|
Title:
|
PREDICTIVE PROBABILITIES IN MARKETING |
|
Author(s):
|
Krzysztof Dzieciolowski |
|
ISBN:
|
978-989-8704-10-8 |
|
Editors:
|
Ajith P. Abraham, Antonio Palma dos Reis and Jörg Roth |
|
Year:
|
2014 |
|
Edition:
|
Single |
|
Keywords:
|
Predictive probabilities, posterior, bias, monotonic cubic splines |
|
Type:
|
Short Paper |
|
First Page:
|
175 |
|
Last Page:
|
179 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Predictive models are increasingly important in marketing due to their demonstrated ability to derive valued information from customer data. The models enable companies to optimize their marketing campaigns and increase profitability. However, model predictions often cannot be considered as probabilities because of oversampling or other forms of bias introduced during data collection or model implementation. When the bias is not recognized, the inferences based on model scores may therefore not be accurate. An commonly used correction for bias is one that is based on Bayesian posterior probabilities, but due to its global rescaling approach, it too may not provide appropriate estimates locally. As an alternative, we propose a novel approach of estimating predictive probabilities by applying monotonic cubic splines to empirically derived probabilities as proportions of an event of interest in quantiles. Our approach ensures that predictive probabilities are monotonic with respect to model scores. In addition, a modified method of obtaining quantiles based on an equal number of events rather than observations has been introduced. During a simulation study, this modified model was found to have smaller average squared error for high model scores than population based bins or posterior probabilities. An implementation of smoothed monotonic predicted probabilities is illustrated with a large marketing dataset. |
|
|
|
|
|
|
