28
The effect of differenal item funconing in anchor items on populaon invariance of equang Anne Corinne Huggins University of Florida

The effect of differential item functioning in anchor items on population invariance of equating Anne Corinne Huggins University of Florida

Embed Size (px)

Citation preview

  • Slide 1

The effect of differential item functioning in anchor items on population invariance of equating Anne Corinne Huggins University of Florida Slide 2 Introduction Population invariance of equating Test score level Differential item functioning Item level Need to connect two facets of invariance Certain types of DIF might cause equating dependence across forms of a test. Due to (1)a differential difficulty, particularly for anchor items (2)a second factor(s) (3)Item parameter drift Slide 3 IRT true score equating A three-step process: Specify a true score on one form Find the corresponding IRT ability estimate Find the IRT ability estimates corresponding true score on the other form Slide 4 Place the latent ability scores on each test on the same scale We can use the probability formula to get the relations. Slide 5 Scaling Mean-sigma Mean-mean 5 Here, measurement errors are ignored. If a and b were biased, A and B will be biased. Q: How to estimate A under Rasch model? Slide 6 The Stocking and Lord method Minimize the difference in test characteristic curves (TCCs) for anchor items The Haebara method Minimize the difference in item characteristics curves (ICCs) 6 Slide 7 Equating invariance Compare the whole group to multiple subgroups Group to overall (compare one subgroup with the whole group) 7 Sigma Q is the unconditional SD of scores in population. Slide 8 The factors that might influence population invariance of IRT true score equating Differential anchor form DIF The magnitude of DIF in anchor items Number of DIF anchor items Direction of DIF in anchor items Ability differences between subpopulations 8 Slide 9 Method Uniform DIF R, BILOG-MG, SPSS Design Nonequivalent groups with anchor test (NEAT) design Two forms contain 50 dichotomous items, respectively, with 20% of common items Ability ~ N (0,1) 4500 participants (3000 in the references group) 3PL model 100 replications 9 Slide 10 Manipulations on DIF in anchor items Magnitude (.3,.6,.9) % of DIF items (20%, 49%, 60%) Directionality of DIF Mean ability differences Differential anchor form DIF Null conditions No DIF in anchor items and no mean ability differences No DIF in anchor items but had mean ability differences 10 Slide 11 ResultsNo DIF (RMSD (x) ) 11 Slide 12 No DIF, Equal mean abilities (RSD (x) ) 12 Slide 13 No DIF, Unequal mean abilities (RSD (x) ) 13 Slide 14 14 Why didnt DIF show the impact here? Slide 15 15 Slide 16 16 Slide 17 17 Slide 18 18 Slide 19 19 Slide 20 20 Slide 21 21 Slide 22 22 Slide 23 23 Slide 24 24 Slide 25 Conclusions Differential anchor form DIF occurs, the magnitude of equating dependence increases. IPD occurs for different subpopulations. When differential anchor from DIF occurs: Direction of DIF in anchor items shows the largest effect on equating dependence. The magnitude of DIF in anchor items Number of DIF anchor items Ability differences between subpopulations 25 Slide 26 Comments It is a good writing sample. Errors: P.9, since the study manipulated the direction of DIF, the statement of magnitude of DIF should be revised properly. The author intended to design the sample size as the true subpopulations in the US, but the actual sample sizes did not reflect its intends. Slide 27 When DIF or item parameter drift occurs in a specific anchor item, this item should not be used as an anchor. We usually do not talk about DIF in anchor items because DIF detections should be done before any equating or linking. When all anchor items are DIF items, the results should be very similar with the ones without DIF anchor items. Slide 28 Future studies Sample size differences across score levels Test length, anchor length, ratio of anchor to test length, mixed item formats 28