level, backward integration ensured
purchase of intermediate inputs against cost, rather than market prices.67 Moreover,
by stimulating vertical integration, the tariff system reinforced the stability of prices
already encouraged by cartels, which – as noted above – reduced the riskiness of invest-
ment in capital-intensive technologies.68 Smaller, non-vertically integrated firms faced
market prices above world level and, therefore, did not gain from protection. As a result,
trade tariffs favored the large-scale, more politically powerful enterprises.69
It is hard to say how the increased costs of intermediate inputs affected comparative
productivity between Germany and the US. As in the case of iron & steel and textiles –
Coal Mining, 1881–1913,” Journal of Economic History Vol. 69, no.2 (2009): 500–527, 502.64. J. Kinghorn, “Kartell or Cartel? Evidence from Turn of the Century German Coal, Iron and Steel
Industries,” Journal of Economic History Vol. 56, no. 2 (1996): 491–492, 492.65. Kocka, “Entrepreneurs and Managers,” 564.66. S. Webb, “Tariff Protection for the Iron Industry, Cotton Textiles, and Agriculture in Germany,
1879–1914,” Jahrbucher fur Nationalokonomie und Statistik Vol. 192 (1977): 336–357, 337.67. S. Webb, “Tariffs, Cartels, Technology, and Growth in the German Steel Industry,” Journal of
Economic History Vol. 40, no. 2 (1980): 309–330, 328.68. Additionally, for the pre-1950 period higher tariffs are associated with lower relative capital-
good prices and because the latter are negatively related to the rate of investment protection canstimulate growth through increased investment. W. Collins and J. Williamson, “Capital-Goods Pricesand Investment, 1870–1950,” Journal of Economic History Vol. 61, no. 1 (2001): 59–94, 80, 81.69. Webb, “Tariff Protection,” 353; Webb, “Tariffs, Cartels, Technology, and Growth,” 323.
i.e. the industries studied by Webb – protectionist policy confers certain cost advantages
to vertically-integrated and large-scale firms. However, these advantages are relative
to other, small-scale firms in German industry. This goes to show that trade tariffs
hurt the competitiveness of small establishments in particular, while larger firms could
only hope to avoid the backlash of protectionism and stand their ground relative to
the US through vertical integration. It is not evident how protectionism improved the
comparative performance of German industries.
Then again, even though domestic firms faced prices above market level as a conse-
quence of tariff walls, protectionist policy may have secured the survival of developing
industries by shutting out foreign competition. Given that many of the German modern
industries were outperformed by their American counterparts, as the comparisons show,
such an infant-industry approach suited these industries well. Indeed, it has been sug-
gested in the literature that in contrast to the post-1950 period, in which – by and large
– growth benefited from free trade, trade tariffs around the turn of the century were pos-
itively correlated with growth.70 As all through the first half of the twentieth century
manufacturing products entering Germany were (increasingly) tariffed, protectionist
policy may explain the growth captured by the German intertemporal benchmark.71
It should be noted, however, that the existence of a ‘tariff-growth paradox’ has been
called into question. Many scholars have used regression analysis to test the hypoth-
esis that economic growth was a function of protection, but different specifications of
the model have led to results both confirming (O’Rourke, 2000; Jacks, 2006) and re-
futing (Capie, 1983; Schularik and Solomou, 2011) the tariff-growth paradox.72 In any
case, the cry for protectionism was fueled in Germany by notions much different than
those set out by the infant-industry argument. The textile industry is a case in point;
not only provided tariffs protection for the strong (instead of the weak), cotton spin-
ning can hardly be described as an emerging industry. To take another example, when
70. P. Bairoch, “Free Trade and European Economic Development in the 19th Century,” EuropeanEconomic Review Vol. 3, no. 2 (1972): 211–245, 242; J.A. Frankel and D. Romer, “Does trade causegrowth?,” American Economic Review Vol. 89, no. 3 (1999): 379–399, 394; K. O’Rourke, “Tariffs andGrowth in the Late 19th Century,” Economic J Vol. 110, no. 463 (2000): 456–483, 473; D. Jacks,“New Results on the Tariff-Growth Paradox,” European Review of Economic History Vol. 10 (2006):205–230, 221.71. V. Hentschel, “German Economic and Social Policy, 1815–1939,” in The Cambridge Economic
History of Europe, ed. P. Mathias and S. Pollard, vol. Vol. 8 (1989), 752–813, 786;C.P. Kindleberger,“Commercial Policy between the Wars,” in The Cambridge Economic History of Europe, ed. P. Math-ias and S. Pollard, vol. Vol. 8 (Cambridge University Press, 1989), 161–196, 180; Broadberry, TheProductivity Race, 141.72. F. Capie, Tariffs and Growth; Some Insights from the World Economy, 1850–1940 (Manchester
University Press, 1994), 42; M. Schularick and S. Solomou, “Tariffs and Economic Growth in the FirstEra of Globalization,” Journal of Economic Growth Vol. 16, no. 1 (2011): 33–70 49, 56. Schularick andSolomou claim that the real paradox is not that free trade was bad for growth, but that changes ininternational economic policies seems to have mattered little to countries’ growth trajectories.
Chapter 2. Catching-Up with the Global Labor-Productivity Leader? 47
cheap foreign grain threatened domestic agricultural production, the politically power-
ful landowners successfully lobbied for tariffs.73 And to add insult to injury, because
tariffs on agricultural imports were on balance higher than for manufacturing goods,
the landowners effectively slowed down the rate of GDP per capita growth by delaying
the shift of employment toward high productive industries.74 Clearly, if protectionism
induces growth, it was an unintended byproduct of an otherwise strictly conservative
policy.
Relative factor costs
Apart from the differences in industrial organization between Germany and the US
described above, Europe’s inability to catch-up in general has been explained by the
Rothbarth-Habakkuk thesis. In Europe, factor and resource endowments as well as de-
mand patterns are said to have favored a labor-intensive way of production.75 Natural
resources were scarce and skilled labor was in ample supply, which provided an incentive
to economize on fixed capital in the form of machinery.76 In contrast, the US was well
endowed with natural resources, while skilled labor was relatively expensive. Machinery
was substituted for skilled labor, resulting in the use of capital-intensive production
techniques. This way, local circumstances determined the initial choice of technology.
Technological progress is subsequently directed toward the particular technological path
a country has chosen, leading to lock-in effects.77 As capital-intensive production tech-
niques are associated with higher labor-productivity levels, Europe could not catch-up
with the US.
A study of capital-intensity levels lies outside the scope of this paper. Chapter 3
returns to this issue and provides a thorough analysis of technological development in
both countries. Nevertheless, some remarks are in place here. The benchmark results
do not fit the Rothbarth-Habakkuk thesis exceptionally well. The German industries
that performed on par with their US counterparts challenge the deterministic nature
of the initial-conditions approach. Some scholars – most notably Rosenberg – have de-
scribed America’s lead as foreordained; US resource endowments acted as a benevolent
73. P. Bairoch, “European Trade Policy, 1815–1914,” in The Cambridge Economic History of Europe,ed. P. Mathias and S. Pollard, vol. Vol. 8 (Cambridge University Press, 1989), 1–160, 76; C.P. Kindle-berger, “The Rise of Free Trade in Western Europe, 1820–1875,” Journal of Economic History Vol.35, no. 1 (1975): 20–55, 46.74. S.N. Broadberry, “How Did the United States and Germany Overtake Britain? A Sectoral Analysis
of Comparative Productivity Levels, 1870-1990,” The Journal of Economic History Vol. 58 (1998): 375–407, 386; Hannah, “The American Mircale,” 201.75. Habakkuk, American and British Technology.76. Temin, “Labour Scarcity,” 162;Field, “On the Unimportance of Machinery,” 379.77. David, Technical Choice, 66
Providence, inevitably setting the stage for America’s edge over Europe.78 However, the
strong German performance in textiles, primary-metals manufacturing, and chemicals
suggests that in these industries either similar production techniques (i.e. capital-to-
labor ratios) were employed by both countries or that higher levels of capital intensity do
not necessarily translate into higher labor-productivity levels. Either way, initial condi-
tions (whatever those were) did not prevent these particular industries from catching-up
and the Rothbarth-Habakkuk thesis – originally suggested as a possible explanation for
19th century Anglo-American, rather than 20th century German-American productivity
differences – sits uncomfortably with the case of Germany.
Some of the cross-industry differences in performance might be explained by the
variance in the degree to which industries rely on raw materials and capital-intensive
production techniques. However, the importance of natural resources in, for instance,
the iron & steel and the textile industries is undeniable, both in the form of raw materials
and as combustibles. A more likely explanation is that factor costs in the ‘successful’
German industries deviated only little from those in the US. In a case study on the pre-
WW1 iron & steel industry, Bob Allen accounted for price differences between German,
American, and British iron products by studying the costs of materials used and the
efficiency of production. Allen finds that in 1910 the price of used raw materials (ore
and scrap) was actually lower in Germany than in both the US and UK.79 Fuel (blast
furnace coke) was more expensive as compared to the US, but cheaper than in Britain.
In line with the benchmark results, Allen shows that productivity in Germany was
comparable to the US and higher than in the UK.80 Iron production in Germany was
characterized by low material costs and high efficiency levels.
Apparently, the costs of using capital in the primary-metals industry did not dif-
fer much between Germany and the US. Does this mean that both countries operated
similar production techniques? In the late nineteenth century, the Bessemer process for
the mass-production of steel from molten pig iron revolutionized the iron & steel indus-
try. Although the large-scale application of the Bessemer process was introduced first
in Britain, the technology was swiftly improved upon in the US so that by the 1880s
the coke-fueled blast furnaces developed in America formed the pinnacle of available
production techniques.81 Hyde shows that when American steel-producing technologies
78. Rosenberg, “Why in America?” 112.79. R.C. Allen, “International Competition in Iron and Steel, 1850–1913,” Journal of Economic His-
tory Vol. 39 (1979): 911–937 932. British iron ore mined in the East Midlands and Cleveland was atleast as cheap as the German ore from West-Phalia, but for some reason Britain mainly used the moreexpensive Spanish ore.80. ibid. 931.81. C. Hyde, “Iron and Steel Technologies Moving Between Europe and the United States, Before
1914,” in International Technology Transfer. Europe, Japan and the USA, 1700-1914, ed. D.J. Jeremy
Chapter 3The Yanks of Europe? Labor Productivity and
Technology in German and US Manufacturing, 1899–1939
3.1 Introduction
The pattern of divergence between German and American manufacturing uncovered
in the previous chapter roughly aligns with the 2:1 ratio suggested by Broadberry for
the transatlantic labor-productivity gap in the early twentieth century.1 Traditionally,
Europe’s inability to catch-up has been partly ascribed to local circumstances, i.e. factor
and resource endowments as well as demand patterns, which favored labor-intensive
production.2 In Europe, natural resources were scarce, while skilled labor was in ample
supply, the combination of which provided an incentive to economize on fixed capital in
the form of machinery.3 In comparison, the US was well endowed with natural resources,
while skilled labor was relatively expensive. Therefore, machinery was substituted for
skilled labor, resulting in a capital-intensive production process.
Furthermore, as the American demand for goods was more homogenous and given
the size of the US domestic market, manufacturers could standardize production, imple-
ment high throughput systems, and thereby raise productivity levels.4 This advantage
was denied to European countries, which faced heterogeneous markets characterized by
a demand for customized goods. Thus, local circumstances determined the initial choice
of capital-labor ratios. If technological progress is directed towards the capital-labor ra-
tios currently in use, local circumstances can lead to technological lock-in.5 Assuming
that the increase of labor productivity achieved at high capital-intensity levels surpass
1. Broadberry, The Productivity Race, 3; Broadberry and Irwin, “Labor Productivity in the UnitedStates and the United Kingdom,” 265.
2. Habakkuk, American and British Technology.3. Temin, “Labour Scarcity,” 162; Field, “On the Unimportance of Machinery,” 379.4. Broadberry, “Technological Leadership,” 291.5. David, Technical Choice, 66.
those realized at low capital-labor ratios, labor-productivity levels will differ across the
Atlantic.
However, three pillars of this explicative model have recently been called into ques-
tion. First, the notion of technological lock-in flies in the face of widespread transatlantic
technology transfer recorded during the early twentieth century. Contemporary industry
periodicals report a good many cases where German manufacturers imported Ameri-
can machinery and incorporated these in domestic production lines.6 Apart from a new
coat of paint, imported American machinery was often installed in its original form; ev-
idence that contradicts the development of dichotomous technological paths.7 Second,
case studies reveal a process of rapid capital deepening over the interwar period in the
German machine-tool industry.8 By the late 1930s the number of machines installed on
the factory floor available per worker was comparable between the US and Germany.
The potential external benefits may be substantial, as developments in the machine-tool
industry spill over to other manufacturing industries that extensively use machinery.
Third, the stereotypical US high-throughput model has been downplayed lately. Only
a minority of American industries actually employed thoroughgoing mass-production
systems; a much larger share of manufacturing focused on specialized, European-type
production processes.9 This begs the question whether the Germans ought to be seen
as ‘the Yankees of Europe’?10
In addition to these case-studies, quantitative research on the level of total manufac-
turing also failed to confirm the alleged importance of capital-intensity differences for the
labor-productivity gap. Decomposing the US/UK and German/UK labor-productivity
gap for years between 1870 and 1950 in effects of comparative total-factor productivity
and comparative capital intensity, Broadberry finds the latter component to explain lit-
tle of the observed labor-productivity differences.11 Using Broadberry’s data and decom-
position framework, only about 25% of the German/US labor-productivity gap in both
1909 and 1937 is explained by differences in capital intensity. Yet in spite of the modest
contribution assigned by the decomposition to the role of capital-intensity differences,
this finding has not led to a reinterpretation of the German/US labor-productivity gap.
Instead, the lack of strong empirical evidence in support of the supposed significance
6. Richter and Streb, “Catching-up and Falling Behind,” 1–2.7. Richter, “Technology and Knowledge Transfer.”8. C. Ristuccia and J. Tooze, “The Cutting Edge of Modernity: Machine Tools in the United States
and Germany 1930–1945,” Cambridge Working Papers in Economics No. 0342 (2003): 1–48.9. P. Scranton, Endless Novelty. Specialty Production and American Industrialisation 1865–1925
(Princeton University Press, 1997).10. Quote obtained from Kindleberger. See C. Kindleberger, Economic Response: Comparative Stud-
ies in Trade, Finance, and Growth (Cambridge, Mass.: Harvard University Press, 1978), 188.11. Broadberry, The Productivity Race, 105,106; Broadberry, “Manufacturing and the Convergence
of variation in capital-labor ratios has been attributed to methodological weaknesses
of the level-accounting exercise based on the standard Solow model.12 Three problems
stand out in this respect.
A first deficit relates to the nature of technological change. In the conventional
Solow-based decomposition framework the effect of technological change is proportion-
ate at any level of capital intensity. New production knowledge increases the labor-
productivity potential by the same factor everywhere along the production possibility
frontier. Technological change was, however, not factor neutral, but localized and capital
biased; Allen shows that ever since the first industrial revolution innovation took place
at the highest capital-labor ratios in use, while low capital-intensive technology saw
little or no improvement.13 This has consequences for the impact of variation in capital
intensity on labor-productivity differences. If innovation was indeed localized, countries
operating at capital-labor ratios unaffected by technological change faced a widening
labor-productivity gap relative to countries that did enjoy the benefits of innovation.
A second inadequacy, related to the implementation of the level-accounting exercise,
concerns the use of total capital-stock data per worker as a measure for the capital
intensity of production. The size of the capital stock is determined largely by stocks
of buildings and inventories and the value of the machinery and equipment stock is
low in comparison.14 Because America’s alleged capital-intensity advantage pertains to
machinery, rather than to buildings, it is inappropriate to use total capital-stock data
to calculate capital intensity for the question addressed here, as I will show later on.
A measure of machine intensity is much better suited for the purpose and more apt to
capture the dynamics in German manufacturing as described above for the machine-tool
industry.
Thirdly, a Solow-based decomposition framework requires information on the shares
of capital and labor in output to proxy the marginal factor returns. Using a weighted
average of Germany, the UK and the US for 1975, Broadberry attributes a weight of 0.23
to capital.15 The capital-intensity gap between Germany and the US must be incredibly
large in this setting to make a substantial contribution to the labor-productivity gap.
Compared to 1975, wage levels were relatively low and capital costs relatively high in
the early twentieth century and, consequently, the effect of capital intensity may be
severely underestimated. Without additional factor price information these shares are
notoriously difficult to pin down.
