The topic of this year's Summer Programme in Data Analysis is structural equation models, “SEMs” for short. The first five days of the Programme, May 2 - 6, provide a complete introduction to SEMs, and will be taught by Professor Doug Baer of the University of Victoria. Days 6 and 7, May 8 – 9, deal with missing data and the application of SEMs to panel data using “growth curves,” and will be taught by Professor Ken Bollen of the University of North Carolina.

In the past 30 years, SEMs have become the statistical tool of choice for analysis of causal relations based on measures with error. SEMs combine ideas of “path analysis,” developed by Sewall Wright in the 1920s to describe the relationships in a causal chain, and factor analysis, developed in the 1930s to conceptualize “traits”, measured imperfectly by a number of “items”, usually in some kind of questionnaire or test. In the late 1960s, K. G. Jöreskog and D. Sörbom combined and generalized these ideas, developing a framework for simultaneous estimation of the causal relations among conceptual variables (usually called “latent” or “unmeasured” variables) and between the conceptual variables and their empirical measures (called “manifest”, “measured” or “observed” variables), using maximum likelihood estimation. Since that time, the application of SEM models has been broadened to cover binary and ordinal variables (with some assumptions), longitudinal data and multiple groups. SEMs have found their widest application in psychology and sociology, but are used routinely in almost every area of social science.

SEMs thus extend familiar regression models, path modeling, factor analysis models, analysis of variance and covariance models, simultaneous equation models and some forms of longitudinal causal models into a general linear model framework (with some extensions for non-linearity) based on the analysis of covariance structures. The models provide estimates of parameters for quite general systems of equations involving mixtures of manifest and latent variables, with both random and systematic measurement errors. Because SEM models are fitted to the covariances of the measured variables, rather than to the underlying observations, computation of SEM parameters is very rapid. Structural equation models are especially useful in the analysis of longitudinal data, where non-spherical error structures are associated with relationships involving variables measured at multiple lag points.

For the computer lab sessions in the Programme, which take place every day in the afternoon, we use AMOS (an SPSS product) and LISREL. We will provide a complete introduction to these programs.

SPIDA 2006 also features a one-day Symposium on Missing Data, presented by Professor Rod Little of the University of Michigan, co-author with Donald Rubin of Statistical Analysis with Missing Data (New York: John Wiley, 2nd Edition, 2002). The Symposium takes place on May 17th, one week after the end of the SEM session, and registration is separate.

While “missing data” have been a ubiquitous and longstanding concern in empirical social research, until the 1980s only rather simple ‘fixes’ were available. It was common, for example, to exclude observations with missing variables or to assume that the missing information could be predicted simply from the non-missing data. Since that time, there has been a growing awareness that such casual treatment of missing data threatens the validity of statistical results. More recently, theoretical work and the availability of much greater computing power have provided the basis for a much more systematic approach, as well as a better understanding of the limitations of any “fix” to empirical data. Professor Little, who has been instrumental in these developments, will survey the field and describe current developments in his Symposium presentation.

The 2006 Programme, coordinated by Professors Robert Cribbie, Bryn Greer-Wootten and Michael Ornstein, is similar to earlier SPIDA programmes, which began in the summer of 2000. In 2007, the final year of current funding from SSHRC and Statistics Canada, our plan is for an integrated treatment of approaches to longitudinal (or “panel”) data, including econometric approaches, structural equation models and “mixed” models.

The proposal for this 2005 – 2007 SSHRC – Statistics Canada grant was developed by a team under the direction of Michael Ornstein, Department of Sociology and Director of the Institute for Social Research, York University. Dr. Ornstein was also the lead organizer of the 2002 - 2004 CISS Data Training Schools, and an organizing committee member of the two CISS pilot projects in 2000 and 2001.

The organizing committee members:

Michael Baker, Department of Economics, University of Toronto and Academic Director, Toronto Regional Research Data Centre.

Robert Cribbie, Department of Psychology, York University and the Coordinator of the Statistical Consulting Service, Institute for Social Research. Dr. Cribbie was a member of the 2002 - 2004 SPIDA organizing committee.

Thomas F. Crossley, Department of Economics, McMaster University. The principal applicant for York’s successful Data Training School application in 2001, Dr. Crossley is a member of the Advisory Committee, Workplace and Employee Survey and of the Advisory Committee on Labour and Income Statistics (ACLIS) at Statistics Canada.

John Fox, Department of Sociology, McMaster University and Associate Coordinator, Statistical Consulting Service, Institute for Social Research, and a member of York's 2002 - 2004 SPIDA organizing committee. Dr. Fox has long-standing and close contact with the Inter-University Consortium for Political and Social Research (ICPSR) at the University of Michigan at Ann Arbor. In addition to teaching ICPSR courses each summer for many years, he is a member of the Advisory Committee to the ICPSR.

Michael Friendly, Department of Psychology, York University and an Associate Coordinator of the Statistical Consulting Service, Institute for Social Research. Dr. Friendly has taught many short courses in the SCS and in the 2000 - 2004 SPIDA programmes, and was a member of York's 2002 - 2004 SPIDA organizing committee.

Bryn Greer-Wootten, (Professor Emeritus) Department of Geography and Faculty of Environmental Studies, York University, Associate Coordinator, Statistical Consulting Service, and an Associate Director in the Institute for Social Research. Dr. Greer-Wootten was the principal organizer of the 2004 and 2005 SPIDA sessions at York.

Georges Monette, Department of Mathematics and Statistics, and past Coordinator of the Statistical Consulting Service, Institute for Social Research. Dr. Monette has organized and taught the short courses offered by the SCS for many years, especially several very successful courses on mixed models and longitudinal data analysis. Professor Monette was the principal organizer of the successful 2000 SPIDA at York University.

Dr. Karen Robson was appointed to the Department of Sociology, York University, in 2004. A specialist in survey research, she has taught at the University of Essex Summer School (including teaching in Bosnia and Albania), since 2001.