Nested data structure


New Member
Hi, I wonder whether the community can help.

The goal is to run a series of regressions with correctly estimated standard errors. Note that the data are panel data with two waves, all independent variables are from wave 1, including a lagged dependent variable.

The data structure is tricky, however. First, I have stacked the dataset x10. By doing this, I get a dummy as dependent (rather than have a nominal variable "which cereal is most preferred", I have dummy "preferred cereal 1/0" once for each cereal). That is, I have 10 units nested in each respondent, which I can account for by computing clustered standard errors (using vce i Stata).

Here comes the tricky part. These respondents ar nested in yet another type of respondent, a max of 3 respondents are nested within the other type of respondent. Say, 1-3 children are nested in one parent. That is:

Level 1: 10 cereals
Level 2: 1-3 children
Level 3: Parent

How can I account for this data structure? Robust standard errors, accounting for the 10 units nested in each child, does not account for the fact that the children are nested in the parent. I have tried to compute multi-level models in Stata, but they never converge, which is, I guess, caused by the fact that only 1-3 children are nested in each parent.

Any suggestions?

Thank you in advance!