Author Topic: six-dimensional model  (Read 1098 times)

hanne72

  • Newbie
  • *
  • Posts: 12
    • View Profile
    • Email
six-dimensional model
« on: December 18, 2015, 08:06:29 AM »
Hi
We've data on a 47-item questionnaire with three defined subscales. We've run a 1-dimensional model and a 3-dimensional model using ConQuest 4.
Using RUMM2030 (PCA/t-test and subtest procedures), all three subscales are multidimensional. Consequently, defining and running the 6-dimensional approach we get an error message saying "the prm file cannot be opened". Our CQ project file is described below.  Do you have any suggestions of what we are doing wrong?

CQ Project file:
title dia6dimtest;
datafile C:\Users\Downloads\EUdia6dim161515RUMM.dat;
format PID 1-3 responses 4-50;
codes 1,2,3,4;
score (1,2,3,4) (0,1,2,3) () () () () () !items (1-6);
score (1,2,3,4) () (0,1,2,3) () () () () !items (7-16);
score (1,2,3,4) () () (0,1,2,3) () () () !items (17-22);
score (1,2,3,4) () () () (0,1,2,3) () () !items (23-31);
score (1,2,3,4) () () () () (0,1,2,3) () !items (32-42);
score (1,2,3,4) () () () () () (0,1,2,3) !items (43-47);
 model item+item*step;
set update=yes,warnings=no;
export parameters >> EUdia6dim161515RUMM.prm;
export reg_coefficients >> EUdia6dim161515RUMM.reg;
export covariance >> EUdia6dim161515RUMM.cov;
import init_parameters << EUdia6dim161515RUMM.prm;
import init_reg_coefficients << EUdia6dim161515RUMM.reg;
 import init_covariance << EUdia6dim161515RUMM.cov;
estimate !method=montecarlo,nodes=2000,converge=0.005;
show !tables=1:2:3:4,estimates=latent >>EUdia6dim161515RUMM.shw;
show cases !pfit=yes,estimates=wle,filetype=xls>> EUdia6dim161515RUMM_PersonFit6dim.xls;
show !filetype=xls, estimates=mle,tables=1:2:3:4 >> EUdia6dim161515RUMM_ItemFit6dim.xls;
itanal >> EUdia6dim161515RUMM_tradAnalyses6dim.xls;
quit;

Kind regards,
Hanne

Eveline Gebhardt

  • Administrator
  • Full Member
  • *****
  • Posts: 103
    • View Profile
    • Email
Re: six-dimensional model
« Reply #1 on: December 21, 2015, 01:08:53 AM »
Hi Hanne

It usually means the prm file is not in the folder ConQuest is looking in. It is often better to type in the full directory of the import and export files. It's possible the reg and cov files are not there either.

If the files do not exist yet (i.e. you have not run this job before, so they have not been exported yet), they cannot be imported as initial values. In that case, you should comment out these lines:

/*import init_parameters << EUdia6dim161515RUMM.prm;*/
/*import init_reg_coefficients << EUdia6dim161515RUMM.reg;*/
/*import init_covariance << EUdia6dim161515RUMM.cov;*/

Best wishes
Eveline

hanne72

  • Newbie
  • *
  • Posts: 12
    • View Profile
    • Email
Re: six-dimensional model
« Reply #2 on: January 14, 2016, 09:40:49 AM »
Thank you! But does this mean that the analysis has to be run twice (one without the import commands, and one where these are included)?

hanne72

  • Newbie
  • *
  • Posts: 12
    • View Profile
    • Email
Re: six-dimensional model
« Reply #3 on: January 14, 2016, 09:48:43 AM »
When running a 12 dimensional model, I get the message: "a parameter index is out of range for an item parameter initial value. Parameter index is: 128. Number of parameters is: 127". When I exclude the import commands, the analysis runs. Do you have any suggestions what I can do to solve this? Is this connected with the problem described in my previous posted question?

