Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: sem bartlett factor scores

From   Stas Kolenikov <>
Subject   Re: st: sem bartlett factor scores
Date   Sun, 4 Dec 2011 12:13:35 -0600

Maria addressed me privately, and told me that she has a model with
about 200 items in 6 factors (which leads to about 20000 covariance
matrix entries, and a little over 400 parameters, and I cannot imagine
what size a data set should be to ensure applicability of the
asymptotic results, such as goodness of fit tests). I agree with Cam
that the best option, methodologically, would be to fit the full model
with both the causal part in the latent variables, and the measurement
part that looks like a CFA model. Just setting this model up would be
an exercise in macro management, though. Alternatively, Maria can fit
each factor separately (assuming that her model has factor complexity
1 for each variable) using -confa-, get Bartlett scores out of that,
and run the regression -- just to see whether it will be far off the
full model, which still serves as the golden standard. (I would only
see any sense in running a factor regression if I had to reproduce
somebody else's results, in case the original paper(s) used this
subpar method; I would do every effort possible to run the full model
in -sem- or -gllamm-.)

On Sat, Dec 3, 2011 at 5:30 PM, Cameron McIntosh <> wrote:
> Maria,
> You don't need to save factor scores and then do a path analysis or multiple regression with those factor scores, if that's what you mean. Besides being rather tedious, this method also leads to underestimation of standard errors unless you use specialized procedures:
> Hoshino, T., & Bentler, P.M. (2011). Bias in Factor Score Regression and a Simple Solution. UCLA Statistics Preprint # 621.
> In the -sem- package in Stata 12, you could just directly do a multiple indicator latent variable regression, where your latent dependent variable is simultaneously regressed on all latent predictors and covariates... quite easily I'm sure.
> Cam
>> Date: Sat, 3 Dec 2011 16:23:19 -0600
>> Subject: st: sem bartlett factor scores
>> From:
>> To:
>> Dear
>> I 'm using SEM in stata 12 and looking through the manual. I'm doing
>> confirmatory factor analysis and I need to predict factor scores by
>> the Bartlett method. How can I do that with SEM?
>> I can not use the factor command because I'm doing CFA.
>> What I want to do is a regression among factor scores (so a latent as
>> a dependent and other factors as explanatory variables ( and other
>> covariates too)
>> Also, I have tried to use confa command but I get some errors and that
>> is why I wanted to try SEM

Stas Kolenikov, also found at
Small print: I use this email account for mailing lists only.

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index