Strategies for the selection of log-linear models

Area of Science:

• Statistics
• Data Analysis

Background:

• Log-linear models are used for analyzing multidimensional contingency tables.
• Model building often involves complex procedures and choices for data analysts.

Purpose of the Study:

• To review and discuss various strategies for constructing log-linear models.
• To compare different model-building approaches, including stepwise methods and standardized estimates.

Main Methods:

• Discussion of established methods for log-linear model selection.
• Examination of Brown's (1976) two-step procedure for screening effects.
• Analysis of alternative model-building strategies.

Main Results:

• Stepwise methods and standardized estimates are presented as approaches for log-linear model building.
• Brown's procedure offers a structured way to screen effects before model testing.
• The choice of model and intermediate data analysis information influence the selection process.

Conclusions:

• Effective log-linear model building requires careful consideration of available strategies.
• The selection process can be guided by specific procedures and the information at hand.
• Understanding alternative methods aids in choosing the most appropriate model.