S-X2 statistic produced NaN for dichotomous items

[Fulya barış]

Hi,

I'm trying to estimate mixed format test statistics (both dichotomous and polytomous) .

I used mirt package and analyzed model according to multidimensional model ( correlated traits) and bifactor model.

ı tried to get fit statistics with this function

"itemfit_multiGP<-itemfit(multiGP, QMC=TRUE)"

when I tried to see item-fit statistics NaN produced for dichotom itemsand  just produced for polythom items for mutldimensional model 

But, bifactor model I get all statistics about item-fit 

What should I do ? how can  I get fit statistics ?

thanx 

[Phil Chalmers]

This can happen due to the collapsing of expected counts in the itemfit() function relative to the number of estimated parameters. But without a reproducible case there isn't much else I can say about what to do without knowing exactly what's going on.

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...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mirt-package/8e752eee-52da-4e5f-a27c-e02dad1db04b%40googlegroups.com.

[Fulya barış]

Hi! again

you mean problem is in my data ?

İf you dont mind, I want to share my data and csv file :)

I use PISA 2018 data for my study.

if problem is in PISA data, I will change my booklet. İF not , dont know what to do :)

This is my syntax;

yeni <- apply(yeni, 2, function(x) {x[x == 9] <- NA;x})

yeni2 <- na.omit(yeni)

library(mirt)

dimensions <- mirt.model('

S1= 1,2,3,4

S2=5,6,7

S3=8,9

S4=10,11

S5=12,13,14

S6=15,16,17

S7=18,19,20

S8=21,22

COV=S1*S2*S3*S4*S5*S6*S7*S8')

multiGP<-mirt(data=yeni2, dimensions , itemtype="gpcm",SE=TRUE,method="MHRM",verbose=FALSE)

LD_multiGP<-residuals(multiGP, type="LD")

#Item fit

fitmultiGP<-itemfit(multiGP)

On Friday, 7 February 2020 00:33:27 UTC+3, Phil Chalmers wrote:

This can happen due to the collapsing of expected counts in the itemfit() function relative to the number of estimated parameters. But without a reproducible case there isn't much else I can say about what to do without knowing exactly what's going on.

Phil

On Wed, Feb 5, 2020 at 5:27 AM Fulya barış <fulya...@gmail.com> wrote:

Hi,

I'm trying to estimate mixed format test statistics (both dichotomous and polytomous) .

I used mirt package and analyzed model according to multidimensional model ( correlated traits) and bifactor model.

ı tried to get fit statistics with this function

"itemfit_multiGP<-itemfit(multiGP, QMC=TRUE)"

when I tried to see item-fit statistics NaN produced for dichotom itemsand  just produced for polythom items for mutldimensional model 

But, bifactor model I get all statistics about item-fit 

What should I do ? how can  I get fit statistics ?

thanx 

--
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]

The item fit statistics actually do not work for a model with this many latent variables that correlate and fewer total scores since the number of parameters for each item and the latent trait distribution are included in computing the degrees of freedom for each item. Hence, while the tables are computable, the itemfit() function just returns NaN since the degrees of freedom are negative since each item has a least 30 free parameter estimated per item (2 or 3 + 28 latent trait pars). With only a small number of total scores to this data provides, which generate the data that provides the degrees of freedom, the df for each item to work with ends up  being around -8 each, and so the S_X2 family becomes nonsensical in a statistical inference sense. Basically, you would either need to fit fewer inter correlations between the latent traits, or increase the number of items and participants to ensure that the total score statistics are 'sufficiently high' to warrant using chi-squared based tables. 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/eaecaf1b-3bb7-43a2-a173-2c64f500bb2e%40googlegroups.com.

[Fulya barış]

Thanx. Phil.

That explanation will help me.

To view this discussion on the web visit https://groups.google.com/d/msgid/mirt-package/eaecaf1b-3bb7-43a2-a173-2c64f500bb2e%40googlegroups.com.