[Rasch] Doubt on Equating and Linking

Hudson Golino hfgolino at gmail.com
Fri Apr 23 11:51:50 EST 2010

Dear colleagues,

I've adopted an procedure, and I would like your opinion: Can it be
considered a valid strategy?

I have administered a 48 items test to a group of 167 people (Form A). After
all the analysis, I changed the majority of items, for different, but
relative ones (Form B), and administered it to a small group of 75 people.
>From 48 items, just 8 remained the same. In spite of this rough test
reformulation, the new items are closed-related to the old ones (same latent
trait, same structure). Being a neo-piagetian like test, I have items
ranging from earlier stages to higher stages. So, the 8 equal items are all
from the earlier stages, the easiest ones.

Aiming to equate and link both forms (A and B), I've adopted this "mixed"

1) Cross-plotted 24 item's measures, being 8 the same for both forms, plus
the 16 less changed items;
2) Analyzed them in terms of the slope of the best fit line, obtaining a
slope of 1.332;
3) I then removed the items whose measure differences between form B and
form A where over 1.25, remaining with 11 items, with a slope equal to
4) I've calculated the equating constant (x-axis' intercept)  and summed it
to form B item's measures. Then, I've examined the scatterplot (new form B
Item's measure on Y-axis, and form B item's measures on X-axis), and find
closed points with the best fit line, for all the 11 items.

Now, I just need to calculate the USCALE ( = the value of 1/slope) and the
UMEAN ( = the value of the x-intercept) and reanalyze Test B.

So, what do you think?

It is a valid procedure?

Thank you,

Hudson Golino
Assistant Researcher
Laboratory for Cognitive Architecture Mapping (LaiCo)
Federal University of Minas Gerais
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://mailinglist.acer.edu.au/pipermail/rasch/attachments/20100422/703363c7/attachment.html 

More information about the Rasch mailing list