12. Broadberry, The Productivity Race, 106–109.13. Allen, “Technology and the Great Divergence,” 6.14. Field, “On the Unimportance of Machinery.”15. Broadberry, The Productivity Race, 105.
mum of assumptions, the decomposition framework itself is firmly grounded in theo-
retical growth models. Particularly relevant in this respect is Basu and Weil’s model
of appropriate technology in which new production knowledge is appropriate only for
a limited range of capital-labor ratios. This setting identifies two channels for labor-
productivity growth depending on a country’s initial position; through innovation for
“leader countries” and for “follower countries” by adopting capital-labor ratios already
explored by leader countries in the past.18 When innovation is confined to high capital-
labor ratios, as Allen demonstrates for my period of study, low-end countries must strive
for higher levels of capital intensity or face an ever increasing labor-productivity gap.19
Yet the process of rapid capital intensification can come at a cost, as empirical studies
point out.20 As much of what one needs to know to employ new production knowledge
is implicit and not available from handbooks, it takes time to assimilate and operate
machinery at the level displayed by countries exploring that capital-labor ratio before.21
According to Los and Timmer, these findings suggest a sequence in which developing
countries first create scope for labor-productivity growth by adopting high capital-labor
ratios and, subsequently, ‘learn’ to operate efficiently at that capital-intensity level, from
which point onwards the latent labor-productivity gains materialize.22
The two sources of labor-productivity growth set out by the model, i.e. capital inten-
sity and efficiency, provide a framework that possibly explains the paradoxical lack of
labor-productivity catch up at a time of capital-intensity convergence between German
and US manufacturing. If innovation was localized and took place at high capital-labor
ratios only, Germany – a follower country – faced a strong incentive to increase capital-
intensity levels, but the associated labor-productivity gains may not have materialized
in the short run as German industries struggled to learn how to operate efficiently at the
the new capital-labor ratios, a process that required time. The next section discusses
the analytical framework necessary to test whether such dynamics can be identified for
German manufacturing over the interwar period. Subsequently, section 3.3 describes the
data, while the results are presented in section 3.4. After positioning these results in the
more qualitative literature on German historical economic development in section 3.5,
the next section (3.6) adopts a long-term perspective and positions the findings of this
chapter in long-run developments. Finally, section 3.7 concludes.
18. Basu and Weil, “Appropriate Technology,” 1036.19. ibid., 1043–1045.20. Los and Timmer, “The ’Appropriate Technology’ Explanation,” 519.21. A. Atkinson and J. Stiglitz, “A New View of Technological Change,” The Economic Journal Vol.
79 (1969): 573–578; David, Technical Choice, 59–60.22. Los and Timmer, “The ’Appropriate Technology’ Explanation,” 529.
In order to see if the model set out above applies to German manufacturing in the early
twentieth century, I follow a two-stage research strategy. First, I look at the level of
capital intensity in German manufacturing in 1909 and 1936 and measure the rate of
labor-productivity growth if machinery was operated at full efficiency throughout the
entire period. In order to do so, I need to know what the maximum labor-productivity
potential for explored capital-labor ratios was. This requires a global best-practice labor-
productivity frontier that captures the highest attainable levels of labor productivity
over the full range of explored capital-labor ratios by any country in the world and
allows for localized innovation. Secondly, I decompose the labor-productivity gap in
manufacturing industries between Germany in 1936 and the US in 1939. This gap is
attributed to differences in both capital intensity and the efficiency at which machinery
is operated. Both components can be measured only by knowing what the full labor-
productivity potential is at the capital-labor ratios adopted by Germany and the US,
for which purpose the global best-practice frontier is used again.
The first part of the analysis quantifies the potential return to the adoption of high
capital-labor ratios and illustrates the (ex-post) incentive for capital-intensification in
German manufacturing. The second part measures how efficiently the machinery stock
was operated in German manufacturing by the late 1930s, following a period of capital
intensification, and indicates the degree to which the transatlantic labor-productivity
gap was sustained by differences in efficiency levels between Germany and the US.
Because both steps in the analysis require a global best-practice frontier, the estimation
of this frontier provides the basis for the analysis of labor-productivity differences.
Therefore, this section discusses the data envelopment analysis (henceforth, DEA) used
to construct the frontier first, before describing the decomposition of the German/US
labor-productivity gap in 1936/39.
The global best-practice frontier
Using DEA-techniques the global best-practice frontier is estimated in four sequential
steps. A first step involves collecting data on the level of capital intensity and labor
productivity for manufacturing industries. Subsequently, manufacturing industries that
produce similar products are sorted into groups. For this purpose the standard industrial
classification of 1945 (henceforth, SIC) is used.23 Thirdly, all industries classified in
23. For a detailed overview of the SIC, see United States Department of Commerce: Bureau of theCensus, Census of Manufactures 1947, vol. Industry Description (Washington: United States Govern-
the same group are placed in 〈k, y〉 space, where k is capital intensity and y is labor
productivity. In a final step, the global best-practice frontier is drawn by enveloping
the data in the tightest-fitting convex cone using linear line segments.24 The upper
boundary of the envelop represents the global best-practice frontier.25 This exercise is
then repeated for several periods to also obtain the movement of the frontier over time.
Figure 3.1: Estimating the frontier for industrial chemicals, 1899–1939
0
1
2
3
4
5 10
y
k
(a) Observations up to and incl. 1909
0
1
2
3
4
5 10
y
k
(b) Observations up to and incl. 1939
0
1
2
3
4
5 10
F(’09)
F(’39)
y
k
(c) The frontiers in 1909 and 1939
0
2
4
5 10
F(’09)
F(’39)
(’09)
(’36)
y
k
ya
yb
(d) German chemical industry, 1909 and 1936
Figures 3.1a until 3.1c capture the procedure described above for the case of indus-
trial chemicals. The top-left pane draws the frontier for 1909 based on data from the
US, Germany and the UK. Although not truly global, the frontier contains the three
leading industrial nations, both in terms of size and labor-productivity levels, of the
ment Printing Office, 1949).24. Kumar and Russell, “Technological Change,” 530.25. For best-practice industries, see W. Salter, Productivity and Technical Change, second edition
early twentieth century. The data on which the frontier for 1909 is based includes all
current and past observations. By including observations from earlier years, knowledge
previously generated is ‘remembered’ over time and remains accessible in the current
period.26 In practice, this means that the cloud of observations in figure 3.1a contains
data from 1899, 1905 and 1909 for the US, 1907 for the UK, and 1909 for Germany.
This information is derived from the published production censuses or other statistical
sources. Section 3.3 provides more detail on the dataset constructed for the DEA.
The top-right pane of the figure shows how the frontier changes over time, in this
case between 1909 and 1939. All observations already included in the frontier estimation
for 1909 are exported to figure 3.1b and show up as the gray dots. In addition, the
figure plots all available observations for the period afterward up to and including
1939, which, as before, are collected from the production censuses. This includes data
on 1936 for Germany, 1930 for the UK and all census years since 1909 for the US.
Subsequently, the frontier is drawn. Leaving out the observations that are not part of
the frontier, figure 3.1c clearly shows the upward shift of the frontier between 1909
and 1939. Throughout the rest of this chapter this outward movement of the frontier is
referred to as technological change. As the frontier captures the highest achieved levels
of labor productivity for contemporaneously or previously explored capital-labor ratios,
the vertical distance between an industry observation and the frontier determines the
efficiency level at which technology is operated. For instance, figure 3.1d positions the
German chemical industry relative to the frontier in 1909 and 1936. Clearly, the level of
efficiency declined with the increase of the distance to the frontier. Linear programming
techniques are used to accurately calculate an industry’s vertical distance to the best-
practice frontier.27
An appeal of the DEA approach is that the shape of the frontier can be revealed
without imposing a specific functional form on technology.28 The convexity of the en-
velop poses the only restriction on the functional form of the frontier. As the param-
eters of the production frontier are obtained from the data, rather than presupposed,
the DEA allows for any form of localized technical change.29 That is, an increase in
the labor-productivity potential at particular capital-intensity levels through innova-
26. The unlikeness of an ‘imploding frontier’ was already noted by Basu and Weil, “AppropriateTechnology,” 1031, 1036 and Kumar and Russell, “Technological Change,” 540, but the first to formallyexclude the possibility of technological degradation were Los and Timmer (Los and Timmer, “The’Appropriate Technology’ Explanation”).27. R. Fare, S. Grosskopf, and K. Lovell, Production Frontiers (Cambridge: Cambridge University
Press, 1994), 1–296; Kumar and Russell, “Technological Change,” 531. For the linear programmingtechniques, see appendix 3.A.28. Fare, Grosskopf, and Lovell, Production Frontiers, 12.29. Los and Timmer, “The ’Appropriate Technology’ Explanation,” 522.
As described in the previous section, for the DEA and the labor-productivity decom-
position industry-level data is required on labor input, capital input and output. The
complete data set used here entails approximately 105 separate industries, which are
classified in 28 SIC industry groups, and in total consists of nearly 1,200 observed
input-output combinations, including US, UK and German observations for years be-
tween 1899 and 1939. US data is available for 1899, 1905, 1909, 1914, 1919, 1929 and
1939, omitting only the census year 1935 on account of unreported capital data.30 For
the UK data is obtained for 1907 and 1930, while the data set includes German ob-
servations for 1909 and 1936.31 Thus, German manufacturing industries in 1936 must
be compared to their American counterparts in 1939 and the labor-productivity gap is
decomposed using the 1939 frontier.
Capital input
The level of technology is measured by capital intensity, but instead of using total
capital-stock estimates, as traditional level-accounting exercises have done for this pe-
riod, I rely on a measure of machine intensity.32 Much of the production knowledge
gained since the second industrial revolution was contained in tangible capital in the
form of machinery installed on the factory floor. So a measure of capital that captures
the stock of machinery is needed. Because total capital-stock data contains other invest-
ment components besides equipment, i.e. inventories and buildings, they do not accu-
rately capture the level of machine intensity; the share of machinery in the total capital
stock is typically small and conclusions regarding the process of machine intensification
drawn from total capital-stock data are bound to mislead.33 Indeed, it has been shown
for the period 1960–1985 that the correlation with GDP growth was much stronger for
changes in equipment than for any other component of investment.34 Although data on
30. United States Department of Commerce: Bureau of the Census, US Census of Manufactures 1910(VIII); United States Department of Commerce: Bureau of the Census, US Census of Manufactures1914 ; United States Department of Commerce: Bureau of the Census, US Census of Manufactures 1920(VIII); United States Department of Commerce: Bureau of the Census, US Census of Manufactures1929 (II); United States Department of Commerce: Bureau of the Census, US Census of Manufactures1935 ; United States Department of Commerce: Bureau of the Census, US Census of Manufactures1940 (II).31. For the UK, Board of Trade, UK Census of Production 1907 ; Board of Trade, UK Census of
Production 1930. For Germany, see chapter 2.32. Broadberry, The Productivity Race, 105, 106.33. Field, “On the Unimportance of Machinery.”34. B. de Long and L. Summers, “Equipment Investment and Economic Growth,” The Quarterly
Journal of Economics Vol. 106, No. 2 (1991): 445–502; B. de Long and L. Summers, “EquipmentInvestment and Economic Growth: How Strong is the Nexus?,” Brookings Papers on Economic Activity
A different source of potential worry when using horse-power statistics concerns the
danger of ‘double counting’. The horse power applied on the factory floor is supplied
by machinery running on either non-electric or electric power. In case of the latter, the
electricity needed to operate the machinery can be internally generated in the factory
(by electricity generators) or purchased from an electrical power network to which the
No. 2 (1992): 157–211.35. As reported by the Census of Manufactures for the US, the Census of Production for the UK,
and the Gewerbliche Betriebszahlung/Gewerbliche Betriebsstatistik for Germany.36. Schurr et al., Electricity in the American Economy, 32; Jerome, Mechanization in Industry, 250,
factory is connected. When the electricity is internally generated, the horse power used
by electricity generators should be excluded in the analysis as it does not contribute
directly to the fabrication of goods. Only for pre-WW1 Germany the data does not
allow a correction for double counting.37 This is not a source of major concern, because
the share of electricity in total horse power was modest before the 1920s and the part of
electric power internally generated even smaller. If anything, the bias provides a lower
bound estimate of machine intensification between 1909 and 1936 and underestimates
the created potential for labor-productivity growth.
A final worry concerns the horse-power data for interwar Germany. The 1936
machine-intensity level is indicated by horse-power data obtained from the employment
census of 1933, a procedure that introduces a bias in the capital-intensity estimates.38
In 1933 unemployment in Germany stood at 36.2%, only slightly lower than the all-time
high level of 43.8% the year before.39 By 1936 the unemployment rate had decreased to
12.0% and the employed labor force in manufacturing was 38% larger than in 1933.40
As 1936 was the first year that saw employment levels equal to those of before the Great
37. Hoffmann, Das Wachstum, 263–264.38. Statistik des Deutschen Reichs, “Gewerbliche Betriebszahlung,” in Volks-, Berugs- und Be-
triebszahlung vom 1933 (Berlin: Verlag fur Sozialpolitik, Wirtschaft und Statistik, 1933).39. T. Pierenkemper, “The Standard of Living and Employment in Germany, 1850-1960: An
Overview,” Journal of European Economic History Vol. 16 (1987): 51–73, 59.40. ibid., 59; Hoffmann, Das Wachstum, 199.
Depression, a part of the installed horse power reported for 1933 may have stood idle or
underutilized on the factory floor. Although the horse-power statstics only report the
frequently used horse power on the factory floor, capital-labor ratios may be spuriously
high.
Alternatively, horse-power data is available for 1938 as well, but these suffer from
three problems and cannot be used. First, the coverage of industries is low as compared
to the 1933 data. Second, by 1938 the German statistical publications hid production
activities related to the war effort. And, third, the build-up for war affected the employ-
ment structure of German manufacturing. Although these impairments have a limited
impact on total manufacturing, the distortions can be quite pronounced on the industry
level and very difficult to identify. Given the importance for this study of data on the
disaggregated level, I prefer the use of the 1933 horse-power statistics. Nevertheless, to
check the sensitivity of the analysis, the decomposition of the labor-productivity gap is
done using German machine-intensity data of both 1933 and 1938. The results do not
differ in any major way (see appendix 3.D) and the findings are robust to variation in
the German level of machine intensity.
Output and labor input
Output is measured by value added as reported by the statistical publications of the
US, the UK and Germany.41 This is necessary to avoid movements of the frontier that
are driven by changes in input prices, rather than improvements of the production
process. German and British output is expressed in US$ using industry-specific output
PPPs.42 Subsequently, nominal value added in US$ is converted to constant prices (with
a 1929 base) by applying price deflators at the industry level. Deflators are calculated
on the basis of Fabricant’s indices of physical output and nominal output series.43
After reclassification to fit the SIC, the modifications and extensions to the indices
of production proposed by Kendrick are incorporated, too.44 Labor input is expressed
in terms of hours worked. The necessity of the hours adjustment has been stressed
41. For the US the Census of Manufactures. For the UK the Census of Production (1907 and 1930)and for Germany the industrial surveys (1909, see chapter 2) and the first industrial census (1936, seechapter 2).42. For Germany/US the PPPs constructed in chapter 2 are used. For UK/US the PPPs are used
of de Jong and Woltjer (interwar period) as well as Veenstra and Woltjer (pre-WW1 period). SeeJong and Woltjer, “Depression Dynamics” and J. Veenstra and P.J. Woltjer, “The Yanks of Europe?Technological Change and Labor Productivity in German Manufacturing, 1909–1936,” XVIth WorldEconomic History Congress (2012).43. S. Fabricant, The Output of Manufacturing Industries, 1899–1937 (New York: National Bureau
Economic Analysis, 1940), 123–321; 605–639.44. J.W. Kendrick, Productivity Trends in the United States (Princeton N.J.: National Bureau Eco-
in chapter 2 and by de Jong & Woltjer in their study on US/UK labor-productivity
differences.45
3.4 Results
The main findings of this chapter can be summarized in four points. First, technological
change at the global production frontier over the period 1899–1939 was decidedly non-
neutral and biased toward capital. Second, in terms of machine-intensity levels Germany
gradually converged on the US over the interwar period. Third, due to the bias in
technological change, the process of adopting machine-intensive technology markedly
increased the scope for labor-productivity growth in German manufacturing. Fourth,
the German/US labor-productivity gap in the late 1930s was mainly due to a relatively
low level of efficiency in factor use, rather than different capital-labor ratios. The created
potential for growth was not realized before WW2 and Germany failed to efficiently
assimilate new production knowledge in the short run.
The global best-practice production frontier between 1899–1939
The movement of the global best-practice production frontier between 1899 and 1939,
as measured by the DEA, contains a bias toward machine-intensive technology. Tech-
nological change did not shift the frontier by the same proportional amount at all
capital-labor ratios. Instead, innovation was localized at the machine-intensive side of
the production frontier.
This finding aligns well with the DEA-literature discussed before. Using similar
techniques, Kumar and Russell concluded for the period 1965–1990 that technological
change has been decidedly nonneutral; outward movement of the frontier was localized
at predominantly high levels of capital intensity.46 Timmer and Los also uncover very
similar dynamics for a broad sample of OECD countries in the last quarter of the
twentieth century; innovation was highly localized and skewed toward the higher capital
intensities.47 Allen shows that the capital bias in technological change was not restricted
to the post-WW2 period. Since the first industrial revolution the global production
frontier shifted upward only at the highest capital-labor ratios in use, while low capital
45. The interwar period saw a substantial drop in the average hours of work for the interwar period. Asthe decrease in hours of work was more pronounced in the US relative to Europe, adjusting for hourswidens the labor-productivity gap between the US and Europe. See Jong and Woltjer, “DepressionDynamics,” 485–488.46. Kumar and Russell, “Technological Change,” 529, 538.47. Timmer and Los, “Localized Innovation,” 55.
tries deviated only little from those in the US. Or if they did, it did not deter German
entrepreneurs from acquiring higher levels of machine intensity.
