Misconceptions about Latent Variable Models

By |2014-08-14T09:23:55-04:00August 14, 2014||

Misconceptions about Latent Variable ModelsWhile giving courses over the years, I’ve noticed that certain misconceptions about latent variable models persist among statisticians and chemometricians.

At the 14th annual ENBIS conference (European Network for Business & Industrial Statistics) in September, I have decided to discuss a number of these during my plenary George Box Medal presentation. Some of the key questions include:

  • Why does PLS give unique models when other regression methods do not?
  • Why do latent variable models handle missing data so much better?
  • Why are these models more useful for monitoring processes and detecting faults?
  • Why can latent variable models be used to optimize processes even though they have been built on historical happenstance data with no DOE?

The ENBIS conference provides a nice forum to discuss these questions openly. And, if you can’t make it to ENBIS ’14, join us at one of our upcoming public courses.

George Box Medal for Outstanding Contributions to Industrial Statistics

This award is particularly special because Dr. George E.P. Box was my PhD supervisor, and more importantly he was perhaps the greatest contributor to industrial statistics in the last century.

George Box spent his life solving real problems. He didn’t plan to be a statistician, but while studying the effects of poisonous gases for the British Army he decided that statistical methods were needed to analyze the experimental data. A statistician wasn’t available, so he learned the methods himself and went on to make practical statistics his life’s work.

In his letter to ENBIS in 2003 he listed one concern: that ENBIS stay focused on solving real problems as opposed to theoretical statistical issues. It is truly an honour to be receiving this award, because ProSensus’ reason for being is to help companies solve real process and product quality issues.

Techniques for Optimizing New Product Formulations

New product formulation is another area where ProSensus is very active in helping European and international companies. In the invited chemometrics session at ENBIS ’14, I will be presenting a talk entitled Using Multi-block PLS Models and Optimization for Rapidly Developing New Product Formulations.

With these techniques, we have helped clients achieve new product formulations much more quickly and at lower cost than with traditional statistical methods. It is an exciting area and I look forward to sharing our experiences with the ENBIS audience.

Dramatically Reduce Experimentation without Compromising Results

Last month, we invited Directors, Vice-Presidents and Managers of new product development to join ProSensus for a 1 on 1 discussion to learn more about our techniques for improving productivity in product development. Back by popular demand, we are extending the invitation and offering some new dates in September. If you haven’t signed up yet, don’t miss your last chance.

About the European Network for Business and Industrial Statistics (ENBIS):

ENBIS is a network of individuals and organizations interested in theoretical developments and practical applications of statistics in European business and industry. 14th Annual ENBIS Conference program.

About the George Box Medal:

The Box Medal is named after George Box and recognises each year an extraordinary statistician who has remarkably contributed with his/her work to the development and the application of statistical methods in European business and industry.

About the Author: John MacGregor

John MacGregor
John MacGregor, Ph.D.
Founder & Chairman
John has dedicated the last 40 years to helping manufacturers improve and optimize their processes using multivariate data analysis. During his tenure at McMaster University, John authored over 200 peer reviewed journal publications in the areas of mathematical modeling of processes, optimization and control, rapid product development and image analysis.John has received many awards for his pioneering work in developing and applying multivariate analysis to solve complex manufacturing problems, including the Shewhart Medal and W.G. Hunter Award from the American Society for Quality, the Herman Wold Medal from the Swedish Chemical Society and the R.S. Jane Award and Century of Achievement Award from the Canadian Society for Chemical Engineering. In 2004, John founded ProSensus to help manufacturers learn more from their big data to increase yield, reduce operating costs, and improve product quality.