Re: Asymmetric

[Phil Chalmers]

Certainly possible, could you provide an article describing these models? Also, this is a great are for users to contribute to the package and send pull requests. That's how the recent GGUM models were added.

Phil

On Sun, Mar 11, 2018 at 3:48 AM, AM <mut...@gmail.com> wrote:

Hello Phil,

I just wondered if you were planning to include asymmetric IRT models (e.g. logistic positive exponent model) to the mirt package at any time in the future.

Best regards,
AM

--
You received this message because you are subscribed to the Google Groups "mirt-package" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[eaez]

Hello Phil,

I tried creating custom item type for dichotomous logistic positive exponent model. When I fit the model, I receive these error messages: "1: In log(computeItemtrace(pars = x, Theta = Theta, itemloc = itemloc,  ... : NaNs produced". Could you provide an insight what might have went wrong?

My code is:

name  <- 'lpe'

par   <- c(a= .5, b = -2, ksi = 1)

est   <- c(TRUE, TRUE, TRUE)

P.lpe <- function(par, Theta, ncat){

  a <- par[1]

  b <- par[2]

  ksi <- par[3]

  P1 <- (1/(1 + exp(-1*a*(Theta-b))))^ksi

  cbind(1-P1, P1)

}

x <- createItem(name, par=par, est=est, P=P.lpe)

lpemod <- mirt(data, 1, itemtype = "lpe", customItems=list(lpe=x))

coef(lpemod, simplify = T)

Also, the reference for the LPE class models: https://link.springer.com/article/10.1007/BF02296149 

Thank you for the great package and your support here!

On Sunday, March 11, 2018 at 12:09:49 PM UTC-4, Phil Chalmers wrote:

Certainly possible, could you provide an article describing these models? Also, this is a great are for users to contribute to the package and send pull requests. That's how the recent GGUM models were added.

Phil

On Sun, Mar 11, 2018 at 3:48 AM, AM <mut...@gmail.com> wrote:

Hello Phil,

I just wondered if you were planning to include asymmetric IRT models (e.g. logistic positive exponent model) to the mirt package at any time in the future.

Best regards,
AM

--
You received this message because you are subscribed to the Google Groups "mirt-package" group.

To unsubscribe from this group and stop receiving emails from it, send an email to mirt-p...@googlegroups.com.

[Phil Chalmers]

Can you think of combinations of the parameters which would cause log( cbind(1-P1, P1) ) to return either Inf, -Inf, NA, or NaN values? If so, special case must be taken to avoid such locations by offering suitable boundary conditions during estimation. I can think of at least two: when k <= 0, and when a = 0 (in which case b -> ±∞ in the M-step).

Phil

To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mirt-package/b85ee1fc-b3e4-49f2-9a26-8fa3c2736d82%40googlegroups.com.

[eaez]

Thanks Phil, for the quick response. Yes, this solution makes perfect sense. Do you have an example showing how to set those parameter location boundaries during estimation?

To unsubscribe from this group and stop receiving emails from it, send an email to mirt-p...@googlegroups.com.

[Phil Chalmers]

The lbound input to createItem() is probably what you want to work with here. For the a and ksi parameters just set the lbound to 0, while for b set the boundary to be -Inf. With any luck that will help the model converge (though I'd recommend using the nlminb optimizer in mirt(...) since it tends to work better with boundary conditions). HTH.

Phil

To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mirt-package/16bd688f-7445-4f87-94dc-0640cf9f6088%40googlegroups.com.

[Ezgi Ayturk]

Will try! Thanks so much!

--

Ezgi Ayturk, M.A.

Ph.D. Candidate,

Psychometrics and Quantitative Psychology

Fordham University

441 East Fordham Road

Bronx, NY 10458

E-mail: eay...@fordham.edu

[Phil Chalmers]

No problem. The other thing to think about is that the classical 2PL model that you've written is probably not very good for estimation purposes due to the a=0 problem. This can be avoided entirely by switching to a slope-intercept format, which is what mirt uses as a default. The classical parameter can then be achieved after convergence via some transformations, and SEs can be obtained via the delta method. HTH.

Phil