Eveline Gebhardt

  • Administrator
  • Full Member
  • *****
  • Posts: 103
    • View Profile
    • Email
Re: six-dimensional model
« Reply #4 on: January 14, 2016, 10:23:18 PM »
No, you only need to run it once (without initial parameter estimates). But, if for any reason you need to run it again, you could use the parameter estimates of the previous run as initial parameters to save yourself some time, but you don't have to.

Eveline Gebhardt

  • Administrator
  • Full Member
  • *****
  • Posts: 103
    • View Profile
    • Email
Re: six-dimensional model
« Reply #5 on: January 14, 2016, 10:27:35 PM »
You have more parameters in your import file than need to be estimates. If you don't need import files (see previous message), you can just run your job without import files. Otherwise, I can have a look at it if you send me your command and import files.

hanne72

  • Newbie
  • *
  • Posts: 12
    • View Profile
    • Email
Re: six-dimensional model
« Reply #6 on: January 18, 2016, 06:16:04 PM »
Thank you very much!

I have also a new question:
When running a 12-dimensional model, I get the following message: 'Element 149 in variance/covariance matrix is negative. Check model indentification' (the same message also for six other elements). The instrument consists of 47 items, and I have 388 responses. I have also tried to run this 12-dimensional model in a sample of 900 respondents, but I still get the same message. The WLE person separation reliability also becomes 1.000 for all 12 dimensions in the sample of 388, but not in the sample of 900.
Is there anything I can do to avoid, or repair this?


The command used is:

title befolkning12dim;
datafile befolkningCQ280915.dat;
format PID 1-6 responses 7-53;
codes 1,2,3,4;
score (1,2,3,4) (0,1,2,3) () () () () () () () () () () () ! items (1-4);
score (1,2,3,4) () (0,1,2,3) () () () () () () () () () () ! items (5-8);
score (1,2,3,4) () () (0,1,2,3) () () () () () () () () () ! items (9-12);
score (1,2,3,4) () () () (0,1,2,3) () () () () () () () () ! items (13-16);
score (1,2,3,4) () () () () (0,1,2,3) () () () () () () () ! items (17-20);
score (1,2,3,4) () () () () () (0,1,2,3) () () () () () () ! items (21-23);
score (1,2,3,4) () () () () () () (0,1,2,3) () () () () () ! items (24-28);
score (1,2,3,4) () () () () () () () (0,1,2,3) () () () () ! items (29-31);
score (1,2,3,4) () () () () () () () () (0,1,2,3) () () () ! items (32-36);
score (1,2,3,4) () () () () () () () () () (0,1,2,3) () () ! items (37-40);
score (1,2,3,4) () () () () () () () () () () (0,1,2,3) () ! items (41-43);
score (1,2,3,4) () () () () () () () () () () () (0,1,2,3) ! items (44-47);
model item+item*step;
set warnings=no,update=yes;
export parameters >> befolkningCQ280915.prm;
export reg_coefficients >> befolkningCQ280915.reg;
export covariance >> befolkningCQ280915.cov;
/*import init_parameters << befolkningCQ280915.prm*/;
/*import init_reg_coefficients << befolkningCQ280915.reg*/;
/*import init_covariance << befolkningCQ280915.cov*/;
estimate !method=montecarlo,nodes=5000,converge=.005;
show !tables=1:2:3:4,estimates=latent >>befolkningCQ280915.shw;
show cases !pfit=yes,estimates=wle,filetype=xls>> befolkningCQ280915_PersonFit12dim.xls;
show !filetype=xls, estimates=mle,tables=1:2:3:4 >> befolkningCQ280915_ItemFit12dim.xls;
itanal >> befolkningCQ280915_tradAnalyses12dim.xls;
quit;

Eveline Gebhardt

  • Administrator
  • Full Member
  • *****
  • Posts: 103
    • View Profile
    • Email
Re: six-dimensional model
« Reply #7 on: February 04, 2016, 12:13:21 AM »
Hi Hanne

I have submitted the command file below and it runs without errors. However, it does not seem to converge to a good solution. I have added a command that exports a log file in which you can see that the deviance is jumping up and down between iterations, instead of slowly becoming smaller. When you look at the maximum change in item parameter estimates, you can see that these stay large until the last iteration and are not converging to your criterion of 0.05. Sometime the problem is just one skewed item or a bad fitting item, but in your case several items keep changing between iteration.

