Marketing Analytics and Big Data (MARK3054) UNSW Business School PDF

Marketing Analytics and Big Data (MARK3054) Topic 1: Introduction to Marketing Analytics Dr. SunAh Kim Senior Lecturer in Marketing, UNSW Business School [email protected] Dr. SunAh Kim UNSW Business School Today: Introduction About Marketing Analytics Course Details Dr. SunAh Kim UNSW Business School Lecturer-in-Charge SunAh Kim – Senior Lecturer in Marketing – Ph.D. in Marketing, Purdue University, US – Research interest: Effects of marketing intervention/strategy on consumer choice and purchase path (e.g., product recommendation, nutrition label) New technology (AI) and public policy on consumer demand and market (e.g., Sharing economy, chatbot) – Expertise related to this course : Quantitative empirical modelling (https://scholar.google.com/citations?user=xHfuYKIAAAAJ&hl=en) Teaching: Marketing Analytics and Big Data (UNSW), Strategic Marketing Planning (Concordia, Canada), Research Methods & Marketing Models (Concordia, Canada), Marketing Management (Purdue, US), Business Statistics (Univ. of Iowa, US) Former database marketer at TESCO Korea Dr. SunAh Kim UNSW Business School Lecturer-in-Charge SunAh Kim – Senior Lecturer in Marketing – Ph.D. in Marketing, Purdue Univ. US #LoveCalculusandStats#LikeArcad eGames#LikeEdSheeran#Beginne rIceSkate#FoodieFoodieFoodie#lo veDogsandCats#likefishing #workinghardonMarketingModels# workingHARDERtohaveFUNwithm ydaughter#HaveGoalsEnjoyALLJo urneys#Benice2yourself Dr. SunAh Kim UNSW Business School Our perception on Marketing Analytics If somebody outside marketing asks you the question, “What is marketing analytics?” how would you answer? Which industry or product categories that you think it has a high demand for marketing analytics? Someone saying that – “I like to be a brand manager or advertising creator, and should I still care about analytics? Probably no…” – “I like to be an influencer marketer owning my own brand and channel, and I want to focus on building creative thinking, not analytical thinking.” – Do you agree or disagree with these perspectives? Dr. SunAh Kim UNSW Business School Some Examples of Marketing Analytics Dr. SunAh Kim UNSW Business School Marketing Analytics: Examples Source: https://youtu.be/lCz0cGIWmlU Dr. SunAh Kim UNSW Business School Marketing Analytics: Examples CRM at Tesco “An opportunity to target, engage and reward loyal Tesco Clubcard customers using Direct Mailings, Reward Mailing Coupons, and Coupon at Till. Through these products, you can deliver tailored, targeted messaging directly to your desired customers.” Source: https://www.dunnhumby.com/wp-content/uploads/2021/11/Tesco-Media-Insight-CRM-OnePager.pdf10 Dr. SunAh Kim UNSW Business School Marketing Analytics: Examples WHERE TO UNEXPECTED PLACEIN VICTORY ORANGE SHOES POLITICAL ? -- ALDO’ ELECTION ML PRACTICE ON CSONSUMER ’S VARIETY SEEKING AND NEW BRAND ADOPTION You can’t spell retail without data: Optimize your chain retail store design with data science DOES PRODUCT ALIGNMENT BY COLOR (VS. PRICE) MATTER IN CONSUMER CHOICE? IF SO, HOW AND WHY? Consumers are less price-sensitive when products are aligned by color (vs. price) Dr. SunAh Kim UNSW Business School Marketing Analytics: Examples Recommendation and promotion based on behavioral data “Of course they know you’re a dog. Your shopping patterns give it away.” – Crowden Satz Dr. SunAh Kim UNSW Business School Contemporary Marketing Analytics Dr. SunAh Kim UNSW Business School Evolution of Marketing & Analytics Orientation Profit driver Timeframe Analytics Production Production methods until the 1950s Minimal Product Quality of the product until the 1960s Minimal Selling Selling methods 1950s and 1960s Minimal Needs and wants of 1970s to the Customer Start to thrive customers (value) present day Keep a long-term 1990s to the Important+ Relationship relationship present day (engagement) Rely more on data to 2000s to the Important++ manage relationship present day Dr. SunAh Kim UNSW Business School A Brief History of Data and Marketing Analytics Michel Wedel and P. K. Kannan (2016), “Marketing Analytics for Data-Rich Environment”, Journal of Marketing https://journals.sagepub.com/doi/full/10.1509/jm.15.0413 Dr. SunAh Kim UNSW Business School Expectation for Contemporary Marketing Analytics As marketing evolves, there is higher and higher needs to: – Understand customer needs deeper – Understand customers at individual level (understanding customer heterogeneity) – Understand customers in a dynamic nature – Take personalized actions – Take actions in real time – Make decisions scientifically (evidence based) – As a