Do not specify MODELS= unless you intend to use the "S" , "F" or "C" models.
|
MODELS =
|
Winsteps chooses the model-family based on MODELS=. Models=R (or MODELS= is omitted) is the rating-scale family of models which includes the Andrich Rating-Scale Model, the Masters Partial-Credit Model, the Grouped Rating-Scale Model and the Rasch Dichotomous Model.
|
MODELS= *file name
|
file containing details
|
MODELS = *
|
in-line list. See Example 2.
|
MODELS = codes
|
codes for item groups, R (standard). S (success), F (failure), D (dichotomy), C (consecutive dichotomy). Items are assigned to the model for which the serial location in the MODELS= string matches the item sequence number. When only one letter is specified with MODELS=, e.g., MODELS=R, all items are analyzed using that model.
|
MODELS = D
|
Rasch dichotomous model. Dichotomous data. Items with only two response categories are analyzed using the Rasch dichotomous model, regardless of what model is specified unless they are grouped with a polytomous item.
|
MODELS = R
|
ISGROUPS= omitted, Andrich "Rating Scale" model
ISGROUPS=0, Masters' "Partial Credit" model
ISGROUPS=11222 or something else, the Grouped Rating Scale Model
Andrich D. (1978) A rating scale formulation for ordered response categories. Psychometrika, 43, 561-573.
Masters G.N. (1982) A Rasch model for partial credit scoring. Psychometrika, 47, 149-174.
Linacre J.M. (2000) Comparing and Choosing between "Partial Credit Models" (PCM) and "Rating Scale Models" (RSM). www.rasch.org/rmt/rmt143k.htm
|
MODELS = S
|
uses the Rasch dichotomous model and the Glas-Verhelst "Success" (growth) model, also called the "Steps" Model (Verhelst, Glas, de Vries, 1997). If and only if the person succeeds on the first category, another category is offered until the person fails, or the categories are exhausted, e.g. an arithmetic item, on which a person is first rated on success on addition, then, if successful, on multiplication, then, if successful, on division etc. "Scaffolded" items can function this way. This is a continuation ratio model parameterized as a Rasch model with missing data on unreached categories. Verhelst N.D., Glas C.A.W. & De Vries H.H. (1997) A Steps model to analyze partial credit. In W.J. van der Linden & R.K. Hambleton (Eds.), Handbook of modern item response theory (pp. 123 - 138) New York: Springer
|
MODELS = F
|
uses the Rasch dichotomous model and the Linacre "Failure" (mastery) model. If a person succeeds on the first category, top rating is given and no further categories are offered. On failure, the next lower category is administered until success is achieved, or categories are exhausted. This is a continuation ratio model parameterized as a Rasch model with missing data on unreached categories. The Success and Failure model computations were revised at Winsteps version 3.36, August 2002.
Recommendation: Instead of an F-type polytomy, model each step to be a dichotomy: 1=succeeded, 0=failed, missing=not reached. This facilitates much more powerful diagnosis of the functioning of the failure process.
Linacre J.M. (1991) Beyond Partial Credit: Rasch Success and Failure Models. www.rasch.org/rmt/rmt52j.htm
|
MODELS = C
|
MSD: the rating scale is analyzed as consecutive dichotomies with one dichotomy for each category boundary. Bradley C., Massof R. (2018) Method of successive dichotomizations: An improved method for estimating measures of latent variables from rating scale data. PLoS One.
|
MSD replaces the rating scale with a set of dichotomous items. This was one of the ways that rating scales were analyzed before David Andrich devised his model in 1977. MSD has recently been reintroduced because of its conveniences. An effect of the MSD dichotomization is to widen the logit range of the measures. In the example analysis below , the Andrich range is 1.4 logits, but the MSD range is 3.2 logits. So, if you are using a rule such as "DIF begins at 0.5 logits" then more instances of DIF will be reported with MSD than with the Andrich model. Accordingly, we need to adjust the DIF criteria to match the new situation.
