Sawtooth Software Conference

Tutorial Sessions

Orlando, Florida – May 3, 2022

Sign Up for Virtual Access

In-Depth Learning

Drill deeper and extend your learning with these 3.5-hour tutorials.  Led by accomplished researchers and speakers, you’re in great hands!

Presentations

Go to Room 2 Presentations

Room 1 – Tuesday, May 3

8:00AM
Breakfast
9:00AM
Why, When, and How to Use Conjoint Analysis in Your Research - Practical Guidance for Beginners and Experts Alike

Learn how to delight your clients and execute great research in this consultative guide to using conjoint analysis.  In this tutorial, we’ll walk through the entire conjoint analysis project life-cycle — from proposal to deliverable — and offer insights around designing your experiment, key questions to ask your clients, pitfalls to avoid, and what to look for in the results. We’ll even cover strategies for reporting that enhance presentations and give your clients confidence in the results. For both newbies and pros alike!

Megan Peitz, Numerious
10:30AM
Break
11:00AM
Why, When, and How to Use Conjoint Analysis in Your Research - Practical Guidance for Beginners and Experts Alike (Continued)
Megan Peitz, Numerious
12:30PM
Lunch
1:30PM
Understanding & Predicting Categorical Outcomes in Marketing Data

Categorical data analysis involves examining data where the target/response variable has been grouped into a set of mutually exclusive categories each of which share a common trait. Some examples include:

• Scoring customers into segments.
• Predicting success or failure of new products.
• Accepting or rejecting admission to an applicant.
• Predicting credit risk category for a person.
• Classifying patients into different categories.
• To predict whether an email is spam or not spam.
• Predicting if a tumor is malignant or not.

This tutorial will explore the application of both statistical, as well as machine-learning based models which may be used for categorical data analysis. Drivers of membership in specific categories will be identified and ranked by importance.

Statistical models such as discriminant analysis and logistic regression will be reviewed. Machine learning algorithms on the other hand will include Support Vector Machines (SVM) as well as tree based recursive partitioning approaches including: Classification And Regression Trees (CART), Random Forest (RF) analysis and eXtreme Gradient Boosting (XGBoost).

In addition to evaluating drivers of categorical group membership, market researchers often wish to predict/score customers into specific categories. For example, researchers may need to score a large client customer-base into clusters identified in a previous study. Model predictive performance will therefore be examined in this tutorial as well. Predictive effectiveness will be evaluated usingcross validation, as well as Out-Of-Bag (OOB), sample prediction.

All data analysis should begin with simple yet insightful data visualization. Visualizations specific to categorical data include such things as: bar charts and mosaic plots, both of which will be illustrated.

“Model” visualization will also be emphasized. This will be accomplished using using appropriate graphical depictions such as permutation importance plots, partial dependence plots, perceptual mapping, etc.

What you will learn in this tutorial:

• How to create useful and engaging visualizations of categorical data.
• Gain an understanding of a variety of categorical modeling techniques along with their advantages and disadvantages.
• Learn how to employ a variety of statistical and machine learning models using R including Automated Machine Learning (AutoML).
• Evaluate model predictive performance with cross-validation and out-of-bag sampling.
• Learn how to perform and evaluate XGBoost models in both R and JMP.
• Learn how to run analyses in Python using the RStudio IDE (via reticulate) and integrate those results into R. Examples of advanced R-markdown will also be presented.

Joe Retzer, Acxiom
3:00PM
Break
3:30PM-5:00PM
Understanding & Predicting Categorical Outcomes in Marketing Data (Continued)
Joe Retzer, Acxiom
Back to Top

Room 2 – Tuesday, May 3

8:00AM
Breakfast
9:00AM
Segmentation & Drivers Analysis

Driver analysis and segmentation are two things marketing research analysts do about as frequently as they run choice experiments. Worse ways of doing driver analysis are easy to identify and also easy to avoid by using demonstrably better methods. The first part of this tutorial covers how and why the poor methods should be avoided and why and how to use better methods instead. Hint:  if you’re running correlation or regression, you really need better methods for driver analysis.      

Less happily, a constellation of mutually reinforcing problems can thwart the segmentation analyst. The longer second part of the tutorial focuses on understanding these several problems and on how to try to avoid them through (a) well-planned research objectives, (b) understanding the nature of your data with diagnostics, and (c) ameliorating as many of the problems as possible using the imperfect tools available to us, namely questionnaire construction, variable selection and clustering algorithms.

Keith Chrzan, Sawtooth Software
10:30AM
Break
11:00AM
Segmentation & Drivers Analysis (Continued)
Keith Chrzan, Sawtooth Software
12:30PM
Lunch
1:30PM
Practical Conjoint Solutions

Sometimes standards are not enough to answer the research questions at hand, and one must deviate from the roads most travelled. We would like to get you on board for this 4-hour tutorial on advanced applications. Here we will focus on several challenging but very interesting extensions of choice modeling, including:

• Custom statistical design techniques (designs using Lighthouse Studio and Excel; e.g. alternative-specific, frequency imbalance, pricing rules) and implications of these customizations
• Showing different coding methods
• Duct-tape solutions for the red-bus-blue-bus problem
• And some other practical solutions

Egle Meskauskaite and Jeroen Hardon, SKIM
3:00PM
Break
3:30PM-5:00PM
Practical Conjoint Solutions (Continued)
Egle Meskauskaite and Jeroen Hardon, SKIM
Back to Top