2019 Advanced Research Techniques (ART) Forum

Registration & Continental Breakfast

Tutorial 1A - Introduction to R and the Tidyverse

R is one of the most popular open-source languages for data analysis. The tidyverse is a collection of powerful packages in R for common tasks, including data cleaning and visualization. The tidyverse provides a consistent and intuitive introduction to R and serves as a foundation for a growing number of packages in R (e.g., see the tutorial on text analysis). In this tutorial, we will cover the basics of using R and the tidyverse.

Before the tutorial, participants should setup a free account at https://rstudio.cloud.

• Marc Dotson, BYU

Tutorial 1B - Introduction to Bayesian Statistics

This tutorial provides an introduction to Bayesian analysis in marketing with a focus on how this paradigm differs from traditional statistical thinking.  We will cover the basics of probability distributions and conditional probability, what a posterior distribution is and why it is important, and how we calculate the posterior distribution.  Bayesian statistics provides a number of practical advantages over traditional statistical techniques and introduces a different way of thinking about models, uncertainty and evidence.  It is especially valuable in marketing analysis, modeling and forecasting, and serves as a foundational skill for many commonly used marketing research techniques.

• John Howell, BYU

Tutorial 2A - Extensions of Discrete Choice Modeling: Situational Choice and Menu-Based Choice

Most researchers are familiar with choice-based conjoint and MaxDiff, two prominent types of choice models which together cover a wide range of research needs. Researchers should realize that there are a variety of types of choice models that fit different sorts of marketing research objectives. This session covers two models we see regularly in our practice, Situational Choice Models and Menu-Based Choice models. Keith and Bryan will introduce the theory behind these choice models and show some applications, hopefully leading attendees to recognize what kinds of marketing needs these tools can meet. Attendees should leave the session with an expanded understanding of choice modeling and a more complete choice modeling tool kit.

• Keith Chrzan and Bryan Orme, Sawtooth Software

Tutorial 2B - Advanced A/B Testing

There is nothing better than a head-to-head A/B test to drive marketing decisions. A/B tests lead to simple analytics that almost everyone in the business can understand and use. If you’ve run a few A/B tests, this workshop will show you how to use more powerful analytics tools like predictive modeling and machine learning to get more insight out of your tests and turbo-charge your A/B testing program. After reviewing the basics of A/B test analysis, we will cover advanced techniques including heterogeneous treatment effects, uplift modeling, blocking, matching, and stratification. If these sound like a bunch of buzzwords to you right now, that’s okay. The course will be organized in terms of testing strategies you can use when you have a very large sample size and strategies that help when your sample size is very small. We will also cover Bayesian approaches to A/B testing including test & roll and multi-armed bandits.

• Elea McDonnell Feit, Drexel

Tutorial 3A - Text Analytics in R

Text analysis has become increasingly accessible in R. The recent development of the tidytext package, built on the foundation of the tidyverse, continues this trend. In this tutorial, we will cover the basics of text analysis in R consistent with the underlying philosophy of the tidyverse, including basic counts, visualizations, sentiment analysis, and topic modeling. Basic fluency in R and the tidyverse is expected (e.g., see the tutorial on R and the tidyverse).

Participants should have setup a free account at https://rstudio.cloud prior to the tutorial.

• Marc Dotson, BYU

Tutorial 3B - Using Machine Learning for Causal Inference

This tutorial provides a hands-on primer to harnessing machine learning methods to measure causal effects from observational data. These methods have important applications in marketing including measuring the effect of advertising and other marketing actions like opening new stores. An active area of recent methodological research, this workshop will overview two distinct methodological areas: (1) Lasso-IV and the broader set of double machine learning methods and (2) synthetic controls and related matrix completion methods. Each of these streams of methods will be illustrated with marketing applications. Simulated data exercises will introduce participants to R code that uses the freely available software packages.

• Mitch Lovett, Rochester

Reception

Welcoming Introduction | Jeff Dotson, BYU

General Session - New Methods I

Conjoint meets AI

Artificial Neuronal Networks (ANN) can be used to approximate any data generating process due to the universal approximation theorem. As the data generating process is known when generating experimental designs for choice models, it is relatively easy to generate a large number of processing units (synthetic respondents) and train the ANN to solve the problem find an optimal experimental design for complex choice models. The paper shows based on synthetic data how to generate experimental deigns using ANNs.

