1. Marketing Analytics Technology: Statistical Analysis Software & R Conjoint Analysis Using R Stephan Sorger www.stephansorger.com © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 1www.stephansorger.com Disclaimer: All images such as logos, photos, etc. used in this presentation are the property of their respective copyright owners and are used here for educational purposes only Some material adapted from: Sorger, “Marketing Analytics: Strategic Models and Metrics” PREREQUISITE: R BASICS

2. Outline: 10 Steps to Conjoint Analysis TopicDiscussion 1. Attribute SelectionSelect relevant product characteristics 2. Attribute Level SelectionIdentify levels of characteristics 3. Product Bundle FormationCreate sample products by bundling 4. Data Collection Method SelectionDecide how to collect data 5. Data CollectionAsk responders to rate products 6. Data PreparationGet data ready for conjoint 7. Importance Weights CalculationAssess importance of each attribute 8. Part-Worth CalculationAssess importance of each level 9. Market Share EstimationPredict preferences of products 10. Data InterpretationCompare against industry averages © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 2www.stephansorger.com

3. Conjoint Analysis Example TopicDiscussion Acme Espresso You are the marketing manager for Acme Espresso Machine Small manufacturer of premium coffee makers Competitors: Breville, DeLonghi, Gaggia, Rancilio © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 3www.stephansorger.com

4. Conjoint Analysis Example TopicDiscussion ObjectivesUnderstand how consumers assess different products Identify specific features desired by consumers Predict market share for different product variants ConjointConduct conjoint analysis to address objectives Use 10-step process Steps 1 - 5 Steps 6 - 10 Review Execute using R © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 4www.stephansorger.com

5. Step 1: Attribute Selection TopicDiscussion AttributesCharacteristics that consumers find relevant when evaluating ResearchSecondary: Product reviews, etc. Primary: Customer surveys Capacity Price Number of cups machine can brew Retail price of espresso machine SpeedBrewing time, in minutes © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 5www.stephansorger.com

6. Step 2: Attribute Level Selection TopicDiscussion LevelsIdentify levels of attributes where consumers see differences Two LevelsIn this exercise, we will identify two levels for each attributes Capacity Price C1: “Single-Cup”: 1 cup at a time C2: “Multi-Cup”: Multiple cups at a time P1: “Budget”: Price under $300 P2: “Premium”: $300 or more Speed S1: “Fast”: Under 1 minute S2: “Slow”: 1 minute of more © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 6www.stephansorger.com

7. Step 3: Product Bundle Formation TopicDiscussion BundlesCreate candidate products by bundling together attributes Vary the levels of the attributes Bundles often called cards Quantity(Levels) ^ (Attributes); 2^3 = 8 Card/BundleSpeed (S)Capacity (C)Price (P) 1111 2112 3121 4122 5211 6212 7221 8222 © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 7www.stephansorger.com

8. Step 4: Data Collection Method Selection TopicDiscussion Pairwise Comparison Respondents compare pairs of options Rank Ordering Respondents place options in rank order: 1 – 100 Rating Scale* Respondents rate each option independently Rank Ordering Rating Scale Pairwise Comparison Data Collection Techniques © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 8www.stephansorger.com

9. Step 5: Data Collection TopicDiscussion Rating ScaleAsk respondents to rate each card on a scale from 1 to 5 1 = poor; 5 = outstanding Card/BundleSpeed (S)Capacity (C)Price (P)Preference 11114 21123 31215 41224 52113 62121 72213 82222 © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 9www.stephansorger.com

10. Step 6: Data Preparation TopicDiscussion DataPrepare the data for use with R Same data we used for Microsoft Excel approach StepData Preparation ADownload R BLaunch R CInstall the R conjoint package DLoad the conjoint library EConvert data set from Excel-friendly format to R-friendly format FLoad the data into R © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 10www.stephansorger.com

11. Step 6: Data Preparation; A: Download R PlatformLink Windowshttp://cran.r-project.org/bin/windows/base/http://cran.r-project.org/bin/windows/base/ Machttp://cran.r-project.org/bin/macosx/http://cran.r-project.org/bin/macosx/ © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 11www.stephansorger.com

12. Step 6: Data Preparation; B: Launch R TopicDiscussion PromptYou will see a “>” prompt in the “R Console” You will be typing commands at the prompt: “ > “ © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 12www.stephansorger.com

13. Step 6: Data Preparation; C: Install R Conjoint Package TopicDiscussion PackagesNeed special functionality to execute our conjoint analysis R extends it functionality through functions called packages (1) ConjointPackage called “conjoint” for conjoint analysis (2) (1)R Documentation, “Install Packages from Repositories http://stat.ethz.ch/R-manual/R-devel/library/utils/html/install.packages.html (2) R Documentation, Package “Conjoint”. August 29, 2013. Bak, Andrej and Bartlomowicz http://cran.r-project.org/web/packages/conjoint/conjoint.pdf Syntax: install.packages(“packagename”) Our case:install.packages(“conjoint”) (See next few slides for screenshots) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 13www.stephansorger.com

14. Step 6: Data Preparation; C: Install R Conjoint Package TopicDiscussion Screenshotinstall.packages(“conjoint”) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 14www.stephansorger.com

15. Step 6: Data Preparation; C: Install R Conjoint Package TopicDiscussion ScreenshotDownloading multiple files in package © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 15www.stephansorger.com

16. Step 6: Data Preparation; C: Install R Conjoint Package TopicDiscussion ScreenshotPackage elements unpacked successfully © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 16www.stephansorger.com

17. Step 6: Data Preparation; D: Load Conjoint Library TopicDiscussion Load fileslibrary(conjoint) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 17www.stephansorger.com

