Martin Ob. posted on Monday, May 07, 2012 - 2:29 am
i would like to do the following analysis with mplus (i have data on two levels: individual level and region level; i have two data sets: individual level data n = 600,000 and region level n = 41)
calculation of region-level partial correlation between x and y, controlled for z1 (control variables on the individual level such as age or race) and z2 (control variable on the region-level such as regional unemployment rate). x is the region mean calculated from the individual level dataset, y is aggregate data from external sources (official regional statistics).
my first question would be, can i calculate such a multilevel random coefficient model with such a large sample size (600,000 individuals) or would it take like weeks to calculate this model?
my second question is: am i right that i have to merge the two datasets (individual level dataset and region level dataset) into a master individual level datasets that includes the region variables (all individuals in one region would get the same score in the region-level variables y and z2)?
If you want to do multilevel modeling, you need to put the data sets together as you have described. If you are analyzing the region data set separately, you are not doing multilevel modeling. Your sample size should not be a problem.
Martin Ob. posted on Wednesday, May 09, 2012 - 6:09 am
thank you very much. i an a response from 2007 you wrote that mplus does not calculate partial correlations. is this still the case or can i set up a syntax for the partical correlation i am interested in (between two aggregate-level variables, controlling vor aggregate level control variables and individual level control variables).