Hi there!

I'm currently analysing a partial credit test which basically consists of two parts: test-takers have to do the same thing twice in different contexts (item a1, a2, b1, b2...). I assigned the same codes to the different contexts and now I want to check whether the items differ significantly according to contexts.
Can DIF / DPF help me there? I tried to work that out but wasn't very successful  

I woud be very thankful for some help!
Thanks

Thank you for your question, JeySee. What software are you using?

Here is how I would do this in Winsteps:
1. Each item-context is a column
2. Each test-taker is a row
3. We can then compare the two matching items in any way we want

Hi Mike,
thanks for your fast answer!

I'm using winsteps.
I've done steps 1&2, but I wonder how I would tell the program which items have the same context (which pairs belong together).
I tried to stack the items and entered a new variable (1 = context 1; 2 = context 2). then I computed the DIF for 1 vs. 2.
But I'm wondering about two things:
1. If I stack the items, then the program doesn't know that the two items were answered by the same person (and test statistics like reliability etc. change) and
2. I treat the context-variable as person-variable which it clearly is not. Can that be right?

I appreciate your help!

JeySee, what comparison between the two versions of the items do you want?

For instance, if you want to compare the item difficulties to discover which pairs have significantly different difficulties, then

1. in the item labels, put the item-context code "a1", "a2", ... "b1", "b2", ....

2. model the two versions of each item to share the same partial-credit structure
ISGROUPS = 12......12.......

3. The analysis will produce a measure, S.E., and count of observations for each item.
So, for each pair of items, we can use Excel to compute Welch's variant of Student's t-statistic: http://www.winsteps.com/winman/index.htm?t-statistics.htm

OK?

Mike,
that's exactly what I need! Thanks a lot!
I'll try this!