Nevertheless, in spite of the rapid increase in machine-intensity levels, German man-
ufacturing still lagged behind its American counterpart. With the exception of electrical
machinery, German industries failed to fully close the machine-intensity gap. The short
time between the hyperinflation and the Great Depression offered only so much room
for extensive revisions to the production process. When the depression hit Europe in
1929 Germany had enjoyed relative stability for less than a decade and many long-term
projects slowed down, stalled, or were canceled all together.50 With the exception of
the machine-tool industry, Germany never reached the level of mechanization displayed
by the forerunners of American industrial development, such as Ford.
It was an often entertained notion that due to labor unions’ increased bargaining
power after WW1 investment was constrained by rising wage costs.51 Although the
consensus view now is that labor costs failed to harm investment worse than it did before
1914, the “roaring twenties” in Germany are broadly agreed to have been confined to
50. M. Nolan, Visions of Modernity. American Business and the Modernization of Germany (OxfordUniversity Press, 1994), 132.51. K. Borchardt, “Zwangslagen und Handlungsspielrame in der großen Wirtschaft der fruhen
dreißiger Jahre,” Jahrbuch der Bayerischen Akademie der Wissenschaften (1979): 85–132; K. Bor-chardt, Perspectives on Modern German Economic History and Policy (1991); A. Ritschl, “Zu hoheLohne in der Weimarer Republik? Eine Auseinandersetzung mit Holtfrerichs Berechnungen zur Lohn-position der Arbeiterschaft 1925–1932,” Geschichte und Gesellschaft Vol. 16 (1990): 375–402.
the brief period between hyperinflation and depression, i.e. between 1924–1928.52 As the
alleged investment boom associated with the armament race in the late 1930s effectively
crowded out investment in both the private and public sectors, all together, the window
of opportunity for capital deepening in interwar Germany was rather small, which helps
understand the lack of full catch-up in machine intensity.53
Creating potential for labor-productivity growth
The specific focus on capital in the form of machinery unveils a tradition of machine
intensification in German manufacturing. As the labor-productivity potential increases
with machine intensity this process created substantial scope for labor-productivity
growth in German manufacturing. Table 3.3 illustrates this potential for growth that
German industries realized as a result of increased capital intensity. The first column
lists the average annual labor-productivity growth at the frontier over the interwar
period at the capital-labor ratios operated by German industries in 1909. As such it
reports the increase in labor-productivity potential had German manufacturing failed
to increase machine-intensity levels after 1909 and provides the counter-factual scenario
of technological lock-in in the strictest sense. The second column captures the average
annual labor-productivity growth at the frontier as a result of the actual change in
machine-intensity levels in German manufacturing industries. The difference between
the columns can be interpreted as the created potential for labor-productivity growth
through machine intensification in Germany.
The first column clearly shows the potential danger of lock-in. For many industries
the frontier changed only little at the capital-labor ratios displayed in 1909, a conse-
quence of the localized and capital-biased nature of technological change. Innovation
and introduction of new technology on the frontier, in this case only by US industries,
took place chiefly at high machine-intensity levels. In the cases of, for instance, metals
and machinery there was practically no movement of the frontier at all at these low
capital-labor ratios. To increase the potential for labor-productivity growth, machine
intensification was a necessity in these industries. Even in textiles, in which technological
change did manifest at low capital-labor ratios, frontier movements were much greater
at higher levels of machine intensity. The second column lists the created potential for
labor-productivity growth at the frontier as a result of the actually realized increase in
machine-intensity levels between 1909 and 1936. The reported growth rates are much
52. H.J. Voth, “With a Bang, not a Whimper: Pricking Germany’s “Stock Market Bubble” in 1927and the Slide into Depression,” Journal of Economic History Vol. 63, no. 1 (2003): 66.53. J. Scherner, “‘Armament in Depth’ or ‘Armament in Breadt’? German Investment Pattern and
Rearmament during the Nazi Period,” Economic History Review Vol. 66, no. 2 (2013): 13.
Table 3.3: Annual labor-productivity growth (ln %) at the frontierbetween 1909–1939 for German capital-labor ratios
Industry At machine intensity of
1909 1909–36
Food, drink and tobacco 0.5 4.7
Textiles and apparel 1.2 3.2
Paper and printing 0.8 3.4
Chemicals, petroleum, coke and rubber 0.6 4.3
Stone, clay and glass 0.7 3.2
Primary and fabricated metals 0.3 3.3
Machinery (incl. electric) 0.1 1.7
Transportation equipment 0.9 6.1
Miscellaneous 0.3 2.9
Total manufacturing 0.6 3.4
For more detail, see appendix 3.C. Sources: see text, section 2.3.
higher than those under the first column, indicating that the potential reward to capi-
tal deepening was very large. If German industries operated fully efficient, the increase
of machine intensity recorded in section 3.4 would have pushed up labor-productivity
levels in manufacturing at an average annual rate of 3.4% (ln), more than five times as
fast as in the counter-factual situation of stagnant machine-intensity levels. For several
industries, such as food, chemicals and transportation equipment, the potential gains
were even larger.
The growth rates in table 3.3 reflect the dynamics of the frontier at German levels
of machine intensity and, as such, capture the effects of innovation appropriate for
German industries. An improved understanding of the displayed patterns can therefore
be obtained by looking at the history of technological change over the interwar period.
Take for instance the transportation equipment industry. Table 3.3 reports almost no
change of the frontier at low levels of machine intensity, but rapid change at high levels
of machine intensity. This aligns with the literature, which has put emphasis on the key
position that the process of mechanization claimed in the development of this industry,
as illustrated by the introduction of Ford’s assembly line in the early 1920s.54 The
movement of the frontier corroborates the notion of rapid labor-productivity growth
induced by increasingly high levels of machine intensity.
The capital bias is less pronounced in the textile industry. Although adoption of
54. R.R. Nelson and G. Wright, “The Rise and Fall of American Technological Leadership: ThePostwar Era in Historical Perspective,” Journal of Economic Literature Vol. 30 (1992): 1931–1964,1944–45.
high machine-intensity levels created substantial additional scope for labor-productivity
growth, the frontier moved outward for low-end technology, too. This development pat-
tern may be explained by the lack of major technological breakthroughs during the
interwar period. The latest technological revolution experienced in textiles dated from
the late nineteenth century with the introduction of the ring spindle, which replaced the
less productive self-acting spinning mule.55 Over the interwar period the ring spindle
was adopted widely and subsequent productivity gains derived from further improve-
ments of the spindle, such as an increase of the spindle’s rate of revolutions.56 This
suggests a tradition of learning-by-doing in textiles, as a result of which the labor-
productivity potential of the ring-spindle technology was exploited to an increasing
extent. Nevertheless, the first half of the twentieth century saw a reduction in the rate
of labor-productivity growth, which suggests that the gains derived from learning-by-
doing fell short of those obtained through the switch from the old self-acting mule to
the ring spindle at the end of the nineteenth century.57
As with the transportation equipment industry, in chemicals technological develop-
ment took place at predominantly high machine-intensity levels. It has been noted in
the literature that in chemicals the subsequent stages of production are closely linked,
resulting in a continuous production line that combines different steps of the produc-
tion chain.58 This promoted not only vertical integration and large-scale production, it
also encouraged mechanization and automation, which accounts for the capital bias in
chemicals. Moreover, the fast pace of change pertains to new technology introduced in
the early 1920s, such as the production of synthetic fuels, rubber, and artificial resins.59
Other industries experienced technological change as well. For example, in primary
metals the open-hearth furnace – a late-nineteenth century innovation – was widely
adopted, in food industries conservation methods revolutionized, and paper machines
were both widened to increase the surface of paper under process and equipped with
multiple engines to improve performance.60 Apart from these industry-specific techno-
logical changes, the whole of manufacturing enjoyed the benefits from electrification.
Although electricity was introduced already in the late nineteenth century, it was the
55. J. Radkau, Technik in Deutschland vom 18. Jahrhundert bis zur Gegenwart (1989), 185–86.56. G. Egbers, “Innovation, Know-How, Rationalization, and Investment in the German Textile In-
dustry During the Period 1871–1935,” Zeitschrift fur Unternehmensgeschichte Beiheft 22 (1982): 234–256, 243.57. ibid., 242 (diagram 3).58. R. Berthold, ed., Produktivkrafte in Deutschland, 1917/18 bis 1945 (Akademie-Verlag Berlin,
1988), 126.59. ibid., 125.60. U. Wengenroth, Enterprise and Technology. The German and British Steel Industries, 1865–
1895 (Cambridge University Press, 1994), 195; Berthold, Produktivkrafte in Deutschland, 141 andibid., 137–38.
first half of the twentieth century that witnessed the widespread application of elec-
tricity in manufacturing industries.61 Among its many advantages, electricity allowed
for a more efficient (and flexible) lay out of factory-floor design as machinery no longer
relied on a single drive shaft for propulsion.62 Production processes characterized by
high machine-intensity by nature profited most from the productivity gains associated
with electricity.
Decomposition of the labor-productivity gap in 1936/39
It is clear that the reduction of the German/US machine-intensity gap over the
interwar period, although far from complete, created a large potential for labor-
productivity growth in German manufacturing. Moreover, this created potential for
labor-productivity growth was larger in Germany than in the US, a necessary condi-
tion for catch-up.63 But the increase of machine-intensity levels proved insufficient to
close the labor-productivity gap. This section presents the labor-productivity gap de-
composition for 1936/‘39, which shows that the potential for catch-up growth created
in German manufacturing was not fully realized, at least not in the short run. A large
German/US labor-productivity gap is still observed at the end of the 1930s, but the
bulk of the gap is ascribed to the inability of German industries to operate machinery
at American levels of efficiency and not to a lack of machine-intensive technology in
German manufacturing as suggested in the literature.64
Table 3.4 reports the results of the decomposition along the lines of equation (3.1)
on page 66. Germany’s labor-productivity gap to the US is decomposed in two elements,
i.e. machine-intensity differences and relative efficiency levels. The table demonstrates
that it was not the choice of capital-labor ratios that kept Germany from catching-up
with America. At the level of total manufacturing, Germany had to increase labor pro-
ductivity by 86% to match the performance of its American counterpart. If Germany
operated machinery at US levels of efficiency, labor productivity would have risen by
62% (which covers 72% of the labor-productivity gap). The complete closing of the
machine-intensity gap would augment German labor productivity by 24% only (which
covers the remaining 28% of the labor-productivity gap), a small increase only in com-
parison with the potential gains attainable through an improved efficiency level.
The relatively small effect of machine-intensity differences may surprise given that
Germany still employed capital-labor ratios about half the level in the US. This lim-
61. Nelson and Wright, “The Rise and Fall,” 1945.62. Schurr et al., Electricity in the American Economy, 32.63. For a comparison between the created growth potential in Germany and the US, see appendix 3.C.64. Broadberry, The Productivity Race, 108, 109.
Yet these possibilities for labor-productivity growth remained largely unrealized. This
raises two questions. First, did German entrepreneurs purposefully create potential for
labor-productivity growth through capital deepening? An awareness of frontier devel-
opments is prerequisite to technological spillover and Germany could expect to create
additional labor-productivity potential only when new production knowledge was imme-
diately available to countries not on the frontier.68 Secondly, if German entrepreneurs
were aware of the potential gains associated with adopting high capital-labor ratios,
what obstructed the efficient use of new machinery stock? In this section I turn to the
literature for an understanding of these issues.
Frontier awareness
In the case of Germany, ‘frontier awareness’ translates to an understanding and ap-
preciation of American production technology among German industrialists. Such an
America-centered orientation is well documented in the literature on interwar Germany.
In a study on German modernization, Nolan notes that American influences on German
entrepreneurship were limited before the 1920s. From the 1890s onwards, the scientific
management of labor as proposed by Frederick Taylor gained a strong foothold in the
minds of American producers. Proponents of Taylorism traveled to Germany, too, but
found their message difficult to sell; partly because of working-class opposition for fear
of reform at the cost of the laborer and partly because Germany’s successful industrial
development before 1914 did not create a necessity for new concepts.69
In the 1920s the situation was different. WW1, the reparation payments demanded
at Versailles, and the hyperinflation of the early 1920s had left the German economy
weakened in general and technological backward in particular. Change was needed and
by that time an attractive alternative to Taylorism was offered by Ford’s achievements in
the Detroit motor-vehicle industry. Rather than improving performance by rationalizing
on the factor input labor only, Fordism stressed the importance of both labor and
technology in the production process.70 As a consequence, the Fordist approach to
production appealed strongly to German entrepreneurs and set the example for future
development in Germany:
“With the end of Germany’s acute postwar dependency and instability,
America came to be seen as an economic model. In the words of one ob-
68. N. Rosenberg, “Economic Development and the Transfer of Technology: Some Historical Perspec-tives,” in The Economics of Technical Change, ed. E. Mansfield and E. Mansfield (Aldershot: EdwardElgar Publishing Limited, 1993), 380, 397.69. Nolan, Visions of Modernity, 45.70. ibid., 48.
Vandenbussche, Aghion, and Meghir, “Growth, Distance to the Frontier and Composition of HumanCapital,” 98.73. Gerschenkron, Economic backwardness, 113, 116; Abramovitz, “Catching-up,” 387.74. Field, “The Most Technologically Progressive Decade.”75. Nolan, Visions of Modernity, 38.
repeated in Germany. Market size, demand patterns, and wage structures differed just
too much between the US and Germany. Nevertheless, it was argued that the principles
of American production technology could be isolated and implemented in Germany as
well.76
Given the widespread enthusiasm about American production technology, it does
not surprise that in the 1920s German manufacturing industries deployed imitating
activities to catch-up with their American competitors.77 Well-known examples of both
imitating strategies and direct technology transfer concern the German machine-tool
industry. Richter and Streb, for instance, quote contemporary sources reporting that
American machine tools were copied by German engineers without any modification to
the original design:
“Information coming from Germany indicates that a number of American
machine-tools are (...) made without the slightest alteration.”78
But there were countless more examples of imitation by German machine-tool man-
ufactures. In the mid-1920s, the American trade commissioner listed over sixty US
machine-tool producers whose export suffered from German firms duplicating their
products.79 They suffered mainly because the changes implemented by the Germans on
American designs were negligible:
“But so far as the central idea and the means of carrying it our [were]
concerned, these tools [were] simply American out and out.”80
Richter concludes that not only thousands of American machine tools were in use in
Germany, but also the same amount or even more German copies of these tools.81 This
invites the question why Germany needed that many American machine tools if the
German production system was locked-in on a technological path essentially different
from the US, as traditionally has been uphold in the literature?82
In a recent paper, Ristuccia and Tooze quantify the prevalence of technology transfer
and imitating activities in the machine-tool industry. Although they base their analysis
on the number of purchased machines in Germany and the US, their conclusion aligns
76. Nolan, Visions of Modernity, 38.77. For a discussion on the channels of technology transfer between Germany and the US, see H.J.
Braun, “The National Association of German-American Technologists and Technology Transfer Be-tween Germany and the United States,” in History of Technology, ed. N. Smith (Mansell PublishingLimited, 1984), 15–35.78. Richter and Streb, “Catching-Up and Falling Behind,” 1007.79. Richter and Streb, “Catching-up and Falling Behind,” 17.80. Richter, “Technology and Knowledge Transfer,” 179.81. ibid., 180.82. ibid., 177.
This reluctance of German producers to fully embrace the American system has
been ascribed to different causes – mostly associated with demand patterns, such as a
relatively small domestic market, a demand for heavy-built, custom-made and expensive
cars or a fluctuating demand for automobiles which encouraged flexible production –
and resulted in low productivity levels; over the year 1921 Daimler produced less cars
than Ford did in one day.95 Still, production was increasingly standardized in Germany,
91. Radkau, Technik in Deutschland, 122.92. Wengenroth, Enterprise and Technology, 195, 243.93. G. Milkereit, “Innovation, Know-How, Rationalization and Investments in the German Mining and
Metal-Producing Industries, Including the Iron and Steelmaking Industry, 1868/71–1930,” Zeitschriftfur Unternehmensgeschichte Beiheft 22 (1982): 159.94. O. Keck, “The National System for Technical Innovation in Germany,” in National Innovation
Systems. A Comparative Analysis, ed. R. Nelson (Oxford University Press, 1993), 129, 131.95. Radkau, Technik in Deutschland, 275, 278.
but instead of using a single production line, as in America, German car assembly
remained divided into different production stages, each with its own assembly belt to
allow for flexible production.96 German industrial organization was aimed at minimizing
operating costs of machinery, rather than maximizing output with respect to labor. So
Germany failed to exploit the full potential of the technology in use and approach
American performance.
