Estimating IRT model parameters when the abilities are known

by Thiago Carneiro   Last Updated August 14, 2019 01:19 AM - source

I've been using mirt and ltm R packages to estimate the parameters (dychotomous 3 PL model) and estimate abilities for any answer pattern.

I've read that estimation is improved when you use the final scores of the people who took the test (I have past scores from people who took the test and the questions they got right and wrong - with 1 and 0's).

I'd like to improve the estimation by using not only the Right/Wrong matrix for the students but also their final grades. How can I do it using one of these packages? If it isn't possible, which package can I use to incorporate the grades of the people who took the test?

Generic illustration:

I already have the matrix: Q1 Q2 Q3 Q4 Q5 Student1 1 0 0 1 0 Student2 1 1 1 1 0 ... Studentn 0 0 0 1 1

I also have the scores for them:

Student1: 450 Student2: 700 ... Studentn: 490

The a(discrimination), b(difficulty), c (guessing) parameters are unknown to me.

I wanna find the parameters (a,b,c) incorporating the scores I know from my n students, and also get the scores of different patterns: Q1 Q2 Q3 Q4 Q5 Student X 1 1 1 1 1 Student Y 0 0 0 0 0 Student Z 1 0 1 0 1

How could I do it effectively?

Tags : r irt

Related Questions