Course. Latent Variables, Structural Equations, and Multivariate Analysis with EQS and REQS

Applied Statistics Week (ASW) is a collection of intensive courses, with a particular theme selected every year. This year 2010, the courses are:

Course 1 (28, 29 June): The R Statistical Environment

Course 2 (30 June): Multilevel models for longitudinal data in R

Course 3 (1, 2 July): Latent Variables, Structural Equations, and Multivariate Analysis with EQS and R EQS

The courses are presented by internationally renowned researchers with excellent teaching skills.

Themes of the past APPLIED STATISTICS WEEK have been:

• "Statistics in the Health Sciences'' (1995),
• "Statistics in Classification and Pattern Recognition" (1996),
• "Design and Analysis of Survey Data'' (1997),
• "Statistics in Environmental Science" (1998),
• "Statistics in Marketing Research" (1999),
• "Statistics in Society" (2000),
• "Statistics in Finance" (2001),
• "Statistical Genetics" (2002)
• "Statistical Modeling for Behavioral and Genetic Studies" (2003)
• "Causal Inference, Missing Data and Measurement Error" (2004)
• "Statistics in Law and Public Policy" (2005),
• "Imputation and Matching: Theory, Practice and Applications" (2006) and
• "Missing Data in Statistical Practice" (2007).

This year's ASW features three courses that will give practitioners access to efficient tools for statistical computing and graphics to the novice and intermediate users, to provide an easy-to-follow introduction to the former, and to expand the horizons of the latter.

The software R is a comprehensive language and environment for statistical computing and graphics. It is Free Software and easy to download from the CRAN website, but the quality of its construction, documentation and maintenance is on par with or better than some of its much more expensive competitors.

Over the last decade or so it has become the principal computational medium in academic and related statistical research. It has no match for its flexibility, based on a carefully researched design, accurate anticipation of the needs of the statistics profession, and a natural and easy-to-learn and compact syntax related to its object orientation.

The wide community of users of R have contributed by numerous facilities as add-ons to its core facilities.

These include many standard or specialized procedures that can be used by analysts with minimum proficiency in R. However, the main kudos of R is the facility to implement virtually any computational algorithm, tailored to the needs of the analyst or its client.

The three courses of this ASW will provide an ocassion for users to develop or enhance skills and proficiency in R in elementary and advanced settings.

The two-day Course 1 is intended for those less experienced in statistical computing in general or in R in particular; it will give a general overview of standard statistical methods in the context of R, centred around real-life examples and applications. The users will become acquainted with the fundamentals of computing and graphics in R.

Course 2, taking one day, will focus on a particular area of statistical analysis, mixed linear models. An algorithm for its analysis will be implemented "live", concurrently with an exposition of its details. The result will be a self-contained routine (function) for fitting multilevel models, accompanied by suitable displays, diagnostics and graphics, applied to a few examples.

Course 3, taking two days, addresses multivariate analysis with latent
variables, a methodology that allows a focus on the constructs underlying error-prone and biased observed measurements. EQS, a standard software for structural equation models, will be used to illustrate topics and applications from business to behavioral science. The new REQS interface between R and EQS is used to show how interesting but difficult topics can be handled in practice.

The three courses will give an opportunity for applied researchers in disciplines such as economics, finances, psychology, education, epidemiology and medicine, policy makers and other professionals who use statistics, to become acquainted with an important new frame for autonomous statistical computing. It will also give an opportunity for all those researchers to get a refresher in statistics that they need in everyday practice.

COURSE 3: LATENT VARIABLES, STRUCTURAL EQUATIONS, AND MULTIVARIATE ANALYSIS WITH EQS AND REQS (1, 2 July):

As a result of inevitable errors of measurement or inadequate sampling of variables, measurement in practice produces observable quantifications that only approximate the true conceptual constructs of real interest. As a result, relations among observed variables rarely accurately describe the true "causal" relations among the underlying constructs of interest. Latent variable modeling allows theory testing and inferences to be made at the level of constructs. In regression, a dependent variable y is predicted from p predictors x's. Latent variable modeling extends regression by allowing (a) latent variables, in which the x's are unobserved factors (a measurement model); (b) latent regressions, in which both y and x are latent variables; (c) multiple equations simultaneously with various dependent variables y, which may be latent; and (d) a dependent variable in one equation may be a predictor in another equation.

