Covariance Structure Analysis of Partially Additive Ipsative Data Using Restricted Maximum Likelihood Estimation.

Area of Science:

• Statistics
• Psychometrics
• Data Analysis

Background:

• Ipsative data, where subject scores sum to a constant, presents unique analytical challenges.
• Existing methods struggle with the degenerate distribution of ordinary additive ipsative data.
• Partially additive ipsative data (PAID) is a generalized form requiring new estimation techniques.

Purpose of the Study:

• To define and analyze partially additive ipsative data (PAID).
• To develop a method for estimating parameters in PAID data, overcoming limitations of ordinary maximum likelihood (ML).
• To evaluate the performance of the proposed estimation method through simulation.

Main Methods:

• Definition of partially additive ipsative data (PAID).
• Development of a data transformation (X* = BX) to achieve a nonsingular density.
• Application of restricted maximum likelihood (REML) estimation to the transformed data.
• Conducting a simulation study to assess estimator performance and goodness-of-fit statistics.

Main Results:

• REML estimates are close to true parameter values but have larger standard errors than ML estimates on underlying nonipsative data.
• The goodness-of-fit test statistic performs well with sufficient sample size.
• Convergence likelihood depends on sample size, factor structure, and degree of ipsativity.
• Statistical decisions based on transformed data (X*) align with those based on underlying data (y).

Conclusions:

• The proposed transformation and REML method offer a viable approach for analyzing PAID data.
• While REML provides consistent results, careful consideration of sample size and model characteristics is crucial for reliable parameter estimation.
• The findings support the use of the developed statistical decisions for models with ipsative data.