Rethinking the Industrial Revolution · Rethinking the Industrial Revolution Liam Brunt Abstract1 In this paper we offer a new and simple definition of the Agricultural Revolution,

Rethinking the

Industrial Revolution

Liam Brunt

Abstract1 In this paper we offer a new and simple definition of the Agricultural Revolution, the event that facilitated and prompted the Industrial Revolution. We define it as an exceptionally rapid increase in Total Factor Productivity (TFP). This generates a rightward shift in the supply curve for agricultural products and results in rising output and productivity. We show that this definition is implicit in the work of most researchers working in this area. The clarity of the new definition enables us to show that the Agricultural Revolution debate cannot be answered by the data currently being generated (output estimates and real rent series). On the basis of the data, we cannot pinpoint the timing of the Agricultural Revolution anywhere between 1500 and 1850 – and so we cannot distinguish between the historical interpretations of Allen (1999) and Overton (1996). We show that instead we need to gather more direct evidence about the supply side and how technology changed over time. This will help us to understand the timing of the Industrial Revolution. Keywords: Agriculture, development, total factor productivity. JEL Classification: N01, N53, O47.

1 This research was funded by the Economic and Social Research Council I am grateful to Lucy White for helpful comments and I would like to thank Bob Allen for making his data publicly available. Any remaining errors are my own responsibility.

6

I. Introduction. During the Industrial Revolution, economic growth and

structural transformation in Britain were dominated by changes in the

agricultural sector. At first sight this statement may appear paradoxical – but, in

fact, it is easy to justify. First, we should note that agriculture was the largest

sector until 1840 (measured as a percentage of total output or total employment)

and wheat was the largest single component of national income.2 Second, the

pre-eminence of the agricultural sector meant that the growth rate in agriculture

was the primary determinant of the national growth rate (of both output and

productivity). Moreover, the downward revision of growth by Crafts and Harley

merely accentuates this effect – because it implies that the agricultural sector

was proportionately larger in 1700 than we had hitherto believed.3 Third, the

defining feature of the Industrial Revolution was the transfer of labour

resources from agriculture to industry. But Britain had to remain more or less

self-sufficient in foodstuffs because there were relatively few exportable

surpluses being produced by other European countries, at least in the eighteenth

century.4 So in order to release labour from agriculture and remain self-

sufficient, Britain required a high level of output per worker in agriculture. In

that sense, an improvement in agricultural productivity was a pre-requisite for

industrialisation to occur.5 Fourth, in international terms Britain was

conspicuous for its high level of agricultural output and productivity.6

In the light of these facts, it is scarcely surprising that historians have

been extremely interested in the timing and causes of output and productivity

growth in English agriculture. Moreover, there is ample qualitative evidence

(from both domestic and foreign sources) that important changes were

2 Crafts N F R, ‘The Industrial Revolution,’ in Floud R and D N McCloskey (eds), The Economic History of Britain since 1700, vol. 1 (Cambridge, 1994), 144-59. 3 Crafts N F R and C K Harley, ‘Output Growth and the British Industrial Revolution: a Restatement of the Crafts-Harley View,’ Economic History Review, vol. 45 (1992), 703-30. 4 That is to say that Britain was a large country (in terms of international trade) and the world elasticity of supply of agricultural products was low. So transferring labour from agriculture to industry in Britain would have resulted in a sharp rise in British agricultural prices and no off-setting inflow of foreign foodstuffs. 5 Crafts N F R, British Economic Growth during the Industrial Revolution (Oxford, 1985), 116-22. 6 Wrigley (Sir) E A, People, Cities and Wealth (Oxford, 1987), 157-193.

7

occurring in the English agricultural sector.7 This included the adoption of

many new techniques and the reorganisation of production and rural

institutions. Together these changes were deemed by Toynbee to constitute

nothing less than an ‘Agricultural Revolution’ which immediately preceded the

Industrial Revolution.8 The Industrial Revolution was thought to have started

quite abruptly in 1780 following a sudden upturn in agricultural output after

1740.9

But the timing of the Agricultural Revolution has become rather

controversial in the last thirty years and has generated a large and contentious

literature. Kerridge was the first dissenter and argued that the important changes

in farming technique occurred in the sixteenth and seventeenth centuries.10 He

was supported in this by Jones and Allen, the latter of whom has argued

strongly in favour of a ‘Yeoman's Revolution’ in the seventeenth century.11

Counter-revisionism has been spear-headed by Overton, who locates the

Agricultural Revolution firmly in the late eighteenth century.12 The desire to

move the Agricultural Revolution in time has necessitated changing the

emphasis on various innovations as engines of growth. Eighteenth century

enclosures, turnips and seed drills were once in favour13; whereas sixteenth

century water meadows and seventeenth century seed varieties have more

recently come into fashion.14

The purpose of this paper is to show that the Agricultural Revolution

debate, in its current incarnation, has a fatal weakness. Quite simply, it will

never be possible to resolve the debate using the type of evidence being

7 For Example, Rochefoucauld F de La (Lord), A Frenchman in England (Paris, 1784; translated by Roberts S C and reprinted in Cambridge, 1933); Young A, A Six Month’s Tour through the Eastern Counties of England (London, 1771). 8 Toynbee A, Lectures on the Industrial Revolution in England (London, 1884); Prothero R E (Lord Ernle), English Farming Past and Present (London, 1912). 9 Deane P and W A Cole, British Economic Growth, 1688-1959 (Cambridge, 1962), 78. 10 Kerridge E, The Agricultural Revolution (London, 1967), 15. 11 Jones E L, ‘Agriculture, 1700-80,’ in Floud R and D McCloskey (eds), The Economic History of Britain since 1700, vol. 1 (1st edition, Cambridge, 1981), 66-86; Allen R C, Enclosure and the Yeoman (Oxford, 1992), 13-21. 12 Overton M, ‘Re-establishing the Agricultural Revolution,’ Agricultural History Review, vol. 44 (1996), 1-20. 13 For example, Chambers J D and G E Mingay, The Agricultural Revolution, 1750-1880 (London, 1966), 54. 14 For example, Allen R C, ‘The Two English Agricultural Revolutions, 1450-1850,’ in B M S Campbell and M Overton (eds), Land, Labour and Livestock (Manchester, 1991), 236-254.

