Using Logistic Regression Models / Evaluation of Multiple Models

Course Number:  MKTG 3509: Customer Data Analytics

Semester: Summer 1, 2020

Homework 5: Using Logistic Regression Models / Evaluation of Multiple Models

Objectives

The objectives of this final exercise are to: 1) practice building and evaluating different models to identify which customers in our database should receive a future marketing communication from our company. You will create a univariate segmentation, and a predictive model using logistic regression analysis, 2) evaluate the lift and gains of each approach and 3) evaluate the profitability of each of the approaches.

Assignment

As a direct marketer of specialty books, the BookBinders Book Club has achieved steady growth in their customer base. Yet while sales have grown steadily, profits began falling when the database got larger and when the company diversified its book selection and increased the number of offers sent to customers. The falling profits have led Dave Lawton, BookBinders’ marketing director, to experiment with different database marketing approaches in order improve BookBinder’s mailing yields and profits.

Dave began a series of live market tests, each involving a random sample of customers from the database. An offer for the current book selection is sent to the sample and then the sample customers’ responses, either purchase or no purchase, are recorded and used to calibrate a response model for the current offering. The response model’s results are then used to “score” the remaining customers in the database and select customers from the full customer database for the ‘rollout’ mailing campaign.

Dave’s team continues to debate which of two methods will best identify potential targets for its new book, ‘The Art History of Florence’. The team is discussing two possible approaches to identify and target customers for ‘The Art History of Florence’:

1. Volume Segmentation (univariate segmentation using Totdol as the segmentation variable), or
2. Predictive Modeling using Logistic Regression Analysis

Dave has a dataset containing the responses of a random sample of 50,000 customers to “The Art History of Florence.” He is eager to assess the potential value of each of three approaches as methods for predicting customer response and has asked you to complete the following analyses.

Note: In order to complete this analysis you will need the two attached files: 1) BBB.sav, SPSS data file which contains all of the data needed to create volume deciles and to conduct the logistic regression analysis, and 2) Homework 5 Lift Gains Profit Analysis.xlsx, which contains 2 pre-formatted worksheets to assist you in your assessment of lift, gains and profitability. You will need to copy and paste data from SPSS output into Excel to complete your assessment.

Due Date:

Thursday, June 18 at 7:00 p.m.

During class we will develop two segmentation models:

• We will segment customers based on volume.

3 equations: (Predicted Probability) p= eB0 + B1(x) / 1+eB0 + B1(x), (Predicted Odds) (p/1-p)=eB0 + B1(x)

**The event something occurring is (p) and the event something not occurring is (1-p) these two divded are the odds of something happening.

• We will identify the best logistic regression model and then use the predicted probabilities from the model to segment customers.

In order to complete the assignment, you will need to:

• Provide answers to the questions on the following pages of this document, and
• Conduct a lift/gains analysis in Excel.

You need to upload both this document and the Excel file which has been provided.

Attention: This assignment requires you to perform another lift/gains/profit analysis. If needed, review the three videos in Fox Video Vault: 1) Evaluating Models: Lift Analysis, 2) Evaluating Models: Gains Analysis, and 3) Evaluating Models: Profit  Analysis.

Question 1: Evaluate the following two Model Comparisons

Comparison 1:

Comparison 2:

Use SPSS output to answer comparison.

The best comparison is #2. Why, support answer.

Question 3:

Using the equation with all 9 independent variables, how are the odds of buying the ‘Art History of Florence’ affected by:

1. # purchases of Art books (Art) – How do the odds of buying change for an additional Art book purchase?

1. # purchases of Do it Yourself books (Do_it) – How do the odds of buying change for an additional Do_It Yourself book purchase?

Question 4:

Use the following cost information to assess the profitability of both modeling methods.

Using the data from the Lift/Gains/Profitability Analysis that you completed in Excel, complete the following table.