Two-Day Workshop: Statistical Mediation, Moderation, and Conditional Process Analysis
Professor Andrew Hayes, Ohio State University & Statistical Horizons
December 13&14, 2012
This workshop is focused on the application of principles of linear modeling, primarily using OLS regression, to exploring questions about mediated (i.e., indirect) and moderated (i.e., interaction) effects. Topics include classic and contemporary approaches to testing mediation and moderation hypotheses, path analysis, indirect and direct effects, probing and plotting interactions, and analytically integrating moderation and mediation analysis. We will spend roughly one-third of the workshop talking about partitioning effects into direct and indirect components and how to quantify and test hypotheses about indirect effects, one-third talking about estimating, testing, probing, and visualizing interactions in linear models, and then one-third discussing the integration of moderation and mediation by discussing how to conceptualize and test for conditional indirect effects (“moderated mediation”) and whether moderated effects are mediated (mediated moderation). Examples will focus on the use of the PROCESS procedure (available for both SPSS and SAS) written by Dr. Hayes, which implements the methods described in this workshop and greatly simplifies computational tasks that historically have been tedious.
Two-Day Workshop: Analyzing Longitudinal Data, with Special Attention to Missing Data: Overview and Guide to Practice
Professor Joseph Hogan, Brown University
December 8&9, 2011
This two-day short course will describe modern methods for analyzing data from longitudinal studies with repeated measures, including methods for handling missing data. The first day will begin with a description of some motivating studies. Next, Dr. Hogan will provide an overview of regression-based approaches to inference about longitudinal data, including generalized linear models and random effects models. Estimation via generalized estimating equations and maximum likelihood will be described, and practical applications will be used to illustrate the methods. The focus of the second day will be on methods for handling missing data. Dr. Hogan will describe a framework for characterizing mechanisms leading to missing data, and give a critique of commonly used methods such as last observation carried forward. He will then describe principled methods such as inverse probability weighting and multiple imputation. In the final part of the course, Dr. Hogan will demonstrate effective use of sensitivity analysis to characterize robustness of inferences to departures from standard assumptions about the missing data mechanism. Throughout the course, concepts and methods will be illustrated using data from studies on which the presenter has collaborated.
Two-Day Workshop: Introduction to Spatial Statistics for Population and Health Research
Professors Mei Po Kwan & Desheng Liu
8:30AM - 4:30PM
1080 and 0140 Derby Hall
IPR will host a two-day workshop for IPR Faculty and Graduate Student Affiliates on September 14-15, focusing on statistical analysis of spatial data. Space is limited; please email Susan Pennington by August 1, and state your affiliation and research interest. You will be notified regarding your application status by August 15. Spatial statistics is a rapidly developing area of statistics that has seen enormous growing interest and applications in a broad range of research fields, including population and health research. This workshop aims at providing participants an overview and introduction to spatial statistical methods and their applications in health and population research. The workshop will include lectures, demonstrations, and hands-on exercises. Lectures will focus on basic statistical concepts and techniques for the analysis of spatially referenced data, including geostatistical data, regional data, and spatial point patterns. Examples from a variety of topical areas in public health and population related research will be used to illustrate various spatial statistical methodologies. Computer laboratory exercises will allow participants to gain hands-on experience in the statistical analysis of spatial data.
Aaron Bramson, University of Toronto
December 9&10, 2010
This workshop will teach participants to build ABMs with the NetLogo through demonstrations, several hands-on exercises, modeling advice to continue moving forward, and an introduction to more advanced modeling capabilities. No previous experience with complexity theory, NetLogo, or agent-based modeling is required. Participants are encouraged to download and install the NetLogo software available for free from ccl.northwestern.edu/netlogo/ before the workshop.