8

produced by the protagonists. Understandably, historians have collected data

which are relatively abundant and relatively easy to interpret (such as output

estimates). But we will show this is neither necessary nor sufficient to establish

the timing and causes of the Agricultural Revolution. Economic historians have

been led astray partly by their failure to define the Agricultural Revolution in

precise economic terms. We start our analysis in the next section by proposing a

clear definition which encompasses the concepts of Agricultural Revolution

embedded (explicitly or implicitly) in the work of the leading researchers in the

area. We then show in section III that the output estimates and price series

currently on offer are neither necessary nor sufficient to establish the timing of

the Agricultural Revolution. In section IV we show that estimates of Total

Factor Productivity (TFP) are both necessary and sufficient – but we cannot

generate reliable empirical estimates using Allen’s real rent technique. And in

section V we discuss what type of evidence will be required to resolve the

debate.

II. Defining the Agricultural Revolution. We have already noted that there

are a number of perspectives on the Agricultural Revolution and these

perspectives stress different aspects of agricultural development. In

consequence, each perspective generates a different description of the timing

and causes of the Agricultural Revolution. Of course, we are trying to

adjudicate between these perspectives in terms of historical relevance rather

than rule out any of them on a priori grounds. So we would like a definition

which encompasses as many of these aspects as possible – it would be pointless

to define the term so tightly that only one characterisation could possibly be

correct. On the other hand, we need a definition which is sufficiently simple and

precise that it is possible in practice to measure changes according to that

definition. It would be no use having a list of ten criteria which were difficult to

measure, such as the rate of institutional change or the level of factor mobility.

We suggest that the Agricultural Revolution be defined as an

exceptionally rapid increase in TFP, which moves outwards the supply curve

for agricultural products and generates higher output and lower prices. Most

researchers regard increases in output and productivity as the key characteristics

9

of the Agricultural Revolution. This is reflected in recent comments by both

Overton and Allen.15 Notice in Figure 1 below that if the demand and supply

curves have the normal shape, then the only way that output and productivity

can rise at the same time is if the supply curve shifts outwards (S to S1). Then

output rises (Q to Q*) and price falls (P to P*): productivity has risen owing to

the shift in the supply curve. By contrast, an outward shift in the demand curve

(D to D1) would lead to an increase in output (Q to Q’) and also an increase in

price (P to P’); productivity has then fallen owing to the existence of

diminishing returns. So an outward shift in the supply curve is the essential

ingredient in the Agricultural Revolution because only then do output and

productivity rise at the same time.

Figure 1. Raising Ouput and Productivity Together.

Notice that the supply curve can shift outwards for two reasons. First,

there could be a fall in one or all of the input prices (land, labour or capital).

Then farmers would be willing to supply more output at any given price

because their costs would have fallen. However, this mechanism for shifting the

supply curve is not particularly interesting for several reasons. Notably, a

change in factor prices could be a temporary phenomenon and so the

agricultural supply curve could easily shift back to its old position - a kind of

Agricultural Counter-Revolution. This kind of cyclical behaviour of factor

prices undoubtedly happened at various times in the early modern period. In

15 Overton M, Agricultural Revolution in England (Cambridge, 1996), 7; Allen R C, ‘Tracking the Agricultural Revolution,’ Economic History Review, vol. 52 (1999), 209.

01 Quantity

Price

D1

D

S1

S

P

Q

P'

Q'

P*

Q*

10

particular, an exogenous increase in population (say, owing to an improvement

in the disease environment) would lead to an increase in the price of output and

a fall in the price of labour. This generates an outward shift in the agricultural

supply curve. Moreover, in a Malthusian model this is partially off-set by a

reduction in food consumption per head and a consequent downward pressure

on the population, thus shifting the supply curve inwards again.16 However, it

seems unlikely that researchers have this kind of mechanism in mind when they

talk about the ‘Agricultural Revolution’. An exogenous fall in factor prices

simply represents more of the same kind of production, rather than something

qualitatively different; and a change in factor prices can be easily reversed.

So we would prefer to have a definition of ‘Agricultural Revolution’

which represents something permanent and demonstrates an important break

with the past. It is for that reason that we define the Agricultural Revolution as

an increase in TFP. Increasing TFP is the other way in which the supply curve

can shift outwards. This shift will generally be permanent because it results

from an increase in the stock of knowledge; and it will result in a new pattern of

production (be it a change in the pattern of outputs or inputs). Hence the new

manner of farming will be qualitatitvely different to the old manner of farming.

This definition makes sense on a number of other grounds. First, it can

encompass all the perspectives which have been proposed on the Agricultural

Revolution. Various agricultural developments have been claimed as more

‘revolutionary’ than others. We can easily assess these claims by considering

how each innovation raises TFP. The size of the increase will be a function of

how much more productive is the new technique than the one which it replaces;

and how widely the technique is adopted. This is a good measure of overall

historical importance. Social and institutional changes fit easily into this

framework. For example, we can consider the effect of enclosure on TFP and

then chart the increase in enclosure over time. Second, our supply curve

definition is closely linked to economic growth - which is a matter of central

16 The Malthusian model of population is complicated because it represents both a supply shock and a demand shock. In general, we can say that the supply shock would be smaller than the demand shock (owing to diminishing returns in agriculture); so eventually the food constraint would bind and off-set some of the exogenous increase in population. Notice, however, that the new low-disease equilibrium could still have more workers than the old equilibrium and hence the supply curve could see a permanent shift to the right.

11

importance in understanding economic development in general and the

Industrial Revolution in particular. An increase in TFP reflects an outward shift

in the Production Possibility Frontier for agricultural products.

We have shown already that most researchers in the Agricultural

Revolution debate have been using the term in a way which is consistent with

our TFP and supply curve definition. We are simply making their concepts

explicit rather than implicit. The main dissenter from this view would be

Grantham and it is important to see how he differs from the others.17 Grantham

views the Agricultural Revolution largely as an intensification in production in

response to population pressure. Recall that rising population creates an

outward shift in the demand curve. Grantham argues that there was very little

change in the knowledge set of farmers between mediaeval times and 1850. For

example, farmers knew very early on that marling would raise grain yields and

total output. Sometimes in early societies it was economic to produce high

yields and output (when population pressure was high) and sometimes it was

not economic to do so (when population pressure was low). So population

growth shifted the demand curve outwards and led farmers to change their

choice of technique in favour of more intensive production. Grantham would

argue that we simply observed this phenomenon on an unprecedented scale in

England in the seventeenth and eighteenth centuries owing to unprecedented

population growth.

