It might depend on which additional fit indices you want to see.
One possibility perhaps is to try to impute the missing values--or just
to substitute sample means for the missing values.
Also, are you sure two observations are so important to your analysis
that you can't just delete them?
It wouldn't be unheard of say something like the following:
"Because LISREL doesn't report all fit indices in the case of missing
data, and because nly two cases had missing data, models were estimated
both including and dropping cases that contained missing data. Without
these cases the [fit index of interest] was xxx with yy df. We then
included the missing the cases for the purpose of evaluating the
parameters of interest..." or something similar.
If any fit index you want is calculated just using the
observed/expected correlation matrices and not raw data, then you could
perform analysis in two stages:
1. Estimate the observed correlation matrix (including the
missing data cases).
2. Supply the correlation matrix to LISREL rather than raw
data.
3. Tell LISREL that the correlation matrix is based on full
data.
Assuming this even makes sense at all for your problem (and, lacking
details, I'm not sure it does) then this might supply the other fit
indices.
Take this all with a grain of salt. These are just ideas to stimulate
your thinking.
If you want a definitive answer, just ask the people on the SEMNET
listerv discussion list. It takes a little effort to subscribe, but
maybe it would be worth it. Another option is to pose the question at
the Mplus discussion group at
www.statmodel.com (if so, you might want
to pose this as a general question and not a LISREL question).
HTH
--
John Uebersax PhD
Burger wrote:
By the way, when I delete the two observations with missing data, the
output
DOES show all the fit indices.
How can I make Lisrel provide all the fit indices without me deleting
the
observations with some missing data from the Prelis data file?
"Burger" wrote ...
I have estimated a structural equation model in Lisrel (version
.
In the output there is the following information on goodness of fit:
Global Goodness of Fit Statistics, Missing Data Case
-2ln(L) for the saturated model = xxx
-2ln(L) for the fitted model = xxx
Degrees of Freedom = xxx
Full Information ML Chi-Square = xxx (P = 0.0x)
Root Mean Square Error of Approximation (RMSEA) = 0.xxx
I would have expected many more fit indices. Does anyone have an idea
why
only these fit statistics are shown? There are only two observations
with
missing data.
