I am interested in testing whether the change on mothers' depression over time influences children (e.g. behavior) over time. so then, I decided to perform a GMM with two clusters one for the mothers' depression, one for the features of children (e.g. behavior). Then I will regress intercept and slopes of mothers (cluster) on those of children (clusters).
However, there may be mothers with more than one child. I am not sure how I can handle this, with the dataset and analysis. My wuestions are:
1. In the dataset, should I repeat the same case (mother) for every children she has in the study? So for instance, in the cases (rows) a mother with two children will represent two cases (two observations).
2. How should I account for the non-Independence of siblings?
3. could you give me a reference of this type of analysis, within the Mplus User guide?
Thank you very much for all you are doing for us Mplus users!
If you want to work with more than one child, then yes responses from those children should be entered as further columns in the data with missing data for those families who have one child. But you may perhaps first want to try out your analysis with one child, say the oldest child.
H Professors, I am still dealing with the analysis described above. For each family (row) I have one mother (and 4 time-point of depression) and from 1 to 4 children (4 time-point measurement of depression for each of the child. I want to estimate the effect of mothers' slope on all the children's depression change (slope). I would like to create a single trajectory for all the children together rather than one trajectory for each of them.
1 - Can I have one trajectory for mothers and one for all the children together? Below what I am using now, the first line is for mothers, and the i1 s1 for the first child i2 s2 for the second child etc.
If you look at one trajectory for all of the children combined, you assume that the children are equivalent. I would have one trajectory for each child. I would estimate each growth model separately as a first step.
Non-independence of observations is taken into account by your multilevel model. There is nothing more to do.
Hi, thank you for the prompt answer. I agree with you, but just one mother has 4 kids, which means that the trajectory is based only on one observation. Mplus gives an error due to the variance = 0 in that curve. So I am trying to estimate the trajectories for each of the kids (first child, second child etc) and then an overall trajectory for all the kinds. But I am having hard time to understand how to put down the sintax for an overall trajectory that includes all the kinds. Can you help me? Thank you again!
Sorry, also, there is a problem with runnuning the analysis because for the 3rd kid, there are too many missing data and I have the following error THE MINIMUM COVARIANCE COVERAGE WAS NOT FULFILLED FOR ALL GROUPS. Sorry to flood with questions. Andrea
Why don't you first run with only the first 2 children of the families as a way to see how this works. You can give a growth model for each child and then test if the growth factor parameters are equal. If so, you can also use the same growth factors for the two children.
Hi, the model seems not to be doable with my data (too many missing data). then, I thought to restructure the dataset where each row represents a kid, to which mother's score is associated. Thus I have a variable (column) for all children and one for mothers' score. The mothers' score are entered 4 times (namely the maximum number of kids for in a family)even if they have 1 kid. So she will have a score for kid 1 and missing on the kid 2 3 and 4. Then I will have a GMM with two processes, one for mothers one for their kids.
Does this sound correct to you? How could I account for non independence of kids' scores? Thank you again for your exceptional work...
Sounds like you are switching from a wide format, single-level setup for the children to long format, two-level analysis. You don't have to enter more than 1 row for the families with only 1 child; it just means that this "cluster" has only 1 "member" (think of children within families as students in a classroom). You have to have a cluster (family/mother) identifier.
But since you are interested in growth over 4 timepoints, it seems you would have 4 x 2 columns in your data. So growth is still done in the wide format using 2 parallel processes.