Influence diagnostics in the Heckman selection models based on EM algorithms

Area of Science:

• Econometrics
• Statistical Modeling

Background:

• Heckman selection models are vital for addressing selection bias in statistical estimations.
• Existing diagnostic techniques may not fully capture influential observations in these models.

Purpose of the Study:

• To develop novel diagnostic techniques for Heckman selection models estimated via the EM algorithm.
• To provide robust methods for identifying influential observations in selection t and normal models.

Main Methods:

• Utilized the EM algorithm for estimating Heckman selection models.
• Developed global and local influence analyses based on the complete-data log-likelihood.
• Explored four perturbation schemes for local influence analysis.

Main Results:

• The proposed diagnostic measures effectively identify influential observations.
• Simulation studies and real-data applications validated the methodology.
• The R package HeckmanEM incorporates the developed algorithms and diagnostics.

Conclusions:

• The new diagnostic techniques enhance the reliability of Heckman selection model estimations.
• The HeckmanEM R package provides a practical tool for applied researchers.
• Accurate identification of influential points is critical for robust econometric analysis.