IRT Modeling Lab

B. Data Preparation for IRT Analyses

Before we can analyze the data, we must make sure that the data is formatted properly for IRT programs such as BILOG or MULTILOG. This is usually done using a statistical program such as SPSS, SYSTAT, SAS or other similar programs. In this tutorial, we will primarily be using SPSS to format the data; although other statistical programs may also be used. The data must eventually be saved in ASCII format as a *.DAT file in order for the IRT programs to access it. We describe the four basic steps in data preparation below.

In order to illustrate the tutorial, we have provided a dataset (raw.sav) for you to prepare. It is a SPSS data file. This data set consists of 20 items with 5 options (1 to 5) and a sample of 3000 subjects. 10 of the items are Agreeableness items (a1 to a10) and 10 of the items are Conscientiousness items (c1 to c10). Below is an example of what the items may look like in a survey.
  • Are you Agreeable?
    Disagree Strongly   Disagree   Neither   Agree   Agree Strongly  
We have also attached the file which contains the complete syntax (syn1.SPS) used for data preperation.

4 Basic Steps to Data Preparation for IRT Analyses

  1. Reverse scoring
    Items in a scale may sometime be scored in a reversed order from other items in a scale. IRT analysis requires that all items be scored in a single direction.

  2. Dichotomization (optional)
    Under some circumstances you may want to have data which is scored dichotomously.

  3. Splitting data into calibration and validation samples
    The data must be split into a calibration sample from which the IRT program will estimate item parameters. The validation sample is used to assess the fit of the model to the data.

  4. Write subscales to separate files
    The data is separated into separate *.DAT files for the IRT program to analyze.
Back To Main