Multilevel Linear Models: May 24-28, 2010
Detailed information about the workshop is below. Registration for the workshop is available here and hotel reservations at the conference center are avaiable here
Overview
Multilevel Linear Models is a five-day workshop focused on the application and interpretation of statistical models that are designed for the analysis of nested data structures. Nesting can arise from hierarchical data structures (e.g., siblings nested within family; patients nested within therapist), longitudinal data structures (repeated measures nested within individual), or both (repeated measures nested within patient and patient nested within therapist). It is well known that the analysis of nested data structures using traditional general linear models (e.g., ANOVA or regression) is flawed, oftentimes substantially so. Not only are tests of significance likely biased, but many important within-group and between-group relations cannot be evaluated. All of these limitations can be addressed within the multilevel model. In this workshop we provide a comprehensive exploration of multilevel linear models with topics ranging from introductory to advanced. A detailed syllabus is presented below. The general structure of each day is lecture-based instruction from 9:00 to 3:30 (with morning, lunch, and afternoon breaks), and separate break-out sessions from 3:30 to 5:00 focused on model estimation using SAS and SPSS. Although there is not a hands-on computer lab component to this workshop, we provide extensive live demonstrations in both SAS and SPSS and distribute the data and code for all examples. Further, participants are welcome to bring personal laptop computers to follow along with the software demonstrations or to work on their own data applications. A summary of participant reviews from prior workshops is available here.
Who Should Attend
Our workshop is designed for graduate students, post-doctoral fellows, faculty, and research scientists from the behavioral, social, and health sciences. Although it is recommended that participants have a working knowledge of the general regression model, these core concepts will be reviewed at the beginning of the workshop. Further, prior knowledge in either SAS or SPSS is useful but not required.
The Goals of the Workshop
Our motivating goal is to provide an intense yet enjoyable instructional experience that focuses on a large number of both introductory and advanced topics in multilevel modeling. We strive to strike an equal balance between core concepts of the underlying statistical model along with the practical application and interpretation of multilevel models fitted to real empirical data. Our workshop is designed to provide participants with the materials and instruction needed to both develop a real understanding of the multilevel model and to be able to thoughtfully apply a variety of basic and advanced multilevel models to their own data.
What is Provided
We provide all participants with a bound set of detailed course notes along with a USB disk key that contains all data sets and computer code. A sample chapter of our course notes is located here and a sample chapter from the computer demonstration notes using SAS is located here; an alternative version is provided that presents the same examples using SPSS. Examples are also coded in HLM6 and Mplus. Finally, tuition includes a continental breakfast, a full lunch buffet, and drinks and snacks throughout the day.
Daily Schedule
Each day begins at 8:00 with a continental-style breakfast provided at the conference center. Instruction then begins at 9:00, there is a mid-morning break, and then concludes at 12:00 for a full buffet lunch provided to all participants. Lecture continues from 1:30 to 3:00, there is a mid-afternoon break, and then two break-out sessions each lead by Dr. Curran and Dr. Bauer are held from 3:30 to 5:00 focused on the demonstration of fitting models in either SAS or SPSS. The day concludes at 5:00 (except on the final day, which concludes at 3:00 to allow for afternoon travel).
Tuition and Registration
Tuition for the five-day workshop is $1800 per registrant. This includes approximately 35 hours of total lecture time, a bound copy of the course notes and a bound copy of the computer demonstrations (approximately 450 pages total), a USB key containing all data and programs, and a continental breakfast and full lunch each day. Tuition is fully refundable (minus a $180 charge imposed by the credit card companies to process refunds) if cancellation is by May 7, 2009. There is no cancellation fee if paid by check if cancellation is made by the deadline. Registration is fully web-based and is accessed here.
Facilities & Accommodations
Our workshop will be held at the Rizzo Conference Center which provides a picturesque, retreat-like atmosphere with superb instructional facilities and is only minutes from historic downtown Chapel Hill. The Rizzo is owned by the Kenan-Flagler Business School at the University of North Carolina at Chapel Hill and is a full service conference center including hotel rooms, dining, exercise facilities, jogging trails, and a free local shuttle service. The Rizzo was ranked #1 for food and accommodations in the 2008 Executive Development survey conducted by the Financial Times.
We have reserved a block of rooms on-site at the Rizzo Center, and participants can make reservations at the special conference rate of $179 per night (note that this includes a full buffet breakfast and full buffet lunch). There are two ways reservations can be made. First, you can call 919.913.2098 and identify yourself as part of the multilevel modeling group with group code MLM10 to receive the special room rate. Second, you can register online here and follow the automated registration instructions. The group code will be automatically populated with “MLM10”; the boxes for Corporate/Promotional Code and IATA number are not needed and can be left blank.
If a participant chooses to not stay at the Rizzo Conference Center itself (or if the reserved block of rooms is filled), there are a number of alternative hotels that are conveniently located near the conference center. A list of these hotels is located here.
Further Information
Please contact us either via email (info@cbanalytics.org) or by phone (919.533.9817) if you need additional information or have any further questions.
Syllabus
Day 1
•When to use multilevel models and why
•Common examples of nested data
•Dependence -- the problem with applying conventional statistics to nested data
•Approaches for analyzing nested data structures:
•Fixed-effects models – sometimes called econometric modeling approach
•Random-effects models – commonly referred to as multilevel modeling
•A Useful Starting Point: the Random Effects ANOVA model
•Partitioning variance into within- and between-groups components
•Computing and interpreting the intraclass correlation (ICC)
•Modeling the effects of predictors
•Level 1 predictors: Random effects regression models
•Random intercepts; Random slopes
•Level 2 predictors: Intercepts and Slopes as Outcomes
•Probing cross-level interactions
Day 2
•Disentangling within- and between-group effects for Level 1 predictors by centering
•Total, within- and between-group effects
•Avoiding errors of inference across levels (ecological fallacy, Simpson’s paradox)
•Separating within- and between-group effects by centering predictors
•With group-mean centering; With grand-mean centering
•Testing mediation in multilevel models: direct and indirect effects
•Types of mediation in multilevel models
•Upper-level mediation (2->2->1 mediation)
•Lower-level mediation (2->1->1 and 1->1->1 mediation)
•Lower-level mediation with random indirect effects.
•Procedures for testing mediation effects.
Day 3
•Estimators, assumptions and diagnostic procedures
•An overview of estimators for the model: ML and REML
•How to choose between estimators
•Accuracy and precision of estimates in small samples
•Flexibility for testing differences between models
•Assumptions of the multilevel linear model
•Procedures for diagnosing and remedying violations of assumptions
•Modeling growth in longitudinal data
•Traditional modeling means with repeated measures ANOVA and MANOVA
•Modeling individual trajectories via multilevel models
•Alternative metrics of time
•Relaxing assumptions about the residuals (homoscedasticity, independence)
•Modeling nonlinear change
Day 4
•Predicting change over time with time-invariant (time-stable) covariates
•Intercepts and slopes as outcomes; Probing cross-level interactions
•When predictors also change over time
•Incorporating time-varying covariates into a growth model
•Time-specific “shocks”; Within- and between-person effects
•Multivariate growth models
•Studying how variables track together through time
Day 5
•Models for intensive short-term longitudinal data
•Advantages in terms of identifying within-person causal processes
•Need for more complex error structures
•Modeling transitions or turning points (piecewise models)
•Overview of advanced topics
•Three level data
•Cross-classified data
•Generalized models for binary, ordinal, count, skewed-continuous or bounded-continuous outcomes