A comparative analysis of Innovation and Productivity: lessons learned and work ahead Chiara Criscuolo Centre for Economics Performance London School of Economics
Dec 18, 2015
A comparative analysis of Innovation and Productivity:
lessons learned and work ahead
Chiara CriscuoloCentre for Economics Performance
London School of Economics
This is a joint effort with ...1. Australia: David (ABS)2. Austria: Martin Berger3. Belgium: Jeoffrey Malek4. Brazil: Bruno Araújo and João De Negri5. Canada: Petr Hanel and Pierre Therrien6. Denmark: Carter Bloch and Ebbe Graversen7. Finland: Mariagrazia Squicciarini Olavi Lehtoranta Mervi Niemi8. France: Stephane Robin and Jacques Mairesse9. Germany: Bettina Peters10. Italy: Francesco Crespi Mario Denni Rinaldo Evangelista and Mario Pianta11. Japan: Tomohiro Ijichi (could not participate for data problems)12. Korea: Seok-Hyeon Kim13. Luxembourg: Anna-Leena Asikainen14. Netherlands: George van Leeuwen, Pierre Mohnen, Michael Polder, Wladimir Raymond15. New Zealand: Richard Fabling 16. Norway: Svein Olav Nås and Mark Knell 17. Sweden: Hans Loof18. Switzerland: Spyros Arvanitis19. UK: Chiara Criscuolo
A big THANK YOU to all!
The challenge
• Comparative analysis of the innovation productivity link using a “structural” model
• Same model across countries: same variables; same estimation method
• So that estimates are “comparable” across countries
• Tools: Innovation Surveys Estimation routines
The difficulties
• Countries data cannot be pooled so each country must estimate model separately (in general no one outside country team can see the data)
• Not all countries could be at all meetings• Innovation surveys similar (CIS “harmonized”)
but not exactly the same across countries both in terms of variables used and presence of “filter” questions
HIGH COORDINATIONS COSTS!!!!!!
Solutions• Use the same model across countries:
– Minimum common denominator set of variables– Use a model that control for selection for all
countries
• “Centralize” implementation of estimation routines:– Lead-country decides model– Modify specific country model so that it is
estimable in all countries (variables included – see also IPR topic; flexibility in using different variable names)
Questions • Was the centralized model the right one?• Should we have used a bottom up approach?• Should have more countries be involved with
the formulation of the model?• How useful was it to have the programming
routine?…Should we change the approach for the follow-
up project?
A brief outline of the model followed…
…To estimate the effects of innovation on productivity controlling for selection and endogeneity
Following the Crepon-Duguet Mairesse “tradition” we estimate a 3 stage/4 equation model:
• 1st innovation equation• 2nd innovation input equation• 3rd innovation output equation• 4th productivity equation
The model• 3 stage with 4 equations • 1st stage explains firms’ decision whether to engage in innovation
activities or not and the decision on the amount of innovation expenditureProb(innovation=1)=f(size; Group; Foreign Market, Obstacles to innovation due to knowledge; costs and market; industry dummies)Ln(innovation expenditure per employee)=f(Group; Foreign Market; Cooperation; Financial Support; industry dummies)
• In the 2nd stage we estimate the knowledge production function where innovative sales depends on investment in innovation. Ln(innovative sales per employee)=f(Innovation expenditure; Size; Group; process innovation; Cooperation with clients; suppliers; other private and public agents; industry dummies; Mills ratio [to correct for selection])
• The 3rd stage we estimate the innovation output productivity link using an augmented Cobb-Douglas production function using IV.Ln(sales per employee)=f(Size; Group; Process Innovation; log innovative sales per employee; industry dummies [Human Capital and Physical Capital])
Obstacles to innovation8.2 During the three years 2002 to 2004, how important were the following
factors for hampering your innovation activities or projects or influencing a decision not to innovate?
