Rasch model is one of the psychometric models. It helps us create measures. The most famous
application of Rasch model can be found in large education data sets, such as TIMSS and NELS, in the forms of student academic
achievement scores. Rasch model can also create attitudinal scales, such as student engagement level
or liking of math. We all need to be aware that Rasch model can get us good scales only when we
design test items and surveys carefully.
My way of Understanding Rasch Model
3 minute Rasch explanation
to someone who at least knows what OLS regression is. Click here.
My Rasch Essay--I show how you may also experience "aha" moment if you see how Rasch model
is related to logistic regression or HLM (actually HGLM or multilevel logistic regresssion).
My Rasch Essay--Do you understand the connection between the Rasch equation and the graphic that carrys a logistic curve? I didn't.
Now I do. Plead read on if you don't.
See my attempts to understand Rasch Model here. I borrowed the idea of Professor Aki Kamata that says Rasch model
can be replicated by HLM. See his work Kamata, A. (2001). Item analysis by the hierarchical generalized linear model.
Journal of Educational Measurement, 38, 79-93. Any mistakes in this presentation file, however, is mine. Download POWERPOINT presentation file (123k), almost the same as above file.
Professor Cook answering
my qustion about why perfect data (a person getting all answers correctly or all wrongly) doesn't
produce valid estimates and he proposes a solution. He wrote, " I looked into the questions
you raised using the attached Excel file to reanalyze John's data in your Appendix 1. I conclude that some statistical
information and a multiplication factor Beta are needed to solve the problem within the logit model framework. All are
described in the attachment. The general logit model does have such a multiplication factor."
One way to strengthen your
understanding of Rasch model: Study a residual file from Winsteps (by specifying xfile=filename) and try to confirm how each
column of information is derived (and understand how each variable in this file is interrelated). Read the description here. You can use these materials to do this mini-project.
sample xfile (residual file) ... I think in the current version ITMMEA is wrong. It is supposed to be item difficulty measure, but it is showing
the same value as MDIFEE (I wrote to Mike about this).
My macro to apply factor analysis based on tetra something correlation matrix. (Please let me know if you have a case where
you succeed using this macro.) The program also creates a correlation matrix (in Excel), which may be also interesting to look at.
Use Mplus to replicate some models and compare model to evaluate model fit, etc. (Currently I am learning it slowly.)
