F2. DIF/DTF questions

lITERLINK questions

Q1. Is there an easy way to feed required information into ITERLINK?

A1: If you are familiar with MS-DOS, you may compose a DOS batch file (.BAT) and a run file (.RUN) to feed required information into ITERLINK. The batch file (a text file) may look like this:

ITERLINK < DIF.RUN

The run file (called DIF.RUN, a text file) may look like this:

Example of Lord's chi-square analysis for 3PL model

3

3plfoc.3pl

0

32

3plref.3pl

0

32

foctran.out

N

3plfoc.cov

3plref.cov

1

You may review the meaning for each line at IRT\Procedure for DIF Detection Using Lord's Chi-Square (.ppt.

Q2. I have to perform a series of simulations. What is the best way to let the computer run one simulation after another?

A2: To make sure that the computer has processed various simulations, you may want to use a DOS batch file, such as run.bat, to make sure that you have executed every simulations in your study. In addition, utilizing "run files" mentioned in the previous question may help you identify where went wrong.

Q3. I heard there are different versions of ITERLINK?

A3: For different analysis purposes, there are different versions of ITERLINK. ITERLINK (2002) may be used for analysis using BILOG. ITERLINK (2000) may be used for analysis using MULTILOG. MULTILOG's estimates for ai contain a scaling factor of 1.7. The only difference between the two versions is the EQUATED item parameter file output (i.e., foctran.out). The linking constants and DIF analysis are the same. In addition, the new ITERLINK06 (2006) is designed to work with BILOG-MG.

Q4. How can I tell which version of ITERLINK I have?

A4: On the downloads page, all executable files are now saved in .ZIP format. Thus, you should be able to tell which version of ITERLINK you have.

Q5. I am getting confused with this 1.7 scaling factor. Can you explain more?

A5: When you select a dichotomous model, say using BILOG for 2PL model, the mathematical form can be written as Picture 1

where D is a scaling constant equal to 1.702 (You may refer to http://io.psych.uiuc.edu/irt/dim_dich1.asp for more details).

When you select a polytomous model, say using MULTILOG for SGR model, if there are only two possible responses to each item (e.g., correct/incorrect), the graded model is equivalent to the 2PL model, which can be written as Picture 2

As you can see, MULTILOG does not include the scaling factor of 1.7 in either the 2PL or the 1PL models. Please refer to page 568 of the book IRT from SSI.