The notion that low levels of efficiency prevented convergence is illustrated most con-
vincingly by machine-producing industries. Table 3.4 reveals that the labor-productivity
gap in these industries was fully attributable to a lack of efficiency on the part of Ger-
many. Indeed, as shown in table 3.2 and more extensively described by Ristuccia and
Tooze, the level of capital intensity was similar between Germany and the US.97 So
differences in labor-productivity performance can only be ascribed to the level of effi-
ciency at which technology is operated. As with the transportation-equipment industry,
it has been noted that German machine-tool industries focused more on flexible produc-
tion than on high-throughput systems.98 The capital intensity did not differ between
both countries, but the composition of installed machinery did. High volume, autom-
atized machinery was underrepresented in Germany, which may explain the relatively
low levels of output per unit of labor input in this industry.99
In addition to these industry-specific factors that kept Germany from exploiting the
full potential of their technology stock, Ristuccia and Tooze suggest that the labor-
productivity gap stemmed from general differences between Germany and the US, such
as the latter’s cheap energy sources and larger scale of production.100 Also, by the
1930s it became evident that due to the emphasis on production and productivity, i.e.
supply-side factors, the productive capacity of industries had expanded much faster than
demand. In effect, many industries were overcapitalized and had excess capacity that
was left unused.101 As the new direction of technological development and industrial
organization was ill-matched to meet demand patterns, the success of the modernization
process was less than what was hoped for. Furthermore, autarkic policies related to the
build-up to WW2 may have acted as a barrier to efficiency, too. For instance, the food
industry saw major changes in the 1930s by government decree to suit the needs of
a country preparing for war.102 Together with protectionist policies that reduced the
incentive to improve efficient production, it may explain the low efficiency levels.
96. ibid., 278, 280.97. Ristuccia and Tooze, “Machine Tool and Mass Production.”98. ibid., 9; Radkau, Technik in Deutschland, 277.99. Ristuccia and Tooze, “Machine Tool and Mass Production,” 9.
100. Ristuccia and Tooze, “The Cutting Edge of Modernity,” 9.101. Nolan, Visions of Modernity, 132.102. Berthold, Produktivkrafte in Deutschland, 143–144.
involves an extensive transformation of the production process, efficiency levels may
be low in the short run. Only after the economy has adjusted to the new situation
and has ‘learned’ to operate technology at its full potential, the labor-productivity gap
to the frontier narrows. The time lag between creating potential and moving toward
the frontier may therefore depend on the speed of capital deepening. For the case of
Germany, this implies that the implementation problems that German engineers and
industrialists encountered in the 1920s and 1930s were not necessarily signs of failed
industrialization. Instead, they were features of progress and inextricably linked to the
initial phase of catch-up growth.
By the 1930s Germany had created a large potential for labor-productivity growth
and, in theory, it should cash in this latent capacity in a later period, i.e. the 1940s, by
means of ‘learning’. It falls outside the scope of this chapter to extend the analysis to
include the post-WW2 period, mostly because the postwar data on capital is not di-
rectly compatible with the horse power measure employed here. Nevertheless, Germany
was not in the position to realize its growth potential during the period 1939–1945 for
obvious reasons. Equally well-documented is Germany’s change of fortunes after 1946,
from which year onwards it rapidly closed the productivity gap with America. In 1980
(West) German levels of labor productivity stood at about 80% of those in the US.105
Furthermore, the process of capital deepening picked up again after 1950 and contin-
ued until the early 1970s, when America had almost lost its lead over Germany.106
Figure 3.7 sets out German/US relative machine intensity against German/US rela-
tive labor productivity for both the pre-WW2 period, based on data from this study,
and the post-WW2 period, based on data from O’Mahony. Both series are not directly
compatible, as the capital data in this chapter refers to machinery, while O’Mahony
measures the total capital stock. Nevertheless, the figure suggests that the unexploited
potential for labor-productivity growth was gradually realized after 1950.
The unprecedented rate of labor-productivity growth in Germany during the early
postwar years has been explained partly by reconstruction dynamics; Vonyo demon-
strates that wartime destruction and dislocation left much of the capacity for growth
unrealized, a potential which was exploited during the late 1940s and early 1950s.107 In
similar vein, Wolf argues that Germany’s direct productive capacity was not severely
damaged during the war, as a result of which the postwar productive capability consid-
erable exceeded actual production.108 The ensuing rebound growth – or “soft growth”,
105. O’Mahony, Britain’s Productivity Performance, 16.106. ibid., 24, 25.107. Tamas Vonyo, “Postwar Reconstruction and the Golden Age of Economic Growth,” EconomicHistory Review Vol. 12, no. 2 (2008): 235, 239.108. H. Wolf, “Post-War Germany in the European Context: Domestic and External Determinants of
In this paper I emphasize the role of technological change as a driver behind the wave
of modernization that marked the interwar period and stress the importance of effi-
ciency behind the German productivity dynamics of the 1920s and 1930s, particularly
in relation to the US. By adopting a data envelopment analysis (DEA), which ap-
plies non-parametric linear programming techniques, I can decompose TFP into two
components: changes in technological efficiency and shifts in technology over time. In
addition, as the DEA does not require the imposition of a particular functional form
on the production frontier, it allows for any type of technological change, be it biased
or factor-neutral.113
In this appendix I will summarize the basic framework behind the DEA, based
primarily on the work of Fare, Grosskopf and Lovell.114 They illustrate that a distance
function can be used to determine the Farrell efficiency indices of a production set
for any number of inputs or outputs. On the basis of the efficiency scores, a (global)
production frontier can be constructed, which in turn allows me to determine the change
in technology over time.115 In this basic example I assume that all inputs and output
quantities are non-negative and that, for each time period t = 1, . . . , T , the production
technology St models the transformation of N inputs, xt ∈ RN+ , into M outputs, yt ∈
RM+ ,
St ={(xt, yt) : xt can produce yt
}(3.2)
The input distance function Dti(x
t, yt) at time t is defined as
Dti(x
t, yt) = min{θ : (θxt, yt) ∈ St
}(3.3)
For the constant returns to scale case and a technology set St, the input distance
113. The main advantage of the Data Envelopment Analysis technique is its flexibility and adaptability.A DEA allows for multiple inputs and outputs, does not require input- or output-prices and does notrequire behavioral assumptions such as cost minimization or profit maximization.114. Fare, Grosskopf, and Lovell, Production Frontiers.115. R. Fare et al., “Productivity Growth, Technical Progress, and Efficiency Change in IndustrializedCountries,” American Economic Review Vol. 84, no. 1 (1994): 68–69.
Chapter 4Industrial Output Growth in Pre-WW2 Germany.
A Reinterpretation of Time-Series Evidence∗
4.1 Introduction
The necessity of re-assessing the state of the German economy, as repeatedly stressed in
the previous chapters, stems for a large part from the difficulty encountered by previous
studies to construct a reliable time series of industrial output in pre-WW2 Germany.
Several series have been proposed, yet a final solution for the issue is not available. The
confusion surrounding the quality of these historical indices invites discussion concerning
the development of Germany’s performance around the turn of the twentieth century.
In particular Germany’s comparative performance in industry relative to the UK prior
to WW1, a time when the effects of modernization were not yet diluted by the economic
dislocations of the World Wars, has been the topic of debate.1
At stake in this debate is the labor-productivity leadership in Europe; if Germany
had already surpassed Britain – the labor-productivity leader of old – this would suggest
a failure on the latter’s part to benefit from the opportunities offered by the innovations
of the second industrial revolution as much as the former did. This line of reasoning
suggests that the barriers to growth frequently attributed to the UK, such as the relative
costs of factor inputs or market conditions discussed earlier, acted less as a constraint on
Germany’s development. The relative standing within Europe may carry implications,
too, for the growth process. If follower countries develop through catch-up mechanisms,
i.e. copying technology operated at the global productivity frontier, Germany could
no longer look to the UK for future growth. A further interest concerns the period
* This chapter is based on joint work with Jan P.A.M. Jacobs (University of Groningen).1. Ritschl, “Spurious Growth in German Output Data”; Broadberry and Burhop, “Comparative Pro-
ductivity in British and German Manufacturing”; Ritschl, “The Anglo-German Industrial ProductivityPuzzle”; Broadberry and Burhop, “Resolving.” Section 4.2 visits the debate in detail.
after WW1. Prerequisite for understanding the impact of the war and the subsequent
upheaval in the Weimar Republic is an assessment of the state of the economy before
these shocks occurred.
The estimates so far presented in the literature differ to such a degree that two
stories can be told of Germany’s comparative performance before WW1 that are in fact
incompatible; Germany either performed on par with the UK or it outperformed the
UK by roughly 25%.2 With an eye to the historical questions touched upon above, the
ambiguity is unsatisfactory. This begs the question whether it is possible to confidently
draw conclusions regarding Germany’s historical growth record in the face of conflicting
data? I think it is and I arrive at that conclusion through application of a new approach
to this debate.
Much of the deviation between the various output series suggested in the literature
results from the use of different output proxies, which are used because data on value
added is unobtainable for the period before WW2. New releases of the German output
series are therefore valued at the accuracy of the proxies applied to estimate industrial
production. In the absence of value added data to evaluate the accuracy of the prox-
ies, the choice between alternative versions of the output index is not straightforward.
Nevertheless, once a revision is deemed more appropriate as a measure of value added
change, it has been custom to discard all other, older alternatives of the output index.
This chapter sets out to solve the time-series issue by casting the debate in a new
framework. Instead of choosing between different output proxies, I acknowledge that
all series estimate output change by studying variables that are assumed to correlate
strongly, but not perfectly with value added. It follows that the behavior of all series is
largely determined by the same underlying component, i.e. value added change, while
deviations in the observed series are contingent on the different correlation between
the employed output proxies and actual output growth. Using state space time series
analysis, I estimate value added change by filtering from all available data an unob-
served common component.3 This way, the analysis makes full and efficient use of all
information available, rather than choosing for one particular alternative only.
A second aim of the chapter is to shed light on the statistical error associated with the
estimation process. Due to incomplete information the estimates of output change are
essentially based on sample data, so the estimates are inaccurate to some extent. In the
debate on German output growth, however, indicators of statistical dispersion are not
2. Broadberry and Burhop, “Resolving,” 932; S.N. Broadberry and C. Burhop, “Resolving the Anglo-German Industrial Productivity Puzzle, 1895–1935: A Response to Professor Ritschl,” Warwick eco-nomic research papers Vol. 848 (2008): 16. See also table 4.1 on page 110.
3. Commandeur and Koopman, An Introduction; Durbin and Koopman, Time Series Analysis.
Chapter 4. Industrial Output Growth in Pre-WW2 Germany 105
provided and point estimates are implicitly treated as true values. As such, important
properties of the data are omitted. By providing an indication of the statistical error
in my estimates, this chapter follows in the tradition of Charles Feinstein and Mark
Thomas, who argued that any new statistical series should be accompanied by a guide
to the associated margins of error.4
The statistical error is important for the reconciliation between time series and
benchmark estimates, i.e direct level estimates. If the output level is known for a par-
ticular year, the index of industrial production can be used to obtain output levels in
other years through extrapolation. The discrepancy between output levels obtained in-
directly through time series and directly by benchmarks is a frequently used measure of
the former’s accuracy. Because benchmark studies calculate output levels for particular
years, these snap shots of industrial performance are a popular measurement technique
in the debate on German output growth; new releases of the output index that do not
reconcile with benchmark estimates are ill-received. Yet conditional on the width of
the confidence interval, the lack of perfect reconciliation between the output index and
benchmark estimates does not necessarily disqualify the fit between both measures.
The innovative feature of my approach is the decomposition of the observed data in
an unobserved value added component and a noise factor resulting from the use of prox-
ies. As state space analysis is designed to uncover the dynamic evolution of time series
when these properties are not directly observable from the data, it is an appropriate tool
of analysis. In my case, the system of observed time series is modeled as a function of
an unobserved common process plus an irregular component containing index-specific
noise. By casting the debate in state space form, I offer a formal framework to sta-
tistically assess the similarity and dissimilarity between output series presented in the
literature. Moreover, because all information available is used, my analysis transcends
earlier contributions to the debate on German output growth by the application of an
integrated, rather than exclusive, approach.
4.2 The time-series debate
Hoffmann’s Historical National Accounts
In the early 1960s, a team of researchers under the auspices of Walther Hoffmann
constructed the German historical national accounts, as a part of which an output index
4. C.H. Feinstein and M. Thomas, “A Plea for Errors,” Historical Methods Vol. 35, no. 4 (2002):155; C.H. Feinstein and M. Thomas, “A Plea for Errors,” University of Oxford Discussion Papers inEconomic and Social History No. 41 (2001): 3.
for German industry was produced.5 Hoffmann’s output index is a weighted average
of the estimated change in output in twelve manufacturing and utility industries for
the period between 1870 and 1938. As for the pre-WW1 period data on value added
change in German industries is not available, output proxies are used instead. Output
change in the majority of the industries is estimated using physical indicators, usually
manufactured tons of goods. In contrast, for metal processing – an industry class that
contains machine building, shipbuilding and electrical engineering – Hoffmann chose
to rely on labor-income data as a proxy of output change. The aggregate time series
for industry is subsequently constructed weighting the twelve industry indices by their
share of value added (the compound output index is plotted in figure 4.1a).
Hoffmann’s index of industrial production has received severe criticism and its reli-
ability has been called into question, in particular by Rainer Fremdling.6 The problems
associated with the Hoffmann series concern two issues. First, using the annual wage-
bill to proxy output change presumes a constant wage-productivity ratio. However,
Borchardt (1979) argued that after WW1 wages rose as a consequence of labor unions’
increased bargaining power, rather than raised labor-productivity levels.7 In light of
Borchardt’s thesis, the assumption of a constant wage-productivity ratio might not be
innocuous. More to the point, the dichotomous development between wages and la-
bor productivity leads to an upward bias in Hoffmann’s output estimates for metal
processing.
Second, value-added data for manufacturing industries is available for 1936 only
(based on the first German census of production). To construct value-added weights
Hoffmann multiplied the level of labor productivity in 1936 by employment in 1933
(derived from the employment census). Although the resulting value-added shares might
function as a weighting scheme for the interwar period, it cannot reasonably be imposed
on periods before WW1. Hoffmann ‘solved’ this problem by using proxy value-added
weights; that is, he multiplied the level of labor productivity in 1936 by employment
in 1882 and 1907 to obtain value-added shares for the years 1871-1895 and 1895-1913,
respectively. However, this assumes comparative levels of value added per employee
across German industries to have remained unchanged over the period 1870-1938, which
it did not.
5. Hoffmann, Das Wachstum.6. Fremdling, “German National Accounts”; R. Fremdling, “German Industrial Employment 1925,
1933, 1936 and 1939. A New Benchmark for 1936 and a Note on Hoffmann’s Tales,” Sonderdruck aus:Jahrbuch fur Wirtschaftsgeschichte Vol. 2 (2007): 171–195.
7. Ritschl, “Spurious Growth in German Output Data,” 202; Borchardt, “Zwangslagen und Hand-lungsspielrame.”
Chapter 4. Industrial Output Growth in Pre-WW2 Germany 107
Revisions of Hoffmann’s output series
As the problems associated with the weighting scheme are not easily solved in the
absence of value-added data, the wage-bill issue has been discussed most extensively in
the literature. The first to address Hoffmann’s output index for metal processing was
Albrecht Ritschl.8 In a comprehensive overview of already existing German time series
he wondered why Hoffmann chose to use the wage bill as a proxy for production in
metal processing while other, possibly less problematic, proxies are readily available for
the interwar period. In fact, production indices of various metal processing industries
were presented in 1933 by Wagenfuhr of the Institut fur Konjunkturforschung (IfK,
see figure 4.1a).9 When Hoffmann’s metal processing time series is compared with the
official production data provided by the IfK, deviations in output growth are manifest
mainly for the machine-building industry, which is part of metal processing. For this
reason Ritschl uses sales data of the Verband Deutscher Machinen- und Anlagenbau (the
German machinery producers’ association) to reassess output change in the machine-
building industry and he records an output growth of only half the magnitude suggested
by Hoffmann. Since metal processing has a weight of 17 percent in Hoffmann’s compound
index of industrial production, the revised data on machine building moderates German
output growth considerably. Ritschl’s revision reports a relatively low rate of growth
particularly over WW1, as can be seen in figure 4.1a.
The modified output index evoked a reaction from Stephen Broadberry and Carsten
Burhop for Ritschl’s proposed changes imply a revision of Germany’s performance rel-
ative to the UK that does not sit well with previous research.10 Combining Ritschl’s
output index with Hoffmann’s time series of employment to obtain the change in la-
bor productivity, the level of labor productivity in 1936 can be extrapolated backward,
as illustrated by figure 4.2. Adjusted to a manufacturing basis (i.e. excluding mining,
construction and utility industries), the extrapolated productivity levels suggest a com-
manding German lead over the UK in the pre-WW1 period, a performance on the
part of Germany much stronger than previously.11 A snap shot of comparative labor-
productivity levels for 1907 constructed by Broadberry & Burhop (table 4.1, first row)
points at an equality in performance between both countries rather than a distinct
German lead; a result seemingly at odds with the extrapolated labor-productivity lev-
els obtained using Ritschl’s output series. Hoffmann’s original index of output, on the
8. Ritschl, “Spurious Growth in German Output Data.”9. Wagenfuhr, “Die Industriewirtschaft.”
10. Broadberry and Burhop, “Comparative Productivity in British and German Manufacturing.”11. R. Fremdling, “Productivity Comparison Between Great Britain and Germany, 1855–1913,” Scan-
dinavian Economic History Review Vol. 1 (1991): 37. See also table 4.1.
Fremdling; this procedure conveniently leads to a German level of labor productivity
prior to WW1 in line with Broadberry & Burhop’s benchmark of comparative perfor-
mance (latest benchmark revisions; table 4.1, third row).13
Table 4.1: Benchmark estimates of comparative labor productivity
Source GER/UK labor productivity
Indus. Manuf.
