An Evaluation of the Effects of Variable Sampling On Component, Image, and Factor Analysis

Area of Science:

• Psychometrics
• Statistical Analysis
• Multivariate Statistics

Background:

• Latent variable procedures are used to infer unobserved variables from observed data.
• Generalizability from sampled variables to a population of variables is a key concern in factor analysis.
• Different analytical methods may yield varying results based on data characteristics.

Purpose of the Study:

• To compare the effects of variable sampling on different latent variable procedures.
• To investigate how factor loading size, variables per factor, and sample size influence generalization.
• To assess the impact of analytical method on the accuracy of variable sampling.

Main Methods:

• Compared Principal Component Analysis (PCA), Image Component Analysis (ICA), and Maximum Likelihood Factor Analysis (MLFA).
• Manipulated independent variables: loading size, average variables per factor, and number of variables sampled.
• Generated multiple correlation matrices and randomly selected variable samples for analysis.

Main Results:

• Method of analysis had minor and complex effects on generalization.
• Degree of saturation (loading size) and average number of variables per factor had clear, dramatic effects.
• Differential impacts were observed on boundary cases and nonconvergence issues.

Conclusions:

• Variable sampling significantly influences the generalizability of latent variable analyses.
• Factor loading size and the number of variables per factor are critical determinants of accurate results.
• Researchers must carefully consider sampling strategies and data characteristics when applying these methods.