I am absolutely a new comer to both the methodology and software.
I would like to know some good references to begin and what example of MPLUS to follow.
I have data on demographic variables on comsumers and their response to whether they are willing to buy a good or not. Based on this data I would like to estimate a Finite Mixture Model to identify some segements based on the demographic variables
There are many references and examples on the website. You might want to start with the following:
Muthén, B. (2001). Latent variable mixture modeling. In G. A. Marcoulides & R. E. Schumacker (eds.), New Developments and Techniques in Structural Equation Modeling (pp. 1-33). Lawrence Erlbaum Associates.
I am a new comer to both the methodology and software.
I have a longitudinal growth data for 7 years with a follow up data of 3 years. There is a gap of 5 years between the original study and the follow up. Form biological point of view, we are not expecting that the outcome continues variable have had a linear pattern during that 5y gap. In fact, a peak in the outcome variable could have been occurred during that measurement gap. I heard that MPLUS using “semiparametric approach with latent variable” can be of help. So, is that possible to predict the values during that gap, even though the pattern is not necessarily linear?
bmuthen posted on Thursday, December 22, 2005 - 4:43 pm
Yes, Mplus can be used for a "semiparametric approach with latent variables". This is typically taken to mean that the growth factors are allowed to have any form of distribution, so not restricted to the usual normal distribution. However, I don't see how this plays an important role in your situation. If you want to predict values during your time gap in observation, this can be done by estimating a factor score for the systematic part of the variation at a certain point in the gap (centering the intercept growth factor at that time). The quality of the prediction, however, would seem to mostly hinge on how well values before and after the gap reflect values in the gap (for instance a peak in the gap period may be inferred if values before the gap are on the increase and values after the gap are on the decrease).
I am not familiar with the “estimating a factor score”. Would you please provide me reference or an example?
Here are some more details about my data. During gap, none of variables (outcome and independent variables) were measured. We only have the chronological age of our subjects during that time. For my subjects there will be three possibilities for outcome variable during the gap: 1: Ascending pattern (those who experience the peak in outcome variable after the gap) 2: Descending pattern (those who experience the peak before the gap) 3: Both ascending and descending patterns (those who experience the gap during the peak) Independent variables are Ht, Wt, and environmental factors such as diet, physical activity levels. For Ht & Wt, we expect to see increase but in environmental factors they might be sustained, increase or decrease. I really appreciate any further comments.
bmuthen posted on Wednesday, December 28, 2005 - 4:59 pm
Factor score estimation is the same as what statisticians refer to as "Empirical Bayes estimation" - it gives estimates of the value of a growth factor for each individual and therefore can be used to predict the outcome. The quality of prediction is however dependent on how many time points you observe for each person. If you expect a peak, you might have a quadratic growth shape and this is not well estimated unless you have several time points that represent the curvature for the quadratic growth curve.
I have a dataset containing responses from 200 surgeons, each of whom rated on a scale ranging from 1 to 6 the appropriateness of a surgical intervention for 16 patients. The descriptions of the patients contained a number of characteristics with two to three levels (represented as dummy variables).
I hope to be able to use MPLUS to estimate a number of regression equations in which the rating of the surgeons is the dependent variable and the patient characteristics are the independent variables and each regression equation represents the results for a “latent class” of surgeons who responded to the clinical characteristics in a similar way. I guess it's a classic market segmentation exercise.
I've tried to find in the user's guide how I could represent that the responses are clustered within the surgeons and that the latent classes should relate to the surgeons.
Ultimately, as output I would like to report the model coefficients for the latent classes of surgeons and the probability / prevalence of class membership.