Hello, I want to use Mplus to cope with the missing data problem and the categorical variables. I would appreciate your advice about the right statistical analysis.
I want to assess the effect of 2 independent Variables (6 and 2 levels) on 4 dependent variables with 16 repeated measures. Thus we get 12=6*2 experimental cells. Moreover the dependent dichotomous variables have a complex structure. 0 | 1 |***first DV missing | 0 | 1 |***second DV missing | missing| 0 | 1 |***third DV missing | missing|miss|0|1|***4th DV
My main interest is about the "main effect" and the interaction over all wavesof the both independent variables. Additionally I am interested if the effect of the IDs is changing over the 16 waves (time effect). Possible moderators like sex and status should be taken into account. I would appreciate any hint! Can I use Mplus to evaluate this and which analysis is appropriate? would be a multilevel model a good idea? What about multinomial logistic regression with repeated measures? More over I wanted to evaluate this by LGC, but then I think I don't know which level or the ID is the best. But my main interest is not how the slope is changing over the time, but the mean effect of the time is different for the 12 experimental cells. THANK you so much for helping me.
I just wanted to symbolize the structure of the DVs. All my dichotomous DVs represeant different kinds of participation behaviour such as "person responded"(DV2). So just if the participant has the value DV2=1 in wave x, the person can have any value for DV2, DV3,DV4 in wave x+1.Otherwise the person has a missing value in DV2 and in all other "nested" DVs. I want to know if there is a main effect of my two IDs and where I can find the difference. My first hypothesis were how the effect of these IDs on my DVs will change over the waves. For that I proposed LGC. But now I am more interested in the "main effect" over the time since I want to be able to make a recommendation which "level" of the IDs is the best to get good participation behaviour. a) I tried multinomial regression but I have a lot of missing values.(just 10 percent of the sample has a value in every DV for all waves) b) is it possible to use LGC for that? It would be nice to have the additional "growth" information, even if it is not my main interest. I am more interested in the "niveau" of the different groups (cells) thant in the difference of "structure". But the intercept in the LGC does just take the first wave into account, doesn't it? Is it possible to evaluate my hypothesis by not restricting the paths from the intercept to 1? Do I miss something? or is LGC the wrong track? THANKS