[R] Structural equation models with R

[John Fox]

Dear Andre,

> -----Original Message-----
> From: André TavaresCorrêa Dias [mailto:atcdias at biologia.ufrj.br] 
> Sent: Thursday, March 10, 2005 6:14 AM
> To: John Fox
> Cc: r-help at stat.math.ethz.ch
> Subject: RE: [R] Structural equation models with R
> 
> Dear John,
> Thanks very much! I don’t know how I didn’t see this before. 
> I checked a thousand of times.... 

(Recall that there was a typo in Andre's model specification.)

> There is one more thing that I 
> can not understand. What are the possible reasons for 
> problems on calculation of confidence interval for RMSEA? For 
> the same model I sent before (after correction of the 
> variable name) the lower CI output was NA:
> 
> Model Chisquare =  10.824   Df =  13 Pr(>Chisq) = 0.62558
>  Goodness-of-fit index =  0.91656
>  Adjusted goodness-of-fit index =  0.76895
>  RMSEA index =  0   90 % CI: (NA, 0.15652)
>  BIC =  -60.425
> 
> I have other models with the same behavior, even when the 
> estimative of RMSEA is different from zero.
> 
> Model Chisquare =  15.165   Df =  14 Pr(>Chisq) = 0.367
>  Goodness-of-fit index =  0.89038
>  Adjusted goodness-of-fit index =  0.71811
>  RMSEA index =  0.053558   90 % CI: (NA, 0.19141)
>  BIC =  -61.564
> 

The method used to get the confidence interval (from Browne and Du Toit,
Multivariate Behavioral Research, 1992, 27:269-300) can produce a lower
bound above the RMSEA estimate or an upper bound below the estimate; when
this happens, the bound is set to NA. One of the nice things about R is that
you can look at the code for a function -- in this case, summary.sem() -- to
see exactly what it does. Take a look at how RMSEA.L and RMSEA.U are
computed.

Regards,
 John