This permits the replacement of data values in your data file with other values, without altering the data file. Data values are in the original data file format, specified in CODES=. If specified as decimals, they are rounded to the nearest integers. Additional persons can be entered beyond those in DATA= or after END LABELS. Values can be in any order, but are more speedily process in person-entry order.

 

 

Suggestion: sort the data lines as "person-entry-number ascending" for faster processing.

 

Its format is:

Ranges are permitted for entry numbers: first-last.

 

Person and item selections must be in quotation marks " ", and follow the selection rules:

 

 

Example 1: In your MCQ test, you wish to correct a data-entry error. Person 23 responded to item 17 with a D, not whatever is in the data file.

 

EDFILE=*

23 17 D     ; person 23, item 17, data value of D

*

 

Example 2: Person 43 failed to read the attitude survey instructions correctly for items 32-56. Mark these missing.

 

EDFILE=* 

43 32-56 " " ; person 43, items 32 to 56, blanks are missing data.

*

 

Example 3: Persons 47-84 are to be given a rating of 4 on item 16.

 

EDFILE=* 

47-84 16 4 ; persons 47 to 84, item 16, data value of 4

*

 

Example 4: Items 1-10 are all to be assigned a datum of 1 for the control sub-sample, persons 345-682.

 

EDFILE=* 

345-682 1-10 1 ; persons 345-682, items 1 to 10, data value 1.

*

 

Example 5: Missing data values are to be imputed with the values nearest to their expectations.

 

a. Produce PFILE=, IFILE= and SFILE= from the original data (with missing).

b. Use those as PAFILE=, IAFILE=, SAFILE= anchor files with a data set in which all the original non-missing data are made missing, and vice-versa - it doesn't matter what non-missing value is used.

c. Produce XFILE= to obtain a list of the expected values of the originally missing data.

d. Use the EDFILE= command to impute those values back into the data file. It will round expected values to the nearest integer, for us as a category value.

 

EDFILE=*

17 6 2.6 ; persons 17, item 6, expected value 2.6, imputed as category "3".

*

 

Example 6: All responses to item 6 for males "M" in column 6 of person label are to be coded as "missing", character ".":

 

EDFILE=*

"?????M" 6 .

*

 

Example 7: We want to do Examples 1, 2, 3, 4, 5, 6 all at once to our dataset:

 

EDFILE=*

23 17 D     ; person 23, item 17, data value of D

43 32-56 " " ; person 43, items 32 to 56, blanks are missing data.

47-84 16 4 ; persons 47 to 84, item 16, data value of 4

345-682 1-10 1 ; persons 345-682, items 1 to 10, data value 1.

17 6 2.6 ; persons 17, item 6, expected value 2.6, imputed as category "3".

"?????M" 6 .

*

 

Example 8: We want person group X (in column 4 of the person label) except for subgroups 11, 24 (in columns 6, 7 of the person label):

PSELECT = "???X?{12}{14}"  ; this selects X and 11, 14, 21, 24

EDFILE=*

"?????14" "?" . ; convert 14 to missing data

"?????21" "?" . ; convert 21 to missing data

*

 

Example 9: Exceedingly unexpected responses are to be coded "missing". (It is easier, but not as exact, to use CUTLO= and CUTHI= to trim the observations).

 

Either
Extract into Excel the list of unexpected observations from Table 6.6 or Table 10.6.

Or
Output the XFILE= to Excel
Sort by unexpectedness (standardized residual)
Delete all rows except for the responses you want to code missing

 

Then
Rearrange the columns: Person Item
In the third column put the desired missing data code.
Copy-and-paste the three columns into a text file.
In your Winsteps control file:
EDFILE = (text file name)
Rerun the analysis

 

Example 9: All data in a separate EDFILE= data file.

 

 

Example 10: Item bank recalibration for computer-adaptive tests (CAT) or similar.

Here is a method is that maintains the accuracy of previously-reported person measures as much as possible:

1.collect up all the relevant data and format the data into a rectangular dataset or equivalent. EDFILE= is useful for this.

2.anchor all the persons at their report measures

3.anchor all items at their item-bank difficulties, and rating-scale structures (if polytomies) SAFILE=

4.analyze the dataset

5.the item displacements tell us which items have drifted by how much.

6.items with displacements of more than 0.5 logits, that are also bigger than the item S.E.s, are candidates for recalibration.