Estimating Incremental Validity Under Missing Data.

Area of Science:

• Statistics
• Data Analysis
• Psychometrics

Background:

• Missing data is a common issue in statistical analyses.
• Selection bias can arise when data are missing based on observed variables.
• Missing At Random (MAR) is a key assumption for handling missing data.

Purpose of the Study:

• To investigate the impact of missing data on statistical estimates, particularly partial correlations.
• To compare the performance of listwise deletion (LD) and maximum likelihood (ML) methods under MAR conditions.
• To identify conditions under which ML estimates for partial correlations may become biased.

Main Methods:

• The study examines statistical biases introduced by missing data.
• It compares listwise deletion (LD) and maximum likelihood (ML) estimation techniques.
• The analysis focuses on data assumed to be missing at random (MAR).

Main Results:

• Listwise deletion (LD) can bias correlation, reliability, and effect size estimates.
• Maximum likelihood (ML) estimates are generally unbiased for MAR data, outperforming LD.
• However, ML estimates for partial correlations can still be biased under MAR, depending on computation methods.

Conclusions:

• While ML generally outperforms LD for MAR data, partial correlation estimation requires careful consideration.
• Even ML estimates may be biased when computing partial correlations under MAR.
• Recommendations are provided for estimating partial correlations when the cause of missingness is unknown.