Techniques for handling missing data in secondary analyses of large surveys

Area of Science:

• Biostatistics
• Public Health Research
• Data Analysis

Background:

• Secondary analysis of survey data is crucial for child health research.
• Incomplete datasets are common, and missing data can introduce bias and reduce statistical power.
• Current practices often fail to address missing data appropriately, with many studies deleting cases or not reporting methods.

Purpose of the Study:

• To evaluate four distinct methods for handling missing data in secondary analyses.
• To determine which technique most accurately estimates true model coefficients under varying missing data proportions.
• To provide evidence-based recommendations for researchers analyzing child health survey data.

Main Methods:

• A simulation study was conducted to compare statistical models.
• Four methods for handling missing data were applied: case deletion, hot deck imputation, reweighting, and multiple imputation.
• Simulations included datasets with 10%, 20%, 30%, and 40% missing data.

Main Results:

• Reweighting and multiple imputation techniques demonstrated superior performance compared to case deletion and hot deck imputation when missing data exceeded 10%.
• Case deletion and hot deck imputation yielded less accurate estimates as the proportion of missing data increased.
• The accuracy of model coefficients was significantly better preserved by reweighting and multiple imputation.

Conclusions:

• Researchers analyzing child health survey data with substantial missing values should exercise caution.
• The practice of dropping cases with missing data is generally discouraged due to potential bias.
• Reweighting and multiple imputation are recommended as more robust methods for handling significant amounts of missing data.