Now, there is an important distinction to be made between the choice of

technique on the one hand, and technological change on the other. First, let us

consider the choice of technique. At a given point in time, farmers have a

number of techniques available to them; and generally we assume that they

choose the technique which minimises the cost of production, given the

prevailing prices of factor inputs (land, labour and capital). Therefore, a change

in relative prices leads to a change in the choice of technique. In Grantham’s

economy, population pressure reduces the real wage and leads to intensification

in production through the employment of more workers (in weeding, ploughing,

drainage and so on). Now let us consider technological change. Economists

17 The clearest exposition of Grantham’s view was posted on the World Wide Web in November 1998. It can be found in the archives of eh.res at: www.eh.net/

12

define technological change as the addition of a new technique to the existing

set of available techniques. If the new technique is economically viable, then it

will lead to a change in the choice of technique even in the absence of a change

in relative prices. This will result in an increase in TFP.

Grantham’s concept of the Agricultural Revolution is rather

unsatisfactory for three reasons.

First, technological change creates economic growth by pushing out the

Production Possibility Frontier. But a simple change in the choice of technique

has no such effect in a neo-classical framework. Grantham’s demand curve

concept of the Agricultural Revolution is therefore unsatisfactory in the sense

that it is not linked to economic growth. This makes it seem rather less

interesting than the TFP definition which we use in this paper. There are

actually two ways in which Grantham’s demand shock can generate a

permanent increase in output and productivity, although he has not explicitly

endorsed either of them. First, the economy could be in a Keynesian

equilibrium characterised by under-employment of resources – that is, operating

within the PPF. Then a demand shock could be validated by an increase in

supply and the economy would move closer to the PPF. This is not a

particularly attractive model because it is not obvious that the economy was

under-employed in the seventeenth and eighteenth centuries and there is no

evidence that markets were failing to clear – which is a prerequisite for

operating within the PPF. Second, we might believe in Boserupian induced

technological change – whereby a demand shock can create its own supply

through technological innovation. Again, this hypothesis has not proved

particularly popular amongst researchers of the Agricultural Revolution. An

obvious question arises as to why demand pressure induced technological

change in England and nowhere else (such as France).

The second problem with Grantham’s definition of the Agricultural

Revolution is that it suffers from reversability. Whereas an increase in TFP will

be permanent, Grantham’s Agricultural Revolution could disappear in response

to a fall in demand. Moreover, this reduction in demand might occur in

response to any number of shocks to the economy - population decrease,

increased foreign competition, a change in taxes, a change in tastes, and so on.

13

Third, Grantham’s demand curve definition throws the responsibility for

both the Agricultural Revolution and economic growth during the Industrial

Revolution wholly onto the industrial sector. A static supply curve implies zero

economic growth in agriculture. Therefore Grantham would have to propose a

very high rate of industrial productivity growth to generate an Industrial

Revolution; and he would have to postulate a large differential between

agricultural and industrial growth rates (since he has just revised downwards the

agricultural growth rate). This can be shown as follows. First, suppose that there

is an increase in agricultural demand through an increase in population. The

increase in output cannot continue unchecked in the face of diminishing returns

in agriculture - because wages will be driven down to subsistence levels (or

lower) in a Malthusian trap. A continuation of the Agricultural Revolution

(intensified production) requires an increase in consumer purchasing power.

But this increase in purchasing power cannot come from the agricultural sector

because there are diminishing returns and there has been no economic growth.

So the industrial sector must be growing very fast in order to drive agricultural

change forward. Second, the absence of productivity growth in agriculture

would drive down the overall rate of productivity growth during the Industrial

Revolution because agriculture was the largest sector. This necessitates a

revision of our growth estimates during the Industrial Revolution in one (or

both) of two ways. Either we need to revise downward further our estimate of

overall economic growth in the Industrial Revolution. Or we need to revise

upwards the rate of productivity growth in the industrial sector. Revising

upwards the rate of industrial growth seems to be the most attractive option, in

the light of the required differential between agriculture and industry. Current

estimates of overall productivity growth and sectoral growth rates in the

Industrial Revolution do not lend much support to this characterisation of

events.

We have now defined the Agricultural Revolution simply and precisely

as an increase in TFP, which is consistent with the usage of most researchers

(particularly Allen and Overton). The next step is to consider how the empirical

evidence can guide us in determining when and why the agricultural supply

curve shifted outwards. We take the recent article by Allen as the basis for our

14

analysis. This is not because our comments are more relevant to Allen than

anyone else in the debate – it is simply because he has offered the most recent

and articulate summary of the current state of knowledge. In the next section we

show that on the basis of aggregate output estimates it is impossible to reach

any sound conclusions regarding the timing of the Agricultural Revolution.

Aggregate output data are neither necessary nor sufficient to resolve the debate.

We show that even if we grant all of Allen’s points (for the sake of argument)

then Overton’s characterisation of the Agricultural Revolution could still be

correct.

III. Demand Equations and Output Estimates. It is obviously important to

estimate the level and growth of aggregate agricultural output. These figures

help to guide our interpretation of agricultural development and feed into our

estimates of national income. In fact, it was in order to refine our estimates of

national income that Crafts introduced the use of demand equations into the

analysis of English agriculture.18 Instead of assuming constant per capita food

consumption (as in the ‘population method’), Crafts simulated changes in per

capita consumption on the basis of prices and incomes. This technique was

ideal for Crafts’ analysis for three reasons. It harnessed much more information

(i.e. wages and prices); the information was relatively easy to come by; and

Crafts could avoid an explicit examination of agricultural development.

Unfortunately, the third of these qualities also makes the demand

equation approach very little use for an analysis of the Agricultural Revolution.

Demand equations are not a magic solution for generating information – we can

only get out as much information as we put into them. Since we put into them

very little information about agricultural production, it is scarcely surprising

that we get very little useful information back out. Researchers have been led

astray because they have not realised that it is historically more interesting to

describe (quantitatively) how production occurred rather than how much

production occurred. We show in this section that aggregate output estimates

18 Crafts N F R, ‘English Economic Growth in the Eighteenth Century: a Re-examination of Deane and Cole’s Estimates,’ Economic History Review, vol. 29 (1976), 226-35.

15

are neither necessary nor sufficient to establish the timing of the Agricultural

Revolution.

