Model with sampling weights

[alexander....@gmail.com]

Dear all,

I am trying to estimate a model which disables the default functionality in lavaan of handling sampling weights (using the argument 'sampling.weights'). By default, lavaan normalizes the weights in such a way that it sums to the total number of observations. More critically, this is also done in every group. Is there a chance to disable the default. At least I would like to see the normalization at the level of all observations and not group by group.

Thank you and kind regards,

Alexander

[Terrence Jorgensen]

Hi Alexander!  

Not sure what the context of your request is, but I recently wondered the same thing.  MuML estimation of MSEMs as multiple-"group" (level) models assumes balanced cluster sizes, and I wondered whether weights could be used as a trick to resolve unbalanced cluster sizes.  I'm sure there are other contexts in which un-normed weights could serve as a useful hack.

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen

[Stas Kolenikov]

Wouldn't lavaan.survey do this properly? I'd be disappointed (in Dan Oberski) if it nonchalantly rescaled weights.

-- Stas Kolenikov, PhD, PStat (ASA, SSC)  @StatStas
-- Principal Scientist, Abt Associates @AbtDataScience
-- Opinions stated in this email are mine only, and do not reflect the position of my employer
-- http://stas.kolenikov.name
 

--
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 post to this group, send email to lav...@googlegroups.com.
Visit this group at https://groups.google.com/group/lavaan.
For more options, visit https://groups.google.com/d/optout.

[alexander....@gmail.com]

Hi Terry, hi Stas,

thanks for your responses.

@Stas: I thought that models with missing data and weights cannot be handled with ML with lavaan.survey. In addition, it (currently) only works for continuous data.

@Terry: My application is local structural equation modeling where I need weights for each fitted model. In these models, I want to have both missing data and multiple groups.

I got a short note of Yves that he will include an option for disabling the default when he finds some time.

Kind regards

Alexander

[Paco Arévalo]

Dear all,

I am trying to estimate some SEM models. I have started with a very simple CFA and a second order and bi a factor model. The specific issue is that I'm working with ordinal variables (1-3 scale) and the data comes from a national survey so it has sampling weights. I know that for now lavaan does not support estimator for ordinal data and sampling weights simultaneously. My data set is of 3700 observatios, 13 ordinal variables and the weighting factor varies from 35.23 to 1698.88. I have tried to analyze the polychoric correlation matrix (weighted), and compare the results to full raw data, the results vary a lot. Also, I have opted for expand rows and evaluate the models with replicated raw data (700k observations) as SPSS does in some of it routines. However, I know that the statistics that I get fail because the variance is reduced by the expansion, also I know that I have problems with chi-square family tests because of artificial N. Do you think there are another way for doing that?