Examination of Parameter Estimation Using Recursive Bayesian
Analysis in Simulated Item Response Theory Applications

\nby Robert
C. Hendrick

For the past several years\, high-stakes testing has b een the predominant indicator used to assess students’ academic ability. S chool systems\, teachers\, parents\, and students are dependent upon the a ccuracy of academic ability estimates\, designated by es\, by item respons e theory (IRT) computer programs. In this study\, the accuracy of 3 parame ter logistic (3PL) IRT estimates of academic ability were obtained from th e BILOG-MG and WinBUGS computer programs which were employed to compare th e use of non-informative and informative priors in e estimation. The ratio nale for comparing the output of these two computer programs is that the u nderlying statistical theory employed in these two computer programs is di fferent\, and there may be a notable difference in the accuracy of 6 estim ation when an informative prior is used by Win BUGS in analyzing skewed po pulations. In particular\, the e parameter estimates of BILOG-MG using tra ditionaiiRT analysis with non-informative priors in each situation and the 6 parameter estimates of WinBUGS using Recursive Bayesian Analysis (RBA) with informative priors are compared to the true simulated 6 value using R oot Mean Square Errors (RMSEs). To make this comparison\, Monte Carlo comp uter simulation is used across three occasions within three conditions giv ing nine comparison situations. For the priors and data generated\, result s show similar 6 estimation accuracy for a normally distributed latent tra it (RMSE = 0.35)\, a more accurate 6 estimation process using RBA compared to traditional analysis (RMSEs of 0.36 compared to 0. 76) when using late nt trait distributions skewed in a similar direction\, and less accurate e estimation using RBA compared to traditional analysis (RMSEs of 1.48 comp ared to 0.80) when using extremely skewed negative then positive distribut ions in a longitudinal setting. Implications for further research include extensions to other IRT models\, developing prior elicitation equations\, and applying Bayesian informative prior elicitations in BILOG-MG.

DTSTART;TZID=America/New_York:20140424T130000 DTEND;TZID=America/New_York:20140424T150000 LOCATION:College of Education\, room 481 @ 30 Pryor Street Southwest\, Geor gia State University\, Atlanta\, GA 30303\, USA SUMMARY:Dissertation Defense – Robert C. Hendrick URL:http://education.gsu.edu/ai1ec_event/dissertation-defense-robert-c-hend rick/?instance_id= X-TAGS;LANGUAGE=en-US:Department of Educational Policy Studies\,dissertatio n defense END:VEVENT END:VCALENDAR