Latent Class Analysis: profile plot by latent classes

[Yingyi Lin]

Hi Statalist,


I have encountered a STATA issue when I was trying profile plot my dataset after I did the LCA (latent class analysis).

Basically, I was following a web post by StataCorp: see https://www.stata.com/new-in-stata/latent-class-analysis/

Everything works perfect except the step that "visually compare the probabilities of participating in these activities across classes." (please see the attached LCA picture from the website).

I have very similar dataset and also would such an visual output by the 3 latent classes.

The codes I used were:

margins 
marginsplot

For -margins-, there was an error saying that "Warning: prediction constant over observations."
For -marginsplot-, I could only get a graph for all classes (y-axis is "mu").
I wonder if anyone has some thought about this.

Thanks in advance

[Weiwen Ng]

Yingyi,

Marcos Almeida documented how to do this in marginsplot an earlier post. I personally went with -coefplot- (available on SSC, by Ben Jann), as I know how to control the syntax better, but it is a rather clunky process.

I would love to see Stata document the marginsplot process better on its blog.

While searching for Marcos' post, I noticed that Paul Lambe asked a similar question earlier. Paul, if that post didn't resolve your question, I hope this does.

[Yingyi Lin]

Thanks so much Weiwen!

[Paul Lambe]

Hi Weiwen and all,
I am still unable to create the same quality of combined bar graphs as advertised on the web post by StataCorp: see https://www.stata.com/new-in-stata/latent-class-analysis/ Anyone able to advise? Perhaps, as Weiwen suggested, Stata Corporation should document the marginsplot better on its blog i.e. give us the syntax. I for one have upgraded to Stata 15 specifically because LCA has been made so much easier in Stata15.
BW
Paul

My syntax:-
gsem ( lev3 stateschool lowpol highimd <-), logit lclass(C 3)startvalues(randomid, draws(30)seed(123321))intpoints(5)iter(1000)
estat lcprob
estat lcmean, post
margins,dydx(*)noesample predict(outcome (1)) predict(outcome (2)) predict(outcome (3))
marginsplot, by(_outcome) xtitle("not title") recast(bar)

[Tamara vanderDoes]

I don't know if your question still holds but the code is on these slides: https://www.stata.com/meeting/uk18/s..._MacDonald.pdf (slides 34-35)