Maximizing the Usefulness of Data Obtained with Planned Missing Value Patterns: An Application of Maximum Likelihood...

Area of Science:

• Statistics
• Survey Methodology
• Psychometrics

Background:

• Survey research often involves missing data, posing challenges for accurate statistical analysis.
• Choosing between collecting limited high-quality data versus extensive lower-quality data is a common dilemma.

Purpose of the Study:

• To evaluate the utility of the 3-form survey design with maximum likelihood methods for estimating missing values.
• To compare the efficiency and bias of various missing data estimation techniques.

Main Methods:

• Simulated data using a 3-form design with missing data.
• Estimated variances and covariances using pairwise deletion, mean replacement, single imputation, multiple imputation, raw data maximum likelihood, multiple-group covariance structure modeling, and Expectation-Maximization (EM) algorithm.
• Applied methods to empirical drug use data.

Main Results:

• Maximum likelihood estimation and multiple imputation yielded the most efficient and least biased estimates for complete random data.
• Pairwise deletion was unbiased but less efficient than maximum likelihood procedures.
• Non-maximum likelihood methods failed when data were not missing completely at random.

Conclusions:

• Maximum likelihood estimation or multiple imputation are recommended for missing data analysis.
• Splitting scale items across forms maximizes the efficiency of maximum likelihood parameter estimates.