Psychoco 2021: Paul Bürkner - Bayesian Item Response Modeling in R with brms and Stan

[@sourabhsharma9830]

Hello! I was just wondering how would i deal with the missing values while model fitting?

[@azeileis]

The brms project web page has a dedicated vignette on the topic of imputation of missing values: paul-buerkner.github.io/brms/articles/brms_missings.html Moreover, a typical type of missing values in IRT models, is handled automatically when fitting the models with brms. Namely, if not every subject sees every item, then the item response _matrix_ would have missing values. However, for brms you do not present the item responses as a _matrix_ but as a _long data frame_ where one row is the response of one person to a specific item. And then you can just omit those rows for person-item combinations that were not observed but you can still fit the model. If you have further questions, it's probably best to ask these on the Discourse page of the Stan project: discourse.mc-stan.org/ It's much more likely that a knowledgable person answers the question there than here on KZbin. (Disclaimer: I'm really no expert in brms.)

[@sourabhsharma9830]

@@azeileis Thanks a lot it helped. I am trying to fit a 2 PL IRT model to data (1000 rows X 50 items) I know it's a very silly question but could you please help me to know how can i get the list of all estimated parameters and measurement error for 1000 persons and 50 items from fit object?

[@azeileis]

@@sourabhsharma9830 Not a silly question at all but KZbin comments are not the best place to work through this. Please consider the accompanying paper (arxiv.org/abs/1905.09501) where you find a few worked examples. The paper also explains that you can use the coef() method to extract the estimated parameters. For further questions I recommend the brms discourse forum: discourse.mc-stan.org/