1907
Broadberry & Burhop (2007) 1.02 1.05
Ritschl 1.25 1.28
Broadberry & Burhop (2008) 1.05 1.08
Fremdling 0.74 . . .
1935/36Broadberry . . . 1.02
Fremdling, de Jong, Timmer . . . 1.07
Sources: see text, section 4.2.
In short, the debate has been fueled to a large extent by the notion that point esti-
mates obtained by benchmarks and time-series analysis should reconcile. However, this
notion defies the literature that emphasized several causes for deviation between both
measures. This topic has been debated in relation with long-span time series projec-
tions and the problems associated with reconciling historical time series with benchmark
estimates are well documented.14 In general, deviations stem from methodological dif-
ferences between both measures. At the root of this inconsistency lies the difference
between weight structures employed in bilateral benchmarks and time series, a problem
solvable only by application of a single aggregation scheme for both spatial and tempo-
ral comparisons.15 As the structural composition of economies changes over time, the
13. Broadberry and Burhop, “Resolving”; Fremdling, “German Industrial Employment.”14. I. Kravis, A. Heston, and R. Summers, “World Product and Income: International Comparisons of
Real Gross Products,” World Bank Report (1982): 326; R. Summers and A. Heston, “The Penn WorldTable (Mark 5): An Expanded Set of International Comparisons, 1950–1988,” The Quarterly Journalof Economics Vol. 106 (1991): 327–368; A. Heston, R. Summers, and B. Aten, “Price Structure, theQuality Factor and Chaining,” 2001, http://www.oecd.org/std/prices-ppp/2425050.pdf and A.Deaton and A. Heston, “Understanding PPPs and PPP-based National Accounts,” American EconomicJournal: Macroeconomics Vol. 2 (2010): 1–35. In the field of economic history this issue has beenaddressed by M. Ward and J. Devereux, “Measuring British Decline: Direct Versus Long-Span IncomeMeasures,” The Journal of Economic History Vol. 63 (2003): 826–851; S.N. Broadberry, “Relative PerCapita Income Levels in the United Kingdom and the United States since 1870: Reconciling Time-SeriesProjections and Direct-Benchmark Estimates,” The Journal of Economic History Vol. 63 (2003): 852–863 and M. Ward and J. Devereux, “Relative U.K./U.S. Output Reconsidered: A Reply to ProfessorBroadberry,” The Journal of Economic History Vol. 64 (2004): 879–891.15. H.J. de Jong and P.J. Woltjer, “A Comparison of Real Output and Productivity for British and
American Manufacturing in 1935,” Groningen Growth and Development Centre Memorandum No. 108(2009): 16; G. Szilagyi, “Procedures for Updating the Results of International Comparisons,” Reviewof Income and Wealth Vol. 30 (1984): 156–157.
Chapter 4. Industrial Output Growth in Pre-WW2 Germany 111
weighting scheme of time series requires updating, which leads to inconsistency with
benchmark comparisons. While forcing a single weighting scheme on time series en-
sures consistency over time and across space, it renders the interpretation of the results
difficult and is therefore undesirable.16
So a perfect fit between time series and benchmarks cannot be expected nor de-
manded, and the time-series revision proposed by Ritschl does not necessarily provide
sufficient grounds to reject Broadberry & Burhop’s 1907 benchmark, or vice versa. In-
deed, the differences between Ritschl’s revised output index and Hoffmann’s original
are, in general, limited at the level of total manufacturing. Even though the data un-
derlying figure 4.1a displays a compound annual growth rate over the period between
1907–1936 of 1.20% for Ritschl’s index versus 1.93% for Hoffmann’s series, which is
substantial because small variations between annual growth rates can and indeed do
lead to large deviations in output levels in the long run, the difference between both
estimates can be almost fully ascribed to the period 1913–1925, while before and after
the annual growth rate hardly differs between the series. Figure 4.1b shows for both
series the compound annual growth rate over this period; whereas Hoffmann’s series
suggest a decade of (small) growth, Ritschl’s data indicate a continuous decline in out-
put. But over the other periods the annual growth rates are very similar. The question
is, then, if the inconsistency between benchmarks and time series can be accounted for
by index-number factors.
This effect can be quantified if the benchmarks and time series are constructed
exclusively on the basis of price and quantity data. This is impossible for pre-WW2
Germany, as the necessary data are not always obtainable; price information is often
unavailable and so are quantities for some industries, in which case proxies are used. The
nature of the data not only hinders a measure of the index-number induced deviation,
by itself it also presents a second source of inconsistency. The proxies employed in the
time series are associated with measurement errors, which introduce inconsistency with
the benchmarks.
Moreover, both the benchmarks and the time series suffer to a different degree from
a lack of representativeness in the data used, as the industry coverage varies between
both measures, an issue already briefly touched upon in chapter 2. The output se-
ries studied here apply to industry and include manufacturing, mining, construction
and some utility industries. Although the 1907 benchmarks include manufacturing and
16. E. Dalgaard and H. Sørensen, “Consistency Between PPP Benchmarks and National Price andVolume Indices,” in 27th General Conference of the International Association for Research in Incomeand Wealth (Stockholm: Sweden, 2002), 4; Jong and Woltjer, “A Comparison of Real Output andProductivity,” 16.
mining, the other two sectors are not captured. This introduces a bias. Ideally, the
construction and utility industries are taken out of the time-series sample, but this pro-
cedure is rendered impossible by the lack of industry weights for Wagenfuhr’s series.17
Dalgaard and Sørenson note that the ensuing discrepancies cannot be accounted for by
index-number formulas and, therefore, pose genuine problems of consistency.18 It is the
inconsistency attributable to these genuine factors that arbitrates the quality of the fit
between benchmark estimates and time-series projections.
Using the state-space form I can assess this fit between both measures. A break down
of the inconsistency between benchmarks and time series in genuine and non-genuine
components is impossible here, but perhaps not necessary. As the model estimates a
common component containing the dynamic properties of the three observed time series,
the different benchmarks presented in the literature are confronted only with my filtered
time-series estimate. Given that all 1907 benchmarks use the same weighting scheme, i.e.
the employment structure obtained from the 1907 census, the deviation with the filtered
time series that is explained by index-number related factors is the same for each match
between benchmark and time series. It follows that variation in inconsistency between
my time series and the presented benchmarks traces back to genuine factors. Assuming,
first, that the estimated time series captures the change in output and, second, the 1936
benchmark from which the time series is extrapolated backward accurately measures
the level of output, the 1907 benchmark that shows the closest fit with the backward
projections suffers least from these genuine consistency problems.
This still leaves the question how much inconsistency one is willing to allow for and
not reject the fit between the filtered time series and the benchmarks? The uncertainty
associated with estimating the unobserved common component provides a yardstick of
measurement error in the time series (although not in the benchmarks). Using the vari-
ance of the model I construct a confidence interval to indicate a range around the point
estimates that contains the true value of the estimated parameter with high probability.
In case a benchmark estimate falls inside that range, the inconsistency can be explained
by measurement error in the time series and while both measures do not reconcile the
fit cannot be rejected. If a benchmark estimate falls outside the confidence interval,
the unexplained inconsistency is caused by additional genuine factors originating in the
benchmark, such as its measurement error or industry coverage, that introduce further
noise and thus impair the quality of the fit with the time-series projections.
17. In the absence of industry weights it is not feasible to aggregate the industry series to the levelof total manufacturing.18. Dalgaard and Sørensen, “Consistency Between PPP Benchmarks and National Price and Volume
Chapter 4. Industrial Output Growth in Pre-WW2 Germany 113
4.3 Methodology
The purpose of state space time series analysis is to uncover the dynamic evolution
of observations measured over time when the dynamic properties cannot be directly
observed from the data.19 As I am interested in the unobserved change of industrial
output, which is assumed to determine the behavior of the observed time series, state
space modeling provides a tool of analysis particularly suited to my design. By using
the state-space form, I build upon a literature that has used such models before in the
field of economic history, in particular the research of Lee & Anderson, Crafts & Mills
and Pfister, Riedel & Uebele.20 All three study the interaction between economic and
demographic variables in early-modern times (the former for England and the latter
for Germany), using the state-space form to estimate the dynamics of, for instance,
technological change, the demand for labor or weather and disease prevalence, none of
which is observed.21
These analyses are all univariate, though, and the application of the state-space
form to filter a common state from multiple observed time series is new to the field of
economic history. Although not applied in the state-space form before, I am not the first
to estimate common components from different time series. When in the face of data
restrictions the dynamics of a particular variable can be extracted from the behavior
of other data that are assumed to relate with the variable of interest, such a procedure
provides a useful research strategy for periods characterized by poor data coverage.
This avenue has been explored by Sarfarez and Uebele, who ‘track down’ business-cycle
movements in Germany before WW1, a period that suffers from data scarcity, through
application of dynamic-factor analysis.22 Similarly, in a paper on market integration,
Uebele employs comparable techniques to estimate a common price change for regional,
national and international markets from multiple time series.23
In my case, using the state-space form to estimate a common component has several
advantages. As explained, by casting the time series of industrial output presented
in the literature in state-space form, I am able to estimate an unobserved dynamic
19. Commandeur and Koopman, An Introduction; Durbin and Koopman, Time Series Analysis.20. R. Lee and M. Anderson, “Malthus in State Space: Macro Economic-Demographic Relations in
English History,” Journal of Population Economics Vol. 15 (2002): 195–220; N. Crafts and T. Mills,“From Malthus to Solow: How did the Malthusian Economy Really Evolve?,” Journal of Macroeco-nomics Vol. 31 (2009): 68–93; U. Pfister, J. Riedel, and M. Uebele, “Real Wages and the Origins ofModern Economic Growth in Germany, 16th to 19th Centuries,” EHES Working Papers in EconomicHistory No. 17 (2012): 1–27.21. Crafts and Mills, “From Malthus to Solow,” 82; Pfister, Riedel, and Uebele, “Real Wages,” 13.22. S. Sarferaz and M. Uebele, “Tracking Down the Business Cycle: A Dynamic Factor Model for
Germany, 1820–1913,” Explorations in Economic History Vol. 46, no. 3 (2009): 368–387.23. M Uebele, “National and International Market Integration in the 19th Century: Evidence from
Comovement,” Explorations in Economic History Vol. 48, no. 2 (2011): 226–242.
process. Moreover, in a multivariate setting information from multiple time series can
be used to improve the estimate of the target series, output in this case, by assuming
that the unobserved component is common to all observed series.24 A second appeal
of the state-space form is that stationarity of the time series is not required, because
it concerns a structural time series model in which the trend, seasonal and error terms
are explicitly modeled. This is an advantage, given that most real series in the field
of economics are non-stationary.25 Thirdly, the state-space framework can deal with
missing observations with relative ease; as the years covering WW1 are not accounted
for in the output indices, this is a benefit, too.26 In short, the state-space form provides
a flexible and easy to work with instrument to analyze the German output series.
Specification of the model
Using matrix notation, all multivariate state-space models can be written in the gen-
eral format of equations (4.1) and (4.2). The model contains two equations. First, the
observed series (yt) are modeled by the measurement (or observation) equation, which
defines the series by two components, i.e. the unobserved dynamic process called the
state (αt) and a disturbance term (εt). Second, the state equation models the unob-
served dynamic process as a function of its value in previous periods plus a disturbance
term (ηt). Both disturbance terms are normally and independently distributed (NID)
around a mean of zero with a variance of σε2 and ση2 , respectively.
yt = Ztαt + εt, εt ∼ NID(0, Ht) (4.1)
αt+1 = Ttαt +Rtηt, ηt ∼ NID(0, Qt) (4.2)
The specification used here is a local linear trend model, which is a special case of the
general state-space framework presented in the set of equations (4.1) and (4.2). Each
of the observed series is modeled as a function of a common state component and an
index-specific observation disturbance. In case of Wagenfuhr’s and Hoffmann’s series,
however, the common state is weighted by a coefficient a, because the literature has
credited Ritschl’s time-series revision with the highest reliability.27 My special case of
24. A. Harvey and C. Chung, “Estimating the Underlying Change in Unemployment in the UK,”Journal of the Royal Statistical Society Vol. 163, No. 3 (2000): 305, 314–315.25. Commandeur and Koopman, An Introduction, 134.26. ibid., 103.27. Broadberry & Burhop accepted Ritschl’s (adjusted) revisions to Hoffmann’s output series, but
combined it with a different employment series, i.e. Fremdling’s instead of Hoffmann’s, to reconcile thebackward extrapolation of labor productivity with their own 1907 German/UK benchmark estimate.See Broadberry and Burhop, “Resolving,” 933.
is also referred to as the drift.29 The behavior of the slope is determined by the slope
disturbance ζt.
The values of the two hyperparameters, i.e. the measurement and state disturbances,
cannot be obtained analytically and the model is therefore estimated using maximum
likelihood based inference. The likelihood function associated with the model is ob-
tained through the application of an algorithm called the Kalman filter.30 In my case,
the estimated unobserved common component refers to the filtered state, which is the
estimate of the state vector based on all past observations and the current observation.
This means that the estimation process involves only a forward pass through the data.
Alternatively, I could have smoothed the state by also performing a backward pass and
thereby using all observations (i.e. past, current and future observations) to estimate
the state vector. As the name suggests, such a procedure effectively smooths the dy-
namics of the state. However, when ‘corners are cut’ the state series takes on a value
for the years before WW1 lower than of the observed series, because the output drop
over the war is already taken into account before it actually happened. With an eye to
the historical context to which the state vector refers, it does not make sense for shocks
to have backward effects and, therefore, I use the filtered state.
The unknown parameters are estimated using the log-likelihood function in Eviews,
which corresponds to the definitions of Durbin and Koopman.31 Estimation involves a
numerical search procedure that starts by choosing a set of starting values for the un-
known parameters and calculating the corresponding value of the log-likelihood func-
tion. Subsequently, the process is repeated, selecting different parameter values that
improve the log-likelihood function. These iterations are executed up to the point that
no further improvements are obtained and the log-likelihood function is optimized.
However, due to the multivariate nature of the model, the optimization process may
produce either a suboptimal or no solution for particular starting values. Following
Van den Bossche, I use a multiple random start procedure that runs the optimization
algorithm repeatedly, each time starting from a different set of initial values for the
unknown parameters.32 The whole estimation procedure is repeated 1,000 times and
the solution reported with the highest log-likelihood value is used henceforth.33
29. Commandeur and Koopman, An Introduction, 21.30. R.E. Kalman, “A New Approach to Linear Filtering and Prediction Problems,” Journal of Basic
Engineering Vol. 82 (1960): 35–45.31. Quantitative Micro Software, Eviews 6 User’s Guide II (2007), 387; Durbin and Koopman, Time
Series Analysis, 138; A. Harvey, Forecasting, Structural Time Series Models and the Kalman Filter(Cambridge University Press, 1989), 126; F van den Bossche, “Fitting State Space Models with Eviews,”Journal of Statistical Software Vol. 41, no. 8 (2011): 3.32. ibid., 10.33. Eviews provides different optimization procedures, i.e. Marquardt and Berndt-Hall-Hausman. I
used the former first derivative technique. For further specification of the program, see appendix.
Chapter 4. Industrial Output Growth in Pre-WW2 Germany 117
Data
It has been noted that before carrying out any estimation, it is important to deter-
mine the nature of the time series in hand.34 It is in particular crucial to examine the
properties of the observed series’ trend and establish whether it is deterministic (sta-
tionary) or stochastic (nonstationary). If the trend’s nature of one of the three series
studied here differs from the others, there is no common trend to estimate. Therefore,
before running the analysis, I test for stationarity using augmented Dicky Fuller (ADF)
tests. If the ADF shows that the series has a unit root, this points in the direction of a
non-stationary trend. Yet the results of a unit-root test do not provide definitive proof
of stationarity or the lack thereof. In the presence of structural breaks, unit root has
difficulty distinguishing stationarity from nonstationarity.35
Table 4.2: Unit-root test (augmented Dicky-Fuller)
Output series Adjusted sample τ -Statistic ρ
Hoffmann (1965) 1872–1938 -1.29 0.92
Ritschl (2004) 1872–1938 -1.10 0.93
Wagenfuhr (1933) 1872–1931 -2.59 0.84ρ Coefficient on the lagged dependent variable.* Significant at either the 0.10, 0.05 or 0.01 level.
A worry in this respect is the inclusion of WW1 in my period of study, as structural
breaks in the twentieth century often occurred at times of war.36 Figures 4.3a, 4.3b
and 4.3c show the logarithms of the observed series fitted with a linear breaking trend
function, where I allow both the intercept and the slope of the linear trend to change
after 1914. All series are clearly upward trending, hinting at the presence of unit roots.
Looking at the regression coefficients, in the case of Hoffmann’s and Ritschl’s series no
trend breaks are detected over WW1. Wagenfuhr’s series, on the other hand, displays
a significant decrease of the intercept at the 1% level and a significant increase of the
slope at the 5% level. This result is driven primarily by the coverage of the series; in
contrast to the other two series, the reconstruction phase after WW1 is included, while
the slump during 1930s is omitted. Nevertheless, as the slope of Wagenfuhr’s series
increases, there is no evidence of a break at which the generating process switched from
nonstationary to stationary. Indeed, table 4.2 suggests that a common trend can be
filtered from the three series.
