Identification and Estimation of Installed-Base Effects for Product Adoption Under Sample Attrition and Homophily

This paper studies identification and estimation of installed-base effects for product adoption using group-level panel data in the presence of endogenous sample attrition and homophily. After exploring conditions under which installed-base effects are identified using group-level panel data in the considered setting, I propose a modified BLP approach for estimation. The proposed approach accounts for endogenously changing composition of remaining group members in the simulation of predicted adoption rates, thereby addressing sample attrition. To address homophily, the proposed method performs first-differencing within a given group and uses lags and lagged differences of the installed base as instruments. I present Monte Carlo results to numerically demonstrate the identification issues as well as the performance of the proposed estimation method.