Lagged dependent variables (LDVs) are commonly used in regression analysis, yet some research argues against them due to potential bias.
➡️ The Problem: Previous studies suggested LDVs negatively bias coefficient estimates if they're part of the data-generating process.
➡️ New Approach: This paper demonstrates that properly accounting for autocorrelation—by including more lagged dependent variables or other lags—actually improves accuracy.
➡️ Simulation Results: Using Monte Carlo simulations, we show this method yields significantly better estimates than alternatives.
➡️ Recommendation: LDVs should be included confidently in robust estimation strategies to avoid bias.
| To Lag or Not to Lag? Re-evaluating the Use of Lagged Dependent Variables in Regression Analysis was authored by Arjun Wilkins. It was published by Cambridge in PSR&M in 2018. |
