Re: Weird standardized regression coefficient values

71 views
Skip to first unread message
Message has been deleted
Message has been deleted

Terrence Jorgensen

unread,
Dec 29, 2014, 11:04:58 PM12/29/14
to lav...@googlegroups.com

Same output from SPSS (with slight differences due to FIML vs. mean substitution methods of handling missing data).  

If you aren't analyzing the same exact data, you can't expect the same exact results. Given how much the results differ (more than slightly), I'm wondering how much missing data there are.  What is the N after listwise deletion?  Are these "Demog..." variables all dummy codes or continuous -- is it possible they are being treated as numeric in SPSS but as ordered categories in lavaan?  Providing your entire syntax (and perhaps a portion of your data if they aren't private) might be more helpful than only snippets of the output.  For example, I notice some parameter labels in your lavaan output -- does this indicate there are some across-group constraints that are not being made in the OLS model with SPSS?

Terry

Message has been deleted

yrosseel

unread,
Jan 7, 2015, 3:36:34 PM1/7/15
to lav...@googlegroups.com
On 12/30/2014 08:06 PM, Dimitri wrote:
> Terry, thanks for your reply. I got the same discrepancy using listwise
> deletion in both lavaan and SPSS. The problem then unexpectedly went
> away when I standardized the dependent variable (subtract mean and
> divide by SD), without any other changes being made.

This makes no sense at all. Could you provide us with a reproducible
example?

Yves.

Message has been deleted

yrosseel

unread,
Mar 25, 2015, 4:58:21 AM3/25/15
to lav...@googlegroups.com
On 03/24/2015 03:18 AM, Dimitri wrote:
> Yves, a follow-up: this happened again with the "MLR" estimator.

What exactly happened again? I went back to see the original posts in
this thread but they were deleted (presumably by you?)

I am not sure what to make from the output in your post: the MLR
estimates (in the 'Estimate') column are different from the ML
estimates. This can only mean that there is something strange going on
here. Since one of the variances (the last one for MLR, 0.001) is very
small, this explains the differences in the standardized values.

The question is: why do you get different (unstandardized!) estimates
between ML and MLR?

I would be happy to investigate this if you send me the data.

Yves.
Reply all
Reply to author
Forward
0 new messages