Test item parameter estimation

Submitted by on Tue, 09/11/2012 - 09:02
Forums: 
General discussion

Hi all,

I have an online testing system and want to make it adaptive, I have a lot of test items in database in order to make the adaptive algorithm work I need to attach difficulty and discrimination parameters to test item.

Can someone explain me how the item calibraiton takes place ? how to estimate difficulty and discrimination parameters ?

Thanks in advance,

Harut.

  • anon

    I agree that the best and usually easiest way to to item parameter estimation is to use special-purpose software, such as Xcalibre.  Such software has the advantage of estimating both person ability and item parameters at the same time.  However, if you already have a measure of person ability for each of your tests, you can estimate item parameters using statistical data analysis methods, such as nonlinear model-fitting procedures in SAS.  We routinely estimate Rasch model parameters using SAS Procs NLIN or NLMIXED, by supplying the previously calculated person ability measure along with each scored item response.

     

    That said, the item parameter estimation software packages are generally preferable, and less prone to technical problems.

    Jim McBride

    James R. McBride

    Vice President & chief Psychometrician

    Renaissance Learning, Inc.

    6430 Elmhurst Drive

    San Diego, California 92120

    USA

    866.988.8048 ext 4700

    Sep 18, 2012
  • anon

    Hi Harut

     
    Sounds like you are looking for a way to calibrate items user the 2pl model. Most IRT software can calibrate this model. Here are some that can be used operationally
    IRTPRO
    PARSCALE
    BILOG-MG
     
    You can also estimate the 2pl model in regular statistical software
    R  see irtoys or ltm
    Stata using gllamm Or openirt
     
     
    And you can estimate the 2pl using sem software
    MPLUS
     
    Within and across these applications, there are different estimation methods and procedures.
    Some examples are 
    MML
    CML
    MCMC
     
    Hope this helps! You may also consider whether the 2pl is the best fitting model to your data. Sometimes a more parsimonious model fits or you may even need the 3pl.

    Dec 06, 2012
  • anon

    Hi Harut

     
    Sounds like you are looking for a way to calibrate items user the 2pl model. Most IRT software can calibrate this model. Here are some that can be used operationally
    IRTPRO
    Parscale
    BILOG-MG
     
    You can also estimate the 2pl model in regular statistical software
    R  see irtoys or ltm
    Stata using gllamm Or openirt
     
     
    And you can estimate the 2pl using sem software
    MPLUS
     
    Within and across these applications, there are different estimation methods and procedures.
    Some examples are 
    MML
    CML
    MCMC
     
    Hope this helps! You may also consider whether the 2pl is the best fitting model to your data. Sometimes a more parsimonious model fits or you may even need the 3pl.

    Dec 06, 2012
  • anon

    Dear Harut, 

    Jmetrik is also a great Rasch/IRT program that is in fact open source. You can obtain your estimates through it. 

     

    Victor

    PhD student in Applied Lingusitics and Technology at Iowa State University

    Jan 18, 2014