Ignoring testlet effects by fitting a unidimensional 2PL model produced mild shrinkage of item intercept and loading estimates toward zero relative to the bifactor model¶

Claim · All claims
Evidence · 1 study · theoretical r3 · q2
Currency · newest study 2009 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Compared to the bifactor model, the 2PL model's intercept and loading estimates were somewhat shrunk towards zero for the most part, with larger discrepancies for loadings. → Rijmen 2009

Evidence¶

Rijmen 2009¶

Rijmen, F. (2009). Three Multidimensional Models for Testlet-Based Tests: Formal Relations and an Empirical Comparison. ETS Research Report RR-09-37. http://www.ets.org/research/contact.html doi:10.1002/j.2333-8504.2009.tb02194.x

q2 · i? · theoretical · r3

Figures 4 and 5 display item intercept and loading estimates with 95% confidence intervals for the 2PL and bifactor models on the English assessment data. The article reports "somewhat shrunk towards zero" for the 2PL estimates, and that "The discrepancies in estimated values is larger than was observed for the intercept parameters."

"The estimates of the item intercepts are quite close together, with the estimates for the two-parameter logistic model somewhat shrunk towards zero, compared to the estimates for the bifactor model for the most part."

Discussion¶

Related Claims¶

• In an empirical comparison on a testlet-based English assessment test, the bifactor model is preferred over the second-order and unidimensional 2PL models by both AIC and BIC — related
• The second-order multidimensional IRT model for testlets is formally equivalent to the testlet model, and both are restricted bifactor models — related
• The assessment violates IRT assumptions of unidimensionality and local independence, which the authors identify as a likely cause of the method discrepancy — related
• In simulation, ignoring local dependence overestimates reliability, and the overestimation grows with the size of the testlet effect (17.0%-21.8% decrease under the larger testlet effect vs. 4.8%-7.4% under the small) — related
• In operational reading-test data, item parameter estimates from a standard 2PL/GPCM that ignores local dependence closely match those from the testlet model (average correlations 0.9893 and 0.9978) — related
• Applying a MIRT model to unidimensional data yields larger difficulty-parameter error but smaller first-dimension discrimination error than multidimensional data — related
• A multidimensional IRT model with simple structure fits testlet-based reading data better than both the general polytomous testlet model and the standard 2PL/GPCM — related
• Most Turkish IRT-based scale development studies did not check the unidimensionality or local independence assumptions — related
• PPMC-N detects local dependence in dichotomous IRT models comparably to the Bayesian PPMC approach — related
• Item parceling at the testlet level substantially improves model fit, allowing partial (Rasch, 2PL) or full (bifactor) recovery of CAT constructs — related
• GLAT shows robust structural validity and reliability under a 2-parameter logistic IRT model — related
• In operational reading-test data, ignoring local dependence inflates reliability estimates, with passage-based alpha 2.3%-4.9% lower than item-based alpha across six test forms — related
• Parceling over-fits models: at the testlet level, data from more complex models fit well with simpler models — related
• Bifactor-(S-1) models of teacher Big Five personality fit student and teacher data better than single-domain or three-facets models — related
• A bifactor model of oral language and reading fits better than one-, three-, and four-factor models in every grade from 4th through 10th — related