"R" in Action - rasch model estimation?

[marten koomen]

Hi Robert,

I am looking to implement an algorithm to estimate rasch model parameters for an online student assessment system. I am thinking of implementing a simple technique in Java based on an algorithm in this Item Response Theory Book. I'm still in the exploratory stages at the moment.

My questions,
1. Does "R" have any features to support Item Response Theory estimation?

2. Does "R" support backend server side estimation, that is, is it easy to integrate into an existing web interface?

Much appreciated.

Thanks

Marten

PS. I got this interesting message when I originally tried to post
The specific error message is: "r" is a silly English abbreviation; use "are" instead.

[Peter Johnson]

1) This might help: http://psy-ed.wikidot.com/item-response-theory (I googled r function item response theory)

2) "R" scripts run within the "R" runtime environment. Essentially in Java you would have to have your web app fire off "R" passing the script, and then collect the data after it is done. Not very integrated, but doable. Not sure if "R" is better integrated with other languages, such as PHP or Ruby.

[marten koomen]

Thanks Peter

I had a quick look at Dimitris Rizopoulos' Latent Trait Model (ltm) package and ltm covers stuff that I am after plus a lot more. The question I need to deal with is which is simpler, 1) understand ltm or similar and write an interface to it, or 2) just write an algorithm myself that is integrated to my application; I suspect the later. If I am to write an algoritm myself would "R" be easier than Java? Again I probably think Java would be better since I know it and as you suggest there would need to be another interface between the "R" and the Java. I am open to suggestions though and that is essentially my reason for engaging this forum.

Thanks for the link and I will contemplate it further.

Cheers

Marten

[Robert Kabacoff]

Hi Marten,

R has a great deal of support for IRT models in general, and Rasch models in particular. The best overview of IRT support in R is the CRAN Task View on Psychometrics. The ltm package is certainly a good place to start.

With regard to creating a back-end, it would be helpful to know more about what you are trying to do. Why are you trying to build Rasch modeling estimation into an existing web interface? R has numerous tools for creating a web interface to R services (including the functionality of the ltm package). Knowing a bit more about what you are trying to achieve would help me to guide you.

[marten koomen]

Thanks Robert

My interest is from developing items, administering to students, and collecting and analysing the data on the fly within a school context. My web app will therefore collect data in a rectangular form of items by students with missing data. For a given rectangular dataset I would like to estimate student (ability) and item (difficulty, point-biserial and a fit statistic) parameters and perhaps one differential item functioning statistic. I want to estimate these on the fly using only the Rasch Model (and considering the Partial Credit Model). More detailed analysis could be done by importing the data into another comprehensive package. I'm actively tryint to steer clear of things like the two and three parameter models, structural equation models, multi levels models etc as I don't need them for my purposes and I find them a bit scary when I see applications devoted to them .

So I'm only interested in some very limited IRT/statistical functionality in the online environment, with my personal particular problem solving focused on integration with other aspects of student assessment.

My IRT/statistics background is not that strong, and currently I only program in Java, JQuery, SQL, so I need to take care with any required learning curve to achieve my aims. I am reading up on IRT at moment. If "R" was to provide an easy way of implementing a simple estimation program that would be great.

Is this at all covered in the book?

[Robert Kabacoff]

Marten,

That sounds like an very interesting application. You should be able to do this in R rather readily. I have added a thread on Embedding R in other applications (including web applications) that may answer some of your questions. Take a look and see if anything grabs you.

BTW, I taught SEM to doctoral students for 10 years and it still scares me