Degree of importance
High Medium Low Factor not
experienced
Lack of funds within your enterprise or group Lack of finance from sources outside your enterprise
Cost factors
Innovation costs too high
Lack of qualified personnel Lack of information on technology Lack of information on markets
Knowledge factors
Difficulty in finding cooperation partners for innovation
Market dominated by established enterprises Market factors Uncertain demand for innovative goods or services
Cooperation partnersPlease indicate the type of co-operation partner and location (Tick all that apply)
CUSTOMERS [Your
country] Other
Europe* United States
All other countries
C. Clients or customers
SUPPLIERS [Your
country] Other
Europe* United States
All other countries
B. Suppliers of equipment, materials, components, or software
PUBLIC [Your
country] Other
Europe* United States
All other countries
F. Universities or other higher education institutions
G. Government or public research institutes
OTHER PRIVATE [Your
country] Other
Europe* United States
All other countries
D. Competitors or other enterprises in your sector
E. Consultants, commercial labs, or private R&D institutes
Extensions/variations of the model ...
• Some countries could add Human Capital (H); Physical Capital (K) and Materials (M) in the productivity equation:– Austria (H,K); Belgium (H,K); Brazil (H,K,M); Canada (H,K); Finland
(H,K,M); Germany (H,K,M); New Zealand (K,M); UK (H)
• Sales per employee is a very rough measure of productivity. Ideally we would want value added per employee (Labour Productivity measure) or Multi Factor Productivity measures
• To do this innovation surveys must be combined with other production datasets. (follow-up of current project?)
…Extensions/variations of the model ...
• Most countries estimate separately for small vs large firms and manufacturing vs services firms (excl. Italy and Norway); Korea and Canada only manuf; Luxembourg serv and small.
• Important to look at differences • Standard size threshold 250 employees but
with some variation• Issue for small countries and/or small surveys
problem of small sample sizes
…Extensions/variations of the model…• Switzerland estimated a slightly different version of
the model (e.g. no group variable; cooperation with foreign counterparts)
• Germany/Netherlands suggested a modification of the model to deal with endogeneity
• Canada: could only estimate on manufacturing and weighted regressions and no information on obstacles to innovation
• Austria: estimate it on CIS3 rather than CIS4• Australia: no information on foreign market; inputed
group information and 2005 rather than 2004 • New Zealand: again differences in variable definitions
…Extensions/variations of the model…• The original model only deals with selection and
endogeneity of innovation sales eq. In the productivity equation but we also wanted to deal with endogenity of innovation expenditure equation in the innovative sales equation:
Ln(innov. sales per employee)=f(Inn. Exp.; Size; Group; process innovation; Cooperation with clients; suppliers; other private and public agents; industry dummies; inverse Mills ratio)
• Option A: use predicted innovation expenditure rather than actual innovation expenditure and bootstrap standard errors
• Option B: instrument innovation expenditure in innovative sales equation (options suggested by Germany/Netherlands)
…Extensions/variations of the model• In current version we use:
– “strict” definition of innovation: firms are innovative if innov. Exp.>0 and innov. Sales>0
– log innovative sales per employee
• Alternative version: – “wide” definition of innovation: firms are innovative
if innov. Exp.>0 and innov. sales>=0– Share of innovative sales: use TOBIT rather than OLS
for estimating innovative sales eq.Tested using data for Canada; UK; Denmark
RESULTS
• Simple “non-structural model”– Probit Innovation– OLS : innovation expenditure eq.– OLS: productivity equation
• “structural” model:– Heckman– Innovative sales eq. – Productivity eq.
