Key Driver Analysis

What are the important variables in a predictive model?

Key Drivers Analysis addresses the questions:

“Which combination of possible explanatory variables best explains the data I see for some question of interest?” and “what is the unique contribution of each predictor?”
 

This question, we are trying to explain, can sometimes be an interval
(e.g. 0-10) scale such as Likelihood to recommend Brand X?

It might be a numeric quantity such as Expenditure on Product Category.

It might be a simple yes/no questions such as “would you Vote for this Party?”

All of these questions can be answered by the use of regression models but the basic forms of these models can be unsuitable for many real applications, where we have many possible highly correlated predictors, limited sample and noisy data. An approach is needed which is robust to these realities and delivers a model which will predict well on new cases, not just those used to build the model (see the Curse of Overfitting).

Our Correlated Component Regression (CCR) approach addresses these issues and enables us to use different types of regression models from all sorts of different data. This and our other Drivers Analysis tools allow us to produce very simple outputs, summarising how well a question is predicted and breaking down the contribution uniquely between predictors.

For scenarios where making future predictions is essential, we can also build simulators in Excel.

Advanced models allow the interpretation of different factors affecting the prediction to provide even greater insight. Our CCR method is the only method that provides factors optimised for key driver problems.

Please contact us for further information.

Case Studies
  • Demographic Drivers

    A Government department wanted to understand the main demographic drivers of renewing a government benefit…

  • Medical Device Drivers

    A project among specialist physicians was commissioned to understand the drivers of satisfaction within and…


Key Factors

  • Consultancy on sensible / well defined content
  • Simple, easy to understand outputs summarising predictor contributions
  • Robust to small samples, and many, highly correlated predictors
  • Different types of prediction (scale, yes/no, ordinal etc.)
  • State-of-the art predictive tools and simulators

Log In

The following log in is for clients who have purchased our Data Visualisation Services.
Enter your log in user or email and password

Click here for more information on our interactive graphs.

CORExpress® Register

Enter your details on our register for more information on CORExpress® Software and training courses.