In marketing research and business analysis, there is often the requirement to compute the 'relative importance' of underpinning predictor attributes on some overall criterion measure (such as Overall Satisfaction).
This is often done using Regression, either by looking at the relative size of coefficients for each predictor, or by looking at the relative contribution of each predictor to R-squared.
However, even assuming that R-squared is a ‘useful’ metric (debateable), this approach is often compromised by the presence of collinearity. The aim of the presentation is to illustrate this dilemma with a small number of test data sets, and suggest a potential way out, using one of the analytical approaches that were originally been developed for the data mining field.
Those looking for a detailed exposition of fiendishly clever R code will be disappointed. Those who are interested in how selected pieces of a massively capable language such as R can be utilised for business purposes, without the need to write, nor to understand, reams of script will hopefully find the presentation of value.
Speaker’s background: Scott MacLean
Scott’s work experience is predominantly in the market research industry (30+ years), including nine years with AGB:McNair/AC Nielsen, seven years with Research International (Melbourne, Frankfurt and London) as Marketing Science/Research Director and two years with Lewers Research (Melbourne) as Head of Advanced Analytics.
For the past four years, he has been Principal of Nulink Analytics, specialising in quantitative input for business decisions.
Scott has advanced level qualifications in statistical analysis and modelling:
Bachelor of Science (Hons) - mathematical and survey statistics, Monash University
Master of Applied Science - transportation planning and modelling, Univ. of Melbourne
He is an elected Fellow of the Australian Marketing and Social Research Society (AMSRS), and has recently presented full-day workshops for AMSRS in Choice Modelling, and Segmentation using Latent Class Analysis.