The results: Innovation equationINNOV=1 LEMP GP FOR_MKT HAKNOW HAMARKET HACOST Observations
Australia 0.0496*** 0.123*** 0.0771*** 0.0688*** 0.116*** 3697Austria 0.0895*** 0.0790** 0.159*** -0.0202 -0.0678 -0.0016 1001Belgium 0.0761*** 0.0650*** 0.168*** 0.0221 -0.0112 0.148*** 2700Brazil 0.0605*** 0.0746*** -0.0811*** 0.0877*** 0.0922*** 0.0226 9384Canada 0.0431*** -0.0353 0.117*** 5355Denmark 0.0846*** 0.0688** 0.211*** 0.0903** 0.00484 0.158*** 1729Finland 0.0842*** 0.0272 0.176*** 0.0736** 0.0869*** -0.00336 2155France 0.0506*** 0.0663*** 0.217*** 0.0594*** 0.0196** 0.0704*** 18056Germany 0.0314*** 0.0609*** 0.198*** 0.00514 -0.0462* 0.0698*** 3258Italy 0.0396*** 0.0573*** 0.123*** 0.0303*** -0.0159** 0.0233*** 15915Korea 0.07*** -0.021 0.0767** 0.012 0.078** 1332Luxembourg 0.0884*** 0.100** 0.114** 0.0772 -0.0373 0.150** 545Netherlands 0.0464*** 0.0433*** 0.133*** 0.0678*** -0.0275** 0.00427 6858New Zealand 0.0208*** 0.0426** 0.129*** 0.0412** -0.0141 0.0701*** 3426Norway 0.0979*** -0.0137 0.231*** 0.156*** 0.0272 0.140*** 1869Switzerland -0.022*** 2050UK 0.0116*** 0.0593*** 0.152*** 0.0973*** 0.0282** 0.0363** 11166
The results: innovation investment equation
ln(Innovation Expenditure per
employee) Group Foreign Market Cooperation Financial Support Observations
Australia 0.423** -0.225 -0.006 1098Austria 0.0798 0.611*** 0.262* 0.798*** 471Belgium 0.0957 0.468*** -0.0225 0.733*** 1055Brazil 0.284*** 0.146*** 0.846*** 0.802*** 9384
Canada 0.124** 0.0960 0.182*** 0.189*** 3488Denmark 0.146 0.342** 0.261** 0.648*** 810Finland -0.0142 0.00129 0.379*** 0.523*** 986France -0.107** 0.478*** 0.397*** 0.760*** 6270
Germany 0.0499 0.556*** 0.384*** 0.574*** 2082Italy 0.0122 0.159*** 0.307*** 0.436*** 5158
Korea -0.22 0.222** 0.498*** 842Luxembourg 0.214 0.447** 0.0399 0.464** 278Netherlands -0.143** 0.288*** 0.304*** 0.491*** 2100New Zealand 0.524*** 0.329*** 0.295*** Confidential 2082
Norway -0.103 0.325*** 0.324*** 0.622*** 919Switzerland 0.081*
Results: the productivity equation
Productvity LEMP Group PROCESS
Ln(innovation Sales per employee) Obs
Australia 0.120*** 0.148** -0.0308 0.248*** 1204Austria 0.132*** 0.381*** 0.107* 0.0670*** 381Belgium 0.00827 0.324*** -0.133*** 0.401*** 861Brazil 0.220*** 0.642*** -0.0652*** 0.130*** 9384Canada 0.0642*** 0.303*** -0.0413 0.219*** 3629Denmark 0.0558*** 0.170*** -0.0913* 0.302*** 701Finland 0.0484** 0.296*** -0.0336 0.223*** 815France 0.0247*** 0.258*** -0.126*** 0.349*** 5050Germany 0.0839*** 0.124*** -0.0646** 0.332*** 1617Italy 0.0429*** 0.242*** -0.0143 0.249*** 2991Korea 0.117*** 0.192*** -0.02 0.476*** 697Luxembourg 0.0518 0.265*** -0.214** 0.375*** 225Netherlands 0.0837*** 0.135*** -0.0447 0.335*** 1680New Zealand 0.0622*** 0.228*** -0.114*** 0.534*** 1041Norway 0.0714*** 0.247*** -0.0794 0.371*** 740Switzerland 0.089*** -0.009 0.353*** 398UK 0.0594*** 0.233*** -0.0992*** 0.453*** 3354
Controlling for Selection: innovation equation (Heckman selection eq.)
