This is for IE7 to hold div open
I'm analyzing a dataset with dyads (couples) and another dataset with families who have different numbers of children per family. Someone suggested analyzing my data using something called a multilevel model. What's a multilevel model and why should I use it?
The data you describe are often referred to as "hierarchical" or "clustered" because subjects (individuals) are nested within clusters or units such as families or couples. Many commonly-used statistical procedures such as ordinary least-squares linear regression assume that every observation is independent of every other observation in the dataset. Obviously, when clusters are present, this assumption is violated.
To address this problem, researchers developed special statistical models to take into account the hierarchical nature of such datasets. As a class, these models are known as multilevel models. Other investigators developed special software programs designed specifically for the analysis of multilevel models.
Among the general purpose software packages that we support, SAS is one of the most commonly used in handling multilevel models. The MIXED procedure can be used to analyze data with continuous distributions whereas the GENMOD procedure can be used for repeated measures with non-normally distributed variables. An additional feature of MIXED that GENMOD lacks is the ability to estimate variances for the cluster level; this feature is useful for descriptive purposes as well as the computation of proportions of variance due to clusters versus individuals in the dataset.
To learn more about using PROC MIXED to fit multilevel models to normally distributed outcome variables, you can download a copy of the paper "Using SAS PROC MIXED to fit multilevel models, hierarchical models, and individual growth models" written by Judith Singer at Harvard University at http://gseweb.harvard.edu/~faculty/singer/. If your outcome variables are non-normally distributed, consider using the GLIMMIX macro available from SAS Institute. GLIMMIX uses PROC MIXED as part of its syntax, so you can use GLIMMIX to obtain variance component estimates for clusters.
Other software packages that provide multilevel modeling analysis are SPSS, LISREL, HLM, MPlus, MLwiN, and AMOS. Details on these packages can be found under the FAQ section of the statistical website of David Garson, Ph.D.: http://www2.chass.ncsu.edu/garson/pa765/multilevel.htm.
Links for the software are given below:
SPSS: http://www.spss.com/advanced_models/data_analysis.htm
LISREL and HLM: http://www.ssicentral.com/workshops/index.html
MPlus: http://www.statmodel.com/features4.shtml
MLwiN: http://www.cmm.bristol.ac.uk/
AMOS: http://www.spss.com/amos/
Multilevel model analysis is complex and a rapidly growing and changing field. To keep up to date, consider joining a Multilevel Internet E-mail list. Also, you can visit the HLM website at http://www.ssicentral.com/ or the Centre for Multilevel Modelling home pages at http://www.cmm.bristol.ac.uk/ to get the latest information about multilevel model workshops, software, and related resources.
If you have further questions, send E-mail to stats@ssc.utexas.edu.