It is clearly not necessary to observe aggregate output in order to

estimate changes in TFP. If we can show instead that a given level of output can

be produced with fewer inputs than previously, then we can measure the

increase in TFP. We could make this calculation with a sample of farms and we

do not actually have to build up national output estimates. We could be forgiven

for concentrating on aggregate output estimates if they were sufficient to

estimate changes in TFP (rather like using a sledgehammer to crack a nut).

Unfortunately, aggregate output estimates are not sufficient. We need to chart

changes in the supply curve due to increases in TFP. But in fact, it is impossible

to quantify changes in the supply curve from documenting changes in the

demand curve. (And even if we could show that the supply curve had shifted,

we would still need to rule out changing factor prices as a cause). This

shortcoming is demonstrated clearly in Figure 2 below.

Suppose for the sake of argument that we knew the position of the

demand curve. We know the market price of agricultural goods and therefore

we can read off the quantity demanded (since we must be somewhere on the

demand curve). At price P1770 in Figure 2 the quantity demanded in 1770 will

be Q1770. Since the quantities supplied and demanded must be equal in

equilibrium, we know that the quantity supplied in 1770 must also be Q1770. If

several conditions are met (such as zero storage from year to year) then output

in 1770 must also be Q1770. So we have now pinpointed one particular spot on

the supply curve for 1770 (marked by the black dot). Now suppose that we

know the position of the demand curve and the market price in every year.19

Then we can estimate output in each year and we generate a graph showing

hundreds of dots, as shown in Figure 2 below.

19 For the period after 1770 we can actually make this calculation using Feinstein’s annual real wage data, provided that certain assumptions are met. See Feinstein C H, ‘Pessimism Perpetuated: Real Wages and the Standard of Living in Britain during and after the Industrial Revolution,’ Journal of Economic History, vol. 58 (1998), 625-58. For the period before 1770 we have to use estimates of per capita national income which are averaged over much longer periods.

16

Figure 2. Output Estimates from Demand Equations.

We must then ask whether we can estimate the supply curve from this

information. The answer is no. In econometric terms, the supply curve is still

not identified. The reason for this is simple. When we observe a move from one

level of output to another, we cannot tell to what extent this is a shift along the

supply curve and to what extent there has been a shift in the supply curve. We

can see this problem in Figure 3 below.

Figure 3. Two Possibilities for Changing Equilibrium Output.

In Figure 3, we observe a movement in the demand curve (D to D1) and

an increase in output (Q to Q1). But this could arise from a move along a

01 Quantity

Price

D

Q1770

P1770

01 Quantity

Price

D1

D

Q Q1

S(flat)S(steep)1

S(steep)2

17

stationary flat supply curve S(flat); or it could arise from a large shift in a steep

supply curve, S(steep)1 to S(steep)2. So if we can information only on the

demand curve, then there is no way that we can distinguish these two cases.

The problem is very clear if we write down the equations for a simple

model of the demand and supply for agricultural goods. The most recent

formulation of the demand curve is presented by Allen.20 He takes a standard

demand function:

Qd = aPeIgMbN

(1)

where Qd is the quantity demanded, a is a constant, P is the market price, e is

the price elasticity of demand, I is per capita income, g is the income elasticity

of demand, M is the price of other goods, b is the cross-price elasticity of

demand for agricultural and other goods, and N is population. All these

variables are in real terms (i.e. deflated by the price index). Then on the basis of

theory and empirical evidence, Allen imposes values on e, g and b. With data

on P, I, M and N, he can then easily calculate Qd.

We could analogously write a simple supply curve:

QS = TPε (2)

where QS is the quantity supplied, T is the level of technology, P is the market

price of output, and ε is the elasticity of supply. Taking natural logarithms

would allow us to express the supply function in a linear form with could be

estimated using regression analysis:

lnQS = lnT + ε.lnP + u (3)

The problem depicted in Figure 2 above is immediately apparent in

equation (3). We have data on P and an estimate of Q from the demand equation

- but we do not know the elasticity of supply, ε, or the level of technology, T.

We have two unknowns and only one equation, so it is impossible to estimate

the values of either of the unknown terms. Notice that in principle we could

rearrange equation (3) to find T.

lnT = lnQ - ε.lnP – u (4)

We have annual data on P and annual estimates of Q. So if we knew the value

of the elasticity of supply, ε, then we could easily produce annual estimates of

18

technology, T. We could then chart the rate of technological progress and the

movement of the supply curve every year and the debate about the Agricultural

Revolution would be over. Putting a value on the elasticity of supply, ε, is the

crux of the problem.

Three potential solutions present themselves at this point. By far the

easiest solution is simply to assume a value for ε. Unfortunately, we have very

little idea what a plausible value might be. The problem is compounded by the

fact that ε is unlikely to be stable over time, especially the long period which

Allen is analysing (three hundred and fifty years). For example, as land

resources become more scarce we might suppose that it will become

progressively more difficult to increase output in response to a rise in price (that

is, the elasticity of supply, ε, will fall).

A more attractive solution would be to assume that T is constant over

short periods (say, thirty years). Then we could estimate ε using the annual data

on output and prices. If our estimates of ε were sufficiently stable, then we

could set a value for ε for longer periods (say, 120 years) and calculate annual

values of T. Unfortunately, this is much more difficult than it appears at first

sight because the production function is rather more complicated than the

simple case which we have so far considered. In particular, agricultural supply

is subject to both annual weather shocks and climatic change.21 So our output

estimate does not enable us to observe the supply curve perfectly – it is

contaminated by a great deal of ‘noise’. The supply curve actually looks more

like this:

lnQ = lnT + ε.lnP + φln.W + u (6)

where W is weather. This is problematic for two reasons. First, remember that

technological change is very gradual (TFP growth of 1 per cent per annum

would be very fast for the eighteenth century). But annual weather shocks are

very large (maybe 40 per cent per annum). So we are trying to measure small

changes in the underlying supply curve whilst observing it through very large

20 Allen R C, ‘Tracking,’ 212. 21 For example, if the index of wheat production were one on average, then it would typically fluctuate between 0.6 and 1.4 on an annual basis. See Brunt L, ‘Estimating English Wheat Production in the Industrial Revolution,’ University of Oxford Discussion Paper in Economic and Social History, no. 29 (June 1999).

19

transient shocks. Even if we controlled quite effectively for weather shocks, it

would be difficult to generate an accurate estimate of ε over a period as short as

thirty years. Notice further that there could be long run changes in the English

climate which move the supply curve outwards.22 This would look like

technological progress because it shifts outward the supply curve but it is really

an unmeasured input (more land).

