I'm trying to fit a longitudinal model with two types of incidental truncated independent variables (also measured over time). The first is 'how long do you have a new partner' obviously depending on 'do you have a new partner'. 1. How can I fit the association of the length of a new relation with the longitudinal dependent. Do I simply enter both the dummy and the time variable in the model at once as time varying independent variables? Since one respondent can have different new partners over time, I think a survival approach might not be appropriate. Or do I need to perform a separate analysis on only those respondents with a new partner? 2. Is it possible to use multiple imputation in this case? I can assume there will be a problem with complete separation while imputing these variables.
The second truncated variable is 'current conflict with the ex-partner' depending on 'contact with the ex-partner'. It's harder for me to consider this as a left-censored variable like the one above since the reason for no contact might be very high conflict. But maybe this doesn't make a difference for the analysis method? Also for this variable, multiple imputation is needed.
For the first variables above (new partner and time since new partner) I think I should make one left-censored variable, impute the dataset and then disentangle the two variables again and analyse them together with there interaction.
For the variable 'contact' and 'conflict with the ex-partner', I'm not planning on using them in one regression anymore. But I would like to use the information of 'conflict' for the MI since it's very valuable for other variables. Is there some sort of two-stage MI available in Mplus, whereby I can impute the cases who have contact with there ex-partner first using the information on conflict and then those without contact with the ex-partner disregarding the information on conflict?
I am aware of this. But I wasn't sure it's reasonable to do it in this case. Am I right that the individuals for which the variable is not applicable will also get imputed values on all other variables partly based on this variable? For these cases I would be imputing values partly based on associations - between the other variables and this variable - that can not exist. Can I disregard this as long as I'm not using this variable in an analysis?