I have only the most general theoretical familiarity with IRT and Rasch models. However, my class files from an excellent instructor address the topic on p.19, If I am not misinterpreting. As a novice, I am interested in learning more about the subtleties between the two methods.

http://web.pdx.edu/~newsomj/pmclass/item%20response%20theory%20and%20rasch%20models%20(full).pdf

The accuracy will depend on (1) how well the Rasch model fits your data and (2) how large the standard errors of the model (i.e. item difficulty) parameters are. Assuming the model is correct for the data, so that (1) holds true, the standard errors will be mostly driven by the sample size--the larger the sample, the smaller the standard errors and the more accurate the prediction made based on the item parameter estimates.

The first point can be examined by looking at tests of model fit such as a chi-square goodness of fit test and other indicators of model fit such as residuals, item fit measures, and/or person fit measures etc. The second point can be examined by looking at the standard errors of the item parameter estimates.

Dear Lawrence Avillano

It depends, Rasch creates linear measurements to assess real intervals of item difficulty and human ability. Item values are calibrated in this procedure, and human skills are tested on a common continuum that accounts for the latent feature.

Rasch modeling, as discussed by Fox and Jones, allows for generalizability across samples and items, accounts for the fact that response options may not be psychologically equally spaced, allows for testing of unidimensionality, produces an ordered set of items, and identifies poorly functioning items as well as unexpected items.

References:

Bruce E. Wampold in "Necessary (but not sufficient) innovation: comment on Fox & Jones", Journal of Counseling Psychology, 1998 45:1 46-49.

Dawis RV. 1987. Scale construction. ibid. 34, 481-489.

Fox CM, Jones JA. 1998. Uses of Rasch modeling in counseling psychology research. ibid. 45:1, 30-45.

Wampold BE The Promising Advantages of Rasch … Rasch Measurement Transactions, 1999, 13:2 p. 695

Kind Regards

Qamar Ul Islam

Hi Lawrence and folks.

Accuracy of the Rasch model depends on the standard error of the responses of each person and the standard error of the difficulty of each item.It is different than classical test theory, the standard error is general for all the persons and for all the items, despite it is possible to obtain a particular std error using classical statistics it is not usual.

On the other side, Christian said (1) how well the Rasch model fits your data, in fact it is the contrary, you must say: how well the data fits the Rasch model. The difference is that the Rasch model does not pretend to fit your data, if you need this you should use Item Response Theory and other logistic models whose purpose is to get the better fit to your data, assuming that data represents the reality and the model should describe the data. In this case you are not intending to get some food measurements but describe your data.

In the Rasch model, the idea is that a set of items may be useful to measurement if the data correspond to the properties defined by the Rasch model. In this case the model is "fixed" and if your data are able to work as the model says, that you are doing a good measurement.

The "correct" answers are sufficient (and then accurate) to obtain the Rasch parameters (difficulty of the items and measures of the persons), this is different for other IRT models.