The third solution to the identification problem is to introduce another

equation into the system. Two equations would enable us to solve out the two

unknown terms, ε and T. The obvious step is to model technology explicitly.

Although modelling the level and change in technology would be difficult, it

would also reveal a great deal of new information. We would not be limited to

saying when T changed; we would also be able to say why T changed. We

would thus be able to offer a much richer interpretation of historical events.

This approach has been noticeably absent in the literature on the Agricultural

Revolution.

So aggregate output estimates have substantial limitations for historical

analysis. We must ask whether this has any real impact on our interpretation of

history. The answer is yes. In Figure 4 below we have plotted the major output

movements highlighted by Allen – the large increase in output and stagnant

prices in the early period, versus the small increase in output and rising prices in

the later period.23 We can see that the large increase in output in the early

period (1520 to 1739) is consistent with a small shift in an elastic supply curve.

The supply curve is elastic because there are still unexploited land resources

which are gradually being brought into production. Then in the later period

(1739 to 1800) the small increase in output is consistent with a large shift in an

inelastic supply curve. That part of the supply curve is inelastic because most

land resources have now been brought into production and the remaining acres

are of low quality. Figure 4 shows that even if we accept all of Allen’s

estimates, it might be the case that the later period saw a rapid increase in TFP

and shift in the supply curve (and hence an Agricultural Revolution). Overton

22 In particular, the mean winter temperature has been getting steadily and significantly warmer in England since records began in 1659. See Manley G, ‘Central England Temperatures: Monthly Means 1659 to 1973,’ Quarterly Journal of the Royal Meteorological Society, vol. 100 (1974), 389-405.

20

contends that the Agricultural Revolution occurred after 1750. So Allen might

have the correct data - but Overton might still be drawing the correct

conclusion.

Figure 4. Major Output Movements, 1520-1800.

The shortcomings of aggregate output data compel us to look at other

data series in order to establish the timing and causes of the Agricultural

Revolution. We will see in the next section that TFP estimates based on real

rent are the crux of Allen’s argument.

IV. Real Rent and Total Factor Productivity. Total Factor Productivity is the

ratio of outputs to inputs. A rise in TFP indicates that we can now produce more

output for a given level of inputs (or analogously, that we can produce the same

output as before using a lower level of inputs). So the crucial step from output

data to an estimate of TFP is the introduction of input data.

The drawback with TFP is that it is very data-intensive. To calculate

TFP at one particular point in time, we need data on all the inputs and outputs

for the each production unit (be it a farm, a county or a country). We then need

to replicate this calculation many times over to generate a time series for TFP –

which is essential if we are to pinpoint the Agricultural Revolution with any

degree of accuracy. Fortunately, if certain conditions are met then we can take a

short-cut by looking at real rent. Economic rent is the return on land generated

23 Allen R C, ‘Tracking,’ 215-6.

01

Quantity

Price

S1520

S1739

S1800

D1800

D1739D1520

21

by ‘that portion of the produce of the earth which is paid to the landlord for the

use of the original and indestructible productive powers of the soil’.24 If land is

in limited supply and there is perfect competition amongst farmers to rent land,

then the landowner can capture all the economic rent. An increase in TFP will

increase the economic rent because farmers will compete away all the extra

profit as higher rent to the landowner. Of course, over time the nominal

economic rent will change in order to keep pace with changes in the price level.

But if we deflate the nominal rent, then the resulting changes in real rent should

reflect changes in TFP. Notice that we do not need explicit estimates of inputs

and output in order to make this calculation – these are already netted out for us

in the rent. Allen produces a real rent series for the period 1540 to 1850 and

proposes to use it as an index of TFP.25 The problem is that real rent will only

reflect TFP accurately if certain conditions are met. The question then arises –

are all the relevant conditions met? The answer is no.

Allen graphs his time series of the level of real rent (his Figures 3 to 6,

pp218-20; they are labelled as TFP). The graphs demonstrate clearly that the

link between real rent and TFP is broken. Figure 3 shows a prolonged and

substantial decline in TFP between 1600 and 1650; Figure 6 shows an even

more impressive decline between 1750 and 1800. We normally think of TFP

growth as an addition to the stock of knowledge – a new ability to combine

inputs in a more efficient way. It is therefore difficult to see how TFP can

decline because a drop in TFP would imply that the body of knowledge open to

English farmers had shrunk. Farmers must have been very forgetful indeed for

TFP to have declined so far and so fast in the seventeenth and eighteenth

centuries. Productivity analyses for modern economies rarely find a decline in

TFP and most economists regard negative TFP as some kind of measurement

error. For example, firms may horde labour during a recession in order to keep

their skilled workers for the up-turn: the temporary decline in output makes TFP

growth look negative because there is no off-setting decline in inputs. This is

known as Okun’s Law. However, it is hard to believe that Okun’s Law explains

Allen’s prolonged negative productivity growth.

24 Allen R C, Enclosure and the Yeoman (Oxford, 1992), 174 (quoting Ricardo).

22

In fact, there are two lines of enquiry which seem more likely than

Okun’s Law to explain the apparent decline in TFP. Either there is a problem

with Allen’s data series; or there has been a violation of the conditions which

are required for real rent to be interpreted correctly as TFP. Let us consider each

of these in turn.

Allen appears to have a solid data series with 1543 real rent

observations for the period 1500-1850, which he presents on an annual basis as

a five-year rolling average. However, closer inspection reveals that this is no

normal time series. The observations are heavily clustered in a small number of

years, as demonstrated in Figure 5 below.

Figure 5. The Number of Rent Observations per annum, 1500-1850.

In fact, 1132 observations (73 per cent of the sample) are found in 17

years; and the remaining 411 observations are spread across the other 334 years

of Allen’s period. So effectively we have 17 benchmarks observed with a

reasonable degree of statistical confidence, and the rest is just interpolation

based on a few observations. So we have very little idea about how the real rent

series was moving between the benchmarks and the series is very erratic

because it is based on so few observations. The real rent series only appears

smooth in the graphs because Allen has taken a five year moving average. This

means that the massive impact of each benchmark is felt for nine years. The

interpolation problem becomes quite severe when we break down the real rent

25 Allen R C, ‘Tracking the Agricultural Revolution,’ Economic History Review, vol. 52 (1999),

0

25

50

75

100

125

150

175

200

1 51 101 151 201 251 301 351Year

Num

ber o

f Obs

erva

tions

23

observations into different categories, as we will need to do below (enclosed

versus open, arable versus pastoral). Then, outside the benchmark years, we

average only one observation of each type of rent every three years.

