I want to conduct path analysis with logistic regression model.
Exposure variables are continuous (body weights at different time points), and mediator variables are also continuous (age etc.).
Outcome is dichotomous variable (existence of a disease).
My interests are the direct and indirect effects of exposure variables on the outcome. So I am trying to do path analysis with logistic regression model.
I use SAS, but I could not find how to apply logistic model with PROC CALIS.
Are there any options to select logistic model? or some other procedures?
This may be done by conducting several regression analyses and re-calculating the output coefficients, but I want to avoid self calculation.
Some websites say it can be done by Mplus. Is it an easier way to use Mplus?
Give up on SAS and move to Stata! Easily done with their SEM procedure.
Thanks you for quick answer! As you advised, STATA seems to support binary outcomes in SEM.
http://www.stata.com/features/structural-equation-modeling/
But I do not have any experience with STATA.
Before I give up on SAS, I would like to wait for other suggestions.
Mplus is definitely easier than SAS from what I know (I can't comment on STATA never having used it). A nice feature to Mplus is that the code for the model is the same no matter what types of outcomes you are using (i.e. you use simple statements like y BY m x; m BY x), and Mplus will estimate the appropriate models depending on you telling it that a variable is dichotomous (you define this in the variable commands). The big caution with Mplus, however, is that the default for dichotomous variables is to estimate a probit regression, but you can add a line to request a logistic model (you simply need to be sure you know what's going on behind the scenes, as Mplus has a lot of assumptions built in, which are often what you want, but you should be sure to check that this is in fact the case). The Mplus user guide is freely available, so I would suggest consulting the examples there to see more about the software: http://www.statmodel.com/download/usersguide/Mplus%20user%20guide%20Ver_7_r3_web.pdf
As for SAS, I've never tried path analysis in SAS with dichtomous outcomes. There may be a way to do it, but I'd imagine it's not immediately straightforward from what limited experience I have with calis/tcalis. Perhaps others can contribute more in that regard.
MPLUS is definitely a good choice. Testing direct and Indirect effects and using bootstrapped for the error terms of the product term is quite easy to specify. If you have access to MPLUS, give it a try out.
Thank you Michael and Diego.
Mplus looks quite good.
After that I found some STATA users in my department, so I am thinking of trying with STATA. For generalized linear model, version 13 seems to be needed.
http://www.stata.com/stata13/generalized-sem/
I am feeling that many SAS users are shifting to other packages.
Anyway, thank you for all helpful answers!
Hi all,
I use both Mplus and STATA. Unless your variables are all continuous then do not use STATA. The reason for this is binary and categorical function has recently been introduced in STATA but the information you need regarding model fit such as CFI, TLI, RMSEA is not available in STATA yet for these variables.
For continuous data, Mplus and STATA give the same answer. Mplus for SEM is easier to use.
Hi, thank you for useful information. Actually, STATA did not provide those indexes. I will check Mplus, if I have a chance.
- Badges
- Science topic
- Similar topics
- Psychology