| Message/Author |
|
|
| Jon Heron posted on Monday, February 29, 2016 - 1:48 am
|
|
|
Hi Bengt/Linda
I've been trying to find some information about TSCORES in the technical appendices but with no success.
can you point me towards any additional info other than what's in the manual?
many thanks, Jon |
|
|
|
|
| I don't think we have anything further. |
|
|
| Jon Heron posted on Tuesday, March 01, 2016 - 1:24 am
|
|
|
that's a shame
is there anything more general? I'm assuming that "tscores" is an Mplus term but the method started out called something else, perhaps in a JASA paper?
best wishes, Jon |
|
|
|
|
| No, I haven't seen this done elsewhere. It is providing the flexibility that the usual repeated measurement software have with individually-varying times of observations in a long format, 2-level approach to growth modeling - but doing this in a single-level model in the more flexible wide format. So getting a random slope for a random time variable. I guess that is pretty unique and maybe we should have written about it to tout it. |
|
|
| Jon Heron posted on Tuesday, March 01, 2016 - 11:56 pm
|
|
|
Thanks Bengt
I am contemplating adding this to a growth modelling lecture I am writing so wanted to make doubly-certain I understood what was happening.
Is that the nub of the issue, it's treating age as a time-varying covariate with a wave-invariant random effect on the repeated measure. |
|
|
|
|
I cannot make this work for me.
I am trying to estimate within person slopes and intercepts for two variables x & y collected at 5 waves. Y'all previously told me I could use the AT command for this as long as the x variables (x1, x2, ....x5) are defined as time scores.
when I define the x vars as Scores and use this code
i s | y1-y5 AT x1-x5;
It says I have to use type random, when I set type random it says.....
Random variances are available only with ESTIMATOR=BAYES. Specify ESTIMATOR=BAYES in the ANALYSIS command.
When I set estimator= bayes it says.....
Random effect variables declared with the AT keyword are not allowed with ESTIMATOR=BAYES.
When I use ON instead of AT, and don 't set TScores, it ends up telling me I can't use Bayes estimation with single level models.
Eventually we want to be able to test whether this relation differs across known groups. |
|
|
|
|
The labyrinth you are describing is because not all features are available for all estimators.
This statement
i s | y1-y5 AT x1-x5;
is equivalent to
i by y1-y5@1;
s | y1 on x1;
s | y2 on x2;
s | y3 on x3;
s | y4 on x4;
s | y5 on x5;
I would say that this model is simplest with two-level estimation but I don't see the model so it is hard to give you more info. In two level the model is this
%within
s | y on x;
%between%
s y; s with y;
What relations are you referring to in the last sentence? |
|
|
| Back to top |
