Question on Item Response Model Vignette

Interfaces brms
1

Hi all,

I’m wondering how, specifically, in Paul’s vignette for the Advanced Item-Response Model estimating guessing parameter (https://cran.r-project.org/web/packages/brms/vignettes/brms_nonlinear.html) we know that including an intercept for eta produces a bias in the guessing parameter? I run the model both with and without an intercept, and the coefficients are obviously different, but how do we know that a bias exists (merely through LOO)? Moreover, Let’s say we only cared about modeling guessing parameter insomuch as we wanted to obtain accurate intercept and slope estimates from the curves, how would we change the model, including the guessing parameter, accordingly?

Thanks!

2

@paul.buerkner wrote the vignette and he’s on the forum so if you’re lucky he may respond in person if he has time for it!

3

We know there is bias because we simulated the data from a known ground truth.

Moreover, Let’s say we only cared about modeling guessing parameter insomuch as we wanted to obtain accurate intercept and slope estimates from the curves, how would we change the model, including the guessing parameter, accordingly?

I am not sure about this. I would have to think of this more but don’t have the time right now. I would hope someone has addressed this in the IRT literature though.

1 Like