There are two further problems with Allen’s sampling procedure. First,

it is frequently the case that a large number of observations from a particular

year are drawn from a small number of villages. For example, the 27

observations for 1542 are drawn from eight villages – and 19 of those

observations come from just four villages. The greatly reduces our statistical

confidence in the estimate of real rent because the sample is not completely

random.26 Second, we are trying to judge the increase in average real rent over

time (in order to draw conclusions about TFP). But as the average real rent

rises, more marginal land will be brought into production. The marginal land

will earn a lower real rent. This will exert downward pressure on the average

real rent and therefore the average real rent series will not rise as fast as TFP. In

fact, the faster the increase in TFP the more marginal land will be brought into

production because it will now be economic to do so. So the increase in average

real rent over time will systematically underestimate the rise in TFP. The best

solution to this problem is to collect rent observations on only those plots of

land which were in production at the beginning of the survey period. Since

Allen does not follow this sampling strategy, we need instead to correct for the

likely downward bias.

So we do not really have an annual series of real rents: in fact, we would

need twenty times the number of observations and a more uniform distribution

to generate a reliable annual series. The econometrically sound way to proceed

209-35. 26 This is known as the Kish Design Effect and requires a corrected confidence interval. The Kish Design Effect arises because we might expect observations drawn from the same sub-sample to share certain characteristics. For example, suppose that we wanted to estimate the average height of the adult male population and we measured 27 people. We would obviously have lower confidence in the result if 19 of our observations were drawn from four families – because there are genetic forces leading to similarity of heights within each family. In the same way, we might wonder if the particular economic characteristics of a village were causing all the rent observations to be correlated. For a more detailed discussion, see Deaton A, ‘Data and Econometric Tools for Development Analysis,’ in Behrman J and T N Srinivasan (eds) The Handbook of Development Economics, vol. 3 (Amsterdam, 1995), 1797.

24

in this situation is to use the clustered observations as benchmarks and discard

the other observations.27 This is the strategy which we pursue here.

Let us turn to the conditions which need to be met in order to interpret

the changes in real rent as changes in TFP. There are (at least) four conditions

which were seriously violated in the period of Allen’s study.

First, recall that TFP is reflected in the economic rent, which is ‘paid to

the landlord for the use of the original and indestructible productive powers of

the soil’. We therefore need to net out the part of the real rent which is actually

a return on the landowner’s investment in physical capital (or we need to

assume that the investment is constant over time). It seems likely that

landowner investment rose and fell over time in response to changing interest

rates, but Allen makes no estimate or adjustment for these fluctuations.

Second, we need to control for economies of scale. When Allen first

adopted real rent as a measure of TFP, he noted explicitly that it was predicated

on the existence of constant returns to scale.28 Otherwise an increase in output

would cause an increase in productivity even though there was no technological

change. But Allen has argued more recently that there were important

economies of scale in agricultural production.29 He makes no adjustment for

this effect, despite the fact that it is normal to do so in productivity analysis.

Notice that Allen finds different economies of scale on arable and pastoral land

– this is not surprising since arable and pastoral production are based on

different technologies. This means that we have to separate arable and pastoral

observations in order to control for each type of economy of scale. Allen also

finds different economies of scale on open and enclosed land.

Third, Allen notes that inputs and outputs should be adjusted for taxes

when calculating TFP. He made this adjustment in his earlier work using a cross

section of farms.30 But remarkably, he makes no adjustment in his time series

real rent analysis. This is likely to be an important omission over the period

27 For another example of this approach, see Brunt L and E Cannon, ‘A Grain of Truth in Medieval Interest Rates? Re-examining the McCloskey-Nash Hypothesis,’ Bristol University Discussion Paper in Economics, No. 98/462 (February 1999). 28 Allen R C, ‘Efficiency and Distributional Consequences of Eighteenth Century Enclosures,’ Economic Journal, vol. 92 (1982), 937-53. 29 Allen R C, Enclosure and the Yeoman (Oxford, 1992), 211-217. 30 Allen R C, Enclosure and the Yeoman, (Oxford, 1992), 176.

25

1500-1850 when there were substantial changes in the taxes levied on land

through the Land Tax, the Poor Rate and the Corn Laws.

Fourth, Allen argues that enclosure caused rents to rise through a

redistribution of income from the tenant to the landowner.31 This is a serious

problem for two reasons. First, note that the enclosure of land would raise the

real rent but would not represent an increase in TFP. It is therefore not

legitimate to pool open and enclosed observations because the estimated

average rent would be affected by the mixture of open and enclosed

observations. If the mixture changed over time then the real rent series would

fluctuate independently of changes in TFP (the ‘composition effect’). Second,

note that if Allen’s characterisation of enclosure is correct then open field

farmers must somehow have been extracting a wedge from their landlord

(otherwise rents could not have risen after enclosure without making the

farmers bankrupt).32 We do not know if this wedge was constant over time. For

example, if population pressure increased the competition amongst open field

farmers to rent land then we might suppose that the wedge would fluctuate in

response to population pressure. The population rose and fell repeatedly

between 1500 and 1850, so the size of the wedge may also have changed.

It is not enough simply to criticise Allen’s assumptions. We need to

consider how the violations would effect his historical interpretation. Hence we

set out to adjust his real rent series and turn it into a TFP series, in order that we

could quantify the impact of the violated assumptions. The following strategy

recommended itself. First, decompose Allen’s data set into the necessary four

categories of land (open arable, enclosed arable, open pastoral and enclosed

pastoral). Second, adjust the average real rent in each category for scale

economies, taxes and marginal land. Third, aggregate the series using constant

proportions of open and enclosed land in order to eliminate the enclosure

composition effect.

31 This is a central plank of Allen’s Yeoman Revolution – enclosure was unnecessary and possibly pernicious. See Allen R C, ‘Efficiency.’ 32 It is not clear how open field farmers were able to withhold some of the value of farm land from the landlord - Allen has never proposed a specific explanation. Presumably there would have to be some kind of imperfect competition amongst open field farmers which prevented them competing away all the agricultural rents to the landlord.

