Multilevel Modeling

This five-day workshop, led by Kelvyn Jones (University of Bristol, U.K.) and S.V. (Subu) Subramanian (Harvard University) is designed to give participants a training experience in the concepts and applications of multilevel statistical modeling, particularly in a spatial and demographic context. More specifically, the workshop has five major objectives: (1) the complex hierarchical and non-hierarchical structures in terms of unit diagrams and classification diagrams, participants are thereby introduced conceptually to a very broad range of designs (e.g., panel, repeated cross-sectional, multivariate, multistage survey, and spatial designs); (2) a thorough consideration of normal theory two-level models; (3) an appreciation of more advanced topics (3-levels structures, multilevel logit models, estimation (including maximum likelihood estimators [(R)IGLS] as well as MCMC); properties of shrinkage estimates; (4) the use of specialist MLwiN software; and (5), the application of multilevel modeling to a social science research problem. Throughout there is a strong emphasis on interpretation, not technical facility per se. To gain the most from this workshop it is expected that participants will have prior experience of linear regression modeling, analysis of variance, statistical significance testing, and familiarity with a Windows environment.

On completion of the workshop students will be able to: recognize a research problem requiring multilevel modeling, outline the technical and substantive advantages of multilevel models in comparison to single-level models; distinguish between fixed and random effects, read and evaluate research papers that apply multilevel models; make the case for adopting an explicit modeling approach to heterogeneity and data dependencies whether these arise from population structures and/or multistage sampling; explain the concepts of: cross-level interactions, variance functions, impact heterogeneity, autocorrelation, design effect, and shrinkage estimates; specify, estimate and interpret 2-level normal-theory linear models that contain both categorical and continuous predictors and predictor variables at each level; interpret the results from a multilevel logit model in terms of odds and probabilities, including differences between subject-specific and population-average models; develop proficiency in the MLwiN software package so as to meaningfully estimate, evaluate, and test a range of models; and apply multilevel models to a research problem according to a well-articulated research strategy.

UCSB requirements for processing stipends.

Laptop Requirement

View the Technical Information Page for requirements for laptops with Microsoft Windows XP, Vista, and 7.


Participants will receive a certificate of completion from the GIS Population Science Training Program in Advanced Spatial Analysis from the Center for Spatially Integrated Science.

GIS and Population Science Advanced Spatial Analysis Workshops Flyer. (42kb)
Please post or circulate this flyer to potentially interested colleagues.