Suggestion: identify DIF in a dataset with the Andrich model. Reanalyze with MSD. You will now be able to construct DIF criteria for the MSD situation which match the Andrich situation. If you do this, please do write a short research note about this for publication in Rasch Measurement Transactions or inclusion in Winsteps Help and, of course, there are many Journals where you could publish a larger paper about this.
Please look at the output in Table 3.2, 3.3, .... to see exactly how the data have been modeled.
Use one character per item with no blanks
NI=8
MODELS='RSRFRSRR' ; this also forces ISGROUPS=0 to be the default
When XWIDE=2, you can also use one character per XWIDE with blanks,
NI=8
XWIDE=2
MODELS=' R S R F R S R R' ; this also forces ISGROUPS=0 to be the default
Example 1: All items are to be modeled with the "Success" model.
MODELS=S ; the Success model
Example 2: A competency test consists of 3 success items followed by 2 failure items and then 10 dichotomies. The dichotomies are to be reported as one grouping.
NI=15 fifteen items
MODELS=SSSFFRRRRRRRRRR ; matching models: ; forces ISGROUPS=0 to be the default
ISGROUPS=000001111111111 ; dichotomies grouped: overriding the default ISGROUPS=0
or
MODELS=*
1-3 S
4 F
5 F
6-15 R
*
MODELS= Demonstration
Score on item
|
RSM Andrich thresholds
|
RSM Thurstone thresholds
|
Consecutive Dichotomy thresholds
|
Success Model thresholds
|
Failure Model thresholds
|
5
|
.00
|
.68
|
1.61
|
.00
|
1.61
|
4
|
.00
|
.24
|
.69
|
-.69
|
1.38
|
3
|
.00
|
.00
|
.00
|
-1.10
|
1.10
|
2
|
.00
|
-.24
|
-.69
|
-1.38
|
.69
|
1
|
.00
|
-.68
|
-1.61
|
-1.61
|
.00
|
0
|
-
|
|
-
|
-
|
-
|
This control and data file demonstrates that the C, S and F models produce the same Andrich thresholds as the equivalent sets of dichotomous items. The standard errors and fit statistics differ because the sets of dichotomous items are treated as independent, but the rating scale categories are dependent.
TITLE = MODELS= Demonstration
ITEM1 = 1
NI=19
NAME1 = 21 ; the score on the item or item set
NAMELEN = 1
CODES=012345
ISGROUPS=0 ; each item has its own rating or other scale
; MODELS=
; R = RATING SCALE MODEL AND SIMPLE DICHOTOMIES
; C = CONSECUTIVE DICHOTOMIZATION
; S = SUCCESS MODEL
; F = FAILURE MODEL
MODELS=RCRRRRRSRRRRRFRRRRR
OFILE=RCSF.txt
PAFILE=*
1-6 0 ; anchor at zero to avoid item interactions
*
TFILE=*
14.1 ; item difficulties
3.2+ ; rating scales
*
CMATRIX = No ; omit confusiion matrix
&END
1. RATING-SCALE MODEL
2. CONSECUTIVE DICHOTOMIZATION
3. 5 CONSECUTIVE DICHOTOMY
4. 4 CONSECUTIVE DICHOTOMY
5. 3 CONSECUTIVE DICHOTOMY
6. 2 CONSECUTIVE DICHOTOMY
7. 1 CONSECUTIVE DICHOTOMY
8. SUCCESS MODEL
9. 5 SUCCESS DICHOTOMY
10. 4 SUCCESS DICHOTOMY
11. 3 SUCCESS DICHOTOMY
12. 2 SUCCESS DICHOTOMY
13. 1 SUCCESS DICHOTOMY
14. FAILURE MODEL
15. 5 FAILURE DICHOTOMY
16. 4 FAILURE DICHOTOMY
17. 3 FAILURE DICHOTOMY
18. 2 FAILURE DICHOTOMY
19. 1 FAILURE DICHOTOMY
END LABELS
551111151111151.... 5
4411110411110401... 4
331110031110.3001.. 3
22110002110..20001. 2
1110000110...100001 1
000000000....000000 0