18. Step 6: Data Preparation; E: Convert Data Set to R Format TopicDiscussion PreferencesMatrix dedicated to preferences stated by respondents Card/BundleSpeed (S)Capacity (C)Price (P)Preference 11114 21123 31215 41224 52113 62121 72213 82222 © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 18www.stephansorger.com

19. Step 6: Data Preparation; E: Convert Data Set to R Format TopicDiscussion ProfilesMatrix dedicated to description of data structure Card/BundleSpeed (S)Capacity (C)Price (P) 1111 2112 3121 4122 5211 6212 7221 8222 © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 19www.stephansorger.com

20. Step 6: Data Preparation; E: Convert Data Set to R Format TopicDiscussion ProfilesMatrix dedicated to description of data structure © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 20www.stephansorger.com

21. Step 6: Data Preparation; E: Convert Data Set to R Format TopicDiscussion ProfilesSave As CSV © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 21www.stephansorger.com

22. Step 6: Data Preparation; F: Load Datasets Into R TopicDiscussion Preferencespreferences<-read.csv(“C:\\Users\\user\\Desktop\\preferences.csv”, header=T) Profilesprofiles<-read.csv(“C:\\Users\\user\\Desktop\\profiles.csv”, header=T) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 22www.stephansorger.com

23. Step 7: Importance Weights Calculation TopicDiscussion ImportanceCalculate importance weights in conjoint Refers to importance of each attribute, such as speed caImportanceR command built into the conjoint analysis (ca) package SyntaxcaImportance(preferences, profiles) caImportance = conjoint analysis function for importance preferences = matrix with preferences for each card profiles = matrix describing each profile or card Commandimportance <- caImportance (preferences, profiles) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 23www.stephansorger.com

24. Step 7: Importance Weights Calculation © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 24www.stephansorger.com

25. Step 7: Importance Weights Calculation TopicDiscussion ComparisonWe compare the Excel results with the R results VariableExcel Coefficient% of TotalR Results Speed1.751.75/3.75=46.67%46.67% Capacity0.75: absolute value0.75/3.75=20.00%20.00% Price1.251.25/3.75=33.33%33.33% Total3.753.75/3.75=100%100% © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 25www.stephansorger.com

26. Step 8: Part-Worth Calculation: Levelnames Matrix TopicDiscussion LevelnamesMatrix to capture names of levels for each attribute Speed: fast, slow Capacity: onecup, twocups Price: cheap, expensive © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 26www.stephansorger.com

27. Step 8: Part-Worth Calculation: Calculate Part Worths TopicDiscussion Read datalevelnames<-read.csv(“C:\\Users\\user\\Desktop\\levelnames.csv”, header=T) Commandpartutil <- caPartUtilities (preferences, profiles, levelnames) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 27www.stephansorger.com

28. Step 8: Part-Worth Calculation: Examine Part Worths TopicDiscussion Part Worthsinterceptfastslowonecuptwocupscheapexpensive 3.1250.875-0.875-0.3750.3750.625-0.625 © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 28www.stephansorger.com

29. TopicDiscussion Part Worthsinterceptfastslowonecuptwocupscheapexpensive 3.1250.875-0.875-0.3750.3750.625-0.625 Step 8: Part-Worth Calculation: Interpret Part Worths Preference = Intercept + (PW_fast) + (PW_slow) + (PW_onecup) + (PW_twocups) + (PW_cheap) + (PW_expensive) Preference = 3.125 + 0.875 * fast - 0.875 * slow - 0.375 * onecup + 0.375 * twocups + 0.625 * cheap - 0.625 * expensive Test machine 1: fast, twocup, cheap: Preference = 3.125 + 0.875 * 1 - 0.875 * 0 - 0.375 * 0 + 0.375 * 1 + 0.625 * 1 - 0.625 * 0 = 5.00 (perfect!) Test machine 2: fast, twocup, expensive: Preference = 3.125 + 0.875 * 1 - 0.875 * 0 - 0.375 * 0 + 0.375 * 1 + 0.625 * 0 - 0.625 * 1 = 4.375 (good) Test machine 3: slow, twocup, expensive: Preference = 3.125 + 0.875 * 0 - 0.875 * 1 - 0.375 * 0 + 0.375 * 1 + 0.625 * 0 - 0.625 * 1 = 2 (fair) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 29www.stephansorger.com

30. Step 9: Market Share Estimation: Maximum Utility TopicDiscussion caMaxUtilitymaxutil <- caMaxUtility (simulations, preferences, profiles) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 30www.stephansorger.com

31. Step 9: Market Share Estimation: Maximum Utility TopicDiscussion caMaxUtilitymaxutil <- caMaxUtility (simul, preferences, profiles) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 31www.stephansorger.com

32. Step 9: Market Share Estimation: Bradley-Terry-Luce (BTL) TopicDiscussion caBTLbtl <- caBTL (simul, preferences, profiles) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 32www.stephansorger.com

33. Step 9: Market Share Estimation: Bradley-Terry-Luce (BTL) TopicDiscussion caBTLbtl <- caBTL (simul, preferences, profiles) © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 33www.stephansorger.com

34. Step 10: Data Interpretation TopicDiscussion ComparisonCompare R results with actual market behavior R ResultsMost important attribute: Speed ArticleSmithsonian article: “The Long History of Espresso Machine” First espresso machines: 1000 cups/ hour ReviewsOnline product reviews, such as those on home-barista.com Many evaluation criteria: - Aesthetic design - Quality of grinder - Degree of automation © Stephan Sorger 2015: www.stephansorger.com; Marketing Analytics; Tech: R Conjoint: 34www.stephansorger.com