Premium
The Effects of Selection Strategies for Bivariate Loglinear Smoothing Models on NEAT Equating Functions
Author(s) -
Moses Tim,
Holland Paul W.
Publication year - 2010
Publication title -
journal of educational measurement
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.917
H-Index - 47
eISSN - 1745-3984
pISSN - 0022-0655
DOI - 10.1111/j.1745-3984.2009.00100.x
Subject(s) - akaike information criterion , equating , statistics , bivariate analysis , mathematics , bayesian information criterion , log linear model , econometrics , likelihood ratio test , deviance information criterion , model selection , goodness of fit , selection (genetic algorithm) , bayesian probability , statistic , information criteria , statistical hypothesis testing , smoothing , linear model , bayesian inference , computer science , rasch model , artificial intelligence
In this study, eight statistical strategies were evaluated for selecting the parameterizations of loglinear models for smoothing the bivariate test score distributions used in nonequivalent groups with anchor test (NEAT) equating. Four of the strategies were based on significance tests of chi‐square statistics (Likelihood Ratio, Pearson, Freeman‐Tukey, and Cressie‐Read) and four additional strategies were based on different evaluations of the Likelihood Ratio Chi‐Square statistic (Akaike Information Criterion, Bayesian Information Criterion, Consistent Akaike Information Criterion, and an index traced to Goodman). The focus was the implications of the selection strategies’ selection tendencies for the accuracy of chained and poststratification equating functions. The results differentiated the strategies in terms of their tendencies to select models with particular bivariate parameterizations and the implications of these tendencies for equating bias and variability .