Deleting or Ignoring data values, observations or elements

There are several ways to drop, delete, remove or ignore data values, observations or elements.

 

1) Ignore elements or element combinations by modeling with an "M", Missing, model.

This method is most flexible, self-documenting, and easiest to undo. Before:

Models=

?,2,R ; Item 2 is on a rating scale (or partial credit)

?,?,D ; All other items are dichotomies

*

 

After:

1,3,M ; Ignore response by person 1 to item 3

4,?,M ; Ignore all responses by person 4

?,2,R ; Item 2 is on a rating scale (or partial credit)

?,?,D ; All other items are dichotomies

*

 

2) Delete the datum from the Data= file.

For single value references:

Before:

Data=

1,2,1

1,3,2 ; Datum to be deleted

1,4,3

 

After:

Data=

1,2,1

1,4,3

 

For multiple-value references:

Before:

Data=

1,2-4,1,2,3 ; Datum values of 2 is to be deleted

 

After:

Data=

1,2-4,1,,3 ; Blank data values are ignored

 

3) Comment out the datum value

For single value references:

Before:

Data=

1,2,1

1,3,2 ; Datum to be deleted

1,4,3

 

After:

Data=

1,2,1

; 1,3,2 ; this datum is ignored

1,4,3

 

For multiple-value references:

Before:

Data=

1,2-4,1,2,3 ; all 3 data values are to be deleted

 

After:

Data=

; 1,2-4,1,2,3 ; all 3 data values are ignored

 

4) To ignore all data values that include a specific element, delete or comment out that element number in the Labels= specification, e.g.,

Labels=

1, Students

1=George

; 2=Mary ; all elements referencing element 2 in facet 1 will be ignored

3=John

 

4B) You can keep highly misfitting elements (persons, items, raters, etc.) in the analysis, but minimize their impact on other elements by giving them a very small weight:

Labels=

1, Students

1-4000   ; 4,000 students

37 = , , , 0.001  ; student 37 has a very small weight, so is effectively deleted

 

5) To ignore facet element references, BUT STILL VALIDATE ELEMENT REFERENCES, leave the facet control character blank.

Model=?,,?,D ; ignore facet element references for facet 2

Data=

4,5,6,1 ; analyzed as 4,0,6,1

 

6) To ignore an entire facet in the data, you can use the Entered= specification.

Entry = 1,0,2 ; the second facet in the data is to be ignored completely. No element number validation occurs.

Model=?,?,D ; model only the two active facets (1st and 3rd)

Data=

4,5,6,1 ; analyzed as 4,6,1

 

or use the "X" facet control character:

Model=?,X,?,D

 

7) To ignore every occurrence of particular response values in the data, you can use Missing=, e.g., data values "9" and "." are to be treated as missing data:

Missing = 9,. ; all 9's and periods are treated as missing.

 

8) To ignore every occurrence of particular response values for a particular Model= specification, e.g, value "3" is missing for rating scale (or partial credit) "Speed".

Model = ?, ?, Speed

*

Rating (or partial credit) scale = Speed, R6

-1=missing,,,3 ; recode 3 as -1, the missing value code.

 

9) To remove all misfitting observations, create a new data file:

1. In your current Facets analysis, output the residual file to Excel from the Facets "Output Files" menu.

2. Sort the Excel file by residual or standardized residual.

3. Delete the residuals you don't want at the top and bottom of the Excel file.

4. Delete all the columns except Observation and the Element numbers

5. Move the observation column after the element number columns

6. Export the Excel file as a text file: gooddata.txt

7. Make a copy of your Facets specification file as backup.

8. Change Data= to Data=gooddata.txt

9. If there is any data in the specification file, delete it.

10. If there are Dvalues= or Delements= in the specification file, delete them. Other changes may be needed.

11. Run Facets with the revised specification file.