| Message/Author |
|
|
| Luke Norman posted on Monday, October 12, 2020 - 1:06 pm
|
|
|
Dear experts,
I am wanting to run a cross-lagged panel model. I have two timepoints, one continuous variable (brain activation) and one zero inflated symptom count. Would this be correct. What about the "with" statement?
VARIABLE:
NAMES ARE symptoms_T1 symptoms_T2 brain_T1 brain_T2;
COUNT is symptoms_T1(i) symptoms_T2(i)
MODEL: symptoms_T2 on brain_T1 symptoms_T1;
symptoms_T2#1 on brain_T1 symptoms_T1;
brain_T2 on symptoms_T1 brain_T1;
symptoms_T2 with brain_T2
I also have a small amount of nesting by family, although the vast majority of subjects came from different families. Nonetheless, would it be possible to account for this in the model should it be deemed desirable to do so?
Thanks in advance! |
|
|
|
|
| The way you are doing this is not correct. Cross-lagged panel modeling has not yet been developed for count variables. But stay tuned for new upcoming developments. |
|
|
| Luke Norman posted on Monday, October 12, 2020 - 4:26 pm
|
|
|
Thanks so much for your quick reply.
Are there any potential (if not entirely optimal) workarounds that you might suggest trying in the meantime? |
|
|
|
|
| What's your sample size? Have you considered RI-CLPM? If the symptom count isn't high, you might consider re-conceptualizing the count variable as an ordinal (categorical) variable in which case WLSMV and Bayes would be suitable using a Y* approach. |
|
|
| Luke Norman posted on Friday, October 30, 2020 - 5:10 pm
|
|
|
Thanks for your help. Sadly we only have two timepoints, which prevents RI-CLPM. The same size around 200-250.
The symptoms range up to 9. So to confirm - you are suggesting treating the data as ordered categorical data (levels: 0-9)? |
|
|
|
|
| Right. |
|
|
| Back to top |
