Kamakura's Analytic Tools for Excel

Data-Driven Cases

In this page you will find a collection (to be extended…) of data-driven cases featuring some of KATE's analytic tools:

  • Each of these data-intensive cases contains at least one database. Many of them require students to submit their predictions or targeting decisions, which the instructor can then grade based on their performance on outcomes that only s/he will know.

  • The text describing each case-setting is fairly short, but the analytical work required is more intense than the short text implies! Each case was designed so that students do all the work using Excel and KATE.

  • Some of the cases require a certain level of Excel proficiency, utilizing functions such as =VLOOKUP, =IFNA, =SUMPRODUCT and =TRIM.

  • If you are a university instructor, you are free to use any of these cases in your classes. If you want a copy of the solution and grading sheet for any of these cases, please send me an e-mail from your university e-mail address containing a link to your professional webpage and indicating the case(s) you are interested in.

Here are the cases available so far:

  • American and British Time Styles – application of K-Means Clustering for market segmentation based on time-use, and demographic profiling of the identified segments.

  • Predicting the prices for pre-owned Audis – students are introduced to the concept of Hedonic Price Regression. They are asked to develop a linear regression model that best explains the pricing of pre-owned Audis in online classified ads. Then, they are asked to predict the posted prices for a holdout set of re-owned Audis. The grade on this case is proportional to the accuracy (RMSE) of these price predictions.

  • Diabetes Among Pima Indian Women – students are introduced to response modeling and propensity scoring by developing their own estimator for Logistic Regression on Excel, starting with results from a standard linear regression.

  • Default on credit-card loansstudents first develop their own estimator for Logistic Regression to learn how response modeling works and how to assess the fit of binary response models. They then use KATE to make optimal (maximum cost savings) targeting decisions on a holdout sample and are graded in proportion to the actual cost savings known only to the instructor.

  • Why are we losing so much human capital?case focused on the interpretation of Linear and Logistic Regression results (fit, predictive fit and parameter estimates) produced by KATE.

  • Caravan Insurance – students are asked to develop a propensity-scoring model on a calibration sample and make targeting decisions on a test sample for which they only see the potential predictors. This case including a scoring sheet that computes the net customer lifetime value for each prospect in the test sample.

  • AFR Lapsed Donors – this is a more advanced case on response modeling than the previous ones. Here, students are asked to make targeting decisions based not only on who is likely to make a donation, but also on how much they are expected to donate.