The Use of Sample Weights in Hot Deck Imputation

Area of Science:

• Survey Methodology
• Statistical Imputation

Background:

• Hot deck imputation is a common method for handling missing survey data.
• This method replaces missing values with observed data from similar respondents.
• The use and impact of sampling weights in hot deck imputation require careful consideration.

Purpose of the Study:

• To evaluate the effectiveness of different approaches to incorporating sampling weights in hot deck imputation.
• To identify the correct method for using sampling weights to avoid bias in survey data.
• To assess the performance of weighted versus unweighted hot deck methods.

Main Methods:

• The study employed simulation methods to compare different hot deck imputation strategies.
• Approaches included ignoring weights, incorporating weights into donor selection probabilities, and using weights as a stratifying variable.
• Bias was assessed in simulations where the outcome was related to sampling weight and response propensity.

Main Results:

• Ignoring sampling weights in adjustment cells leads to bias by imputing unweighted distributions.
• Weighted hot deck methods, where weights influence donor selection, do not correct for bias under specific conditions.
• Simulation results indicate that these weighted approaches fail when outcomes correlate with sampling weights and response propensities.

Conclusions:

• The naive approach of ignoring sampling weights in hot deck imputation is biased.
• Weighted hot deck methods that incorporate weights into donor selection probabilities do not resolve bias effectively.
• The correct method involves using sampling weights as a stratifying variable within adjustment cells for unbiased survey imputation.