Multiple imputation of continuous data via a semiparametric probability integral transformation

Area of Science:

• Statistics
• Biostatistics
• Data Science

Background:

• Missing continuous data is a common challenge in statistical analysis.
• Existing imputation methods often rely on strict distributional assumptions.
• Accurate imputation is crucial for unbiased statistical inference.

Purpose of the Study:

• To develop a flexible semiparametric imputation method for continuous data.
• To handle variables with any marginal distribution.
• To improve the accuracy of statistical analyses with incomplete datasets.

Main Methods:

• A semiparametric approach combining multiple imputation under normality assumption and empirical cumulative distribution function (eCDF) calculations.
• Data transformation to normality, imputation, and back-transformation to original scale.
• Multivariate number generation for imputation.

Main Results:

• The proposed method demonstrated promising results in imputing continuous data.
• Effective handling of variables with diverse marginal distributions.
• Successful application to both simulated and real-world datasets.

Conclusions:

• The semiparametric imputation method offers a robust solution for missing continuous data.
• It provides a flexible alternative to traditional parametric imputation techniques.
• The method shows potential for enhancing the reliability of statistical findings.