Structural Equation Modeling for Dynamic Panel Data: Addressing Endogeneity and Lagged Effects

Recent management literature emphasizes the need to model complex dynamic relationships, including autoregression, endogeneity, and lagged effects, and often turns to panel data as a solution. This article proposes a structural equation modeling (SEM) framework for panel data that accommodates these complexities through simultaneous equations, latent variables, and covariation between endogenous regressors and disturbance terms. We illustrate the approach using empirical data on how firms’ research and development (R&D) investment—a potentially endogenous regressor—affects firm performance. Results show that parameter estimates are highly sensitive to how endogeneity and lagged effects are specified. To aid interpretation, we compute dynamic quantities—response functions, cumulative effects, and long-run multipliers—directly from SEM estimates. Monte Carlo simulations further demonstrate the estimation bias that arises from misspecification, such as omitted lags or failure to account for endogeneity.

JEL CLASSIFICATION: M10, M21, C15, C33, C36, O32