Interval censored regression with fixed effects

Area of Science:

• Econometrics
• Statistical Modeling

Background:

• Many real-world datasets contain interval-censored dependent variables, where only the range of the variable is known.
• Existing methods may struggle with identification and estimation in such scenarios, particularly with fixed effects.

Purpose of the Study:

• To develop and evaluate methods for identifying and estimating fixed-effects models with interval-censored dependent variables.
• To address both parametric (logistic errors) and semiparametric (unspecified error distribution) models.
• To investigate the direct estimation of causal effects.

Main Methods:

• Proposed a conditional-logit-type composite likelihood estimator for the parametric logistic fixed-effects model.
• Developed a composite maximum-score-type estimator for the semiparametric model.
• Allowed for heteroskedasticity across units and stationarity within units; the semiparametric model also accommodates serial correlation.

Main Results:

• The proposed estimators identify the scale of coefficient parameters, enabling direct estimation of causal effects.
• Monte Carlo simulations demonstrated the performance of the parametric estimator.
• An empirical application to birthweight outcomes validated the practical utility of the parametric approach.

Conclusions:

• The developed estimators provide a robust framework for analyzing interval-censored data in fixed-effects models.
• Direct estimation of causal effects is achievable, enhancing the interpretability of results.
• The methods are applicable in various fields, including health economics and social sciences.