I am using "traj" module in Stata. What would be the right way to store the data for further use. I mean, in SAS it is something like this:
PROC TRAJ DATA=A OUT=OF OUTPLOT=OP OUTSTAT=OS
But what would be the analogue in Stata? I get an error when trying:
traj, out(of) outplot(op) outstat(os) var(measure*) indep(age*) model(cnorm) min(0) max(10) order(2 1)
The error is either "variable op not found" or "option outstat() not allowed"
Any suggestions and references are appreciated.
The syntax for the traj command in Stata differs from that in SAS. Please read the command's help (type help traj) in Stata or its extended version with guided examples (http://repository.cmu.edu/cgi/viewcontent.cgi?article=1405&context=heinzworks) to see which options are available.
If you need to store group assignment and probabilities, the traj command in Stata outputs them in the variables _traj_Group and _traj_ProbGn, where n is group number. These are overwritten by the next call to traj, so if you execute several commands in a row and want to save the estimated assignments and probabilities for all of them, you have to rename the variables, e.g.:
traj ...
ren (_traj*) (m1_traj*)
traj ...
ren (_traj*) (m2_traj*)
You can then export these variables (together with subject ids) or store them with the data by saving (a portion of) the dataset by the appropriate Stata commands.
If you run a more advanced version of the model with time-constant and/or time-varying factors, the estimates of these factors are not saved as variables with the dataset (which is common for estimation commands). If you want to store these results, you can e.g. use the est* family of commands (eststo, esttab, estout). Type net install estout to get the package and then read help for its commands. Alternatively, you can use the standard estimates command after estimation with the relevant arguments and options (e.g. estimates save).
Dear Alexander. Could you, please, extend your advice a little? Just remember, I am not a statistician.
I have, btw, found a very nice site with a procedure for calculating posterior probabilities after traj:
https://andrewpwheeler.wordpress.com/2016/10/06/group-based-trajectory-models-in-stata-some-graphs-and-fit-statistics/
It is exactly as you said - Andrew Wheeler uses this _trajGroup
What I still need is the possibility to compare the subpopulations at baseline (chi square etc.). How can I store the traj results in Stata and then compare the groups? I am running a censored normal model with two subpopulations. I would really appreciate a Stata code example or a reference. Sorry, for bothering you. I think, it can be useful for other members as well. This group-based trajectory modelling is going to be a very popular procedure when it is not limited anymore to R.
Actually, I think I've found it. Sorry.
If we want just to see the data then e.g.:
. by _traj_Group, sort: tabulate gender
If we want calculate some tests then (if ordinal) e.g.:
. tabulate gender _traj_Group, chi2
or (if continuous) e.g.:
. kwallis age, by(_traj_Group)
I think, the problem was that I was expecting something much more complex. :-) Bobby L. Jones did really good job with this traj. _traj_Group is very easy to use.
Hi,
thanks for this thread, especially the link from Andrew Wheeler. There may be an obvious answer to this question but I am also not a statistician. I have noticed from Nagin’s book, chapter 5 and 7, that CIs are useful for diagnostics but that bootstrapped CIs are required. With more and more journals expecting CIs, I would like to report them for my thesis. I can see from the code that I can get the CIs on the graph itself but I cannot see them in the output. Am I missing something?
I have found the answer to my question, the CIs are stored in the estimates and can be retrieved using the e(plot1) matrix command.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size