Research Framework - Statistical Analysis
Model Fit
Model fit research quantifies and describes the degree to which the estimated psychometric model accurately describes the observed data, especially in regard to accuracy of intended consequences. That is, model fit analysis should focus on evaluating the assessment outcomes in the context of how they affect the assessment's ability to achieve its intended purpose(s).
Examples of model fit analyses include correspondence between observed data statistics and their model-predicted counterparts, statistical tests of residuals, pattern analysis of residuals, and log-likelihood statistical comparisons of the fit of alternative (especially nested) models.
Research
Evaluation of the Effect of the Proportion of Examinees Used for Nonparametric Dimensionality Assessment (DIMTEST), by EunYoung Lim and Louis Roussos (2007), Paper presented at the American Education Research Association Annual Meeting, Chicago, IL
Formulation of an Effect Size Measure for DIMTEST, by Minhee Seo and Louis Roussos (2007), Paper presented at the American Education Research Association Annual Meeting, Chicago, IL
Assessing the Relative Performance of Local Item Dependence Indices, by Doyoung Kim, Ralph DeAyala, Abdullah Ferdous and Michael Nering (2007), Paper presented at the American Education Research Association Annual Meeting, Chicago, IL
Missing Data Treatment Methods in Parameter Recovery for a Mixed-Format Test, by Thakur Karkee and Matt Finkelman (2007), Paper presented at the American Education Research Association Annual Meeting, Chicago, IL
Fusion Model "Fit" Indices, by Robert Henson, Louis Roussos, and Jon Templin (2005), Unpublished ETS Project Report, Princeton, NJ.
