Messages

16

Questions and Answers / Re: Model comparison with zero correlations???

« on: June 30, 2012, 03:29:51 AM »

Best to post your control file so I can review it.  In general a model is a sub-model of another model if the sub-model can achieved by constraining parameter values of the super-model. Typically (but not always) the sub-model will be a special case of the super model with some parameters fixed at zero

17

Questions and Answers / Re: Person parameters in within-item multidimensional models

« on: June 30, 2012, 03:23:32 AM »

This looks like a bug or an identification problem.  Please email me your data and control file and I will diagnose the problem.

18

Questions and Answers / Re: zero correlation

« on: June 30, 2012, 03:22:30 AM »

It constrains the correlations to be zero, that is they are no longer free to be estimated but they are set to a zero value.

19

Questions and Answers / Re: sufficient statistic

« on: May 18, 2012, 05:31:34 AM »

Depends upon the "estimator" you use.  For WLE and MLE, the results are independent of the regression variable.  For PVs and EAP, the regressors mean that cases with the same raw score, but different background characteristics can indeed obtain different estimates.

ray

20

Questions and Answers / Re: Dimension variances

« on: May 18, 2012, 02:14:24 AM »

ConQuest variance estimates are unbiased estimates of the variance of the latent variable.  If you want to report the variance of a latent variable, then yes it is the value you should use.

Ray

21

Questions and Answers / Re: Dimension variances

« on: May 16, 2012, 03:53:03 AM »

Which person parameters did you export?   The variance of MLEs and WLEs will be greater than the ConQuest direct estimate.  The variance of EAPs willl be less than the ConQuest direct estimate and the variance of PVs will be equivalent to the ConQuest direct estimate.

You say "the variances reported by ConQuest are not the true variances".  In fact the reverse is the case.  The ConQuest direct estimate is unbiased whereas the MLE, WLE and EAP-based estimates are all distorted by measurement error, they are the "not true" variances.

Ray

22

Questions and Answers / Re: interpret

« on: May 16, 2012, 01:51:29 AM »

The third column is the posterior standard deviation.  You could use it as a standard error for teh EAP estimate.  The fourth column is the reliability of the case estimate.

The theta metric is a log-odds metric. A person's estimated theta is the log-odds of their success on an item with zero difficulty

Ray
 

23

Questions and Answers / Re: confidence interval

« on: May 08, 2012, 06:59:20 AM »

Sanei,

for the MNSQ confidence intervals normality is assumed because the degrees for freedom is (typically very)large.  Under the null hypothesis of model fit the MNSQ has an expecation of 1 and a variance that is computed internally, but not actually printed by ConQuest 3.0.   To obtain this number you could subtract the lower 95% CI value from one and divide by 1.96 to get the SE.  The to get an alternative find the appropriate nornmal deviate (eg 1.645 for a 90% interval) multiply that by the CI obtained above and substract from 1.0 for lower bound and add to 1.0 for the upper bound.

OK?

Ray

24

ConQuest News / ACER ConQuest Manual and Command Reference

« on: May 02, 2012, 12:52:01 AM »

The manual for ACER ConQuest 5 is available online:

https://conquestmanual.acer.org


..... <old content below>
We do not plan on providing a separate printed ConQuest 3.0 manual. All functionality as described in the v2.0 manual still holds.  For the additional functionality of ConQuest 3.0, the ConQuest 2 manual is supplemented by the ACER ConQuest 3.0 command reference and the notes and tutorials available from: http://www.acer.edu.au/conquest/notes-tutorials