34. J.P.A.M. Jacobs and J.P. Smits, “Historical Time Series Analysis: An Introduction and SomeApplications,” Jahrbuch fur Wirtschaftsgeschichte / Economic History Yearbook (2006): 5.35. ibid., 7.36. ibid., 10.
Idem, lower bound 101.4 76.1 76.9 . . .∗Extrapolated backward from a German/UK comparative level of 105.4, obtained fromFremdling, de Jong, and Timmer, “British and German Manufacturing ProductivityCompared,” 353.Sources: see text, section 4.2.
In a final step I take these estimates to the issue of reconciling time series and
benchmark estimates. As described in section 4.2, the different sides in the debate on
German/UK comparative labor productivity tried to ensure a close fit between their
time series projections and benchmark comparisons. However, the confidence interval
around the filtered state, and thus around the levels of comparative performance, already
showed that point estimates of time series estimates are associated with considerable
uncertainty. So I move away from the notion that benchmarks and time series estimates
need to align closely. Instead, I let the measurement error of my estimated value-added
change determine the deviation between both measures that I am willing to allow for.
The question is, then, which of the benchmarks presented in the literature (if any at all)
I am compelled to reject on the basis of the uncertainty associated with the estimation
Chapter 4. Industrial Output Growth in Pre-WW2 Germany 127
Second, over WW1 there was a statistically significant change in labor productivity
leadership with Germany dropping below the UK. And, third, given Fremdling, de Jong
and Timmer’s 1936/35 German/UK benchmark comparison, Britain’s lead evaporated
again in the 1930s and both countries performed roughly on par shortly before WW2.
In view of earlier research, these findings confirm the trend over the last two decades
in the literature on German-Anglo productivity differences. Fremdling’s estimate of a
German/British comparative level of 74% seems very low now, but deviated much less
from other estimations presented in the literature in the 1980s and 1990s. For instance,
Bairoch (1973) placed Germany at 93% the level of Britain, Crafts (1983) suggested a
German performance of 87%, Dormois & Bardini (1994) found a comparative level of
82% and Burger (1994) indicated a level of 79%.40 Compared to these earlier works,
the findings of Broadberry & Burhop and Ritschl correspond much better with the
contemporary perspective on German-Anglo industrial relations. Arthur Shadwell, who
traveled the UK, Germany and the US shortly after the turn of the twentieth century in
order to compare the qualities of industrial life in the three leading industrial countries
of the time, wrote that:
“[Germany] built up (. . . ) a great edifice of manufacturing industry which
for variety and quality of output can compete in any market with most of
the finest products of Great Britain.”41
4.5 Conclusion
Several attempts have been made in the literature to quantify output and labor-
productivity growth in German industry for the period before 1950. Given that on
the basis of different time series of output two incompatible stories can be told of
Germany’s comparative performance before WW1, the uncertainty is uncomfortable.
This begs the question whether it is possible to confidently draw conclusions regarding
Germany’s historical growth record in the face of these conflicting data?
I contribute to this debate by casting the time-series discussion in a new framework.
All output series presented in the literature set out to measure the change in output,
but value-added data is not obtainable in which case proxies are used to estimate
output change. While output proxies are assumed to correlate strongly with value-
added change, they cannot do so perfectly. The underlying data used to construct the
40. J. Dormois, “The Impact of Late-Nineteenth Century Tariffs on the Productivity of EuropeanIndustries, 1870–1930,” in Classical Trade Protection, 1815–1914, ed. J. Dormois and P. Lains (London:Routledge, 2006), 179.41. Shadwell, Industrial Efficiency, 14-15.
output indices differs little between the series, but for some industries different output
proxies are employed, which drives the deviation between the series.
Because in the absence of output data it is impossible to determine which of the
proxies captures output growth best, the choice between time series is arbitrary to some
extent. Therefore, I argue that it is inappropriate to choose between the series, discard-
ing information provided by the rejected series. As all series employ proxies that are
correlated with value-added change, the dynamic properties of the three observed series
must be contained by a common unobserved component. Using time-series analysis, in
a first step I filtered this common component from the series, which is then interpreted
as the actual change in value added.
In a second step I looked at the uncertainty associated with the process of estimating
the common component. This means that I look at point as well as interval estimates.
Although this seems obvious, it has been tradition in the literature addressed here to
exclude information on statistical error and implicitly treat the point estimates as the
‘true’ value of the parameter. Using an upper and lower confidence limit, I indicate
a range around the estimated common component which contains the true value of
value-added with 99% certainty.
In a third step the estimated change in output is combined with data on employ-
ment to get the change of German labor productivity, which I then compared with its
British counterpart. Extrapolated backward from a robust benchmark of comparative
labor productivity in 1935/36, the level of comparative labor productivity in 1907 is
obtained. This exercise is repeated thrice, replacing the filtered common component
with the upper and lower confidence bound, respectively. This way, I identify a range
of comparative labor productivity containing the true value of the estimated parameter
with a high probability.
With this approach I deviate from the traditional notion that benchmarks and time
series estimates need to align closely. Faced with the different time series of output pre-
sented in the literature, scholars have previously employed the 1907 labor-productivity
benchmarks to test the accuracy of the time series estimates. The idea is simple; if
the benchmark estimate does not provide a tight fit with the backward projections,
the latter must be flawed. Criteria for the fit between benchmark estimates and time
series projections are loosely defined and not supported by a theoretical justification
thereof.42
In this chapter, I move away from that notion. Instead, I let the measurement error
42. Broadberry suggests a range of 10% around the point estimates. See: Broadberry and Burhop,“Comparative Productivity in British and German Manufacturing,” 326 in which the authors refer toBroadberry, “Manufacturing and the Convergence Hypothesis.”
Chapter 4. Industrial Output Growth in Pre-WW2 Germany 129
of my estimated value-added change determine the deviation between both measures
that I am willing to allow for. The question is, then, which of the benchmarks presented
in the literature (if any at all) I am compelled to reject on the basis of the uncertainty
associated with the estimation process. Using a conservative 99% confidence level, all
benchmarks fall within the interval around my point estimate, while at the 95% level
Broadberry and Burhop’s 2007 benchmarks falls outside the confidence bounds. These
findings suggest a comparatively strong performance on the part of Germany.
The interval around the point estimates are fairly large and I am willing to accept a
broad range of German/UK comparative labor-productivity levels. The message to take
away from this is that measurement error must be considered when benchmarks are used
to check the accuracy of time series. Still, if I project the margin of error around the point
estimate (at the 95% confidence level) on the reliability scheme of Chapman, my series
falls into the B-category of “good estimates”.43 Moreover, the width of the intervals
dooes not prevent me from drawing conclusions regarding Germany’s comparative labor-
productivity development during the first half of the twentieth century. If anything, I
draw such conclusions with increased confidence. It is clear that Germany had a lead
over the UK before WW1 around the range of 10%–20%. The situation reversed over
WW1, when Germany fell behind. Although the German economy managed to catch-up
again by the late 1930s, it did not regain the advantage over the UK enjoyed before
WW1.
43. A. Chapman, Wages and Salaries in the United Kingdom, 1920–1938 (Cambridge: CambridgeUniversity Press, 1953), 231; Feinstein and Thomas, “A Plea for Errors,” 158; Feinstein and Thomas,“A Plea for Errors,” 16.
Chapter 5Did a European Convergence Club Exist Before
World War 1? Comparative Labor Productivity in
Northwestern Europe, 1875–1913
5.A Introduction
Looking back on the previous chapters, much of the presented and discussed evidence
hints at the possibility of a common growth experience for European countries in the
period before WW1. First, chapter 2 showed that the US enjoyed a commanding labor-
productivity lead in manufacturing over Europe. This finding broadly aligns with other
manufacturing benchmarks presented in the literature to the degree that they demon-
strate an inability on the part of Europe to close in on America.1 Subsequently, in
chapter 3 it was noted that on the basis of capital-intensity data for the pre-WW1 pe-
riod the possibility could not be ruled out that Europe’s backwardness resulted from the
use of relatively labor-intensive technology, possibly induced by a skilled-labor abun-
dance. Furthermore, chapter 4 demonstrated that within Europe Germany and the UK
operated at roughly similar levels of labor productivity.
These findings suggest that while the preconditions for growth in the period run-
ning up to 1910 differed across the Atlantic, they may have been similar between Eu-
ropean countries. Indeed, according to Stephen Broadberry, the US’s substantial lead
over Europe in manufacturing labor productivity showed a great degree of stationarity
of comparative performance in manufacturing, which suggests the prevalence of dif-
ferent long-run growth paths across the Atlantic.2 This begs the question whether in
1. US/UK: Broadberry and Irwin, “Labor Productivity in the United States and the United King-dom”; Jong and Woltjer, “Depression Dynamics.” Germany/UK: Broadberry and Burhop, “Compar-ative Productivity in British and German Manufacturing”; Ritschl, “The Anglo-German IndustrialProductivity Puzzle”; Broadberry and Burhop, “Resolving.”
2. Broadberry, “Manufacturing and the Convergence Hypothesis,” 788.
the decades leading up to WW1 European countries converged on a common level of
manufacturing performance?
Although the incapability of the UK and Germany to match US performance levels
has received most attention, several of the arguments presented in the literature aimed
at explaining the transatlantic labor-productivity gap are in principal easily extended
to include other countries as well. First, America’s advantage over Europe has been as-
sociated with its uniquely abundant supply of industrial mineral supplies and a scarcity
of skilled labor, the combination of which favored capital-intensive production.3 The
analysis in chapter 3 indeed uncovered a large capital-intensity gap in the pre-WW1
period in line with David’s and Broadberry’s view regarding differences in the choice of
technology. Given America’s unique supply of natural resources, all European countries
suffered from the same disadvantage as the UK and Germany did. The same holds for ar-
guments concerning market size and demand preferences; if these prevented Britain and
Germany from catching-up, they may well have constrained labor-productivity growth
in other European countries in a similar fashion.4
If European countries were indeed similarly affected by these local conditions in the
period before WW1, convergence with the US was unattainable for all. Instead, until
the catch-up mechanism described in chapter 3 kicked in after WW1, at least in the
case of Germany, countries may well have followed a European labor-productivity path,
characterized by low levels of performance as compared to the US. The notion of such
conditional convergence was introduced in the literature in response to the lack of em-
pirical support for unconditional convergence, as originally suggested by Solow.5 Since
the preconditions for unconditional convergence – i.e. countries are identical in levels of
technological knowledge, savings rates, population growth and depreciation rates – exist
in theory only, Solow’s model provides a bad fit with observed historical growth pat-
terns.6 However, controlling for differences across countries with respect to particular
parameters, such as the savings rate, human-capital formation or government consump-
tion, research revealed an inverse relation between initial per capita levels of income and
3. N. Crafts, “Forging Ahead and Falling Behind: The Rise and Relative Decline of the First Indus-trial Nation,” Journal of Economic Perspectives Vol. 12, No. 2 (1998): 202–203. See also chapter 3 fora more detailed discussion.
4. The lack of large-scale production has featured prominently, for instance, in explanations for theslow development in Dutch industry during a large part of the nineteenth century. See J.P. Smits,“The Determinants of Productivity Growth in Dutch Manufacturing, 1815–1913,” European Review ofEconomic History No. 2 (2000): 223–246.
5. R. Solow, “A Contribution to the Theory of Economic Growth,” Quarterly Journal of EconomicsVol. 70 (1956): 65–94.
6. L. Pritchett, “Divergence, Big Time,” Journal of Economic Perspectives Vol. 11 (1997): 1034–1052; W. Easterly and R. Levine, “It’s Not Factor Accumulation: Stylized Facts and Growth Models,”The World Bank Economic Review Vol. 15, no. 2 (2001): 177–219
Chapter 5. Did a European Convergence Club Exist Before World War 1? 135
subsequent rates of growth predicted by Solow.7 If the conditions under which countries
develop are similar, which may be argued for turn-of-the-century Europe, convergence
is expected.
A factor that promoted similar conditions for growth in Europe at the start of
the twentieth century concerns the relative openness of European economies between
1870–1913.8 Trade theory (Heckscher-Ohlin-Samuelson) predicts that differences in rel-
ative factor prices and thus in the mix of factor inputs used in production disappear
over time under conditions of free trade.9 Openness to trade and perfect competition
induces a country to specialize in the commodities whose production requires the inten-
sive use of the country’s relatively abundant, and thus cheap, production factor. When
engaging in international trade, a labor-abundant country specializes in labor-intensive
production processes. Consequently, the demand for labor increases, wages rise and the
wage/interest ratio goes up, too, which in turn erodes the the country’s comparative ad-
vantage in the production of labor-intensive commodities. As capital-abundant countries
experience a change of the wage/interest ratio in the opposite direction, relative factor
costs equalize between countries. For these dynamics to occur, barriers to trade ought
to be minimal. Around 1900, Europe showed the potential for such a well-integrated
market.10
In fact, Williamson already documented strong patterns of convergence in Europe
in the period 1870–1913 on the total-economy level. Using Maddison’s GDP-per-capita
data as well as his own data on real wages, Williamson shows that in the pre-WW1
period present OECD countries converged at a steady pace, a pattern particularly dis-
tinct when the US and Canada are left out of the sample.11 Broadberry notes, however,
that in the case of Germany, Britain and the US convergence was stronger on the
total-economy level than for manufacturing only.12 This suggests that convergence was
fueled mainly by compositional effects (reallocation of labor from agriculture to either
industry or services) rather than driven by the use of increasingly similar production
techniques induced by relative factor-cost equalization between countries. This might
be because the equalization of relative factor costs was, perhaps, thwarted; although
European economies were relatively open, trade tariffs did persist throughout the pe-
7. Barro, “Economic Growth”; Barro and Sala-I-Martin, “Convergence”; Fagerberg, “Technology.”8. For globalization and catch-up, see Williamson, “Globalization,” 295. For globalization in general,
see K. O’Rourke and J.G. Williamson, Globalization and History: The Evolution of Nineteenth CenturyAtlantic Economy (Cambridge, 1999).
9. Heckscher, “The Effect of Foreign Trade”; Ohlin, Interregional and International Trade; Samuel-son, “International Trade”; Samuelson, “International Factor-Price Equalization.”10. Hannah, “Logistics, Market Size, and Giant Plants.”11. Williamson, “Globalization,” 284.12. Broadberry, “Manufacturing and the Convergence Hypothesis,” 780–781.
riod 1870–1913, particularly in countries such as Germany and France.13. Alternatively,
even if relative factor costs differed little between countries, they may not have oper-
ated the same technology. Because the social competence necessary to exploit the most
advanced technology was still limited in the period before WW1, notes Abramovitz,
technology transfer left a weak mark on convergence.14 Therefore, the strong conver-
gence measured on the total-economy level before 1914 might not be visible when the
focus is on manufacturing only.
This chapter studies (sigma) convergence in manufacturing between five northwest-
ern European countries, i.e. the UK, Germany, France, the Netherlands and Sweden, in
the period leading up to WW1. First, I look at levels of manufacturing labor produc-
tivity in 1910 by constructing five bilateral industry-of-origin benchmarks. These are
needed because the time series of long-run productivity performance suffer from the
drawback that they do not adequately account for shifts in sectoral output and changes
in product prices, particularly when they are projected from a certain benchmark-year
into distant periods. In recent years, economic historians have stressed the need for new,
more detailed, comparisons of welfare and productivity for earlier periods, particularly
for the pre-WW1 era.15 As the previous chapters have emphasized, direct benchmark
comparisons between countries are a much wanted addition to the long-span projections.
Moreover, the best-known comparisons of long-run performance, i.e. those of Maddi-
son, are unsuited for the purpose of this chapter, as they capture developments at the
total-economy level only.16 Although I am not the first to measure comparative labor
productivity between pre-WW1 European countries, previous studies deviate from the
approach applied here in that they do not use the ICOP-technique to convert output
of different countries in a common currency.17
13. Dormois, “The Impact of Late-Nineteenth Century Tariffs,” 187.14. Abramovitz, “Catching-up,” 395. Williamson underlines this point, too, and ranks technology
transfer as a minor player in nineteenth-century convergence. See Williamson, “Globalization,” 299.15. E. Frankema, J.P. Smits, and P. Woltjer, “Comparing Productivity in the Netherlands, France,
UK and US, ca. 1910: A New PPP benchmark and its Implications for Changing,” Groningen Growthand Development Centre Memorandum no. 113 (2010): 1–34; L. Prados de la Escosura, “InternationalComparisons of Real Product, 1820–1990,” Explorations in Economic History Vol. 37 (2000): 1–41;K. Fukao, D. Ma, and T. Yuan, “Real GDP in Pre-War Asia: A 1934-36 Benchmark PurchasingPower Parity Comparison with the U.S.,” Review of Income and Wealth Vol. 53 (2007): 503–537; J.van Zanden, “Rich and Poor Before the Industrial Revolution. A Comparison Between Java and theNetherlands at the Beginning of the Nineteenth Century,” Explorations in Economic History Vol. 40(2003): 1–23.16. Maddison, Phases of Capitalist Development ; Maddison, Dynamic Forces in capitalist develop-
ment ; Maddison, Monitoring the World Economy.17. J. Dormois and C. Bardini, “Branch Comparisons of Manufacturing Labour Productivity for Eight
European Countries, Ca. 1910–1913,” Paper for N.W. Posthumus seminar on comparative historicalnational accounts for Europe in the 19th and 20th centuries (1994): 1–29; Dormois, “The Impactof Late-Nineteenth Century Tariffs”; J. Dormois, La Defense du Travail National? L’Incidence duProtectionnisme sur l’Industrie en Europe, 1870–1914 (Presses de l’Universite Paris-Sorbonne, 2009).
