Estimating Causal Installed-Base Effects: A Bias-Correction Approach

New empirical models of consumer demand that incorporate social effects seek to measure the causal effect of past adopter's behavior the installed-base on current adoption behavior. Identifying such causal effects is challenging due to several alternative confounds that generate correlation in agents' actions. In the absence of experimental variation, a preferred solution has been to control for these spurious correlations using a rich specification of fixed effects. The authors show that fixed-effects estimators of this sort are inconsistent in the presence of installed-base effects; in simulations, random-effects specifications perform even worse. The analysis reveals the tension the applied empiricist faces in this area: a rich control for unobservables increases the credibility of the reported causal effects, but the incorporation of these controls introduces biases of a new kind in this class of models. The authors present two solutions: a modified version of an instrumental variable approach and a new bias-correction approach, both of which deliver consistent estimates of causal installed-base effects. The empirical application to the adoption of the Toyota Prius Hybrid in California shows evidence for social influence in diffusion and reveals that implementing the bias correction reverses the sign of the measured installed-base effect. The authors also discuss implications of the results for identification of models in marketing involving state dependence in demand, and incorporating discrete games of strategic interaction.