0
$\begingroup$

I am working with a data set to model student performance with numerous variables from the class/school/district/provincial level. Student performance is extremely low though -- 70% of reading performance scores are 0.

I have no experience working with non-parametric models, but assume I should use one for this non-normal distribution. I am playing around with a logit model and using an above/below threshold for student performance.

What other models should I explore with this kind of data? Here is the score distribution:

Score Number
0 630
1 2
2 1
3 1
5 1
8 1
9 2
14 1
19 1
20 2
21 1
23 2
28 3
29 2
Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

How are "scores" measured? If it's numeric with a small number of levels, you could try ordinal logistic regression. If it's a count (number of items correct) then try a zero-inflated count model (probably zero-inflated negative binomial). Hurdle models are also a possibility. If it's a 0 to 100 rating, the count models might still work, but a more general class of models is finite mixture models. If you are using SAS there is now (in SAS 9.3) PROC FMM which will allow quite a bit of flexibility. In R there is the package flexmix - but I haven't used it.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thank you for the input. The scores for each task have potential scores around 50 or 100, that are not really reachable due to the time allotted. There are theoretically 50 levels on the test--only 19 in the data. Regardless think that is too many for a ordinal logistic model. I will give a zero-inflated count model a go on R and post back on how things work out. Thanks again. $\endgroup$
    – user2168478
    Commented Mar 14, 2013 at 12:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.