Chapter 5. Did a European Convergence Club Exist Before World War 1? 137
In a second step, I study the growth of manufacturing labor productivity in these
northwestern countries over the period 1870–1913. The notion of convergence implies a
time dimension and it is necessary not only to look at comparative levels of performance
in 1910, but also at the increase or decrease of labor-productivity differences over time.
If the level comparison constructed in the first step demonstrates very similar levels
of performance between these countries, labor productivity in all probability converged
in the period running up to WW1. In case of widely different labor-productivity levels
in 1910, the time dimension may still provide evidence of a decreasing variation in
performance levels, but – in the light of Broadberry’s findings – it is not necessarily
anticipated. In addition, as in the case of America versus Europe, between European
countries the conditions for growth differed also. While the five countries studied here
may be the same in that they all faced less favorable conditions for labor-productivity
growth as compared to the US, although arguably to a different degree, dissimilarity
prevailed in many other respects.
5.2 Methodology
For the construction of benchmarks this chapter employs the approach set out in chap-
ter 2. The only differences are that, first, output is measured by value added (rather
than gross output) and, second, employment is not corrected for hours worked. With
respect to the latter, this choice is induced by the inability to find the necessary data
on hours worked on the industry level for all countries studied here. This is, however,
unlikely to introduce a bias in the results as chapter 2 already showed that such an
adjustment makes little difference for the pre-WW1 period. This means that relative
productivity at the industry level is estimated by the value of net output per employee
(in national currency), translated into a common currency with an industry-specific
PPP-adjusted price ratio based on factory-gate data.18
Previous work on economic performance in pre-WW1 European countries has been
conducted along different lines. The choice for a different strategy has been fueled by
the lack of data on the early twentieth century. In some cases, the limited availability of
data for the pre-WW1 period introduces difficulty implementing the analysis along the
lines of the ICOP approach. With the exception of the US and the UK, no other coun-
18. The benchmark method is formally defined in section 2.2 on page 20 of chapter 2. See also: D.Paige and G. Bombach, A Comparison of National Output and Productivity of the United Kingdomand the United States (Paris: Organisation for European Economic Co-operation, 1959), 1–245; vanArk and Timmer, “The ICOP Manufacturing Database”; Fremdling, de Jong, and Timmer, “Britishand German Manufacturing Productivity Compared”; Fremdling, de Jong, and Timmer, “CensusesCompared”; Jong and Woltjer, “Depression Dynamics”; Jong and Woltjer, “A Comparison of RealOutput and Productivity.”
try published a census of manufactures and the quantity and quality of the data made
available through other sources, such as statistical yearbooks, lack full-manufacturing
coverage, as already demonstrated for Germany. Consequently, sometimes even the con-
struction of output conversion factors based on factory-gate prices can be difficult. In
such circumstances, one may proceed in different ways to convert output. First, as
is done here, PPPs can be constructed using the factory-gate prices that are available,
which possibly provide poor coverage of the output that is compared between countries.
Alternatively, price information can be obtained from other sources, for instance
from wholesale, retail or trade data. This approach perhaps increases coverage, but it
introduces an unknown bias in the PPPs, as these expenditure prices are determined
partly by factors outside the production process. Third and finally, when no price infor-
mation is available or the quality of the data is ill-regarded, the official exchange rate
presents a last resort. There are arguments against and in favor of each of these three
options and the choice of technique primarily depends on the type of question that is
confronted with the data. As the benchmarks presented here are valued for their break-
down of manufacturing in underlying industries, the exchange rate, which captures a
total-economy average relative price level, provides a poor instrument of analysis in this
case. As a result, either factory-gate or expenditure PPPs should be used.
Both measures of relative price do not necessarily return the same value and as
the PPPs form a main ingredient in the calculation of comparative labor productivity,
the choice between factory-gate or expenditure PPPs may affect our understanding of
historical development. This sensitivity of comparative labor productivity with regard to
relative price levels is clearly demonstrated by the debate between Broadberry and Ward
& Devereux in the Journal of Economic History. Ward & Devereux have constructed
expenditure PPPs – in line with the methods applied by scholars such as Gilbert, Kravis
and Maddison in the United Nations International Comparison Project – to obtain seven
benchmark estimates of US and UK income per capita and output per worker between
1872 and 1930.19 Their expenditure PPPs deviate markedly from conventional estimates
of US/UK relative price levels and imply a revision of America’s overtaking of Britain
in GDP-per-capita levels. While traditionally the view was held that the UK pertained
a lead up till 1900, the results obtained by Ward & Devereux suggest that the US had
overtaken Britain already before the 1870s.20 In this instance, the new PPPs change
our perception of the past.
19. Ward and Devereux, “Measuring British Decline”; Ward and Devereux, “Relative U.K./U.S. Out-put Reconsidered”; A. Maddison, The World Economy: a Millennial Perspective (Paris: Organisationfor Economic Cooperation / Development, 2001), 1–383.20. Broadberry, “Relative Per Capita Income Levels.”
Chapter 5. Did a European Convergence Club Exist Before World War 1? 139
As the expenditure PPPs establish a direct link between comparative income levels
and consumption possibilities, those estimates are particularly suited for international
comparisons of income and living standards, as in the case of Ward & Devereux. How-
ever, for international comparisons of productivity and economic performance in gen-
eral, which is the purpose of this research, a direct comparison of output at an industry
level is preferable.21 Whereas expenditure PPPs take the impact on consumer prices
of imports, trade margins, transport costs and taxes into account, factory-gate PPPs
exclude such factors and thus produce a more refined comparison of labor productivity
levels. This is not to say that a factory-gate approach is a superior methodology, it is
suggested only that the choice for expenditure or factory-gate prices primarily depends
on one’s research objective: living standards as measured by real income or economic
performance as measured by real value added? Given that the focus in the current
chapter is on the latter, factory-gate prices are favored.
Previously, however, researchers studying comparative labor productivity in pre-
WW1 Europe have opted for one of the two alternative strategies for the purpose of
converting output values. Most relevant in this respect is the research conducted by the
French economic historian Jean-Pierre Dormois, who released several vintages of a pre-
WW1 European labor-productivity comparison, also on the disaggregated level. His first
attempt, together with Carlo Bardini, employed the price comparison approach using
PPP-adjusted price ratios to convert output, as advocated here. The PPPs, however,
are based on expenditure prices obtained from either wholesale, retail or export data
and calculated on the level of total manufacturing only.22 The problems with these
PPPs, duly acknowledged by the authors, not only concern the nature of expenditure
prices and the lack of detail on the industry level, but also the selection bias of the
commodities included in the basket of goods compared between countries; Dormois and
Bardini selected semi-finished products only, while finished goods are left out. They do
so because semi-finished products are of universal quality and thus easily comparable
between countries.23
In more recent years, Dormois introduced a new release of his European labor-
productivity comparison. Here, the construction of PPPs involves a two-step approach.
First, he takes the crude ‘real’ exchange rates published by the Economic and Financial
Department of the League of Nations in 1926, based on factory-gate prices, which in
a next step are extrapolated backward to 1910 using country-specific price indices.24
21. van Ark, International Comparisons of Output and Productivity; van Ark and Timmer, “TheICOP Manufacturing Database.”22. Dormois and Bardini, “Branch Comparisons,” 7.23. ibid., 8.24. Dormois, “The Impact of Late-Nineteenth Century Tariffs,” 177.
Chapter 5. Did a European Convergence Club Exist Before World War 1? 141
5.3 Purchasing power parities for pre-WW1 Euro-
pean countries
This chapter’s main contribution to the literature is the application of the ICOP ap-
proach to construct industry-of-origin benchmarks for pre-WW1 European countries.
The PPPs constructed for the countries studied here are reported in table 5.1. The
table shows the official exchange rate and three PPP-variants; the Laspeyres, Paasche
and Fisher PPPs. Respectively, these refer to PPPs obtained by use of base-country
weights, non base-country weights and the geometric average thereof. Comparing the
Fisher PPP with the exchange rate, some deviation is observed, but in all cases to a
small extent only. For the UK and Germany, the official exchange rate slightly overes-
timates the domestic currency’s strength, while the reverse applies in all other cases.
The fact that the manufacturing PPPs resemble the exchange rate fairly closely signifies
that the former heavily influences the latter. Given that the exchange rate captures,
by and large, the relative price of traded goods, it suggests that trade consisted of
manufacturing products in the main, which – indeed – it did.27
Table 5.1: Purchasing power parities for total manufacturing, ca. 1910
UK GER FRA NL SWE
(£/$) (Mark/$) (Ffr/$) (Dfl/$) (Skr/$)
Exchange rate 0.21 4.20 5.18 2.49 3.73
PPP – Laspeyres 0.22 4.33 5.44 2.66 4.17
PPP – Paasche 0.18 3.49 5.60 1.99 3.98
PPP – Fisher 0.20 3.89 5.52 2.30 4.07
Sources: see section 5.2.
As compared to other pre-WW1 star comparisons presented in the literature, the
use of factory-gate prices in the construction of PPPs sets this study apart from earlier
work. So how do my results compare to the conversion factors used by others? Table 5.2
reports the PPPs used here and those introduced before in the research discussed in the
previous section. The first two rows of table 5.2 sets out my results against Dormois’
latest PPPs. As already explained, this last batch of PPPs by Dormois are obtained
by taking the relative factory-gate prices in the 1920s and then extrapolating these
27. By 1913, northwestern Europe was a net exporter of manufactured goods and a net importer ofprimary products, such as food and raw agricultural materials. Moreover, the bulk of the trade betweennorthwestern European countries studied here (imports plus exports) involved manufactured products,while primary goods were imported from other, less-developed parts of the world. See O’Rourke andWilliamson, Globalization and History, 412.
Sources: J. Dormois, “The Impact of Late-Nineteenth Century Tariffs on the Productivity ofEuropean Industries, 1870–1930,” in Classical Trade Protection, 1815–1914, ed. J. Dormoisand P. Lains (London: Routledge, 2006), 178, J. Dormois and C. Bardini, “BranchComparisons of Manufacturing Labour Productivity for Eight European Countries, Ca.1910–1913,” Paper for N.W. Posthumus seminar on comparative historical national accountsfor Europe in the 19th and 20th centuries (1994): 9 and A. Burger, “A Five CountryComparison of Industrial Labour Productivity, 1850–1990,” Paper for N.W. Posthumusseminar on comparative historical national accounts for Europe in the 19th and 20thcenturies (1994): 5.
Chapter 5. Did a European Convergence Club Exist Before World War 1? 145
matches for each bilateral comparison. In general, a high number of product matches
indicates a broad coverage of production and ensures that the PPP is representative
for the branch it applies to. In contrast, when industry PPPs rely on a few product
matches only, the ensuing PPP may not reflect accurately the relative price level of a
branch, especially when the products included in a branch are highly divers, such as in
chemicals. With this in mind, table 5.4 invokes confidence in the British, German and
Dutch comparisons, as they provide a much higher coverage of products as compared
to, for instance, the earlier star comparisons of Dormois & Bardini and Burger.28 The
product coverage of the French and Swedish comparisons is similar to that in these
earlier studies.
Yet there are mitigating circumstances for the comparisons with few product
matches only. The coefficient of variation, which captures the spread between the branch
PPPs of a country, reported in the last row of table 5.4, is reassuringly low for France
and Sweden. Although in combination with a low number of matches, a variation of rel-
ative prices larger than for countries with better coverage may suggest that the branch
PPPs are based on an unrepresentative sample of products, the spread of the branch
PPPs in neither France nor Sweden points in that direction. Even if we adjust the co-
efficient of variation for all countries to exclude metals & machinery, as in the case of
Sweden, the country displaying the largest variation is the Netherlands.29 Given the
high number of product matches for the Dutch/US comparison, I am fairly confident
that this reflects actual differences in relative price levels between branches.
This belief is strengthened by the fact that manufacturing branches by 1910 were, in
terms of product variation, much less complex than in later periods and a large share of
total output was covered by fewer products. This means that a low number of matches
does not necessarily pose problems concerning the reliability of the comparison. Lastly,
both in the case of France and Sweden, the total-manufacturing PPP relate to the
expenditure PPPs presented before in the literature and the formal exchange rate in a
manner very similar to the countries with much higher coverage.
5.4 Comparative productivity around 1910
Using the PPPs introduced above to convert the labor-productivity data of the countries
studied here to a common currency, I obtain a measure of comparative performance
28. Dormois and Bardini, “Branch Comparisons”; A. Burger, “A Five Country Comparison of Indus-trial Labour Productivity, 1850–1990,” Paper for N.W. Posthumus seminar on comparative historicalnational accounts for Europe in the 19th and 20th centuries (1994): 1–27.29. Excluding metals & machinery the coefficient of variation for the UK, Germany, France and the
Netherlands is, respectively, 0.09, 0.16, 0.22 and 0.24.
relative to the US. Table 5.5 reports the comparison of single deflated value added
per employee.30 Clearly, none of the European countries were able to catch-up with
America, a finding which does not come as a surprise. Germany approached American
productivity levels closest, but still faced a big gap. Moreover, the German/US gross
output per employee comparison presented in chapter 2 attributed a stronger relative
performance to Germany, i.e. a comparative level of 57%, which suggests that the share
of intermediate inputs in gross output was larger for Germany than for the US. The
country lagging behind furthest was the Netherlands, which attained a performance
of only a third the level realized across the Atlantic, while the other three countries,
i.e. France, the UK and Sweden, were evenly spaced in between these two European
extremes.31
Table 5.5: Comparative labor productivity (US = 100%), ca. 1910single deflated value added per employee
Branch UK GER FRA NL SWE
Food, drink & tobacco 47 33 38 40 38
Textiles, leather & clothing 48 75 46 29 44
Chemicals 49 54 32 10 39
Metals & machinery 38 72 45 18 36
Miscellaneous 42 51 49 40 44
Manufacturing 41 50 38 32 36
Sources: see section 5.2.
The size of the US lead differed between manufacturing branches, with each Euro-
pean country having relatively strong and weak points. The most pronounced differ-
ences in comparative performance between manufacturing branches are observed for the
Netherlands. The Dutch economy displayed extremely low levels of labor productivity
in heavy industries, a finding which was already anticipated by the PPPs reported in
table 5.3. The high PPPs for these industries indicate that the Netherlands proved un-
able to produce at low costs, which, among other things, could result from low levels of
productive efficiency. The Dutch performance in more traditional and light industries
was much stronger. Germany showed a mixed experience, too, which has already been
pointed out in chapter 2. The UK, France and Sweden are characterized by less di-
verging levels of comparative performance, although even in these cases branches with
30. In case of single deflation, the purchasing power parities are based on final products only andnot corrected for possible deviations between German/US price relations of intermediate and finalproducts. See also section 2.2 in chapter 3.31. Sweden’s relative distance to the US has been calculated using a two-step procedure. See ap-
Prado’s figures imply a UK/US level of 61%, which seems unreasonably high set out
against the direct estimates composed by Broadberry & Irwin and Woltjer, he seems to
overstate the British performance prior to WW1. This results partly from Prado’s use
of census-definition UK data. My results are closer to Broadberry’s estimates, which
put Sweden and the UK on parity in 1913 and points at a Swedish performance of about
80% the level in Britain by 1909.
5.5 Change of comparative labor productivity, 1870–
1910
The notion of convergence implies a time element and a study of the levels of relative
performance in one year only cannot answer the question whether European countries
gravitated toward a common path typified by a performance about half the level of
the US. Even though the previous section recorded substantial differences between the
manufacturing performance of European countries, the spread of comparative labor-
productivity levels may still have been less by 1910 than in periods before. Having
established the relative levels of labor productivity for European countries around 1910,
these can be extrapolated backward using time series of output and employment for
each country. This is done by, first, calculating the change in labor productivity per
country and, second, multiplying the relative change of labor productivity between two
countries by the level of comparative labor productivity in a base year, which is the
benchmark year 1909 in this case, as described in equation (5.1):
yeurt
yust=
(yeurt /yeur09
yust /yus09
)· y
eur09
yus09(5.1)
with yt as a country’s level of labor productivity in period t and yt/y09 the change
of labor productivity between period t and base-year period 1909. As the level of la-
bor productivity is unobtainable, except for the benchmark year, the change of labor
productivity is derived from the change in output and employment:
yty09
=ot/o09lt/l09
(5.2)
where ot and o09 capture output in period t and base-year period 1909, respectively,
while lt and l09 refer to employment in these periods. The time-series data necessary
for this exercise is taken from the existing literature.33
33. UK and US: Broadberry, The Productivity Race – SWE: S Prado, Aspiring to a Higher Rank:Swedish Factor Prices and Productivity in International Perspective, 1860-1950 (University of Gothen-
Figure 5.1 plots the projections. It does not reveal obvious signs of convergence. At
the start of the 1870s, the northwestern European countries studied here were divided
in two groups. The UK and France performed at slightly less than half the level of the
US, while the Netherlands and Sweden trailed further behind, operating at roughly a
third of American performance levels. For this period no information on Germany is
available. Shortly after 1880, when Germany does enter the sample, matters had started
to change in Europe. France’s comparative performance steadily declined on account of
a stagnant level of labor productivity at home. During this process France fell behind
of the UK and by 1890 joined ranks with the Netherlands and Sweden. Germany, which
in 1882 still trails behind the UK, appears to catch up with Britain around 1890 and,
particularly in the period 1900–1905, managed to move away, but by a small margin
only. During the same years, Sweden started to take over first the Netherlands and then
burg, 2008) – FRA: J.P. Dormois, “Tracking the Elusive French Productivity Lag in Industry, 1840–1973,” Hi-Stat Discussion Paper Series No. 152 (2006): 1–41 – NL: J.P. Smits, E. Horlings, and J.L.van Zanden, Dutch GNP and its Components, 1800–1913, GGDC Monograph Series 5 (Groningen:Groningen Growth / Development Centre, 2000), 1–246.
a trend.34 In contrast to total-economy developments, the first era of globalization saw
no convergence in manufacturing labor productivity between northwestern European
countries.
