Taking a deeper dive into your survey data with key driver analysis Diana Allen, Head of Statistics, ORC International
TODAY’S AGENDA 01. 02. 03. 04. DEFINE THE BUSINESS CONTEXT THE ‘KEY DRIVER ANALYSIS’ TOOLKIT BEYOND THE STATISTICS Q&A
WHAT’S THE BUSINESS QUESTION WE’RE TRYING TO ANSWER? How do we keep loyal customers? What matters the most to our customers? WHICH FUTURE CUSTOMER ATTITUDES AND BEHAVIOURS ARE WE TRYING TO AFFECT? Likelihood to recommend Overall satisfaction Likelihood to switch
TYPICAL CUSTOMER RESEARCH SURVEY Customer research into perceptions of the brand/provider Survey includes target outcome question(s), plus specific attributes, eg Easy to understand
REPORTING RESULTS FROM CUSTOMER RESEARCH Summary statistics showing positive, neutral and negative perceptions
KEY DRIVER ANALYSIS OVERVIEW Key Driver Analysis (KDA) is a statistical technique that helps us focus in on what thing or things (‘inputs’) have the biggest or strongest influence on others (‘outputs’) This analysis helps take the guesswork out of determining what inputs we need to change or take action on in order to make a desired change in the output(s) by pinpointing which one or more of the inputs is going to have the biggest effect The underlying principle is that if you do something that causes a change to these ‘key driver’ inputs, you are much more likely to experience a change in the outputs than if you made a change to something that is not a key driver Input INPUT ? Output(s)
INDIRECT EFFECTS / INTERACTIONS TRADITIONAL STATISTICAL TECHNIQUES OVERVIEW OF METHODS SIMPLE COMPLEX BIVARIATE MULTIVARIATE INPUT-OUTPUT MODEL INDIRECT EFFECTS / INTERACTIONS TRADITIONAL STATISTICAL TECHNIQUES MACHINE-LEARNING
Independent variables CORRELATION Dependant variable Independent variables
MULTIPLE LINEAR REGRESSION Modelling likelihood to recommend as a linear combination of potential drivers Those drivers that are found to have a statistically significant effect are considered to be key drivers
ONE ISSUE WITH REGRESSION ANALYSIS MULTICOLLINEARITY highly correlated predictor variables in a multiple regression model Volatile findings in tracking studies One driver with a very large effect 9 4
HOW DO YOU SOLVE A PROBLEM LIKE MULTICOLLINEARITY? FACTOR ANALYSIS Fees are reasonable Proactive in support Knowledgeable staff Product offers value for money Problem resolution is effortless Easy to understand Product meets my needs Application is easy Timely communication Product and price perceptions Processes Communications Likelihood to recommend Direction and strength of the relationships between Product and Price, Processes and Communications, and Likelihood to recommend
RELATIVE IMPORTANCE REGRESSION (“SHAPLEY VALUE”) Proportionate contribution each predictor makes to R2, considering both its direct effect (i.e. its correlation with criterion) and its effect when combined with the other variables in the regression equation Utilises the R package relaimpo (Relative importance of regressors in linear models) created by Urlike Groemping
MODELLING COMPLEX RELATIONSHIPS Performs simultaneous estimation of multiple equations in order to understand a system of complex relationship Models relationships between individual attributes, underlying core dimensions, and ultimately the dependent variables of interest
MACHINE LEARNING APPROACHES TO KEY DRIVER MODELLING DECISION TREE TECHNIQUES BAYESIAN NETWORKS Classification and Regression Trees (CART) Random Forest Trees Capture nonlinearities, thresholds and interactions in the data Based on the inference of probability distributions from the data
BEYOND THE STATISTICS Data Visualisation
“The goal is to turn data into information and information into insight.” / Carly Fiorina, former CEO of HP/
Thank you. Diana.Allen@orcinternational.com