Estimating dynamic panel data discrete choice models with fixed effects

Abstract

This paper considers the estimation of dynamic binary choice panel data models with fixed effects. It is shown that the modified maximum likelihood estimator (MMLE) used in this paper reduces the order of the bias in the maximum likelihood estimator from OðT À1 Þ to OðT À2 Þ, without increasing the asymptotic variance. No orthogonal reparametrization is needed. Monte Carlo simulations are used to evaluate its performance in finite samples where T is not large. In probit and logit models containing lags of the endogenous variable and exogenous variables, the estimator is found to have a small bias in a panel with eight periods. A distinctive advantage of the MMLE is its general applicability. Estimation and relevance of different policy parameters of interest in this kind of models are also addressed. r

The effects of parental leave on child health and postnatal care: Evidence from Australia

“…The two approaches that are now commonly used to control for unobserved individual characteristics using panel data include a xed-eects estimator (Chamberlain, 1980;Ferrer-i Carbonell & Frijters, 2004;Carro, 2007;Jones & Schurer, 2011), and a random-eects estimator (Mundlak, 1978;Contoyannis et al, 2004;Contoyannis & Li, 2011). The former approach eliminates the individual unobserved heterogeneity factor α i via means such as rst dierencing whilst the latter assumes that α i is uncorrelated with other covariates.…”

Persistence of innovation: stylised facts and panel data evidence

“…Honoré and Kyriazidou (2000) proposed a semiparametric estimator for the FE logit model, but their estimator is extremely data demanding and cannot be used here. Carro (2003) suggested a modified maximum likelihood estimator for the dynamic probit model, but the estimator is only consistent when T goes to infinity. 17 Therefore, I decide to apply a RE model.…”

Understanding Bias in Nonlinear Panel Models: Some Recent Developments

“…In a likelihood setting it is also possible to form an estimate of b i (θ) that uses expected rather than observed quantities, giving rise to alternative score-corrected estimators, such as those considered by Carro (2004) and Fernández-Val (2005) for binary choice models.…”

“…After all, Arellano's procedure is numerically identical to Woutersen's, which does not require such assumption. Indeed, Carro (2004) has shown that Arellano's moment equation reduces the order of the score bias regardless of the existence of an information orthogonal reparameterization.…”