This workshop describes this path analysis, simultaneous equation, confirmatory factor, and LISREL-type methodology using the EQS Structural Equations Program. We review introductory and overview material first, followed by a variety of intermediate and advanced topics that are encountered in real-world practice.

A variety of important topics, such as power analysis, are best studied in an environment where the latent variable capabilities of EQS are merged with the statistical and graphical capabilities of R. The package REQS is an interface between EQS and the R environment. The package consists of three main functions that read EQS script files and import the results into R, call EQS script files from R, and run EQS script files from R and import the results after EQS computations. We show how to use the package by means of several examples.

Lectures will be primarily conceptual in nature, punctuated by practical implementation using EQS and its R interface. EQS for Windows and REQS will be made available to participants to install and use on their computers.

Prerequisites: The course is designed for a broad audience with varying levels of statistical background, but it would be helpful if the participants are acquainted with the basics of correlation, regression, and factor analysis.

Program:

1st July (morning)

• Introduction to latent variable modeling and EQS
• Model specification with the Bentler-Weeks model
• Mean structures, growth curve models, multiple group, and mixture models
• Missing data: Case-wise ML, two-stage ML, auxiliary variables

1rst July (afternoon)

• Model tests and standard errors with normal and nonnormal variables
• Model modification with Wald and Lagrange Multiplier tests
• EQS Interface with R in REQS package

2nd July (morning)

• CFI, RMSEA and related fit indices, Rho and related reliability coefficients
• Correlation structures for continuous and categorical variables
• Monte Carlo simulation with REQS for studying effects of non-normal data

2nd July (afternoon)

• Case weighted robust methods and methods for complex samples
• Multilevel models
• Sensitivity analysis with REQS

Instructors:
Peter M. Bentler, and Eric Wu created the EQS Structural Equations Program and distribute it for teaching and research through Multivariate Software, Inc. A prodigious developer of the mathematical and statistical theory of structural equations modeling as well as its applications, Peter has been an elected president of the Division of Evaluation, Measurement, and Statistics of the American Psychological Association, the Society of Multivariate Experimental Psychology, the Psychometric Society, and the Western Psychological Association. Karl Jöreskog and he were the joint 2007 recipients of the American Psychological Association's Distinguished Scientific Contribution Award for the Applications of Psychology. Currently, he is Distiniguished Professor of Psychology and Statistics at the University of California, Los Angeles.

Eric Wu, Is the developer of the graphical and computational environment of EQS, including its interface to R. He is the main computational statistics resource person at the UCLA - National Institute on Drug Abuse Center for Collaborative Research on Drug Abuse, and has conducted workshops for EQS users in many countries. He is currently leading a development team to create a user-friendly Item Response Theory program under a grant from National Cancer Institute, US National Institute of Health.

Course 3. Latent Variables, Structural Equations, and Multivariate Analysis with EQS and REQS (1, 2 July)

Timetable

From 9.30 to 13.30 and from 15.00 to 18.00 hours.

IDEC-Universitat Pompeu Fabra
Balmes, 132-134
08008 - Barcelona
Tel.: +34 93 542 18 50
info@idec.upf.edu

Limited number of seats in each course

To enable unhindered interaction between delegates and instructors, coffee breaks and a one-and-half hour lunch break is scheduled as integral parts of the course. The class sizes for each course are limited to 30. Applications will be processed on the 'first come first served' basis.

Fees

Registration fee for course 3: Latent Variables, Structural Equations, and Multivariate Analysis with EQS and REQS (1, 2 July): 630 €, including teaching materials and meals.

For more than one course:

Registrations for more than one course will be eligible to a discount of 20% on the total amount of the programme.

The policy on discounts, which are non-accumulative, is as follows:

• For prompt payment:
  Fees paid one month before the start of the program will be eligible to a 10% discount on the original cost of the programme.
• IDEC AA and Executive Club:
  Members of IDEC's Alumni Association and Executive Club will be eligible to a discount of 20% on the basic cost of the programme.
• More that one participant per company:
  Companies which register more than one participant on the programme will be eligible to a 10% discount on the first participant and a 20% discount on the second and subsequent participants.

Cancellation policy:

• In the event of being unable to attend the programme once registered, we will return 90% of the fees provided that a request for cancellation is made over a week prior too the beginning of the programme. The individual registered may be substituted by another from the same company as late as the day before the programme is due to start.

Admission requirements and registration

Registration process