Item Response Theory
Posted on October 19, 2007
Paul A. Pilkonis (bio) explains IRT, a latent variable approach to measurement.
Item response theory is a latent variable approach to measurement. The assumption, there are a couple of key assumptions in using item response theory models. One is that the latent constructs that we're measuring are unidimensional. It's important to try to measure only one thing at a time. If you're confident that you're looking at a single latent construct, the task then becomes to develop the very best manifest indicators of that latent construct and what the models allow us to do is then to calibrate these manifest indicators in terms of severity, whether they measure the construct at low, medium, high ranges of severity. And these models can be implemented with data from large samples.
One of the constraints is that you probably need 2-, 3-, 400-person samples in order to be able to do this kind of model fitting. But it gives us statistical tools for increasing the precision of the measures that are available to us out there and it opens the door, perhaps an even greater innovation, to computerize adaptive testing. It becomes possible if you have an item bank that you're confident is assessing the severity of a single construct at different levels, to then, with a smaller subset of those items, to peg where a person is in terms of severity of depression, anxiety, anger on the basis of just a smaller subset of those items.
Ideally you want to be able to identify items that provide the most information about the construct and most IRT models will allow you to estimate at least two parameters; one parameter has to do with the extent to which the item does discriminate people who are higher and lower on the construct. The second parameter is a location parameter where in terms of the thermometer; the scale of severity is this assessment of more versus less located.
Ideally what this allows us to do is to develop an item bank, a set of items that assesses the construct across the entire range of severity with good discriminating capacity. Then I can take this pool of items and with, even if I have an item bank of 100 items, use just four or five or six to characterize where you are in terms of depression, anxiety, anger, other psychological constructs.
Excerpted from interview with researcher at the 2006 Career Development Institute for Psychiatry in Pittsburgh, PA.
Please note that the feedback is viewed only by 4researchers staff and is not intended for communication with individual contributors.
Use the form below to submit feedback about this article. If you would like a response, please be sure to include your e-mail address.