Environmental Interest Groups and Authoritarian Regime Diversity – Online Appendix To assess the robustness of my findings, I changed a variety of specifications and estimated the empirical models again.1 Clarke (2005) shows that the inclusion of control variables may actually increase estimation bias instead of decreasing it. Moreover, some of the control variables may undercut the significance and size of my key explanatory variables. As shown in the paper, however, including or excluding these control variables does not affect the main results. Second, as indicated, discrete time data are very similar to a binary dependent variable in a time-series cross-section format (Beck et al. 1998). Therefore, I estimated all models again using a Cox duration setup. Again, this did not change my core findings. Finally, as indicated in the main paper, autocratic regime typologies vary in terms of their behavior toward the establishment of ENGOs, making ENGO occurrence unlikely to be a random set. In other words, autocracies with (more) ENGOs should differ in important and predictable ways from those autocracies that are characterized by only a few ENGOs. Ultimately, we face a selection problem that – if not addressed – may either underestimate or exaggerate the findings, leading to biased estimates if selection is an issue. Previous research dealt with this problem either through an instrumental variable approach or the use of selection estimators. Following, e.g., the approach in Beardsley (2008: 731), I therefore calculated bivariate probit models in addition, using the specifications outlined in Greene (2003, p. 710) and Maddala (1983, p. 122). For these models, I had to define two different dependent variables: one for the outcome equation and one for the selection equation. Consequently, the dependent variable in the latter equation is whether an autocracy is characterized by ENGOs or not. The second equation using IEA ratification as the outcome 1 All robustness checks can be replicated with the replication files. 2 variable is then estimated simultaneously, while taking into account the correlation in the equations’ error processes (Greene 2003). My estimate for the parameter in either model is positive and significant (for the full model: 2 (1)=15.1417; Prob>2=0.0001). The estimate of can be highly sensitive to model specifications and should be expected to be negative if unobserved features that increase the likelihood of selection actually decrease the probability of IEA ratification. However, there is evidence of a positive correlation in the disturbances between the selection and outcome equations as is positive. This implies that unobserved features that make ENGO creation in autocracies more likely also induce a higher likelihood of IEA ratification. Despite some signs that point to a selection issue, though, my core results remain virtually unchanged even when controlling for this problem. References Beardsley, K. (2008). Agreement without peace? International mediation and time inconsistency problems. American Journal of Political Science, 52(4), 723–740. Beck, N., et al. (1998). Taking time seriously: Time-series-cross-section analysis with a binary dependent variable. American Journal of Political Science, 42(4), 1260–1288. Clarke, K. A. (2005). The phantom menace: Omitted variable bias in econometric research. Conflict Management and Peace Science, 22(4), 341–352. Greene, W. H. (2003). Econometric analysis. 5th ed. Upper Saddle River, NJ: Prentice Hall. Maddala, G. S. (1983). Limited-dependent and qualitative variables in econometrics. Cambridge: Cambridge University Press.