Modelling dichotomously scored items with the Rasch Model

[Johanna Skillen]

Hello,

I'm new to ConQuest and trying to figure out how to run a Rasch Model analyses on a test with 33 dichotomously scored items. I followed the ConQuest tutorial on modelling a dichotomously scored multiple choice test with the Rasch Model but I get warnings and it seems to me that the analyses does not run properly.
And this leads to my first question. My Items already contain only two options (correct answer coded with 1 and incorrect answer coded with 0). Do I follow the right commands when following the ConQuest tutorial that provides a multiple choice test as example?

Second question -  when running the analyses I get the following warning:
"Parameter (7) for item 7, is not identified. If anchors or initial values are used the referencing must reflect removal of this parameter." Item 7 is then excluded from the analyses. Unfortunately, I don't know what this warning means and what I can do to get Item 7 included.

Last but not least I looked at the provided itanal statement and it seems to me that something does absolutly not fit! E.g. with item 1 it says that 11 students responded with 0 and 88 responded with 1. But in fact, 99 students responded this item correctly which is coded with 1 for all students (the coresponding collum in the data file does only contain 1, no 0!) What has happend here?  Is it possible that ConQuest mixed up the collums? And can I trust the provided parameter estimates and the map of item difficulty parameters if the count of students who responded to the items does not fit?

Thank you for help!
Kind regards, Johanna

[dan_c]

Hi Johanna,

It's hard to know what is going on without your data, syntax, and output. If you are uneasy about uploading your actual data, have a look at this thread about creating a minimum working example:

https://conquest-forums.acer.edu.au/index.php?topic=1239.msg1475#msg1475

[Johanna Skillen]

Thank you for this advise!

I created an example data set with 20 participants and the same 33 dichotomous items that fits my actual data. I will upload the data, label, syntax and output files.
Also with this example data set I get the same warnings and the count of participants who responded to the items does not fit. So, my questions remain the same... ;-)

[Johanna Skillen]

And here come the output files!

Thank you and kind regards!

[dan_c]

Hi Johanna,

I found a few small issues in your data (example1.dat) and command file (example1.cqc) that were probably causing the issue.

See the attached.

1. Your data file had some tabs in it (between the id and the responses). It is best to keep to just spaces in your data file.
2. Your format statement was pointing to the wrong columns for the values. I changed it to read

Code: [Select]

format id 1-8 responses 13-45;
This appears to produce the correct frequencies in the repose categories.

[Johanna Skillen]

Hello again,

and first of all thank you very much for your last advices! Afterwards my analysis work pretty well and I got much further.

Nonetheless I'm coming with another question. I managed to include cases in the item-person-map. It says that "each x represents 0.2 cases". In the working example ist even says that x represents 0.0 cases, what seems quiet odd to me.
I attach the examples.
Now I have the following questions: First how can I interpret this and second can I somehow define that x represents 1 person?

Thank you and kind regards!
Johanna

[dan_c]

I managed to include cases in the item-person-map. It says that "each x represents 0.2 cases". In the working example ist even says that x represents 0.0 cases, what seems quiet odd to me.
...
Now I have the following questions: First how can I interpret this and second can I somehow define that x represents 1 person?

there is an option to show, "xcases=n", which sets the number of cases to be represented by each 'X' in Wright maps. The default value is a value that ensures that the largest bin uses all available bin space. The value is replaced by the default if the display would not otherwise fit in the available space. So you could use:

Code: [Select]

Show ! estimate = latent, x=1 >> example1.shw;
to force ConQuest to have each X represent 1 case in the dataset. With small datasets this will make it hard to interpret the shape of the distribution (it will look very flat). With very large datasets it will result in many cases not being displayed (they won't fit on the screen).

NOTE: there is a small bug in ConQuest versions below 4.16 where the option is `x` instead of `xcases` - thats what I've included in the example above.

You can interpret the Wright map as a visual display of the alignment of the abilities of the students to the difficulty of the items. Check out some further reading here:

https://www.rasch.org/rmt/rmt253b.htm

[Johanna Skillen]

Hi,

I'm not sure whether my last reply was send, so here a second try.
Thank you for your quick answer to my last question. Unfortunately, the command does nor seem to work when I enter it. I tried x=1 and xcases=1 but I get the following error massages (attached).

Did I miss/forgett anything?

Thanks again!

[dan_c]

Unfortunately, the command does nor seem to work when I enter it. I tried x=1 and xcases=1 but I get the following error massages (attached).

your error message is referring to an option word for `estimate`, which makes me think that maybe you are running an older version of ConQuest. Can your run `about;` at the ConQuest command line or use the GUI to let me know what version you are running? One possible work around for older versions of ConQuest is to delete the spaces around your equals signs. For example `estimate=latent` rather than `estimate = latent`.