ResearchTopics Publications In progress publications Communications
PackagesAnaQol Project PRO-online R Packages Online R-package
Life of the unitProjects Collaborations PhD thesis Traineeships Traineeships propositions Seminars
21 janvier 2019
20 janvier 2019
07 janvier 2019
02 janvier 2019
01 décembre 2018
Updated23 janvier 2019
The Stata module "Backrasch"
Backrasch realizes a backward procedure on a Rasch model: the items are removed one per one if they have a bad fit to the Rasch model. The fit of the items is valuated by a first-order statistics (test R1c, R1m or Q1). It is possible to build several sub-scales of items, the second sub-scale is build with the items unselected in the first sub-scales, the third one with the items unselected in the two first sub-scales, and so on... By default, the parameters of the Rasch model are estimated by conditional maximum likelihood (CML), but it is possible to estimate them by marginal maximum likelihood (MML) or generalized estimating equations (GEE).
Type "findit backrasch" or "ssc install backrasch" directly from your Stata browser.
Syntax (version 2)
backrasch varlist [, method(cml/mml/gee) test(R/Q) p(#) nbscales(#) nodetail noautogroup ]
This program requires an access to the following program(s):
- method(cml/mml/gee): defines the method of estimation of the difficulty parameters among conditional maximum likelihood (cml - by default), marginal maximum likelihood (mml) or generalized estimating equations (gee).
- test(R/Q): defines the first order statistics to use between R-type test (R1c or R1m - by default) or the Q1 test of Van den Wollenberg (Q)
- p(#): defines the level of signification who define a significantly bad fitted item (0.05 by default)
- nbscales(#): defines the maximal number of sub-scales to build. By default, the program builds only one sub-scale
- nodetail: does not display the description of the algorithm
- noautogroup: forces the program to compute the first order fit statistics with the groups defined by the value of the score. by default, the scores are grouped to obtained groups of 30 individuals or more
backrasch itemA* itemB*, p(0.2) method(mml) nodetail
backrasch itemA* itemB1-itemB7 , p(0.1) nbsc(5) noautog