Our 1995 paper relied on simulations of three-level variance-components models with moderate and large random effects at levels 2 and 3 using the same nested structure as the actual survey data from Guatemala.
The datasets available here represent 100 simulations of the structure with large (s2=1) random effects at both levels 2 (family) and 3 (community).
Each dataset has 8 variables on 2449 cases. The variables are
|1||child id||2449 kids, ids in 1..14684|
|2||family id||1558 families, ids in 3..2782|
|3||community id||161 communities, ids in 1..242|
|4||binary dependent variable||0,1|
|5||corresponding latent variable||logistic|
|6||child-level covariate||mean .0955621, range -.452557 .541957|
|7||family-level covariate||mean -.083816 , range -1.485043 to 2.284106|
|8||community-level covariate||mean -.6857591, range -1.818267 to -.0097808|
All datasets were generated using true parameter values equal to 0.665267 for the constant and equal to 1 for each of the three covariates, and using normal variates with variance 1 for the family and community effect.
The 100 simulated datasets are named
and are available in plain text ASCII format, zipped into four archives with 25 datasets each:
Note that only variables 4 and 5 vary across datasets. In view of this, we are also making the data available in a much more compact format, a single zipped file rg3bbpack.zip that includes five files:
s3bbx.dathas variables 1,2,3,6,7 and 8; in other words, the variables that are constant accros simulations.
s3bby4.dathave 25 columns each, representing the outcomes of simulations 1-25, 26-50, 51-75 and 76-100, respectively, for all 2449 cases
You may find this format more suitable for some applications.