Missing data in experiments: Challenges and solutions

Area of Science:

• Psychology
• Statistics
• Data Science

Background:

• Missing data is prevalent in experimental psychology datasets.
• Traditional methods for handling missing data often violate assumptions and bias results.
• This impacts the validity of random assignment and effect size estimates.

Purpose of the Study:

• To provide guidelines for handling different classes of missing data in experimental datasets.
• To introduce methods that make realistic assumptions for causal inference.
• To enhance the accuracy of effect size estimates in psychological research.

Main Methods:

• Categorization of missing data types in experimental datasets.
• Application of inverse probability weighting (IPW) for mild missingness.
• Utilization of double sampling and bounds for severe missingness.

Main Results:

• Demonstration of how different missing data classes affect causal inferences.
• Explanation of how IPW and double sampling/bounds improve estimate accuracy.
• Provision of R code for practical implementation.

Conclusions:

• Adopting realistic methods like IPW and double sampling/bounds is crucial for robust causal inference.
• Accurate handling of missing data leads to more reliable effect size estimates.
• The provided guidelines and code empower researchers to address missingness effectively.