• Peter Kurtz – Managing Partner, Innovation & Methods

Modeling Consumer Engagement Across Multiple Social Media Platforms

With increased availability of social media, an issue that needs understanding is whether messages on one social medium affects the consumer engagement on the other social media. This issue is particularly relevant in social media marketing because the number of exposures to a message is not just under the control of the marketer. Due to the ease of sharing information, consumers often post messages they see on one social medium on to another social medium. This study investigates both the direct efforts of marketers on one social media platform and the indirect effects of its spread on to other social media platforms.

• Vasu Unnava and Ashwin Aravindakshan, University of California, Davis

General Session - New Methods II

Benefit Formation and Enhancement

The distinction between attributes and benefits is at the heart of product development. Attributes represent the physical and psychological inputs that comprise a product from which consumers may or may not realize benefits. Some attributes are benefit enabling in the sense that their presence signals that an offering is potentially responsive to a benefit being sought, while their absence indicates that the product does not provide the benefit. Not all brands, for example, are seen as offering luxury and specific brand names may be needed for consumers to consider an offering as luxurious. Benefit formation hinges on these types of attributes being present, while other attributes may only enhance a benefit that has already been formed. We develop a model to identify attributes that are critical to a benefit formation and apply it to two conjoint datasets. We show that identifying product features needed for benefit formation is important to product development activities.

• Hyowon Kim and Greg Allenby, Ohio State University

General Discussion

Introduction to Afternoon Sessions | Elea Feit, Drexel University

Advances in Choice Modelling I

MDRG’s Pricing Playbook

MDRG’s Pricing Playbook delves into the research behind creating optimal pricing strategies for a retail brand client against its major competitors. Over 2,000 survey respondents participated in the Adaptive Choice-Based Conjoint (ACBC) methodology. From the thousands of data points, MDRG created a practical solution that managers across the organization utilized to effectively price their products in the market.

• Heath Gregory, MDRG

Product Launches with New Attributes: A Hybrid Conjoint-Consumer Panel Technique for Estimating Demand

We propose and empirically evaluate a new hybrid estimation approach that integrates choice-based conjoint with repeated purchase data for a dense consumer panel, and show that it increases the accuracy of conjoint predictions for actual purchases observed months later. Our key innovation lies in combining conjoint data with a long and detailed panel of actual choices for a random sample of the target population. By linking the actual purchase and conjoint data, we can estimate preferences for attributes not yet present in the marketplace, while also addressing many of the key limitations of conjoint analysis, including sample selection and contextual differences. Counterfactual product and pricing exercises then illustrate its managerial relevance.

• Paul Ellickson, Mitch Lovett and Bhoomija Ranjan, University of Rochester

Advances in Choice Modelling II

Informing Preference Heterogeneity with Stated Preferences or Passive Geolocation

How informative is passive geolocation for preference heterogeneity in choice models? Using a unique dataset where survey panelists have opted into passive geolocation tracking, we compare models where standard stated preferences or passive geolocation are used as covariates in the upper-level model. The aim of this research is to help determine which data, or combination of data, is more effective for informing preference heterogeneity and improving predictive fit.

• Marc Dotson and Edward Johnson, BYU and Dynata

Round-Table Discussions

Introduction to Session | Greg Allenby, OSU

New Data I

Leveraging Limited Loyalty Programs Using Competitor Based Targeting

For many firms, the location of an individual store with respect to both their customers and their competition will strongly influence the level of customer engagement with the firm. We illustrate through machine learning techniques the advantage of incorporating complex spatial information into the targeting decision. Our approach uses readily available software and highlights the potential gains of considering location effects on targeting decisions. This research allows managers to easily asses the influence of rich spatial information onto customer behavior and in turn guide their decisions on which customers to target for marketing.

• Wayne Taylor (SMU) and Brett Hollenbeck (UCLA)

How to design a better diary study 

This paper will take a fresh look at how to more effectively conduct diary research based on an extensive series of primary research experiments and a ground-up rethink about the whole process of in the now data capture. It will center around a real-life case study, applying modern survey design and techniques in practice to demonstrate the impact that re-thinking and making simple changes and improvements can make to the end results.

We will offer everyone who has ever had to conduct any form of repetitive research the chance to learn how to adopt a modern design approach in conducting diary studies more effectively in the future, a huge opportunity to update traditional research in the face of a fast-changing industry and consumer landscape.

• Mitch Eggers, Kantar Profiles division, previously Lightspeed

New Data II

Multivariate Analysis with MaxDiff

MaxDiff is a popular method for sorting lists of attributes, but is it appropriate as an input to multivariate techniques like regression analysis and cluster analysis? We’ll share the results of a real study where survey respondents were split between a MaxDiff exercise and a select any of J question.

• Jeff Brazell and Jake Lee, Blue Owl AI