Stratified partial likelihood estimation

Area of Science:

• Econometrics
• Biostatistics
• Demography

Background:

• Statistical models often assume independence between observations from the same unit.
• However, unobserved unit-specific variables can introduce correlation across multiple durations.
• These unobserved factors may be linked to both the durations and observed explanatory variables.

Purpose of the Study:

• To propose a novel estimator addressing concerns of omitted unit-specific variables.
• To develop a specification test for detecting unobserved unit-specific effects.
• To assess the impact of unobserved heterogeneity on child mortality concentration in Malaysia.

Main Methods:

• Development of a new statistical estimator designed to account for unobserved unit-specific effects.
• Creation of a specification test to identify the presence of such unobserved effects.
• Application of the proposed methods to Malaysian family data on child mortality.

Main Results:

• The concentration of child mortality within families is not fully explained by observed variables alone.
• Failure to control for unobserved heterogeneity significantly biases parameter estimates.
• The proposed estimator and test provide a more accurate analysis of child mortality determinants.

Conclusions:

• Unobserved unit-specific effects are crucial for understanding child mortality concentration.
• Statistical models must account for unobserved heterogeneity to avoid biased results.
• The developed methods offer a robust approach for analyzing complex family-level data.