Working Papers

Estimating Conditional Probabilities from Binary Panel Data

Miller and Sanjurjo (2018) highlight a bias in estimating conditional probabilities from binary dependent variables which appears to resemble the Nickell bias in panel data settings (Nickell, 1981). In this paper, I demonstrate that dynamic panel data estimators mitigate this bias and can be used to estimate unbiased conditional probabilities from panel data with binary dependent variables. Using Monte Carlo simulations, I find that a bias-corrected linear method of moments estimator outperforms commonly used linear and non-linear dynamic panel estimators. To supplement this analysis, I re-estimate findings from the seminal paper on the hot hand effect. I conclude by providing theoretical applications of applied domains in which these biases may arise and would benefit from the use of such estimators.