logit scale

[Balal Ezanloo]

Hello Phil

when i convert estimated parameters of mirt for one factor that are in logit metric to the tradiationa parametes by "coef" function the results for "g" and "u" converted in logit and for "a" and "b" converted to traditional IRT parameters. i need to report them in one metric. how can i solve this problem? 

Regard

[Phil Chalmers]

I assume mean that you are seeing the names logit(g) and logit(u) in the coef() output? Just take these coefs and apply a plogis(value) transformation to them to obtain the original g and u parameters. Cheers.

Phil

--
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.

[Phil Chalmers]

On Sun, May 14, 2017 at 1:26 AM, balal izanloo <b.ez...@gmail.com> wrote:

Greeting again

1) As i find from your previous answers in the mirt forum, you are against the standard errors of item parameters in traditional metric but are agree with standard errors as estimated by mirt function. If I've understood correctly? 

No. I don't like the parametersization of the traditional metric because it's problematic. That doesn't have a lot to do with the large-sample SEs. But, because I don't recommend that parameterization, SEs are harder to get in mirt without defining your own IRT model or using the (often problematic) delta method.

 

2) if i convert "logit(g)" and "logit(u)" by "plogis(value)", what should i do to there standard error, whether they shoud be converted by "plogis(value)" also?

Use the delta method. 

Phil 

Cheers

On Mon, May 1, 2017 at 11:49 PM, Phil Chalmers <rphilip....@gmail.com> wrote:

You can't go those from mirt directly, you'll have to apply the delta method manually. See this post: https://groups.google.com/forum/#!searchin/mirt-package/delta$20method%7Csort:relevance/mirt-package/qlHrupGOl2A/p9f_3y3SBQAJ. But be warned, you're getting an SE for a bounded parameter, so it might not make a whole lot of sense. I have a paper coming out on this very problem, and currently advice against it. 

Phil

On Mon, May 1, 2017 at 3:16 PM, balal izanloo <b.ez...@gmail.com> wrote:

I did it but the problem exist. i just want item parameters in traditional metric along with their standard error. changing one option results to changing some other characteristics that is not related for it. 

On Mon, May 1, 2017 at 11:31 PM, Phil Chalmers <rphilip....@gmail.com> wrote:

Remove simplify = TRUE here.

Phil

On Mon, May 1, 2017 at 3:00 PM, balal izanloo <b.ez...@gmail.com> wrote:

Thanks

I think the option "printSE" in "coef" function do not work properly, as defined for it. compare the results for below two syntaxes. the only difference is in "printSE". I think it is worthwhile to check it. 

> coef(out4pl, CI = 0.95, printSE = FALSE,

+ rotate = "none", Target = NULL, IRTpars = TRUE, rawug = FALSE,

+ as.data.frame = FALSE, simplify = TRUE, unique = FALSE,

+ verbose = TRUE)

$items

              a      b       g        u

X156 1.895 0.572 0.000 0.999

X157 1.375 2.003 0.000 0.766

X158 2.021 1.735 0.000 0.965

X159 2.665 1.518 0.015 1.000

X160 0.896 4.207 0.000 0.950

.....

> coef(out4pl, CI = 0.95, printSE = TRUE,

+ rotate = "none", Target = NULL, IRTpars = TRUE, rawug = FALSE,

+ as.data.frame = FALSE, simplify = TRUE, unique = FALSE,

+ verbose = TRUE)

$items

              a     b           g        u

X156 1.895 0.572  -8.801  6.960

X157 1.375 2.003  -9.670  1.184

X158 2.021 1.735  -9.950  3.315

X159 2.665 1.518  -4.181  7.968

X160 0.896 4.207  -9.549  2.937