Daniel posted on Thursday, February 26, 2004 - 6:38 am
Good morning. I understand that when one has a non-linear trend, the interpretation of the effects of time on growth is confusing. I ran a non-specified LGM, with four repeated measures, and the final two factor loadings unspecified. I also have covariates in the model. When I regress my trend factor on the covariates, is the resulting effect affected by the shape of my growth curve? More specfically, my trend has the form: 0, 1, 1.809, & 2.230. The observed repeated measure (smoking) is an ordered categorical variable. The effect from gender (0=male, 1=female) to smoking trend is negative. Is it possible to interpret this effect in terms of log odds difference between females and males, or is the interpretation of the effect of gender on the trend completely confused due to the non-linear trend?
The issue here is that the significance of the regression of the mean and the variance of the slope growth factor refers to the linear part of the growth trajectory, that is, from time score 0 to 1 but not to the rest of the development which does not follow the linear slope. Although the covariates may predict the rest of the trajectory, you can't determine the significance with a non-linear trend of this kind.
Daniel posted on Thursday, February 26, 2004 - 8:17 am
Thanks very much.
Rich Jones posted on Thursday, February 26, 2004 - 3:50 pm
Could you get around this problem in inference by re-scaling the loadings by a factor of 1/2.230 ... therefore 0, .448, 0.811, 1 used as fixed or starting values for the loadings?
bmuthen posted on Thursday, February 26, 2004 - 4:01 pm
If you are interested in the significance of the influence of covariates for other time periods, say the last, why don't you use time scores of 0, *, *, 1, where * means estimated?
Jahun Kim posted on Monday, February 28, 2011 - 5:47 pm
1) I am running GMM with covariates. In my input file, I wrote,
inter slope on age gender; c#1 on age gender; c#2 on age gender;
In output file, I saw... intercept (and slope) on age XXX gender XXX
Categorical Latent Variables C#1 on age XXX gender XXX
How can I interpret them? (e.g. when age significantly influences on intercept, but not on slope and C#1..)
2) When I include outcome variables in my model, in the output file, I can see means of each outcome variables based on each class. Is there any way to know whether the mean difference of class significant?
1. C31 ON x is a multinomial logistic regression. It describes how the x's predict class membership. Within each class the linear regression of the growth factors on a set of covariates describe how growth factor variances are explained by the covariates.
2. You can use MODEL TEST to test these differences.
I have run a conditional 3-class GMM model of life satisfaction (LS).
My predictor (traumatic experiences) is significantly able to distinguish among the three classes of LS. For example, for a one unit increase in my predictor, the chances of being classified in Class 2 are 70% higher than being classified in my reference class.
However, when looking closer at the within-class variations, my predictor does not affect latent growth parameters of Class 2. In other words, my predictor does not affect significantly neither the intercept, nor the slope or quadratic term.
What could this mean? My idea is that even though people in Class 2 might have more traumatic experiences than people from the reference class, there are other factors not accounted for that could shape their trajectory of life satisfaction. Traumatic experiences per sŤ is not enough in explaining it.
Is this interpretation correct? Any clarification would be very much appreciated.