czopluoglu / copydetect Goto Github PK

Add an argument to the function so the users can supply their own item parameter matrix instead of using the one estimated of by the mirt package.

From: my email [email protected]
Sent: Thursday, August 20, 2020 9:14
To: Cengiz Zopluoglu [email protected]
....
Also, the function takes quite a long time to run. I am currently running all pair calculations for a sample of 200 examinees, and haven't got results yet. I am wondering why it could take such a long time. Do you have any information on that?
....
Jin

For really tiny p values, think about a better way of printing out the p values. They printed as 0 right now.

This is also problematic during internal computations when there are so many small probabilities that are multiplied for large number of items due to precision.

Think about better way of handling small p-values (e.g., Maynes discussed log-transforming p-values).

p = 0.0005 * 0.000025

log(p) = log(0.0005) + log(0.000025)

When you report the outcome for many pairs, consider reporting more information (such as the number of matching responses, number-correct score, etc.) in the matrix in addition to p-values.

Consider adding this feature as an optional argument.

https://www.tandfonline.com/doi/abs/10.1080/15305058.2016.1217418

Just a suggestion, but would be handy to disable the verbose mode in the mirt NRM fit.
To include the outputs on a markdown file is much cleaner.
On similarity2, line 123:
nrm <- mirt(resp, 1, 'nominal',verbose=FALSE)
Thanks for the work, very useful!

From: my email [email protected]
Sent: Thursday, August 20, 2020 9:14
To: Cengiz Zopluoglu [email protected]

First, I noticed that the similarity2 function can only handle cases where all items have the same number of response options. I am wondering if you would plan to expand on this function so that it can handle cases where different items have different number of response options?
....
Jin

Reconsider similarity1 so common missing responses are not counted as an identical incorrect responses.

Handling missing data for pairs of response vectors for all these indices should be reviewed and reconsidered. There is nothing in the literature about this.

consider adding indices for response time similarity or item preknowledge statistics based on the response time.

Consider adding a function to fit the Deterministic Gated IRT model for detecting item preknowledge based on raw item responses.

Hi Professor Zopluoglu,

I am wondering what flagging criteria does CopyDetect similarity1 function use for M4 statistics when I run multiple pairs? Does it use the M4 similarity index that Maynes uses? Or does it uses Bonferroni adjusted alpha value?

Thanks a lot!
Jin