Geographic Information Systems Workshop
Professors Mei-Po Kwan and Desheng Liu
Remote sensing provides a synoptic view of the Earth environment, offering important spatial-temporal context of human activities. Recent advances in remote sensing technologies have enabled the measurement of biophysical variables that are critical to the understanding of various social and environmental processes. This workshop aims to provide the participants an overview of remote sensing technologies and their applications in social and environmental studies. Lectures will focus on basic concepts and techniques in remote sensing data acquisition and analysis. Examples from a variety of topical areas such as deforestation, urbanization, epidemiology, and population research will be used to illustrate how the information derived from remotely sensed data can be used to analyze social and environmental changes. Computer laboratory exercises will allow participants to gain hands-on experiences on the analysis of remotely sensed data.
Missing Data Workshop
Dr. Paul Allison
December 11-12, 2009
This 2 day course provides an in-depth look at modern methods for handling missing data, with particular emphasis on maximum likelihood and multiple imputation. These methods have been demonstrated to be markedly superior to conventional methods like listwise deletion or single imputation, while at the same time resting on less stringent assumptions. Although the course is applications oriented, it also covers the conceptual underpinnings of these new methods in some detail. Maximum likelihood is illustrated with two programs, M-Plus and LEM. Multiple imputation is demonstrated with two SAS procedures (MI and MIANALYZE) and the ICE command in Stata.
An Interaction on Interactions (Statistical, that is)
Robert Kaufman - Sociology, Ohio State University
Bayesian Data Analysis
Dr. Xinyi Xu, Ohio State University
Bayesian analysis is emerging as one of the leading paradigms of statistical inference. It has been widely used in numerous fields, such as policy making, financial modeling and information theory. A major difference between Bayesian analysis and conventional inference is the adoption of some prior knowledge on the parameter of interest. The talk will give a brief introduction to the basic framework of Bayesian analysis and its applications in hierarchical linear regression, model selection and model averaging, and missing data analysis. It will be shown through some examples that Bayesian methods sometimes provide better results than conventional approaches by allowing a more complete account of all possible sources of uncertainties.
Dr. Herbert Smith
Matching is a "natural" way to do covariance adjustment in the observational studies that typify the social sciences. It was once fairly common, but fell out of favor because of social scientists' desire to control for more and more covariates. Regression and other generalized linear models that seemed better suited to a "high dimension" world took the place of matching. The revival of matching as a method of covariance adjustment dates primarily to the introduction of matching on propensity scores, which reduce multiple dimensions to a single dimension. Interestingly, this revival -- which began primarily in statistics -- has a direct lineage in sociology. Matching is not the holy grail. Like other methods of covariance adjustment, it is only as good as one's scientific ability to specify and measure the covariates for which we need to adjust. But it has several advantages over regression and regression-like approaches to covariate adjustment. It focuses the researcher on making inferences -- or statements about how the world works -- in places where there are data (in technical terms, "support"). It helps us recognize that there is rarely "a" or "the" effect of some treatment on a response (in technical terms, "heterogeneity of treatment effects"). There is a lot to be said for this, as for other "homely" methods of data analysis, such as low-dimension contingency tables. Regression tends to make us a bit silly; we do not recognize the extent that we are substituting assumptions for facts. This is especially problematic for population sciences such as sociology, where we tend to lack real theories in any meaningful sense of the term.
Geographic Information Systems (GIS): Introduction and Use in Population Research
Drs. Alan Murray and Ningchuan Xiao
Random and Fixed Effects Models
Dr. Randy Olsen - Director of IPR and Professor of Economics, The Ohio State University
Hierarchal Linear Models
Dr. Chris Browning, Ohio State University
Hierarchical Linear Models provide a useful statistical framework for the analysis of clustered data. Analysts are often faced with data that are hierarchically organized (e.g., students within classrooms, individuals within neighborhoods, employees within organizations, time points within individuals). This workshop will introduce participants to the basic logic of the hierarchical model and provide an overview of the standard two-level linear case (highlighting the key statistics relevant to the interpretation of the model). I'll also review more advanced applications and provide an example of a three-level model based on some recent research.