Selection Eq. GP FOR_MKT LEMP HAKNOW HAMARKET HACOST rho* Observations P-value LR test
Austria 0.213* 0.454*** 0.253*** -0.0765 -0.182 -0.00122 0.223 1001 0.226Belgium 0.198*** 0.617*** 0.267*** 0.0427 -0.0500 0.455*** 0.41 2695 0.0012Brazil 0.424*** -0.264*** 0.123*** 0.152*** 0.131*** 0.0320 2.019*** 9384 0.000
Canada -0.105* 0.290*** 0.140*** 1.005*** 5355 0.000Denmark 0.186** 0.637*** 0.253*** 0.243** 0.0288 0.391*** 0.324** 1729 0.0202Finland 0.0649 0.532*** 0.254*** 0.190** 0.259*** -0.0266 0.477*** 2155 0.00178France 0.227*** 0.778*** 0.204*** 0.201*** 0.0678*** 0.227*** 0.643*** 18056 0.000
Germany 0.144*** 0.529*** 0.0884*** 0.0144 -0.107 0.173*** 0.256** 3242 0.0656Italy 0.203*** 0.478*** 0.185*** 0.110*** -0.0680** 0.0908*** 0.753*** 15915 0.000
Luxembourg 0.267* 0.314** 0.248*** 0.191 -0.101 0.359* 0.192 545 0.701Netherlands 0.164*** 0.546*** 0.213*** 0.175*** -0.111** 0.0123 0.727*** 6858 0.000New Zealand 0.113** 0.349*** 0.0785*** 0.0892* 0.0270 0.138*** 1.337*** 3426 0.000
Norway -0.0724 0.643*** 0.320*** 0.301*** 0.0478 0.301*** 0.739*** 1852 0.000United Kingdom 0.174*** 0.464*** 0.0468*** 0.287*** 0.0883** 0.0883** 0.189 11162 0.261
Heckman outcome equation: innovation expenditure eq.
Outcome eq. GP FOR_MKT COOP FINSUP Observations
Austria 0.161 0.737*** 0.408*** 0.746*** 1001Belgium 0.233* 0.524*** -0.0205 0.714*** 2695
Brazil 0.875*** -0.204* 0.384*** 0.332*** 9384Canada 0.145* 0.448*** 0.173** 0.183* 5355
Denmark 0.477*** 0.762*** 0.182 0.735*** 1729Finland 0.260** 0.361* 0.495*** 0.460*** 2155France 0.231*** 1.158*** 0.427*** 0.683*** 18056
Germany 0.0538 0.610*** 0.402*** 0.469*** 3242Italy 0.268*** 0.511*** 0.310*** 0.412*** 15915
Luxembourg 0.212 0.434 0.102 0.352 545Netherlands 0.247*** 0.675*** 0.389*** 0.569*** 6858New Zealand 0.664*** 0.740*** 0.225*** Confidential 3426
Norway -0.0436 0.706*** 0.354*** 0.657*** 1852United Kingdom 0.0508 0.513*** 0.377*** 0.537*** 11162
Careful: group and foreign market are not marginal effects
Innovation Sales eq.GP LEMP PROCESS MILLSstrict COOP_client COOP_supplier COOP_private COOP_public Obs
Austria 0.32 0.0996 0.32 0.629 -0.142 0.0573 0.0828 -0.533 359Belgium 0.244 0.0432 0.0374 -0.0760 0.230* -0.150 0.307** -0.273** 411Brazil 0.748*** 0.136*** 0.0580 0.747** -0.0878 0.164 0.0340 0.313*** 1954
Canada 0.0717 0.0214 0.263*** 0.244 -0.0676 -0.0250 0.281** -0.145 2273Denmark 0.631*** -0.101* 0.0522 -0.0849 0.184 -0.111 -0.0260 0.0547 584Finland 0.264* -0.172* 0.297** -0.632 -0.189 0.00830 0.250 -0.0291 698France 0.552*** -0.0117 0.155** 0.0785 0.116* 0.0437 0.147** -0.0437 2511
Germany 0.0723 0.0367 0.0593 -0.469** 0.00700 0.127 -0.0507 -0.0601 1390Italy 0.0851 -0.0378 0.528*** -0.443 0.041 0.109 0.197* 0.0659 747
Luxembourg 0.183 -0.284** 0.509*** -1.130* 0.144 -0.127 0.0246 0.224 207Netherlands 0.320*** -0.0301 -0.0125 -0.186 0.0481 -0.000654 -0.0957 0.0889 1374New Zealand 0.300*** -0.0885*** 0.000126 0.548* -0.278*** 0.190* -0.103 -0.0765 993
