Estimation of the parameters of undimensional and bidimensional IRT models
geekel2d varlist [, coef(matrixname) nbit(#) critconv(#) ll quad(#) novar]
varlist is a list of two existing dichotomous variables or more.
geekel2d estimates, by Generalized Estimating Equations (GEE), the parameters of the model defined by Kelderman (1994) with one or two dimensions and dichotomic items. This model includes the Rasch model and the One Parameter Logistic Model (OPLM) for the unidimensional models, the Multidimensional Generalized Rasch Model (MGRM) and the Multidimensional Completely Sufficient Rasch Model (MMSRM) for the two-dimensional models.
coef is the name of a matrix which contains the coeficients B. This matrix relies the items and the latent traits. Each row represents an item and there is as many colmuns than the supposed number of latent traits (one or two). The coefficients are choosen, in general, among the first intergers, but geekel2d allows using real coefficients. By default, the Rasch model is supposed (the matrix coef is a vector of 1).
nbit defines the maximal number of iterations in the estimation algorithm. By default, this number is fixed to 30.
critconv is the value of the convergence criterion, calculated as the square of the cross-product of the vector containing the difference between two successive iterations of the parameters estimations. By default, this criterion is fixed to 1e-15.
ll estimates the marginal log-likelihood and the Akaike Information Criterion (AIC) by Gauss-Hermite quadratures.
quad defines the number of nodes of quadratures.
novar avoids to compute the standards errors of the estimators (faster).
For detailed informations on the Kelderman model, see Kelderman and Rijkes (1994) or Adams and al. (1997).
geekel2d don't allows using of polytomous items.
The ghquadm Stata module is needed (use findit ghquadm to obtain it).
. geekel2d item1 item2 item3 item4 /*Rasch model*/
. matrix B=(1,0\1,0\0,1\0,1)
. geekel2d item1 item2 item3 item4 , coef(B) nbit(50) critconv(1e-30)
Kelderman H. and Rijkes C. P. M., Loglinear multidimensional IRT models for polytomously scored items. Psychometrika, 1994, 59, 149-176.
Adams R. J., Wilson M. R. and Wang W., The multidimensional random coefficient multinomial logit model. Applied Psychological Measurement, 1997, 21, 1-23.
Jean-Benoit Hardouin, Regional Health Observatory (ORS) - 1, rue Porte Madeleine - BP 2439 - 45032 Orleans Cedex 1 - France. You can contact the author at firstname.lastname@example.org and visit the