Message/Author |
|
|
I think this may be a very simple question, and it may just be about output interpretation, but I'm not sure. I am using the auxiliary (r) option to explore the extent to which a covariate predicts latent class membership. i have done this successfully for boys and girls separately, looking at the pseudo-class draws. However, what I really want is to be able to compare across as well as within gender. Can I use the knownclass option in combination in order to examine gender differences? When I run a model with: classes = sex (2) ldf(4) ; knownclass = sex (kz021=1 kz021=2); Either I don't get all comparisons I expect to, or I do, but I don't understand the output! Many thanks |
|
|
Auxiliary (r) should not be used with more than one latent class variable. |
|
|
Many thanks for your swift reply - I'm confused by it, though, since sex is a defined, known variable, not a latent class...? Thanks |
|
|
Even though one categorical latent class variable is a known class variable, you still have two categorical latent variables in the analysis so AUXILIARY (r) should not be used. |
|
|
Sorry Linda, just to check, you mean that the gender variable is treated as a latent variable even though it's known? So, second question then - is there a way to look at gender differences using the auxiliary (r) command? I'm after comparisons between and within gender for my 4 classes (grouping doesn't work, of course, because these are mixture models). Thanks again, Bonny |
|
|
Yes, the known class categorical latent variable is the same as gender. No, AUXILIARY (4) cannot be used in this case. Instead do not use gender as a known class variable. Instead regress the categorical latent variable on gender. |
|
|
Following up on a similar question as posed in this thread. I am new to Latent Class/Profile Analysis and am currently attempting to get an LPA to run off of 5 continuous variables while also distinguishing differences between a treatment and control condition. I have used both Auxiliary and knowngroup variables to model this treatment/control, but appear to get slightly different results. What exactly is the mathematical difference between these two? Thanks, Andrew Iverson |
|
|
You mention auxiliary r. There is now a better approach in auxiliary R3STEP. See our web note 15. Knownclass is basically the same as having a binary covariate. I assume that you include c ON cg; where cg is the knownclass variable. Including the covariate in the model implies that the classes are formed also based on information from this covariate, not only using the information from the LPA indicators. |
|
|
My goal is to examine if the classes/profiles created by the LPA indicators differ across the knownclass covariate (in this case, a variable which specifies if the participant is in either a treatment or control group). So ideally, I want to see the program create c (the profiles), and then examine the differences between the knowngroups. So my current input is: VARIABLE: NAMES ARE x y1-y5; CLASSES = c(4) xclass(2); knownclass = xclass( x=1 x=2); Plot: type is plot3; series is y1 (1) y2 (2) y3 (3) y4 (4) y5 (5); ANALYSIS: TYPE = MIXTURE; STARTS = 100 20; OUTPUT: TECH1 RESIDUAL; SAVEDATA: File is Teststhis2.out; save is cprob; format is free; Does this leave the knowngroup as an auxiliary variable, or is it being included in the creation of the profiles? Andrew |
|
|
xclass will be included in the creation of the latent classes. Also, the default is that the two latent class variables are unrelated - you may want to say c on xclass. Alternatively, if you don't want xclass to be part of creating the latent classes you can specify xclass as an R3STEP auxiliary covariate of the latent class variable. |
|
|
I am following up on the questions in this thread. I am running a growth mixture model and would like to determine whether the 2 trajectories we have identified are related to a known class (3-level categorical variable of drinker status). I do not want the known class to be part of creating the classes. Here is the important syntax: AUXILIARY = drinker (R3STEP); MODEL: %OVERALL% i s | ang_7@0 ang_6@1 ang_5@2 ang_4@3 ang_3@4 ang_2@5 ang_1@6; My question is whether or not it is possible to obtain information about which known class levels are different from each other in the output? We have the C#1 ON Drinker significant output, but do not have the information about which levels are different. Can this be obtained? |
|
|
If you have Drinker as a Knownclass variable with 3 categories then there should be 2 parameter estimates: C#1 ON Drinker#1 c#1 ON Drinker#2 |
|
|
We are only getting the following output: C#1 on DRINKER C#2 on DRINKER We have not specified DRINKER as a knownclass variable. We had included it only as an AUXILIARY variable, because we do not want it included in the model calculation. Do we need to also specify it as a knownclass to get the full output? |
|
|
You need both dummy variables for Drinker in your Auxiliary statement. Mplus does not automatically break up the3-categoy Drinker variable for you. |
|
Back to top |