1 Online Appendix for Networks and Innovation: Accounting for Structural and Institutional Sources of Recombination in Brokerage Triads Sarath Balachandran Exequiel Hernandez This appendix presents a descriptive table and some additional analyses mentioned in the main paper. Table A1 displays the distribution of countries across the different clusters, used for the analyses presented in tables 3 and 4 of the paper. Table A2 presents the results for two of the robustness tests described in the paper. Models A1 A4 of this paper show the results for the analyses with the largest firms dropped and models A5 A8 present the placebo tests where we compute the network measures using non-r&d alliances. Table A3 shows results from the seemingly unrelated estimation where we examine whether firms differentially select into different triads over time based on the technological diversity of their innovation activities. Table A4 shows the results from analyses where we interact the global spread of a firm s inventors (inventor dispersion) with the triad count variables. Some additional robustness checks are also described below. Additional Robustness Tests We mentioned in the paper that the counts of domestic, mixed, and foreign triads exhibit high correlations with each other. This could lead to multicollinearity. Given the very large sample size, there are sufficient observations in which pairs of triad types do not coincide so as to reduce collinearity problems that would arise in smaller samples. To verify this, we attempted to ensure that the coefficients were stable in magnitude, direction, and significance. We introduced each triad count variable individually into the model and found that the magnitude and significance of the coefficients did not vary substantially with the addition of the others. We also verified that none of these coefficients are particularly sensitive to the addition or removal of control variables. Further, we operationalized our independent variables in different ways to reduce the level of correlation between them, such as normalizing the triad counts by the total number of ties possessed by the firm or using non-logged
2 versions of the variables. Using these alternative measures, the coefficients displayed similar relative magnitudes and we found similar support for the hypotheses compared to the main results. Our study is built on the notion that knowledge is more novel or distinct across than within institutional boundaries. This has been suggested in many studies already cited, but we attempted to validate this assumption in our data. To do so, we calculated the average technological distance of each firm to (a) every other firm from its own country and (b) every other firm from other countries. We expected the latter to be statistically larger than the former. We made the calculation separately for two different ways of measuring technological distance patent classes and disease areas. There are seven broad patent classes representing biotechnology (based on Rothaermel & Hess, 2007) and 38 disease areas (e.g. cancer vs. diabetes) in which biotechnology firms can perform R&D (per Recap). We calculated the Euclidean distance between each pair of firms in the sample along each vector of patent classes and disease areas (separately). The results of t-tests reveal that technological differences to firms from the same country are statistically smaller than distances to other countries (p < 0.001 in both cases, two-tailed). There is an important caveat to this test. Quantifiable differences in patents or technological areas are only one aspect of the types of knowledge differences we theorize about. There are also tacit, unseen differences in knowledge having to do with mental models, ways of doing science, or approaches to problem solving that make a significant (and probably more important) difference to innovation volume and radicalness. For instance, two countries may share similar profiles in terms of the patenting or disease categories in which biotechnology firms are conducting R&D, but tacit cultural or legal differences lead the firms across countries to go about conceptualizing and solving problems in those observable domains in very different ways. Indeed, this hard-to-measure tacitness is the gist of the knowledge-based view of the firm (Kogut and Zander 1992, 1993) and the work on knowledge differences across institutional boundaries (Nelson 1993, Vasudeva et al. 2013).
