Adjusting for attrition in event-history analysis.

Area of Science:

• Sociology
• Statistics
• Demography

Background:

• Panel attrition, where participants drop out of a study over time, poses a significant challenge in longitudinal research.
• Event-history models are sensitive to how sampling weights are adjusted to account for participant attrition.
• The Survey of Income and Program Participation (SIPP) is a key source for U.S. household economic data, making attrition a critical issue.

Purpose of the Study:

• To investigate the impact of different weighting procedures on event-history models concerning panel attrition.
• To compare the effectiveness of initial selection probability weights, panel weights, and attrition-adjusted weights in a divorce model.
• To assess whether attrition is misclassified as divorce in the SIPP data.

Main Methods:

• Utilized event-history models to analyze divorce data from the 1986 U.S. Survey of Income and Program Participation (SIPP).
• Compared three distinct weighting procedures: initial selection probability weights, 1986 SIPP panel weights, and monthly attrition-adjusted weights.
• Contrasted weighted estimates with those from a structural model treating attrition as an error-correlated competing alternative to divorce.

Main Results:

• Different weighting procedures yield varying estimates in event-history models when addressing panel attrition.
• A notable finding is that divorces within the SIPP dataset are frequently recorded as instances of panel attrition.
• The choice of weighting method can substantially alter conclusions regarding divorce rates and patterns.

Conclusions:

• Careful consideration and appropriate adjustment of sampling weights are crucial for accurate analysis of event-history data with attrition.
• The misclassification of divorces as attrition in the SIPP highlights potential data quality issues and biases.
• Future research should focus on robust methods for distinguishing between true attrition and substantive events like divorce.