0

I'm hoping to run an IRT analysis where I have a couple families of items, and each person has answered one question from each family. Are there R packages for doing something like this? I've looked at options like hIRT but this seems to only account for individual level covariates, not item level covariates. I'm expecting the data to look something like this:

1a 1b 1c 2a 2b 2c
1 0
0 1
1 1

where each person answers one of 1a, 1b, and 1c, and one of 2a, 2b, and 2c.

Matt
  • 126
  • 10
  • You can use Stan or JAGS in R to model such data. – Amin Shn Jan 10 '23 at 20:01
  • Thanks - I was thinking of doing something like that, though I wasn't sure if a bayesian approach was the only choice. (That would be a fine option, just wasn't sure.) – Matt Jan 10 '23 at 20:04
  • Using a Bayesian approach you can model the missing values and put some priors on them. I am not sure how else you can deal with the missing values. – Amin Shn Jan 10 '23 at 20:14
  • some IRT-R packages, such as `ltm`, `mirt` and `TAM` implement maximum likelihood estimation, that can handle missing values. What is the relationship among the items within a family? Might be a relationship that is typically modeled by testlet models, linear logistic test models or copulas... – Tom Jan 11 '23 at 09:47
  • @Tom - I'm basically hoping to model questions like "what's the derivative of ax^2" where a is picked from a list, so I expect that there will be correlation between those items. I'm not familiar with some of those types of models you mentioned, so I'll look into them, thanks@ – Matt Jan 11 '23 at 13:47
  • @Matt - ok, so we're talking about automated item generation. – Tom Jan 12 '23 at 09:06
  • @Tom - I guess maybe eventually, but right now I'm thinking about something like "what's the derivative of ax^2" where a is selected from e.g. [-1, 1, 5, 3.5] so there would just be 4 versions of that problem. – Matt Jan 12 '23 at 13:58

0 Answers0