"openirt" not working in STATA 16

[Gaurav Dhamija]

Dear All,

I am not able to run the openirt command in STATA 16.

openirt, id(id) save_item_parameters("items_math_oc2.dta") save_trait_parameters("traits_math_ocr2.dta") item_prefix("math")

Though it worked in STATA 13.





Is there any specific reason that this command is not working in STATA 16?

[Mead Over]

-openirt- still works in Stata 16.1 for the examples provided by its author

-openirt- is an item response analysis program written in Stata by Tristan Zajonc and posted on GitHub in 2014. Here is his profile on GitHub.

https://github.com/tristanz

While I am not an expert on -openirt-, I advised colleagues on installing it in 2016. I discovered that the installation script posted on GitHub has a small issue. It correctly installs the program -openirt-, but unfortunately it drops the example DO file and DTA files into Stata's reserved PLUS directory, a directory that users should not normally visit. So I wrote a program called "install_openirt" which can be found with "search install_openirt". If that does not work, you can describe my installation package with:

view net describe install_openirt, from("http://digital.cgdev.org/doc/stata/NonCGD")

Choose the install option.

Once you have installed the package, change to the Stata working directory in which you intend to experiment with -openirt-. Then for instructions type:

help install_openirt

To install -openirt- correctly, so that -openirt-'s example DO and DTA files drop into your current working directory, type:

install_openirt

You will find in your working directory the DO file called: -openirt_examples_update.do-.

Executing the -openirt_examples_update.do- file today in Stata 16.1, I find that the DO file runs to completion, estimating a 3-parameter logistic model without any problems.

However, since -openirt- was posted in 2014, Stata has substantially enhanced its IRT suite of commands to include e.g. the 3PL model. See for example:

help whatsnew15to16##irt

So perhaps you should try Stata's -irt- commands.

[Mai Chinh]

Dear Mead Over

I tried to run the openirt on Stata 17.0 and also got the same error as Gaurav Dhmijia although I had installed your packages and run the DO file called -openirt_examples_update.do-.
Do you have any suggestions to fix this error?

Thanks a lot!

[ericmelse]

I have no experience with using openirt and not much of a help with that.

But, besides using Stata's various irt commands, I can recommend the user community contributed package uirt, by Bartosz Kondratek, that can be installed from the ssc server, like:

ssc install uirt , replace
h uirt

Besides running rather faster than irt (on my system) with identical (or improved) results, uirt has more to offer with model options as well as postestimation commands for various plots.

[Mai Chinh]

I have no experience with using openirt and not much of a help with that.

But, besides using Stata's various irt commands, I can recommend the user community contributed package uirt, by Bartosz Kondratek, that can be installed from the ssc server, like:

ssc install uirt , replace
h uirt

Besides running rather faster than irt (on my system) with identical (or improved) results, uirt has more to offer with model options as well as postestimation commands for various plots.

Dear ericmelse

Thank you for your helpful advice.
I took a look at the package uirt that you suggested.
According to the help menu, this package only provide the expected a posteriori (EAP) estimator of the latent trait (theta). However, I would like to obtain the Maximum Likelihood estimate of theta.
I would be grateful if you can suggest any package in Stata that provides Maximum Likelihood estimates of theta besides openirt.

Thanks!
Chinh

[Bartosz Kondratek]

I would be grateful if you can suggest any package in Stata that provides Maximum Likelihood estimates of theta besides openirt.

Chinh

Dear Mai Chinh,

I am not sure what these ML estimates mentioned in openirt help file stand for. What comes to my mind are estimates from Joint Maximum Likelihood (JML) by Lord, but this approach is long depreciated (reasons explained, for example, in Bock and Aitkin, 1981). Ability parameter is treated as an unobserved random variable in nowadays IRT, thus you get an a posteriori distribution of ability of the test taker (obtained via EM algorithm or MCMC) and from this you can obtain the EAP estimator (mean of the a posteriori distribution) MAP estimator (mode of the a posteriori distribution) or draw PVs (draws from the a posteriori distribution).

Maybe you are looking for the MAP estimator? This is something that resembles ML the most, in the sense that you get the point where the a posteriori distribution is maximum in value. If so, you can obtain it with the conditional(ebmodes) option of predict after fitting the IRT model with default Stata irt command.