group analysis error for missing category

106 views
Skip to first unread message

mattw...@gmail.com

unread,
Aug 27, 2014, 12:43:18 PM8/27/14
to lav...@googlegroups.com
Hi,

I am conducting a multiple group analysis among survey respondents and receive the following error:

"lavaan ERROR: some categories of variable `C1d' are empty in group 2; frequencies"

When I checked responses to question 'C1d' I found that none of the survey respondents in group 2 chose category 4 - "Strongly disagree", whereas
other groups chose answers across all categories.

Is this question unusable or is there a way to get lavaan to work through this?

Thanks for any assistance

Matt

Terrence Jorgensen

unread,
Aug 27, 2014, 1:35:36 PM8/27/14
to lav...@googlegroups.com

When I checked responses to question 'C1d' I found that none of the survey respondents in group 2 chose category 4 - "Strongly disagree", whereas
other groups chose answers across all categories.

Is this question unusable or is there a way to get lavaan to work through this?


I've run into the same problem using other software as well (Mplus, LISREL).  As far as I know, you will need to collapse the categories in the other groups so they have the same number of categories as in group 2.  Something I haven't tried, though, is adding a single observation to group 2, which would have C1d == 4 and missing values for all other variables.  That might trick the program into recognizing 4 categories when it estimates thresholds and subsequently the polychoric correlation matrix, but I'm not sure it will work.

Terry

Seongho Bae

unread,
Aug 27, 2014, 3:21:58 PM8/27/14
to lav...@googlegroups.com
I'm not sure. You may fill NA values with imputeMissing() function in mirt() package.


--
Seongho Bae

2014년 8월 28일 목요일 오전 1시 43분 18초 UTC+9, mattw...@gmail.com 님의 말:

mattw...@gmail.com

unread,
Aug 28, 2014, 12:16:33 AM8/28/14
to lav...@googlegroups.com
Thanks for the response Terry.  I will give that a try.  I could collapse the question into a binary Agree/Disagree.

Regards,
Matt

mattw...@gmail.com

unread,
Aug 28, 2014, 12:18:14 AM8/28/14
to lav...@googlegroups.com
Thanks for the response Seongho.  I do not have any missing responses.  It is a missing category because none of the respondents chose that particular response.

Regards,
Matt

Seongho Bae

unread,
Aug 28, 2014, 3:50:39 AM8/28/14
to lav...@googlegroups.com
Matt,

If particular responses were 'Strongly agree' to 'Strongly disagree', You may can convert numerical (integer) vector; Like Likert Scale, 1 to 5 or 1 to 7.

If NA value appeared within converted vectors, You may fill NA values.

But, Without NA values, You may try to collapse Agree or Disagree.

Question: Your main advices were collapsing the first category into the second, and the last category into second last. So, I have a one more question. Can I convert this item responses to dichotomous type to save degree of freedom?
 
Answer: You can do this of course, but you'll lose the property of you test being polytomous and ordered (and as a consequence lose precision when trying to estimate θ, if that's your end goal). How you try and dichotomize the data will also affect the fit statistics as well as the estimated parameters, so it's not entirely clear what the best way to dichotomize is. I would recommend trying to preserve the nature of your test to the best of your ability, since in the end that's likely what you are trying to understand. Collapsing from 7 to 5 categories isn't unheard of in the literature, but reviewers may have a bigger issue when going from 7 to only 2. Cheers. 
https://groups.google.com/d/msg/mirt-package/0fG60G8chzY/4jWwdOlUimoJ

Seongho Bae

2014년 8월 28일 목요일 오후 1시 18분 14초 UTC+9, mattw...@gmail.com 님의 말:

Phil Chalmers

unread,
Aug 28, 2014, 6:27:21 PM8/28/14
to lav...@googlegroups.com
This seems like an interesting phenomenon, and is something I thought about a while back as well (I'm somewhat surprised to hear it exists in Mplus and LISREAL though). What happens when estimating intercept parameters in multiple group settings where one group has no instantiations of a relevant category is that no information about the parameter estimate can come from that respective group (e.g., when maximizing the likelihood, all the likelihood information comes from only one group instead of two). This isn't to say that it doesn't exist, it's just that it doesn't exist in isolation when considering one group at a time, and this can cause issues for estimators which need intercept and covariance parameters to be identified for each group. Basically, the idea is to set the parameter estimates equal across group so that they can be uniquely identified, but for the group without the category of interest the parameter is really just a placeholder since all the information comes from other sources. I've attached an example which doesn't require collapsing the data arbitrarily to get around this, and I imagine lavaan could do something like this as well (at least with the MML estimator, not sure about the others). Cheers.

Phil
multiple-group missing category.R
Reply all
Reply to author
Forward
0 new messages