simul_confusion/Modules/ado/personal/r/raschfit.hlp

{smcl}
{* 208December2005}{...}
{hline}
help for {hi:raschfit}{right:Jean-Benoit Hardouin}
{hline}

{title:The Raschfit procedure}

{p 8 14 2}{cmd:raschfit} {it:varlist} {cmd:,} [{cmdab:ker:nel}({it:#}) {cmdab:nbsc:ales}({it:#}) {cmdab:items:order}({it:keyword}) {cmdab:nofast}]

{title:Description}

{p 4 8 2}{cmd:raschfit} realizes the Raschfit algorithm defined by Hardouin and
Mesbah (2004). This method selects sub-scales of items which fit a
Rasch model. The method begin with a kernel of items (two or more items)
defined by the user. At each step, the method uses a new item and verifies if this
new item is explained by the same latent trait than the already selected items.
If not, the items is not selected. The former version of the Raschfit algorithm
is based on the comparison of two model: A Rasch model and a Multidimensional 
Marginally Sufficient Rasch Model (MMSRM). These two models are compared by the
Akaike Information Criterion (AIC). A faster version of the algorithm (Raschfit-Fast)
compares the Rasch model and an adapted version of this model where the response 
to the "new" item is not explained by the latent trait. Raschfit-Fast is executed
by default.

{title:Options}

{p 4 8 2}{cmd:kernel}({it:#}) defines the # first items of {it:varlist} as the kernel 
of the first sub-scale (by default with {cmd:itemsorder}({it:order}), this number is 
fixed to 2).

{p 4 8 2}{cmd:nbscales}({it:#}) defines the maximal number of sub-scales (by default,
only one sub-scale is selected).

{p 4 8 2}{cmd:itemsorder}({it:keyword}) defines the order of the items. If you type 
{it:order}, the kernel is composed of the first items defined in {it:varlist},
and the items are tested in the same order than in {it:varlist}. 
If you type {it:msp} or {it:mspinv}, a Mokken Scale Procedure is run under 
the items (the {it:msp} and {it:loevH} Stata programs are necessary) and the 
items are selected from the first order selected by this procedure to the last 
one (with {it:msp}), or in the inverse order (with {it:mspinv}). The method {it:msp} 
is generaly more robust, but is longer to run. By default, the program uses {it:msp}.

{p 4 8 2}{cmd:nofast} allows to execute the former version of the algorithm.

{title:Example}

{p 4 8 2}{cmd:. raschfit itemA* , itemsorder(order)}

{p 4 8 2}{cmd:. raschfit itemA1-itemA7 , itemsorder(msp) kernel(4) nofast}

{p 4 8 2}{cmd:. raschfit item* }

{title:References}

{p 4 8 2} Hardouin J.-B. and Mesbah M. {it:Clustering binary variables in subscales using an extended Rasch model and Akaike Information Criterion}, Communication in Statistics <20> Theory and methods}, {cmd:33}(6), pp. 1277-1294, 2004

{p 4 8 2} Hardouin J.-B. {it:Construction d'<27>chelles d'items unidimensionnelles en qualit<69> de vie (Item selection in unidimensional scale applied to the Quality of Life)}, PhD thesis of the University Ren<65> Descartes - Paris 5, 
France, 201 pp, 2005

{p 4 8 2} Hardouin J.-B. and Mesbah M. {it:Fast Clustering of Binary Variables in Subscales}, Unpublished document, 2005.

{title:Author}

{p 4 8 2} Jean-Benoit Hardouin, Regional Health Observatory (ORS) - 1, rue Porte
Madeleine - BP 2439 - 45032 Orleans Cedex 1 - France. You can contact the
author at
{browse "mailto:jean-benoit.hardouin@orscentre.org":jean-benoit.hardouin@orscentre.org}
and visit the websites {browse "http://anaqol.free.fr":AnaQol}
and {browse "http://freeirt.free.fr":FreeIRT}