In Stata, how do I estimate mixed or multilevel linear models?

In Stata 11, you can use the .xtmixed command to estimate multilevel mixed-effects linear models, also known as mixed-effects, multilevel, or hierarchical models. Mixed models have both fixed effects and random effects, and are appropriate for cases when observations are clustered in some manner (e.g., students within schools, voters within districts, or workers within firms).

As an example, suppose you want to predict GRE verbal scores based on a student's GPA and hours spent preparing for the exam. If you have observations on students from multiple colleges, you may also explore how the expected outcome (average GRE score) and slopes of the predictors change across schools. This hypothetical data set would have five variables:

The following commands estimate different mixed models for this data set:

A dependent variable gre is listed first and followed by the independent variables in the .xtmixed command. The || characters are followed by a grouping variable college, a colon, and a list of variables whose effects you wish to model as random. The first example estimates a random intercept model only. The second adds the random coefficients for the GPA variable, while the third model allows for the intercept and both slopes to change across groups.

For help with the .xtmixed command, enter help xtmixed in Stata, or consult "Longitudinal/Panel Data" in the Stata 11 manual.

For multilevel non-linear models, see the following commands: .xtlogit, .xtprobit, .xttobit, .xtpoisson, .xtnbreg, .xtmelogit and .xtmepoisson.

For more about statistical and mathematical software, email the UITS Stat/Math Center, visit the center's web page, or phone 812-855-4724 (IUB) or 317-278-4740 (IUPUI). The center is located in Bloomington at 410 N. Park Avenue, and is open for consultation by appointment Monday-Friday 9am-5pm.

Comments/Questions/Corrections