26

Unfortunately, this strategy immediately ran into serious problems. In

the period 1624 to 1806 almost all the arable observations are drawn from open

land: so even for benchmark years it was impossible to get an estimate of

average real rent on enclosed arable land. Remember that in order to control for

economies of scale we need to treat the four categories separately (open and

enclosed arable, open and enclosed pastoral). Then to control for the

composition effect we need to re-aggregate the data in fixed proportions

throughout the period of study. So unless we placed a zero weight on enclosed

arable land, we could not re-aggregate the data over the period 1624 to 1806

(because we have no adjusted real rent figure for enclosed arable land before

1806). The reverse problem occurred after 1806 (i.e. virtually all the arable

observations pertain to enclosed arable land and there are not enough

observations of open arable land). We were confronted by a similar problem for

pasture land. In the period 1727 to 1844 almost all the pastoral observations are

drawn from enclosed land and it is impossible to generate an estimate for

pastoral rent on open land.33

In the face of this problem, we confine ourselves to presenting a series

for Open Arable Real Rents for the period 1624 to 1806 (Figure 6 below); and a

series for Enclosed Pastoral Real Rents for the period 1727 to 1844 (Figure 7

below). We then make the following comparisons and adjustments.

First, we demonstrate the composition effect by comparing our Open

Arable Real Rent series (OA RR) to Allen’s pooled All Arable Real Rent series

(AllA RR) in Figure 6; and comparing our Enclosed Pastoral Real Rent series

(EP RR) to Allen’s pooled All Pastoral Real Rent series (AllP RR) in Figure 7.

33 To get a reliable estimate of (say) enclosed arable rents, we required at least 20 observations of rent on enclosed arable land in each benchmark year. Before 1806 there were no benchmark years with 20 enclosed arable observations - in fact, the best we can manage is 1804 (ten observations), 1727 (eight observations from three villages) and 1791(six observations).

27

Notice that the upturn in the Open Arable series occurs later than in

Pooled Arable series. This is because Allen’s sample switches from being

almost all low rent open fields in 1806 to almost all high rent enclosed fields in

1811. According to Allen, this increase in real rent should not be interpreted as

a rise in TFP. Similarly, the Pooled Pastoral series grows faster up to 1770 than

the Enclosed Pastoral series – because after 1770 there are no more low rent

open pasture observations. Again, we should not confuse this pastoral

composition effect with TFP growth, which is what we do in Allen’s Pooled

Pasture Series.

Figure 6. Arable Real Rent, 1624-1806.

Figure 7. Pastoral Real Rent, 1727-1844.

0.44

0.94

1.44

1624 1661 1698 1735 1772 1809Year

TFP

AllA RROA RROASA RROASTA RROASTMA RR

1

1.5

2

2.5

3

3.5

1727 1750 1773 1796 1819 1842Year

TFP

AllP RREP RREPSA RREPSTA RREPSTMA RR

28

Second, Allen argues that the rate of TFP growth was the same on open

and enclosed land.34 For the period 1500-1739, Allen demonstrates the

similarity by graphing the pooled series (his Figure 3, p218) and the open field

series (his Figure 4, p219). The graphs are almost identical. However, it should

be noted that 75 per cent of Allen’s observations in period 1500-1739 come

from open fields. If we exclude the year 1727, then 81 per cent of Allen’s

observations come from open fields. So the reason that Allen’s Figures 3 and 4

look so much alike is that Figure 4 is simply Figure 3 reproduced with a few

missing observations. In order to examine the hypothesis that TFP growth is the

same on open and enclosed land we need a large sample of both types. In the

case of arable land, the only years for which this is possible are 1806 and 1811.

A comparison of real rents for open and enclosed arable over that period shows

that real rents on enclosed land were rising twice as fast as real rents on open

land. Of course, there is no reason to suppose that technological progress would

be the same on open and enclosed land – in fact, there is substantial historical

literature suggesting the exact opposite. The discrepancy in real rent behaviour

on open and enclosed land suggests that we should be sure to control for this

effect carefully in our overall real rent series.

Third, we then further adjusted the Open Arable series and Enclosed

Pasture series for economies of scale. We took Allen’s estimates of the size

distribution of farms and the effect of scale economies.35 This enabled us to

estimate the overall productivity effect of changing economies of scale. Again,

this is demonstrated above in Figure 6 (OASA RR) and Figure 7 (EPSA RR).

The trend towards land concentration over the period 1500 to 1850 implies a

downward revision to rate of productivity growth, particularly towards the end

of the period (when Allen finds a sharp increase in the level of TFP). It also

eliminates the apparent TFP growth on enclosed pasture land in the early

eighteenth century, during Allen’s ‘Yeoman Revolution’.

34 Allen, ‘Tracking,’ 219. 35 Allen R C, Enclosure and the Yeoman (Oxford, 1992). For the distribution of farm sizes, see p73; Allen gives data for 1600, 1700 and 1800 and we supplemented this with data for 1850 from Grigg D, ‘Farm Size in England and Wales, from Early Victorian Times to the Present,’ Agricultural History Review, vol. 35 (1987), 179-90. For the effect of economies of scale on Ricardian surplus per acre, see Allen p214.

29

Fourth, we made a further adjustment for the growing burden of the

Poor Rate on landowners. The poor rate was levied on each acre of land (in

accordance with the rateable value and the local burden of poor relief). In order

to make a proper adjustment for the changing tax burden, we would need to

know the rate levied in each village in Allen’s sample at the time the rent was

observed. This could potentially have a very large effect on the rent series

because the local variation in the cost of poor relief was large (so a high rent

might be due entirely to the high local cost of poor relief). We have avoided the

task of detailed adjustment here because it would obviously be a very onerous

undertaking. In fact, for many of the individual rent observations it could be

impossible because the relevant information may not exist in the local archives.

So instead we have focussed on the change in the average burden over time.

We estimated the average annual burden by taking the national expenditure on

poor relief in the benchmark years and dividing that sum by the number of acres

in cultivation.36 We then added that sum to the nominal rent. Remember that the

land owner is absorbing the poor rate – so an increase in the poor rate leads to a

lower nominal rent. Hence we need to add the tax to the nominal rent. The main

effect of this calculation is to revise upwards the estimate of real rent around

1800, when the poor rate was most burdensome. See Figure 6 (OASTA RR) and

Figure 7 (EPSTA RR).

