|
|
Missing by design in multilevel |
|
Message/Author |
|
|
Hi, I have a dataset that contains data from four different studies that examined the relation between a team process and team performance over multiple measurement points. I decided to use a threelevel multilevel approach to examine the relation (level 1 time, level 2 team, level 3 study). However, I am not sure how to deal with the missing data in this case. First, two of these studies have four measurement points, while the other two only have three. Secondly, we aim to examine a moderator variable that is only measured in two of these studies. Should I just use the default mplus settings for this (and for example including the variances in the model so that mplus estimates the missing data) or do I need to become more specific? I read about the missing by design option but I am not sure what it exactly does with the missing data? Thank you! |
|
|
With only 4 studies, you shouldn't do threelevel analysis but instead twolevel, multiple-group analysis with 4 groups. Different number of time points is not an issue when time is level 1. |
|
|
Ok, thank you very much. A follow-up question: We are using the Define command to compute interaction terms. Mplus, however, appears to compute the interactions prior to estimating the missing data. I therefore also requested the variances of the interaction variables to let Mplus estimate the missing data. Is this approach comparable to doing multiple imputation prior to the analysis? |
|
|
Conceptually, they are the same. |
|
Back to top |
|
|