3 TableA1: Country Clusters based on Cultural, Administrative, Geographic and Economic distance Cultural Administrative Geographic Economic Country Cluster Country Cluster Country Cluster Country Cluster Canada 1 United States 1 Canada 1 Ireland 1 New Zealand 1 Canada 2 United States 1 New Zealand 1 Australia 1 Israel 2 Denmark 2 Spain 1 United States 1 Italy 3 Germany 2 Israel 1 Ireland 1 Ireland 3 Croatia 2 Croatia 1 United Kingdom 1 Japan 3 Poland 2 Germany 2 Switzerland 2 Germany 3 Czech Republic 2 United States 2 Germany 2 Austria 4 Slovak Republic 2 Russian Federation 3 Japan 3 Thailand 4 Austria 2 Poland 3 Norway 4 United Kingdom 4 Hungary 2 Venezuela 3 Finland 4 Mexico 5 Italy 2 Argentina 4 Latvia 4 Bahamas 6 Belgium 3 Canada 5 Netherlands 4 Finland 7 France 3 Italy 5 Belgium 5 Chile 7 United Kingdom 3 India 6 Poland 5 Australia 7 Switzerland 3 China 6 France 5 Ukraine 8 Netherlands 3 Australia 7 Austria 6 Spain 8 Japan 4 Netherlands 8 Sweden 7 Netherlands 8 Bahamas 5 Austria 8 Denmark 7 France 9 Australia 6 Sweden 8 China 8 Belgium 10 China 7 Belgium 8 Hong Kong 8 Poland 11 Finland 8 Czech Republic 9 Italy 9 Denmark 11 Latvia 8 South Africa 9 Hungary 9 Hong Kong 11 Norway 8 Slovak Republic 9 Taiwan 10 China 12 Sweden 8 Hungary 9 South Korea 10 Sweden 13 Ireland 9 Switzerland 10 Israel 11 Iceland 13 South Korea 10 Cyprus 11 Singapore 12 South Korea 14 Cyprus 11 Greece 11 Argentina 13 South Africa 14 Israel 11 Singapore 12 Czech Republic 13 Switzerland 14 Taiwan 12 Hong Kong 12 Spain 13 Norway 15 Singapore 13 United Kingdom 13 Croatia 14 Taiwan 16 Spain 14 France 14 Russian Federation 14 Singapore 17 Portugal 14 Denmark 15 Romania 14 Argentina 18 Argentina 15 Norway 15 Portugal 15 Russian Federation 19 Russian Federation 16 Romania 16 India 16 Portugal 21 South Africa 17 Latvia 16 Venezuela 17 New Zealand 22 New Zealand 18 Finland 17 Greece 18 India 19 Portugal 18 Slovak Republic 19 Venezuela 20 Japan 19 Romania 21 Greece 21 Countries with the equal cluster numbers occupy the same cluster Cultural clustering based on Kogut and Singh's (1988) composite index of Hofstede (1980) cultural dimensions Institutional clustering based on World Bank governance indicators (Control of Corruption, Rule of Law, Voice and Accountability, Regulatory Quality, Political stability and absence of violence, Governance quality) Geographic clustering based on great circle distance between national capitals Economic clustering based on per capita Gross National Income (GNI)
