Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • uirt - a new package to fit unidimensional Item Response Theory models is now available from SSC

    Dear fellow Stata users.

    Thanks to Kit Baum a new package uirt is now available for download from SSC.

    Description: uirt is a Stata module for estimating variety of unidimensional IRT models (2PLM, 3PLM, GRM, PCM, GPCM). It features multi-group modelling, DIF analysis, extended graphical item fit analysis and generating plausible values (PVs) conditioned via latent regression. uirt implements the EM algorithm (Dempster, Laird & Rubin, 1977) in the form of marginal maximum likelihood estimation (MML) proposed by Bock & Aitkin (1981) with normal Gauss-Hermite quadrature. LR test is used for DIF testing and model based P-DIF effect size measures are provided (Wainer, 1993). Generating PVs is performed by adapting a MCMC method developed for IRT models by Patz & Junker (1999). Unconditioned PVs are used to plot observed response proportions against the item characteristic curves allowing for detailed graphical item fit analysis.

    To install, in Stata type: ssc install uirt. For detailed description of options end references type help uirt after installing.

    I hope this package to be especially useful in the following instances:
    1. If you are a Stata 13 or a Stata 12 user and want to perform IRT analysis (I plan to enable it to run on Stata 10 in future releases)
    2. If you are a Stata 14 user and:
    • want to fit a multi-group IRT model
    Code:
    use http://www.stata-press.com/data/r14/masc2.dta
    *2PLM/GRM are default models
    uirt q*,gr(female)
    • want to analyze for differential item functioning with IRT P-DIF effect measures
    Code:
    *using the above masc2.dta
    uirt q*,gr(female) dif(q*)
    *to view a compact summary of DIF analysis:
    mat l e(dif_results)
    • want to graphically inspect item fit in your IRT model
    Code:
    use http://www.stata-press.com/data/r14/alike.dta
    *this will save ICCs with observed proportions in your working directory
    uirt v*,icc
    *you can see that the “v3” item exhibits some misfit:
    Click image for larger version

Name:	ICC_v3_alike.png
Views:	1
Size:	13.0 KB
ID:	1360249
    • want to generate plausible values
    Code:
    *using the above alike.dta
    uirt v*, pv(2) theta
    *we can see a lack of shrinkage in comparison with EAP estimate:
    sum pv* theta
    • have a large dataset
    This is an ongoing project. Any comments, questions or bug reports are welcome. Contact information is included in the help file.

    Regards,

    Bartek

  • #2
    Speaking of bugs…
    I spotted a problem with generating starting values for PCM and GPCM. For a quick fix you can edit one line in the beginning of the uirt.ado file (line #8) form:
    Code:
    version 12
    to:
    Code:
    version 10
    Sorry for that, will fix it in next release after more improvements come along.
    Last edited by Bartosz Kondratek; 14 Oct 2016, 06:58.

    Comment

    Working...
    X