CSP 2017 Online Program

Two-Step Logistic Regression Model for Predicting Phone Campaign Response (303409)

Keywords: phone campaign, two-stage, logistic regression, lift

AARP Social Security Campaign is a standard phone campaign. We aim to collect members' responses about social security questions operator reads from the script. For targeting as many as possible members who are positive (answering ‘Yes’) to the topic, our traditional way was to build one binary logistic regression model based on the response of all the members we called. The new method we present considers the campaign process as two steps. First, whether the member answers the phone; second, under the condition of answering the phone, whether the response is 'Yes'. Therefore, the first model is to predict member's likelihood to answer the phone; the second model is to predict the respondent's positiveness only among members who answered the phone. The multiplication of probabilities from both models is the new campaign response likelihood. Validation shows the new two-step method significantly improves the accuracy of the model and brings a bigger lift compared to the traditional one-step way and can be applied to other AARP phone campaigns.