xluo11 / xxIRT

R package for item response theory
25 stars 15 forks source link

Implementing number-correct (NC) scoring for routing #8

Open ckoenig2 opened 5 years ago

ckoenig2 commented 5 years ago

Dear xluo11,

Great package!

I was wondering, would it be possible to implement a number-correct (NC) scoring approach for routing decisions? More specifically, a NC-scoring approach where you would have multiple thresholds for one route and multiple associated randomizers.

As an example: suppose you have a stage 1 module which has route to two stage 2 moduls (2A and 2B). Suppose you want to control module/item-exposure and you choose suboptimal routing, but with multiple thresholds. So for instance, if the test taker has a number-correct score of smaller than 4, he has a probability of 90% to go to module 2B (the easier one), and a 10% chance to to to module 2A (the more difficulty one). Additionally, if a test taker as a number-correct score between 4 and 8, he has a 50% chance to go to either stage 2-module. Lastly, if a test-taker has a number-correct score larger than 8, he has a 90% chance to go to module 2A, and a 10% chance to go to module 2B (the suboptimal, because easier one).

In sum, you have three thresholds (smaller than 4, between 4 and 8, larger than 8) and three randomizers (.9, .5, .9). Would that be possible?

Thanks!

Best wishes Christoph

xluo11 commented 5 years ago

Hi Christoph,

That's an interesting idea! I'm going to come back to this thread in the next round of development, probably in the fall. Currently, I am busied with operational work. I think a better way to implement it is to separate the routing rule from the general MST framework so that users can write and plug-in different routing rules.

Thanks for sharing your idea. As I said before, I'll come back to this thread in a couple of months.

Xiao