Null/baseline model

[Helene]

Hi,

I am estimating a SEM (model M1) with WLSMV estimation method.
I would like to know exactly how is defined the null / baseline model in lavaan. I tried to estimate a sem with no regressions at all, only the measurement model. But I don't get the same chi-square, nor df, with this "base" version as the one reported in baseline.chisq or baseline.df (scaled or not) for the model M1 I want to estimate.

Best regards,

Hélène

[Terrence Jorgensen]

I would like to know exactly how is defined the null / baseline model in lavaan. I tried to estimate a sem with no regressions at all, only the measurement model.

I think you are confusing different uses of the ambiguous term "baseline model".  It sounds like you are taking about a best-fitting model against which to compare more restricted models, as in comparing a restricted structural model among common factors (i.e., your latent regression model), compared to a CFA with unrestricted covariances among common factors (i.e., your measurement model).  

The "baseline" model reported in summary() or fitMeasures() output is instead the "null" model that is used to calculate incremental fit indices such as CFI and TLI.  This is a worst-fitting model, not a best-fitting model.  For continuous data, the default null model in lavaan (and all other SEM software I am aware of) is the "independence" model, which constraints all covariances to zero, and only freely estimates means and variances.  To see the default version of the independence model for categorical data, you can inspect the parameter table lavaan uses. 

fit <- lavaan(...)

lav_partable_independence(fit)

Terrence D. Jorgensen

Postdoctoral Researcher, Methods and Statistics

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

UvA web page: http://www.uva.nl/profile/t.d.jorgensen

[Reubs Walsh]

Hi Terrence and Helene,

Late addition I know. I'm trying to compute an approximate bayes factor. Blavaan gives me BICs for the specified model. How do I get the BIC for the independence model? Presumably I need the Lavaan SEM output (when run via blavaan and therefore via jags) for the independence model and specify that as I otherwise would by putting e.g.:

R> fitMeasures(fit1)

(where 'fit1') is the lavaan object outputted from the SEM estimation? So something like

R> fitMeasures(baseline_model(fit1)) 

#or

> fitMeasures(fit1.baseline)

Thanks in advance.

[Mauricio Garnier-Villarreal]

Reubs

blavaan calculates the Laplace approximate log-bayes factor, with the function blavCompare

For example, you have your blavaan objects fit1. Then you need to estimate the null model, such as

fit_null <- bcfa(lav_partable_independence(fit), data=dat)

blavCompare(fit1, fit_null)

There you will have the log-bayes factor. If you want it in the regular bayes factor metric do exp(log-bayes factor)

hope this helps