Re: Questions regarding information and reliability in MIRT package

[Phil Chalmers]

Hi Shiyang,

On Mon, Aug 15, 2016 at 7:33 PM, Shiyang Su <suxx...@umn.edu> wrote:

Hi Phil,

Thanks for creating the great MIRT package! I have two questions regarding your functions of information and reliability. I'll greatly appreciate your help with them.

1) When I checked your code of iteminfo, I am not sure of your ways of calculating the item information. You use the function ItemInfo and ItemInfo2. But I didn't find the source code for these two functions. I would love to know how you calculate the item information. If it is OK, could you please share your code or a link to the source code of the function ItemInfo and ItemInfo2? I'd greatly appreciate that.

All packages on CRAN are completely open source by definition, so either download the tarball from there and extract, or download the development version from Github.

 

The second question is regarding your ways of calculating the marginal/empirical reliability. Do you have a reference of the reliability function you used in the MIRT package? My understanding of reliability is calculated as 1 - average squared SE(theta)/var(theta), where var(theta) is the variance of the estimated theta. This doesn't seem to be the same as your function of empirical_rxx. Could you please let me know which formula you are referring to when calculating the reliability?

Yes, that looks correct, and is essentially what the function is doing. Though marginal reliability is different (requires integration across a pdf rather than using secondary estimates). Cheers.

Phil 

 

Thank you very much for your help in advance!

Shiyang