Norway 0.239** 0.169** 0.147 0.219 -0.336* 0.304 0.0609 -0.00662 672United Kingdom 0.225*** -0.0349* 0.125** -0.163 0.0138 0.0660 0.00828 -0.00909 2989
LRTOTPE LRTOTPEstrict_hat
Austria 0.0501 0.768**Belgium 0.274*** 0.425**
Brazil 0.354*** -0.111Canada 0.367*** -0.0422
Denmark 0.141*** 0.297*Finland 0.218*** 0.495France 0.214*** 0.312***
Germany 0.340*** 0.112Italy 0.210*** 0.0689
Luxembourg 0.211*** 0.475Netherlands 0.260*** 0.267**New Zealand 0.516*** 0.573***
Norway 0.471*** 0.121United Kingdom 0.263*** 0.0217
Productivity equation
GP LEMP PROCESS MILLSstrict LISPE Lispehat Observations
Austria 0.137 -0.0236 -0.0103 -1.077** 0.401 0.0577 359Belgium 0.289*** -0.0235 -0.125* -0.240 0.467*** 0.374*** 411Brazil 0.183** 0.140*** -0.211*** -0.315 0.647*** 0.207** 1954
Canada 0.250*** 0.0772** -0.122** -0.00113 0.436*** 0.183 2273Denmark 0.186** 0.0732*** -0.0405 0.146 0.345*** 0.296* 584Finland 0.244*** 0.0859** -0.0677 0.135 0.314*** -0.198 698France 0.232*** 0.0536*** -0.129*** 0.089 0.474*** 0.226** 2511
Germany 0.0838** 0.0625*** -0.116*** -0.259** 0.500*** 0.185 1390Italy 0.093 0.00391 -0.192** -0.312* 0.485*** 0.0669 747
Luxembourg 0.434*** 0.0349 -0.142 0.170 0.226* 0.742*** 207Netherlands 0.0219 0.0902*** -0.0440 -0.0765 0.409*** 0.0746 1374New Zealand 0.128** 0.0662*** -0.135*** -0.200 0.682*** 0.835*** 993
Norway 0.256*** 0.0407 -0.0716 -0.168 0.344*** 0.396** 672United Kingdom 0.150*** 0.0580*** -0.121*** -0.272*** 0.550*** 0.116 2989
Summary of results
• When significant, coefficients are surprisingly similar• Serving a foreign market; being large and being part
of a group are generally associated with higher probability of being innovative and financial support and cooperation activity with higher investment in innovation
• Using a selection model is appropriate for most countries (exc. Austria; Luxembourg and UK)
• In the innovation sales eq. the elasticity of innovative sales to innovation expenditure is mostly between 0.2 -0.35
• In productivity eq. 0.3-0.6
Some counterintuitive results
• Obstacles to innovation have mostly positive coefficients. More innovative firms try harder and therefore find more obstacles?
• Process innovation is mostly negative in productivity equation. Measurement issues? Adjustment costs? (Possibly future work?)
Issues with the model
• Use of log innovative sales per employee• Use of detailed cooperation variables in
innovative sales eq. And cooperation in innovation exp. Eq.
• Use of process in both innovative sales eq. And productivity eq.
• Perhaps a simpler model? E.g. only cooperation in innovative sales eq. And process in productivity eq.?
Lessons learned and steps aheadWhat have we learned?• 18 countries!• Interesting results• High coordination costs/model more suited for some
countriesWhat next?• Match CIS with production data both for better
productivity measure and longitudinal dimension?• Alternative “organisational/coordination” model?