|
|
LPA with Ipsative Data/Log Ratios |
|
Message/Author |
|
|
Hello, I have a two part question: (a) Can Latent Profile Analysis in Mplus handle ipsative data where the predictors/indicators in each observation/row sum to a constant(1)? Each indicator is a proportion/percentage to be exact. If so, do I do anything different in specifying the LPA model other than transforming before I run test models with varying number of classes? (b)If my LPA indicators are in log ratio form, will LPA handle this? Are there any limitations with this approach in Mplus? This was suggested by a colleague, but I wanted to confirm here. |
|
|
(a) I would recommend that you don't use one of the variables, say the last proportion which doesn't have any or bring any additional information. (b) This is the preferred model. Any linear regression would directly translate into multinomial logistic regression result. Also using LPA directly on the proportions is an inconsistent model because the model would imply that negative proportions are possible. |
|
Back to top |
|
|