This often happens for models with many (about 6 or more) dimension, especially -- as in your case -- when you have a small sample and only few items per dimension. You probably need to simplify your model to make it converge to a good solution. Have a look if there are any bad fitting items that you can delete and decrease the number of dimension. You could also try to run two separate models with only part of the items and dimensions.

============================================================================
title 12dim;
datafile 12dim.sav !filetype=spss, responses=q1 to q47;

codes 1,2,3,4;
score (1,2,3,4) (0,1,2,3) () () () () () () () () () () () ! items (1-4);
score (1,2,3,4) () (0,1,2,3) () () () () () () () () () () ! items (5-8);
score (1,2,3,4) () () (0,1,2,3) () () () () () () () () () ! items (9-12);
score (1,2,3,4) () () () (0,1,2,3) () () () () () () () () ! items (13-16);
score (1,2,3,4) () () () () (0,1,2,3) () () () () () () () ! items (17-20);
score (1,2,3,4) () () () () () (0,1,2,3) () () () () () () ! items (21-23);
score (1,2,3,4) () () () () () () (0,1,2,3) () () () () () ! items (24-28);
score (1,2,3,4) () () () () () () () (0,1,2,3) () () () () ! items (29-31);
score (1,2,3,4) () () () () () () () () (0,1,2,3) () () () ! items (32-36);
score (1,2,3,4) () () () () () () () () () (0,1,2,3) () () ! items (37-40);
score (1,2,3,4) () () () () () () () () () () (0,1,2,3) () ! items (41-43);
score (1,2,3,4) () () () () () () () () () () () (0,1,2,3) ! items (44-47);

set constraints=cases;
export logfile >> 12dim.log;

model item+item*step;

export parameters >> 12dim.prm;
export reg_coefficients >> 12dim.reg;
export covariance >> 12dim.cov;
import init_parameters << 12dim.prm;
import init_reg_coefficients << 12dim.reg;
import init_covariance << 12dim.cov;

estimate!method=montecarlo,nodes=2000,converge=.005, stderr=quick,iter=1000;
show !tables=1:2:3:4:5,estimates=latent >>12dim.shw;
/* note that next command will produce many warnings because of so few items per dimension */
show cases !pfit=yes,estimates=wle,filetype=xls>> 12dim_PersonFit12dim.xls;
show !filetype=xls, estimates=mle,tables=1:2:3:4 >> 12dim_ItemFit12dim.xls;
itanal >> 12dim_tradAnalyses12dim.xls;
============================================================================

Please let me know if you need any further assistance.

Eveline

hanne72

  • Newbie
  • *
  • Posts: 12
    • View Profile
    • Email
Re: six-dimensional model
« Reply #8 on: February 08, 2016, 10:16:21 PM »
Thanks a lot again! Now the 12-dimensional model runs.
I ´ve also tried to run a multidimensional within-item model on the same 47 items (3x4 dimensions). But I get a warning of convergence trouble on almost every case. I ´ve tried to reduce the number of dimensions, using a 3x3 model, but still get the same warning. I ´ve used Figure 12 in Tutorial 7 and just added your recommendations export logfile and estimate!method=montecarlo,nodes=2000,converge=.005, stderr=quick,iter=1000; in the second step of the analysis.
Maybe I have too few cases (388) to run this analysis?

Best regards
Hanne

Eveline Gebhardt

  • Administrator
  • Full Member
  • *****
  • Posts: 103
    • View Profile
    • Email
Re: six-dimensional model
« Reply #9 on: February 19, 2016, 06:42:01 AM »
Yes, I am not sure if you will be able to converge such a complicated model with relatively few cases.

hanne72

  • Newbie
  • *
  • Posts: 12
    • View Profile
    • Email
Re: six-dimensional model
« Reply #10 on: March 03, 2016, 11:28:47 AM »
Thank you so much for your help!

Best regards, Hanne