ISFILE= item structure output file

Do not use this file for anchor values. Instead use IFILE= (becomes IAFILE=) and SFILE= (becomes SAFILE=).

 

Communicating the functioning of a rating scale is challenging, especially if your audience think of its categories as separate and equally-spaced points on the latent variable.

 

If you want to communicate the categories as points, then the best points for the intermediate categories are the locations on the latent variable at which the probability of observing each category is the highest. These are also the points where the expected score on the item is the category value. In Winsteps these are the "AT CAT" measures in the ISFILE= output file.  These points are at infinity for the extreme categories, so Winsteps reports the  measures for expected scores of "lowest category + 0.25" ( = CAT +0.25) and "highest category - 0.25" (= CAT - 0.25).

 

The Rasch-Thurstonian thresholds (50%PRB in ISFILE=) dichotomize the rating scale at each category boundary into 50% probability of being observed below the category and 50% probability of being observed in or above the category.

 

The points on the latent variable where the expected scores are 2.5, etc., are called the CAT-0.5 and TOP-0.5 points in ISFILE=.

 

The Rasch item difficulty (in IFILE=) is the point on the latent variable at which the highest and lowest categories are equally probable.

 

ISFILE=? opens a Browse window

 

ISFILE=filename produces an output file containing the category structure measure (Andrich threshold) information for each item. All measures are added to the corresponding item's calibration and rescaled by USCALE= and UDECIMALS=. This file contains 4 heading lines (unless HLINES=N or ROW1HEADING=N), followed by one line for each item containing:

 

DISFILE=, DISOPTION=Score shows the contents of each ISFILE= category.

 

 

The "AT CAT" values in the ISFILE= are based on the Rasch-model. They are the points on the "expected score" ogive for the rating scale (also called "the model ICC") at which the expected score = the category number. This is also the point at which the probability of observing the category is highest.

 

For extreme categories (top and bottom of the rating scale), the model values are infinite, so an adjustment is made. The "AT CAT" values correspond to expected scores bottom+0.25 score points and top-0.25 score points. These provide reasonable estimates for performance in the extreme categories of the rating scale. The adjustment of 0.25 can be changed with LOWADJ=  and HIADJ=. The "AT CAT" values are plotted on Table 2.2.

 

Since the ISFILE= has the same number of category entries for every item, the repeated fields are filled out with "0" for any further categories up to the maximum categories for any item.

 

When CSV=Y, commas separate the values with quotation marks around the "Item name". When CSV=T, the commas are replaced by tab characters.

 

When STKEEP=YES and there are intermediate null categories, i.e., with no observations, then the Rasch-Andrich threshold into the category is set about 40 logits above the previous threshold. The threshold out of the category, and into the next category, is set about 40 logits above. The exact values depend on the category frequencies of the observed categories. Thus:

 

 

---

 

Meanings of the columns

There are several ways of conceptualizing the category boundaries or thresholds of a rating (or partial credit) scale item. Imagine a rating (or partial credit) scale with categories, 1, 2, 3:

 

From the "expected score ogive", also called the "model item characteristic curve"

 

From the "category probability curves" relative to the origin of the measurement framework (need not be ordered)

 

 

From the "cumulative probability curves" (preferred by L.L.Thurstone) (must be ordered)

 

 

Example 1: You wish to write a file on disk called "ITEMST.FIL" containing the item statistics reported in Table 2.2, for use in constructing your own tables:

ISFILE = ITEMST.FIL

ISGROUPS = 0 ; each item has its own "partial credit" scale

LOWADJ = 0.25 ; the standard for the low end of the rating scale

HIADJ  = 0.25 ; the standard for the high end of the rating scale

 

For column definitions, see above.

 

; ACT ITEM-STRUCTURE FILE (not for anchoring: use SFILE=) FOR LIKING FOR SCIENCE (Wright & Masters p.18) Mar 24 20:55 2015

;ENTRY STAT  MAX  CAT BOT+.25  CAT  ORD  THRESH I+THRSH    S.E. CAT-0.5  AT CAT   PR50%  CAT  ORD  THRESH I+THRSH    S.E. CAT-0.5 TOP-.25   PR50% ACT

     1    1    2    0   -2.68    1    1   -1.03   -1.50     .14   -1.76    -.47   -1.61    2    2    1.03     .55     .13     .81    1.74     .66 Watch birds

     2    1    2    0   -3.01    1    1   -1.03   -1.83     .14   -2.09    -.80   -1.94    2    2    1.03     .23     .13     .49    1.41     .33 Read books on animals

 

Example 2: To produce a Table of expected measures per item-category similar to Pesudovs, K., E. Garamendi, et al. (2004). "The quality of life impact of refractive correction (QIRC) questionnaire: Development and validation." Optometry and Vision Science 81(10): 769-777, write the ISFILE= to Excel. Then delete or hide unwanted columns.

 

Example 3: To plot the operating range of each item using Excel.

 

"Output Files", "ISFILE=", output to Excel.

"Output Files", "IFILE=", output to Excel.

 

Paste the "MEASURE" column from the Excel IFILE= into the Excel ISFILE=

Arrange the columns: TOP-.25, BOT+.25, MEASURE

If some items have fewer categories, then paste their highest values into the TOP-.25 column

Draw a "hi-lo-close" plot with TOP-.25, BOT+.25, MEASURE for each item

 

If some items have fewer categories, then use their highest values in ISFILE=

 

This is one I have drawn from Exam12.txt with Exam12lo.txt+Exam12hi.txt

 

 

Example 4: To produce ISFILE= without the addition of item difficulties (similar to Table 3.2 for a rating scale)

 

We  can do this by writing out the SFILE=sf.txt from the original analysis. Then do the analysis again with SAFILE=sf.txt and IAFILE=*

1-number of items 0

*