WLSMV better fit with non-ordered variables

241 views
Skip to first unread message

Emanuele

unread,
Aug 2, 2020, 5:19:02 PM8/2/20
to lavaan
Hi,

I am running CFA with a set of items on a 5-point Likert, with WLSMV and ordinal variables ("ordered = c"()), but model fit is not satisfactory. Similar model fit with MLR and MLM. However, when I run WLSMV without declaring variables as ordered, model fit is excellent. Would it make sense to accept model fit with WLSMV as run on a continuous variables?

Thank you
Emanuele

Henrique Oliveira

unread,
Aug 3, 2020, 3:40:31 PM8/3/20
to lavaan
Hi,

My data was 24 itens, all using a five points Likert scale. I used the option 'ordered=TRUE' in cfa command, instead of 'ordered = c()':
myModel.v1.fit <- cfa(myModel.v1, data=my.data, std.lv=TRUE, ordered=TRUE)
The estimator DWLS was automatically chosen (there is no need to specify it)

Best,
Henrique

Emanuele Fino

unread,
Aug 3, 2020, 4:31:31 PM8/3/20
to lav...@googlegroups.com
Hello,

Thank you very much! I tried and I've got the same result as with ordered = c(). I've read other posts and apparently it's not a good idea to assume continuous data with WLSMV, so I guess I've got no other options but to accept that. It's a shame because my model fits great with just estimator = WLSMV. I would be interested to understand what does lavaan compute when neither ordered = c() nor ordered = TRUE are provided, anyway. 

Best wishes 
Emanuele 

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/af8aff5b-d480-49f4-874a-43e921ec21ddo%40googlegroups.com.

Guido

unread,
Aug 4, 2020, 3:25:15 AM8/4/20
to lavaan
I think this article may be of interest for you concerning your question: https://pubmed.ncbi.nlm.nih.gov/26174714/
By the way if you don't declare your variables as ordered it is my understanding that Lavaan automatically uses ML in stead of WLSMV. 

Cheers, G

Henrique Oliveira

unread,
Aug 4, 2020, 7:03:08 AM8/4/20
to lavaan
Hi,

Very nice reference. Thanks.
I first used SPSS with ML but the reviewer refused these approach and told me to use Lavaan.
Fortunately for me, I have now best fit results using DWLS in Lavaan.

All the best,
Henrique

Emanuele Fino

unread,
Aug 4, 2020, 7:19:03 AM8/4/20
to lav...@googlegroups.com
Hi,

Thank you very much for the useful reference. I totally agree with Henrique, it is really great that we can have a range of estimators with Lavaan, including DWLS / WLSMV! Sorry for my delay in replying, I went back to my analysis and I tried what Guido said. My results suggest that if you don't declare your variables as ordered but include estimator = "WLSMV" you won't get ML, though, as results differ from when I use estimator = "ML".Interestingly, if I declare my indicators as ordered and use WLSMV, I get a very high CFI.scaled (.997) but unsatisfactory RMSEA.scaled (.137). The same did not happen when I (erroneously, I guess) ignored the ordered command.

Best wishes
Emanuele



--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages