IRT Modeling Lab

3. Splitting Data Into Calibration and Validation Samples


When performing IRT analysis, it is desirable to cross-validate your data by splitting the data into a calibration sample and a validation sample. The 'Calibration sample' is used for estimating item parameters. The 'Validation sample' is used for checking model-data fit by looking at empirical fit plots and chi-squares.

In order to split your data into the calibration and validation samples, we have used the SPSS program to assign a value of 1 to all even numbered records and a 2 to each odd numbered record. Other strategies for splitting data may also be used. Below is the SPSS syntax used to number the records in the polytomous data set (raw_reversed.sav).

SPSS Syntax:
  * Write subscales to separate files, calibration and validation samples separately.

* Polytomous file.
GET
FILE='C:\IRTtutorial\raw_reversed.SAV'.
EXECUTE .

*Create split variable.
COMPUTE SNO = $CASENUM.
Execute.

*mark all odd numbers as 1.
IF (MOD(sno,2)) split = 1 .
EXECUTE .

*mark all even numbers as 2.
RECODE
split (SYSMIS=2) .
EXECUTE .

SAVE OUTFILE='C:\IRTtutorial\raw_reversed.SAV'.
Execute.
 		 
 
Continue to next step