Fifth, we revised the real rent series in order to control for changes in

the average quality of arable and pastoral land (OASTMA RR and EPSTMA

RR respectively). Without a full production model it is difficult to know by how

much the output was lower on marginal land. We know that in the case of wheat

production, yields on marginal land were around one third lower.37 So in the

absence of more precise data, we simply assume that output per acre was one

third lower on both marginal arable land and marginal pasture land. We can see

in Figure 6 above that the effect of marginal land on average real arable rent

was quite modest; this is because the proportion of arable land going into and

36 For poor relief expenditure, see Slack P, The English Poor Law (Basingstoke, 1990), 30; and Fowle T W, The Poor Law (London, 1881), 156. Acres in cultivation are taken from Brunt L, ‘Estimating English Wheat Production in the Industrial Revolution,’ University of Oxford Discussion Paper in Economic and Social History, no. 29 (June 1999). 37 Brunt L, ‘Estimating English Wheat Production in the Industrial Revolution,’ University of Oxford Discussion Paper in Economic and Social History, no. 29 (June 1999), 16.

30

out of production was quite small. By contrast, we can see in Figure 7 that the

effect of marginal land on average real pastoral rent was very large. This is not

surprising because, according to Allen’s data, the acreage of pasture land

increased from 10 million in 1700 to 17.5 million in 1800.38 In fact, if output

were only 20 per cent lower on marginal pasture land (rather than the 33 per

cent which we are currently assuming), then this would be sufficient to

completely eliminate the suspicious decline in real pastoral rents which Allen

finds for the late eighteenth century.

The overall effect of these revisions is to make the Open Arable and

Enclosed Pastoral real rent series noticably flatter, and to eliminate almost all

the puzzling drops in the pasture series. It is clear that these adjusted real rent

series cannot yet be interpreted as indices of TFP because there are still violent

fluctuations and substantial declines in the series, particularly in the case of

arable rents. It is likely that the extreme peaks and troughs are caused by the

lack of data: if we had more observations then we would have more precise

estimates of average real rent and the series would probably be smoother. It is

also likely that correcting for the violated assumptions would eliminate more of

the long upswings and downswings in the series. Of course, the crucial question

is - what upswings would remain and when would they be most rapid? The

substantial revisions proposed thus far (inadequate as they clearly are) suggest

that it is difficult to form a confident opinion about TFP on the basis of the

current real rent evidence. For example, if we postulate that output on marginal

pasture land was 20 per cent lower then we are left which an upturn in real rents

starting in 1806 (supporting Allen’s argument). But if we postulate that output

on marginal pasture land was 50 per cent lower then the upturn in real rents

starts around 1776 (supporting Overton’s argument). Clearly more precision is

required before we can reach any firm conclusions using real rents. A fact

which gives this interpretation added bite is that rents are likely to reflect

lagged productivity change. The qualitative evidence suggests that landlords

and farmers were constantly surprised in the eighteenth century by the growth

of prices, ouput and productivity. If leases were renewed every 20 years, then

38 Allen R C, ‘Agriculture during the Industrial Revolution,’ in Floud R and D N McCloskey (eds), The Economic History of Britain since 1700, vol. 1 (Cambridge, 1994), 96-123.

31

the rent increase between contracts reflects the increase in TFP which occurred

over the last 20 years. So in line with Overton’s argument, Allen’s evidence

could imply that the increase in pastoral TFP started in 1756.

Again it is important to ask whether these revisions could lead to any

change in our interpretation of history. Let us return to the debate between

Allen and Overton. We showed in the previous section that Allen’s estimates of

output growth are consistent with Overton’s interpretation of a large shift in the

supply curve in the late eighteenth century. Now we can further show that

Allen’s estimates of output and real rent growth are together consistent with

Overton’s interpretation of a large shift in the supply curve in the late

eighteenth century.

Recall that Allen finds that there was more rapid growth in real rents in

the early period than in the late eighteenth century. On the one hand, there are

several reasons to suppose that the rise in real rents was unrelated to

productivity (the switch from open to enclosed land, the growing tax burden). In

that case, Figure 4 in the previous section already captures all the essential

features of Allen’s evidence. On the other hand, supposed that there was some

increase in productivity in the early period as Allen suggest. Then it is still

possible that there was no shift in the supply curve in the early period because it

could have been due to economies of scale. This is a genuine increase in

productivity which would push up real rent (although not TFP). If the supply

curve were very elastic in the early period then the existence of economies of

scale could imply a downward sloping supply curve over some range of output.

This is reflected in Figure 8 below. The increase in demand between 1520 and

1739 could create not only higher output but also higher rents through

productivity increases. So Figure 8 can capture the finer points in Allen’s

description of events. But Figure 8 also shows that the supply curve could have

shifted much more dramatically between 1739 and 1800 than in the preceding

230 years. So Overton could still be correct in placing the Agricultural

Revolution at the end of the eighteenth century. We simply do not have enough

data to discriminate between the contrasting perspectives of Allen and Overton.

Despite the research effort of the last 30 years, we are not yet in a position to

pinpoint the timing of the Agricultural Revolution within a period of 350 years.

32

Figure 8. Major Output and Productivity Movements, 1520-1800.

V. Still Seeking the Agricultural Revolution. The uncertainty over the timing

of the Agricultural Revolution cannot be resolved using the data which has so

far been presented in the debate. We have discussed in some detail the

limitations of the two main types of evidence (output estimates and real rents).

The other types of evidence are no more conclusive. For example, grain yields

are often taken as a indicator of productivity growth but the evidence is open to

many interpretations. In particular, crop yield is a choice variable which farmers

adjust in response to the prevailing input and output prices (as well as the state

of technology). One of Grantham’s insights is that a rise in crop yields does not

necessarily represent a shift in the supply curve – it could just be a best

response to a shift in the demand curve.

If we want to ascertain what happened to the supply-side of the

agricultural sector then it behoves us to study the supply-side. The application

of modern quantitative techniques has been notably lacking in that area. A

careful reconstruction of the production function for agricultural goods would

achieve much more than pinpointing an increase in TFP and hence the

Agricultural Revolution. It would enable us to quantify the role of different

techniques and we could then extend the analysis by modelling the invention

and adoption of new technology. We could then postulate counterfactuals

examining how British agricultural development might have been different and

01 Quantity

PriceD1520 D1739

D1800

S1520

S1739

S1800

33

why it differed from other European economies. These are more fundamental

and interesting issues than simply dating the Agricultural Revolution. We

should not let an obsession with the Agricultural Revolution deflect us from a

more complete understanding of the sources of economic growth in developing

economies.