Subtest scoring and weighting

A test or protocol may consist of a series of subtests. You want each person to have a separate measure on each subtest, but in the same frame of reference as the overall test measures.

 

First, code each item in its item label with what subtest it belongs to.

 

A. Using DPF.

 

1. Analyze all the data together.

 

2. Reporting of subtests.

Reporting subtest items: In an overall analysis, the items of individual subtests can be reported after applying ISELECT= from the Specification pull-down box.

 

Reporting subtest person measures: a measure for each person on each subtest can be obtained by specifying the subtest identification column in the item label in DPF= and producing Table 31 or the DPF plot.

 

B. Using separate analyses.

 

1. Analyze all the data together.  Produce an IFILE=if.txt. Produce a PFILE= for global measures.

 

2. Anchor all items IAFILE=if.txt.

Select one subtest: ISELECT=subtest code

Run Winsteps

Produce IFILE= and PFILE for the subtest.

 

3. Repeat for the other subtests.

 

---

 

Subtest Weighting

 

The different subsections have different maximum scores, but you want each subsection to be equally influential in estimating the final measures.

 

1) Do a standard Rasch analysis to investigate coding problems, misfit, reliability, etc.

 

2) Weight the items in the different sections so that the overall weights for each section are the same.

Let's suppose that the items are all dichotomies scored 0,1, then

 

IWEIGHT=*

1-40 1 ; the 40 items in section 1 are given a weight of 1, so weight of section 1 = 40*1 = 40

41-70 1.333 ; the 30 items in section 2 are given a weight of 1.333, so weight of section 2 = 30*1.333 = 39.99

71-120 .8 ; the 50 items in section 3 are given a weight of 0.8, so weight of section 3 = 50*0.8 = 40

*

 

We want to maintain close to the original overall weight, so that the final reliabilities and standard error have reasonable values.

 

3) Perform the weighted analysis to report the final person measures.

 

There are other effective approaches. For instance, perform a separate Rasch analysis for each section, and then combine the subtest measures using www.rasch.org/rmt/rmt83f.htm:

 

combined measure = sum(section measure/S.E.(section measure))/sum(1/S.E.(section measure))