I'm a geek

I'm really excited for this next week!

But I have a bunch of reading to do in preparation.

HLM workshops

Location: Gleacher Center, Chicago
Instructors: Steve Raudenbush, University of Chicago & Tony Bryk, Stanford University

Research applications of HLM analysis of hierarchical data structures continue to expand as new models are incorporated in the computing procedures. The models make possible a highly efficient and informative analysis of complex survey data, including those from two- and three-stage unbalanced sampling designs, possibly with explanatory variables and random effects at all three stages. For example, in studies of factors that influence year-to-year gains in the achievement of school students, the level-1 model may describe the course-of-growth in the achievement of individual students over a number of years, the level-2 model represents differences between classrooms, and the level-3 model incorporates effects between schools.

From variables specified at each level, the program generates the linear model with the respective explanatory variables that account for response variability at each level. The hierarchical linear analysis not only estimates the model coefficients at each level, but it also predicts the random effects associated with each sampling unit at each level.

These analyses can now be carried out by full-information maximum likelihood methods using a combination of EM and Fisher scoring algorithms for fast and stable convergence to the solution optimum. These procedures provide standard errors for both the fixed effects and the variance-covariance components that describe the random effects.

Previously, these analyses were available only for data in which the errors of prediction at each level could be assumed approximately normally distributed. New statistical developments now allow Bernoulli and binomial models for binary data to be analyzed with a logit link function, and a Poisson model allows count data to be analyzed with a log link function.

The session will cover the principles and assumptions of hierarchical linear analysis, applications to a wide variety of practical problems, and the interpretation and reporting of results. The presentation will include many worked examples and computer demonstrations. A prerequisite for the workshop is a working knowledge of simple regression analysis.