4 Table A2: Robustness Checks In models A1 A4 the top 10% of firms in terms of patents over the study period are dropped from the analysis. Models A5 - A8 replicate the analysis using non- R&D alliances instead of R&D alliances for the same firms as in the original sample. Heteroskedasticity robust standard errors clustered by firm in parentheses. ^ - Logged Variable. # - p values from one tailed Wald test. + < 0.1, *<0.05, **<0.01, ***<0.001 Largest Firms Dropped Placebo Test Model A1 Model A2 Model A3 Model A4 Model A5 Model A6 Model A7 Model A8 Dependent Variable Volume Volume Radicalness Radicalness Volume Volume Radicalness Radicalness Domestic ^ 0.1003** 0.1396*** 0.0345 0.0914 0.0409 0.0668-0.0617 0.0650 (0.0361) (0.0415) (0.0792) (0.0892) (0.0384) (0.0410) (0.1231) (0.0877) Mixed ^ -0.0319-0.1520* -0.2006-0.2860+ -0.0105-0.0389-0.0457 0.0392 (0.0434) (0.0745) (0.1505) (0.1734) (0.0585) (0.0512) (0.1089) (0.1274) Foreign ^ 0.0659* 0.0466 0.1936** 0.2954* 0.0562 0.0488 0.1244 0.0578 (0.0327) (0.0449) (0.0742) (0.1189) (0.0430) (0.0326) (0.0810) (0.0888) Network Efficiency 0.3843+ 1.2648 0.8714 4.1816-0.0489-0.2174-0.3901 0.1864 (0.2205) (1.7410) (0.5302) (3.0397) (0.1819) (0.1996) (0.7217) (0.5463) Network Centrality -0.3990-3.4691* -3.8359-10.7611* -0.4650 0.9762 0.0196-8.5086 (0.9285) (1.7233) (2.3534) (4.8868) (0.7604) (1.0385) (2.8954) (5.7486) Closed Domestic ^ -0.0259 0.3140+ 0.0199 0.5268* 0.0950-0.0719 0.0722 0.0490 (0.0721) (0.1692) (0.2079) (0.2626) (0.0641) (0.0752) (0.2703) (0.3112) Closed Mixed ^ 0.0590 0.4594* 0.4246+ 0.1920 0.0010-0.0394 0.2831 0.4240+ (0.0671) (0.1861) (0.2460) (0.2923) (0.0957) (0.1129) (0.2084) (0.2553) Closed Foreign ^ 0.0529 0.0977 0.1631 0.1183-0.0302 0.0160 0.1586 0.2919 (0.0929) (0.1453) (0.1310) (0.3089) (0.0893) (0.0892) (0.2011) (0.2834) Within Triad Technological 0.0157 0.0418 0.0483 0.0397 0.0313 0.0325 0.0114-0.0105 Distance (0.0245) (0.0306) (0.0901) (0.0695) (0.0250) (0.0277) (0.0691) (0.0773) Triad Portfolio Balance 0.0476 0.2387* -0.2909 0.6086* 0.0120 0.0420-0.2015-0.3547 (0.0943) (0.1176) (0.2684) (0.2643) (0.0984) (0.0907) (0.2895) (0.2191) Technological Base ^ -0.0554-0.1348** -0.1157-0.0253 0.0883* -0.0399-0.1294+ 0.1191 (0.0439) (0.0442) (0.0947) (0.2516) (0.0389) (0.0461) (0.0754) (0.2491) Partner Technological -0.0065 0.0026 0.0251 0.0458 0.0031 0.0027-0.0331-0.0090 Base ^ (0.0104) (0.0099) (0.0342) (0.0308) (0.0119) (0.0123) (0.0550) (0.0330) Avg Technological Distance 0.1018* 0.0173 0.2577 0.2783 0.0501-0.0256-0.0377 0.0591 to Partners (0.0399) (0.0427) (0.1602) (0.1759) (0.0335) (0.0354) (0.1100) (0.1201) No of firms from same -0.1524-0.2112* -0.5172* -0.4191 0.0060 0.0184-0.0319 0.0112 country ^ (0.1135) (0.1067) (0.2522) (0.4816) (0.0199) (0.0198) (0.0471) (0.0639) Percentage Foreign 0.0827-0.0838-0.0147-0.2732 0.0481 0.0609 0.0606-0.0401 Partners (0.0669) (0.2005) (0.1960) (0.5137) (0.0605) (0.0553) (0.1695) (0.1705) More than two ties 0.0279-0.0631-0.1485-0.4045* -0.0735-0.0854+ 0.1380 0.0572 (0.0499) (0.0515) (0.2052) (0.2020) (0.0505) (0.0518) (0.1805) (0.2060) Firm country IP protection 0.1404 0.3158 1.0858* 0.1323-0.0018 0.1825 0.6327+ -0.0776 (0.1682) (0.2208) (0.4834) (0.8151) (0.1523) (0.2227) (0.3799) (0.6799) Avg Partner country IP 0.0408-0.0137 0.1198-0.4674 0.0025 0.0123 0.1206-0.0695 protection (0.0998) (0.1253) (0.4586) (0.6609) (0.0414) (0.0443) (0.1955) (0.1351) Average Within triad 0.0458-0.0050 0.3201* 0.3495* 0.0581-0.1768-0.4132-0.3709 prior ties (0.0900) (0.0745) (0.1583) (0.1660) (0.1169) (0.1266) (0.3157) (0.2687) Firm Fixed Effects Y Y Y Y Y Y Y Y Year Fixed Effects Y Y Y Y Y Y Y Y Strata Fixed Effects Y Y Y Y Num of Obs. 10,039 6,841 10,039 6,841 11,993 7,995 11,993 7,995 R Squared 0.2067 0.0305 0.1138 0.0085 0.2166 0.0091 0.1197 0.0064 AIC 13939.9 7668.4 41262.4 26463.6 18667.4 9667 49604.9 30667 Log - Likelihood -6930.9-2390.2-20592.2-11787.8-9294.7-3389.5-24763.4-13889.5 P (Domestic = Mixed)# 0.007 0.112 0.002 0.046 0.264 0.469 0.089 0.445 P (Domestic = Foreign)# 0.221 0.061 0.028 0.044 0.394 0.126 0.366 0.475 P (Mixed=Foreign)# 0.039 0.020 0.029 0.011 0.230 0.104 0.097 0.462
5 Table A3: Effect of Technological Diversity on Counts of Triad Configurations Results based on seemingly unrelated estimation. All covariates from the main models included in the estimation. Standard errors in parentheses. Technological diversity is a Herfindahl-based characterization of the extent to which a firm s R&D is spread out across different patent classes or disease areas, respectively. + < 0.1, *<0.05, **<0.01, ***<0.001 Domestic Foreign Mixed Coef. SE Coef. SE Coef. SE Technological Diversity 0.0348 (0.0444) 0.0492 (0.0503) -0.0551 (0.0331) + (Patent Classes) Tech. Diversity n.s. across Domestic, Foreign (p = 0.82) Technological Diversity 0.0698 (0.0336) * 0.1055 (0.0345) *** -0.0142 (0.0272) (Disease Areas) Tech. Diversity n.s. across Domestic, Foreign (p = 0.43) Table A4: Effect of Inventor Dispersion Inventor dispersion is a Herfindahl based characterization of the extent to which a firm s inventors are spread out across different countries. The higher the value, the more spread out the firm s inventors. Heteroskedasticity robust standard errors clustered by firm in parentheses. ^ - Logged Variable. + < 0.1, *<0.05, **<0.01, ***<0.001 Model A9 Model A10 Model A11 Model A12 Dependent Variable Volume Volume Radicalness Radicalness Domestic ^ 0.0672* 0.0876* 0.0531 0.0772 (0.0319) (0.0410) (0.0820) (0.0986) Mixed ^ -0.0062-0.0005-0.2565-0.2924 (0.0445) (0.0661) (0.1586) (0.1935) Foreign ^ -0.0359-0.0835+ 0.1868* 0.2893* (0.0402) (0.0436) (0.0747) (0.1196) Inventor Dispersion -0.0478-0.0752 0.3717 0.2514 (0.1433) (0.1442) (0.3312) (0.5526) Domestic x Inventor Dispersion 0.1501 0.1318-0.3802-0.4446 (0.1469) (0.2181) (0.2585) (0.3644) Mixed x Inventor Dispersion -0.0937-0.4719** 0.5394+ 0.1975 (0.1094) (0.1459) (0.2964) (0.4428) All Foreign x Inventor Dispersion 0.2333* 0.3442** -0.0275 0.0565 (0.1036) (0.1218) (0.1408) (0.2753) Controls Y Y Y Y Firm Fixed Effects Y Y Y Y Year Fixed Effects Y Y Y Y Strata Fixed Effects Y Y Num of Obs. 11,764 7,748 11,764 7,748 R Squared (Within) 0.2323 0.0286 0.1295 0.0076 AIC 18,087.8 9,293 47,719.5 29,261.4 Log - Likelihood -9,000.9-3,202.5-23,816.7-13,186.7