I am currently performing a latent profile analysis in Mplus and I have two questions:
1) In the output of the LPA analysis for 3 classes are the mean values and variances for each class. Is it normal that some variances are not estimated and marked with ********?
2)My second question is about outliers in the LPA analysis. Do I have to exclude a few outliers before performing the analysis or is there a big risk that they distort the results?
Thanks in advance!
1) This seems odd and not normal. Are you getting any warning messages before the MODEL FIT output in Mplus? Do you have variables in your data set with very large or very small variances?
2) I would run the analysis with and without the outliers to see how they impact the results (number of classes, class profiles)
Thanks for the answer! No I not getting any warning messages before the Model fit Output in Mplus. I will check again if there are very small or large variances. Could this result in the non possibility to estimate the variance within a latent subgroup? The mean value for this variable is in the output.
Thanks in advance!
Hi
(1) the asterisk means your number could not fit in the ouput. It probably requires rescaling it
(2) possibly many answers to this. But if the outlier has serious issues, why not remove it? It is not jus about how it affects the analysis; but how it is a pre-analysis decision-making
Also, do the variance estimates come out fine (without ***) when you use the Mplus default settings (according to which the variances are set equal across classes)? Sometimes, estimates become unstable when the variances are freely estimated across classes.
Otherwise, I would suspect the same reason as Gerard Chung (a variance that is too large to be displayed in the output)--since you're not getting a warning message about estimation problems.
Thanks for the answer! yes exactly, also if I choose the default setting of Mplus then the variances for some variables can also not be estimated and are marked with (****). I think also this could be the reason, because I have very large variances in these variables and I have indicator variables with different scales. I was just wondering if I rescale these variables is this then necessary for all variables? Otherwise, the relationship of the variables to each other will change. I have in my data inflammatory markers with different scalings. Some markers occur in a much higher concentration than others. I read on the Mplus site that i should rescale the variance by dividing the variable by a constant, which should result in variances between 1-10.I just don't quite understand yet how this is meant concretely and whether this refers only to the affected variables or to all indicator variables (even those with smaller variances)? Thanks in advance!
Yes, you could rescale the variables with large variances by dividing by a constant (the other variables can be left as is). The rescaling can be done using the DEFINE command in Mplus. With that command, you can create new variables, e.g.:
DEFINE: X_new = X/10;
where X_new is the new variable that is equal to the original variable X divided by 10.
The newly defined variables are then used in the model. They need to be listed in the USEVARIABLES list instead of the original variables.
Thank you very much! Should I choose a constant that results in variances between 1 and 10? (so I read it on the MPlus page). I have partly very large variances of e.g. 286171.987. For this variable the smallest value is 55.03 and the largest value is 6336.24. Accordingly, I would have to choose a very large value for the constant to get variances between 1-10. Should I still proceed in this way? Thanks in advance!
An alternative to the DEFINE command in Mplus is to rescale the variables prior to the analysis in other software, e.g., SPSS. I sometimes use this to automatically rescale variables to a certain range (e.g., to values 1-10). Although it should not affect the class formation, I find that it helps with the interpretation of the results when all variables are on the same scale.
- Badges
- Science topic
- Similar topics
- Psychology