On the disaggregated level it was not possible to extrapolate the benchmark levels
backward. The data necessary for such an exercise are not available.35 Nevertheless,
looking at the dispersion of labor productivity at the industry level between European
countries reported in the last column of table 5.8, it seems unlikely that a break down of
the aggregate time series would show different results. Only in food, drink & tobacco the
dispersion is lower than on the total manufacturing level. Because the composition of
miscellaneous differs between the countries, the CoV thereof is difficult to interpret and
may not reflect the dispersion of performance between similar industries. That leaves
textiles, chemicals and metals, all of which show a high degree of variation. There is no
reason to expect industry-level patterns different from the lack of convergence observed
for total manufacturing. Rather, the disaggregated results clearly reject the notion of a
similar labor-productivity path across countries in northwestern Europe before WW1.
5.6 Manufacturing and convergence at the country
level
The time-series extrapolations demonstrate a stationarity of Europe’s comparative per-
formance relative to the US in the long run, a conclusion very much in line with Broad-
berry’s earlier work on the US, UK and Germany. Although the ranking of European
countries according to their comparative performance changed from time to time, e.g.
France’s relative decline between 1870–1890 and Sweden’s growth spurt after 1900, the
dispersion of performance in manufacturing remained unaltered on average. With re-
gard to a European convergence club, no evidence was found of a common long-run
equilibrium for northwestern European countries. The stationarity of European manu-
facturing performance applied to productivity differences both relative to the US and
between European countries.
At the same time convergence did take place on the total economy level. Table 5.10
reports for the five European countries included in this study the levels of GDP per
capita relative to the US and the dispersion of productivity across countries measured
34. The slope coefficient on the fitted linear trend is not statistically different from zero. Moreover,augmented Dicky Fuller tests do not suggest unit root, so the series appears stationary.35. For the US, the UK, Germany and the Netherlands time-series evidence is available at the level of
industries, for France and Sweden it is not. It may be hazardous to draw conclusions regarding Europe’sgrowth experience on an even further reduced sample of European countries (only 3 countries).
Figure 5.4: Dispersion of comparative performance, 1875–1909(coefficient of variation)
.12
.16
.20
.24
.28
1875 1880 1885 1890 1895 1900 1905
Manuf. labor productivity GDP per capita
Average (GDPcap)
Average (manuf. lp)
Countries: UK, Germany, France, the Netherlands and Sweden.Sources GDP per capita: J. Bolt and J.L. van Zanden, “The First Update of the MaddisonProject; Re-Estimating Growth Before 1820,” Maddison Project Working Paper 4 (2013).Sources manufacturing labor productivity: this study.
gap to the US was much smaller in services. Moreover, particularly in the UK compara-
tive levels of labor productivity in agriculture were higher than in other sectors. Also in
the Netherlands agriculture had not fallen as far behind the US as manufacturing, even
though agriculture had experienced a relative decline in the decades running up to 1910.
Midway the nineteenth century the Dutch level of labor productivity in agriculture was
at 85% of the British level.38 And in services this figure was even as high as 92% (es-
pecially due to the strong performance of the Dutch trade sector, which had a level of
labor productivity which was 30% higher than in the UK). Both agriculture and services
witnessed a steady decline in comparative productivity rates vis-a-vis the United States
as well as the United Kingdom throughout the second half of the nineteenth century.39
In France, agriculture operated at low labor-productivity levels relative to other
38. Frankema, Smits, and Woltjer, “Comparing Productivity,” 21.39. ibid.
Chapter 5. Did a European Convergence Club Exist Before World War 1? 157
Table 5.11: Comparative labor productivity in sectors of the economy(US = 100%), ca. 1910
UK FRA NL
Agriculture 56 37 47
Mining 38 39 10
Manufacturing 41 38 32
Services 84 68 85
Source: Frankema, Smits, and Woltjer, “Comparing Productivity.”
productive activities, including manufacturing. Surprisingly, during the phase of indus-
trialization after the 1850s France maintained a large labor force in agriculture. The
limited migration from rural to urban areas has been ascribed to a persistent cultural
belief in and adherence to small, traditional farming, which defied modernization and
suppressed agricultural labor productivity until well into the twentieth century.40 In
contrast, agriculture in the UK attained high levels of performance already early in the
nineteenth century. Whereas in France the move out of agriculture was delayed until
after the turn of the century, the share of agricultural employment was comparatively
small in Britain. These different dynamics help explain why the gap to the US in terms
of GDP-per-capita levels was much smaller than the manufacturing labor-productivity
gap for the UK and the Netherlands, but less so for France.
The comparative productivity levels obtained in this study also carry implications
for our understanding of the period after 1909 and may answer questions concerning
economic development in the interwar period. Van Ark’s data show that by 1950 the
UK, Germany, France and the Netherlands operated much closer together in terms
of manufacturing labor productivity than before WW1. Furthermore, over the period
1950–1989 the dispersion thereof did not reduce further.41 Given the lack of convergence
(or, for that matter, divergence) between 1875–1910, forces must have been active during
the period 1910–1950 that drove together levels of manufacturing performance between
European countries. An example of which is the Netherlands, which showed the lowest
levels of labor productivity before WW1, but closed the gap to Germany entirely over
the interwar years.42 In view of the findings in chapter 4, it seems likely that such
40. J.P. Dormois, The French Economy in the Twentieth Century (Cambridge: Cambridge UniversityPress, 2004), 102.41. Van Ark reports two series of comparative labor productivity (Sweden is not included), taking
first the US and then the UK as the base country. The CoV of the former increased between 1950–1989from 0.09 to 0.12, while in the latter’s case it remained constant at 0.10. See van Ark, InternationalComparisons of Output and Productivity, 290–291.42. H.J. de Jong, Catching Up Twice: the Nature of Dutch Industrial Growth During the Twentieth
Century in a Comparative Perspective (Berlin: Akademie Verlag, 2003), 66.
Chapter 5. Did a European Convergence Club Exist Before World War 1? 159
of labor as the US and France about a sixth only.44 These pronounced differences at the
start of the twentieth century between European countries and America suggest that
variation in capital-labor ratios contributed substantially to the US’s lead over Europe.
France is a case in point. According to Dormois, in the traditional consumer indus-
tries – which employed a large part of the French manufacturing labor force – it failed
to achieve a strong labor-productivity performance because of a limited adoption of
newly available technology, a feature common to much of French manufacturing.45 The
relatively weak Dutch performance during the late nineteenth century has also been
explained by the slow adoption of steam power.46 Traditional sources of energy, like
wind, water and peat prevailed. These technologies had remained unchanged from the
seventeenth century until about the 1850s.47 Levels of aggregate domestic demand were
so low that traditional types of production (i.e. based on the use of wind- and water
power) retained their cost advantage over the introduction of steam engines, a process
that induces high initial fixed costs.48
So among other factors, the Netherlands retained obsolete technology because it
was cost efficient, a point which has also been made repeatedly for nineteenth-century
Britain.49 Broadberry argues that for Britain, the increased competition from abroad
between 1870–1914, particularly from America and Germany, led to an efficiency in-
crease in flexible and labor-intensive production.50 It was a rational response to compe-
tition to cut-back on relatively expensive factor inputs in an attempt to minimize pro-
duction costs. In the face of small domestic markets, heterogeneous demand patterns,
scarcity of natural resources and abundance of skilled labor, this process of competition-
induced cost minimization discouraged the UK from acquiring machine-intensive tech-
nology and encouraged the further improvement of the technology already in use.
“Most industries were characterized by a high degree of competition, which
acted as a spur to efficiency, with existing rivals or new entrants ready to
44. Hannah, “Logistics, Market Size, and Giant Plants,” 71.45. Dormois, The French Economy, 14; Dormois, “Tracking the Elusive French Productivity Lag,” 8.46. Smits, “The Determinants of Productivity Growth in Dutch Manufacturing, 1815–1913,” 239–240.47. M. Jansen, De Industriele Ontwikkeling in Nederland (Amsterdam NEHA, 2000).48. Smits, “The Determinants of Productivity Growth in Dutch Manufacturing, 1815–1913,” 235–
238; Horlings and Smits point at the importance of demand constraints in the Dutch economy and itsimpact on the timing of modern economic growth, see: E. Horlings and J. Smits, “Private ConsumerExpenditure in the Netherlands, 1800–1913,” Economic and Social History in the Netherlands No. 7(1996): 15–40.49. See also: D. McCloskey, Economic Maturity and Entrepreneurial Decline: British Iron and Steel
(Cambridge, Mass.: Harvard University Press, 1973), C.K. Harley, “Skilled Labour and the Choiceof Technique in Edwardian Industry,” Explorations in Economic History Vol. 11 (1974): 391–414, L.Sandberg, Lancashire in Decline (Columbus, 1974).50. Broadberry, The Productivity Race, 158–159.
take up opportunities neglected by incumbent producers.”51
Other than considerations of cost advantages, it has been suggested that countries
may refrain from adopting advanced technology due to a lack of necessary social ca-
pabilities.52 Yet before WW1 educational attainment differed little between countries
in the developed world, although some deviations were evident. Germany enjoyed a
lead over the US, UK and France in terms of average years of secondary schooling,
while the percentage of population gaining access to a college education was largest
in America.53 However, given the small share of science-based industries in manufac-
turing before WW1, the modest differences in educational attainment were of limited
consequence for convergence.54 What appears to have mattered more, at least on the
total-manufacturing level, are specialization patterns. An emphasis on mature industries
delays the wide-spread adoption of technologies introduced in modern industries. For
instance, in the industries in which the Dutch economy had strongly specialized, such
as food processing, the use of steam power proved difficult for technological reasons.55
The upshot of this literature is clear; instead of adopting high capital-labor ratios,
European countries improved their competitiveness by increasing the labor-productivity
performance of the technology already in use. It implies that variation in labor-
productivity levels within Europe stemmed from the degree to which countries success-
fully explored or even enhanced the potential of the machine-intensity level at which
they operated. This provides a useful perspective to address questions of relative stand-
ing that remained unanswered by a study of capital-intensity differences only. How could
Germany attain higher labor-productivity levels than the UK with a lower capital-labor
ratio? Why did France trail the UK at relatively close distance only, while it employed
half as much horse power per unit of labor?56 What prevented Sweden, which enjoyed a
machine-intensity level not dissimilar from the US, from outperforming all other Euro-
pean countries?57 If European countries indeed experienced labor-productivity growth
through a process of learning-by-doing, rather than by slavishly copying advanced tech-
51. Broadberry, The Productivity Race, 209.52. Abramovitz, “Catching-up,” 395.53. Nelson and Wright, “The Rise and Fall,” 1947–1948.54. ibid., 1942, 1949.55. W. Lintsen, Geschiedenis van de Techniek in Nederland. De Wording van een Moderne Samen-
leving, 1800–1890 (Zutphen: Walburg Pers, 1992), 269–271.56. Broadberry reports a French machine-intensity level of 77% of the UK. As, according to Broad-
berry, machine intensity in the UK stood at a level 47% of the US, France’s comparative machine-intensity relative to the US was 36%, i.e. similar to Germany’s level of machine intensity. Moreover,he reports a France/UK comparative labor-productivity level of 65%, which is much lower than myestimate of 91%. Broadberry, The Productivity Race, 109. Without going into detail about the differ-ences in the estimates, Broadberry’s figures imply a different question: why did France trail Germanyat considerable distance, while it employed just as much horse power per unit of labor?57. Hannah, “Logistics, Market Size, and Giant Plants,” 71.
hand, was rapidly increasing and by 1930 relatively high in most Dutch industries as
compared to their British and German counterparts.60 Clearly, there was a continuity
over the interwar period for some constraints to labor-productivity growth, but machine
intensity does not appear to be one of them.
A possible explanation for the change over WW1 relates to the disintegration of mar-
kets. The decision to not adopt American production technology before WW1 seems
related to competition, which forced countries to produce cost efficiently, even though
innovation took place at predominantly high capital-labor ratios and an awareness of
these frontier movements provided an incentive for capital intensification. The outbreak
of WW1 ended the long period of openness and globalization between 1870–1914 and
countries resorted to protectionist policies after WW1. International markets disinte-
grated rapidly.61 As protectionist policies raise domestic prices above the world-market
level, they relieve the downward pressure on production costs. Firms are allowed to
produce with some degree of inefficiency without losing their domestic market share
to foreign competitors, which was not possible before WW1. Under these condition in-
dustries can adopt new technology, even when it is initially operated at low efficiency
levels.
This line of reasoning presents a variation on the infant-industry argument and
may shed new light on the differences in accumulation strategies over WW1.62 But
this perspective is at odds with the literature that associates the move away from
competition during the interwar years with a distortion to ‘adjustment mechanisms’,
which provided the possibility to retain old technology longer than would have been
feasible under conditions of competition.63 These adjustment mechanisms, however,
prevented capital intensification in the period before WW1 and, once blocked, may have
created the necessary opportunity for a move toward high machine-intensity levels. But
this assertion requires supporting empirical evidence, which calls for further research.
60. Jong, Catching Up Twice, 79.61. Broadberry, The Productivity Race, 210.62. H.J. Chang, Kicking Away the Ladder. Development Strategy in Historical Perspective (London:
Anthem Press, 2002).63. Broadberry, The Productivity Race, 291.
an implicit Swedish/UK level. Table 5.12 presents such a gross-output based compari-
son for the UK, Germany and Sweden. For France and the Netherlands no gross-output
data is available. Clearly, adjusting our output definition from value added to gross out-
put has a substantial effect on the comparative performance of the UK. The increased
British performance suggests that the share of intermediate inputs in gross output is
far lower in the US than in the UK.
If I now calculate for Sweden indirect levels of comparative productivity relative
to the UK (shown in the last three columns of table 5.12), Sweden’s performance is
adjusted downward from 110% to 88%.65 The Swedish performance is still substan-
tially higher than Prado’s estimates, but his figures imply a UK/US level of 61%, which
seems suspiciously high set out against the direct estimates composed by other re-
search.66 As with the case of France, the difference between Prado’s direct Swedish/UK
benchmark and the implicit relative level indicated here results mainly from Prado’s
use of census-definition UK data, which overestimates British performance. Also, the
PPPs constructed here favor Sweden. Sweden’s comparative performance reported in
table 5.8, is then obtained by projecting the Swedish/UK ratio presented above on the
UK/US relative levels of table 5.8.
65. For Germany I have included in table 5.12 an indirect productivity level relative to the UK usinggross-output definitions, too, which turns out slightly lower than in the 22% German advantage overBritain obtained using value added (see table 5.8).66. Broadberry and Irwin, “Labor Productivity in the United States and the United Kingdom”;
Frankema, Smits, and Woltjer, “Comparing Productivity”; Jong and Woltjer, “Depression Dynamics.”
Notes to table A.1:a Output values are measured in Goldmarks.b Weighted average of the UVs of products produced in that industry.c Output is deflated using industry-specific PPPs; comparison based on real output values.d Output is not deflated; comparison based on nominal output values.
Notes to table A.2:a Output is measured in hectoliter.b Output is measured in metric ton.c Output is measured in kilogram.d Output is the quantity of raw tobacco used in the production of tobacco.
Notes to table A.3:a Output is measured in hectoliter.b Output is measured in metric ton.c Output is measured in kilogram.d Output is the quantity of raw tobacco used in the production of tobacco.
Notes to table A.4:a Output is measured in hectoliter.b Output is measured in metric ton.c Output is measured in kilogram.d Output data is derived from the ’Report of commissioner of internal revenue’. The numbers
of cigars, cigarettes and other tobacco products are converted to the input of raw tobacco.
This was feasible as the internal revenue reports the number of pounds of tobacco needed for
the production of 1,000 cigars or cigarettes or other products. Subsequently, the use of raw
tobacco for the production of cigars, cigarettes and other tobacco manufactures are summed.
Employment is obtained from the census of manufactures 1909 and represents the total number
of people working in the tobacco manufactures industry (so both workers and proprietors).e For cotton yarn and thread production, only the number of spinners are reported by the
census. Here, I added to the spinners an estimate of wage earners also working in this industry.f These data refer to ‘Rubber, not elsewhere classified’. Tire production is the main activity
of this industry. For 1909 I do not have information about the share of the value of tires in
total production. For 1914, tires formed 65% of the total output value of ‘Rubber, n.e.c.’.