Modification index for threshhold in ...
Message/Author
 Xu, Man posted on Monday, March 28, 2011 - 5:02 am
Dear Dr. Muthen,

I have two repeated factor in a MIMIC type model where I constrained the factor loading and threshhold for items across the two repeated factors. However, I am not sure which section in the mod(all) output I can check if the threshhold are invariant across waves.

Many thanks!

Kate
 Linda K. Muthen posted on Monday, March 28, 2011 - 6:35 am
These modification indices would be under means, intercept, or thresholds. If you can't find them, please send your output and license number to support@statmodel.com.
 Xu, Man posted on Monday, March 28, 2011 - 7:07 am
I see them now. yes, they are at where you suggested. thanks very much!
 Xu, Man posted on Tuesday, April 12, 2011 - 6:41 am
Could I ask a follow up question please? The items were binary and in the modification indicse, I see this:

Means/Intercepts/Thresholds

[i1t1 ] 144.703 0.390 0.390 0.390
[i1t2 ] 144.684 -0.390 -0.390 -0.390
[i1t1\$1] 144.703 -0.217 -0.217 -0.217
[i1t2\$1] 144.684 0.173 0.173 0.173

i1t1 and i1t2 are the same item asked twice. I am not sure what is the [i1t1] and [i1t2] mean here as they are probably not referred to as threshold as [i1ti\$1] and [i1t2\$1]? Many thanks!
 Linda K. Muthen posted on Tuesday, April 12, 2011 - 9:24 am
Note that the modification indices are the same for the two. I would only look at the ones for the thresholds as they are your model parameters.
 nina chien posted on Friday, July 13, 2012 - 1:37 pm
I'm conducting a growth curve model with three time points. The fit is pretty poor because the curve is not linear (it dips/spikes). However, the modindices say:

Means/Intercepts/Thresholds

[ DRINKW1 ] 567.281 -0.975 -0.975 -0.865
[ DRINKW3 ] 567.281 0.450 0.450 0.300
[ DRINKW4 ] 567.281 -0.836 -0.836 -0.599

Therefore, in desperation I added [DRINKW12] to my model statement and now the fit is acceptable. (the model is

i s | DRINKW1 DRINKW3 DRINKW4 )

However, I have no idea what it means for my model to have now called out the mean. I am guessing this is not an acceptable way to improve model fit, but just wanted to verify. Thanks so much.
 Bengt O. Muthen posted on Friday, July 13, 2012 - 4:23 pm
That shows that the growth model doesn't fit. If you had many time points and say one such time point needed the mean deviation from the growth curve, that might have been defensible, but with only 3 time points to begin with it is not defensible.
 Shiko Ben-Menahem posted on Friday, May 06, 2016 - 8:40 am
Dear Bengt,

Under what conditions would you say it is admissible/defensible to free up an intercept for a deviating time point in a growth model with observed variables?

For the piecewise growth model

i s1 | B1@-1 B2@-0.5 B3@0 B4@0 B5@0;
i s2 | B1@0 B2@0 B3@0 B4@0.5 B5@1;

modindices suggest adding [B3c]--which makes the difference between bad fit and good fit--but I'm wondering about the interpretation of doing so.

And if not for a 5-wave growth model, would you say freeing up an intercept could make sense in a 9-wave growth model?

Thanks!
 Shiko Ben-Menahem posted on Friday, May 06, 2016 - 9:12 am
sorry, that should be [B3], not [B3c].
 Bengt O. Muthen posted on Friday, May 06, 2016 - 12:54 pm
So that's a deviation from the linear growth. Unorthodox but seems reasonable if you have a strong substantive reason for it. If you had a time-varying covariate, it might explain it.