If SFILE=filename is specified, a file is output which contains the item and category information needed for anchoring structures. It has 4 heading lines (unless HLINES=N or ROW1HEADING=N), and has the format:
Example for dichotomous items:
; STRUCTURE-THRESHOLD MEASURE ANCHOR FILE
; CATEGORY Rasch-Andrich threshold
1 .00 ; this is a dummy threshold to tell Winsteps that 0 is the bottom category
2 .00
Example for the Andrich Rating Scale model:
; STRUCTURE-THRESHOLD MEASURE ANCHOR FILE
; CATEGORY Rasch-Andrich threshold
0 .00 ; this is a dummy threshold to tell Winsteps that 0 is the bottom category
1 -.86
2 .86
Example for the Partial Credit model with ISGROUPS= 0
STRUCTURE-THRESHOLD MEASURE ANCHOR FILE
; ACT CATEGORY Rasch-Andrich threshold MEASURE
1 0 .00 ; this is a dummy threshold to tell Winsteps that 0 is the bottom category
1 1 -1.64
1 2 1.64
2 0 .00 ; this is a dummy threshold to tell Winsteps that 0 is the bottom category
2 1 -.30
2 2 .30
3 0 .00 ; this is a dummy threshold to tell Winsteps that 0 is the bottom category
3 1 -1.13
3 2 1.13
....
Example with CMLE=Yes with Rating Scale Model.
; STRUCTURE-THRESHOLD MEASURE ANCHOR FILE
; CATEGORY Rasch-Andrich threshold
0 .00 ; .00 ; CMLE
1 -.86 ; -.80
2 .86 ; .80
......
JMLE threshold estimates are shown in the second column.
CMLE threshold estimates are shown in the third column. CMLE estimates are usually slightly more central than JMLE estimates.
Example with CMLE= Yes and null (unobserved) category with ISGROUPS= 0
; STRUCTURE-THRESHOLD MEASURE ANCHOR FILE FOR ....
; ACT CATEGORY Rasch-Andrich threshold MEASURE
1 0 .00 ; .00 ; CMLE
1 1 -3.94 ; -3.62
1 2 -.92 ; -.74
1 3 3.65 ; 3.65
1 4 40.74 ; 40.67 ; NULL
1 5 -39.52 ; -39.96
Columns:
1. The item sequence number (I6), if required.
Shown only if there are multiple groupings in ISGROUPS= or SAITEM=Yes, otherwise this is omitted.
2. The category number (I3) (STRU) - the higher numbered of the pair of categories on each side of the Andrich threshold.
3. Andrich threshold value (structure calibration) (F7.2) (user-rescaled by USCALE=) (number of decimals by UDECIM=) (MEASURE)
Use this for anchoring, together with IFILE=
If CSV=Y, these values are separated by commas. When CSV=T, the commas are replaced by tab characters.
SFILE=? opens a Browse window
Andrich Thresholds for Unobserved Categories
When STKEEP=YES and there are intermediate null categories, i.e., with no observations, then the Rasch-Andrich thresholds are infinite. To approximate infinite Andrich thresholds in computer arithmetic, the Andrich threshold into the unobserved category is set 40 logits above the highest threshold calibration. The Rasch-Andrich threshold out of the unobserved category, and into the next category, is set down by the same amount. Thus:
Category Structure Calibration
Rasch-Andrich Threshold
|
Category
|
Table 3.2
|
In SFILE
|
0
|
NULL
|
0
|
1
|
-3
|
-3
|
2
|
NULL
|
40 + 2 = 42
|
3
|
1
|
1 - 40 - 2 = -41
|
4
|
2
|
2
|
TOTAL:
|
0
|
0
|
This is also an attempt to model unobserved categories for the next time, when they may be observed.
If categories will not be observed next time, then please specify STKEEP=NO, which automatically drops the category from the rating (or partial credit) scale.
If categories may be observed next time, then it is better to include a dummy data record in your data file which includes an observation of the missing category, and reasonable values for all the other item responses that accord with that missing category. This one data record will have minimal impact on the rest of the analysis.
Example for high-stakes examination:
We have a 12-category rating scale, 1-12, and are using the Partial Credit Model. Some items were not observed in the low categories (1-5). Consequently a standard PCM analysis treats these as items with a shorter rating scale. We don't want this. We want every item to have the full range of categories.
We apply some Bayesian-style thinking: There must be some severe cases somewhere who we may meet later, so let's include them now. An approach is to add two dummy data records that look like this:
12121212......12 dummy-1
21212121......21 dummy-2
give them a small weight so they don't mess up other statistics:
IWEIGHT=*
entry-number-for-dummy-1 .001
entry-number-for-dummy-2 .001
*
and keep unobserved intermediate categories:
STKEEP=Yes
Now, every item will be reported with the full range of categories 1-12. The easy items (originally without the low categories) will be reported correctly as the easiest items.
If we want to rerun the analysis without the dummy persons, then, from the analysis with the dummy persons:
SFILE= thresholds.txt
Remove the dummy persons:
SAFILE= threshold.txt ; anchor the thresholds
and reanalyze everything.
