On 28 June, UIC’s Applied Psychology programme (APSY) and Statistics (STAT) programme invited Prof Peter M. Bentler to come to UIC and organized a talk for him to give that was titled, “Regularized GLS and Monte Carlo Probability for Improved Structural Equation Model Evaluation”. Prof Bentler is an American psychologist, statistician, and distinguished professor at the University of California, Los Angeles. He received the award of UIC Honorary Fellowship on 29 June.

Prof Peter M. Bentler

In multivariate analysis and psychometrics, Prof Bentler is the developer of the structural equation modeling software EQS. His publications have over 190,000 citations as of 2017. He was awarded the Psychometric Society Career Award in 2014 and was elected Fellow of the American Statistical Association in 2015.

Associate Professor and Programme Director of Applied Psychology, Dr Ghee Ho, listed Prof Bentler’s achievements before introducing UIC’s President, Pro Ng Ching-Fai to participate in rewarding a souvenir to Prof Bentler. Following that, Prof Bentler then began his talk by dividing it into five parts. The first part was a short appreciation to his students and colleagues for their valuable contributions followed by a short overview of Structural Equation Modeling. Prof Bentler followed up by talking about regularized generalized least squares for improved testing with normally distributed data. Prof Bentler spoke about general models and methods for testing causal and latent variable hypotheses on non-experimental data, which have been a huge success story for the behavioural sciences. He especially focuses on path analysis, confirmatory factor analysis, and general mean/covariance structure models that have become easily accessible via computer programmes such as EQS, Lavaan, LISREL, and Mplus. However, even with 50 years of technical statistical developments, the scientific conclusions reached from such analyses can be inadequate when ideal data conditions such as multivariate normality and large sample sizes are not available.

At the end of the talk, Prof Bentler took questions from the audience

Prof Bentler provided an overview of Regularized Generalized Least Squares for normally distributed data and Monte Carlo simulated probability values for arbitrarily distributed data, which are two new model testing methodologies that seem to be a substantial improvement over existing methods in small to medium sample sizes. He spoke about the Monte Carlo Probability for improved testing with arbitrarily distributed data before finishing with his conclusions. Some of his conclusions were that estimates are always consistent with normal theory, while further studies were needed on other aspects, such as standard errors when considering RGLS test statistic, as well as concluding that other tests with direct Monte Carlo simulation should be explored further. The talk ended with a Questions and Answers session where a mix of faculty members and students asked more questions about the reliability of various equation models.

Reporter/Photographer: Samuel Burgess
Editor: Deen He
(from MPRO, with special thanks to the ELC)