result, optimize marketing-mix and improve ROI Dr. SunAh Kim UNSW Business School Reality in Companies Some are leading the development of big data tech. Many others: – Cumulated tons of data but don’t know what to do with it – The top-level managers (e.g., CEO, CMO) are eager to embrace the digital age. – The middle level managers may not be sure about what analytics can bring them. – The operational level staff don’t know what analytics are or don’t know how to do it. Dr. SunAh Kim UNSW Business School Reality in Companies (Mela & Moorman 2018) Mela, Carl F. and Christine Moorman (2018), "Why Marketing Analytics Hasn't Lived up to Its Promise," Harvard Business Review. https://hbr.org/2018/05/why-marketing-analytics-hasnt-lived-up-to-its-promise Two important analytics trends – Marketing budgets allocated to analytics: 5.8% -> 17.3% (198% increase in last 3 years) – Effects of analytics on performance: 3.8 -> 4.1 (modest increase in the last 5 years, seven-point scale) Two competing forces explain the discrepancy – Data used in analytics: lack of a good design – Analyst talent producing it: lack of the right talent Dr. SunAh Kim UNSW Business School Reality in Companies (Mela & Moorman 2018) What is the RIGHT talent? – Clearly define the business problem – Understand how algorithms and data map to business problems – Understand the company’s goals – Communicate insights, not facts – Develop an instinct for mapping the variation in the data to the business questions – Identify the best tool for the problem – Span skill boundaries Dr. SunAh Kim UNSW Business School Implication for You Global challenges, your opportunity – Smarter customers and markets – Information technology and more data – Employers are struggling to find enough qualified people to make the best possible use of this avalanche of valuable data. The world needs a new breed of marketers who understand marketing AND at the same time, know how to handle massive data and use them to make informed marketing decisions. – Highly sought after and well paid! Dr. SunAh Kim UNSW Business School Implication for You “Only 1.9% of marketing leaders reported that their companies have the right talent to leverage marketing analytics.” (Mela & Moorman 2018) Dr. SunAh Kim UNSW Business School Why? What is NOT Marketing Analytics Marketing analytics talent ≠ people know marketing + people know math/stats/data mining – One needs to build an analytical way to re-investigate marketing problems “Analytical” ≠ “mathematical” or “statistical” – Mathematics and statistics are basic knowledge – Analytics is about using them in practice – Example: different views of correlation coefficients Dr. SunAh Kim UNSW Business School Spurious correlation Dr. SunAh Kim UNSW Business School Views of Correlation Coefficients Where the effect size of correlation is large enough to be useful (e.g., profitable) Statistical test tries to verify whether the correlation exist (i.e., non- zero) r -1 0 1 Dr. SunAh Kim UNSW Business School Views of Correlation Coefficients Take away – We are not trying to degrade the value of statistical test – On the contrary, you have to know mathematics and statistics to be analytical, not knowing the correct statistical concept is not acceptable, you will not go too far if you don’t know that – However, just knowing the math or stats is not enough. We are moving from class to the real world. You need a deeper understanding of how to use them in practice. Dr. SunAh Kim UNSW Business School This Course Intends to Prepare You to: Be able to design an analytical study (LO1) Be able to implement analytical tools using data collected by marketers (LO2) Be able to translate analytical results into managerial insights (LO3) Be able to communicate the findings (LO4) Be able to work in teams and develop collaboration and leadership skills in a team environment (LO5) Dr. SunAh Kim UNSW Business School About marketing analytics What is the main difference between data and information? Analysing Data Information Decisions Neither data, nor information are making decision(s), this is still a job for managers! * Don’t Let Metrics Undermine Your Business. HRB Dr. SunAh Kim UNSW Business School More about the Course Positioning Try to give you the ability to bridge analytics methods and marketing practices The emphasis is on how to use the methods to solve problems, but not to make you a statistician. That said, – You have to have a certain understanding level of the methods to be able to use them and communicate to others. – You are encouraged to take sophisticated statistics courses to get to the bottom of the methods. Dr. SunAh Kim UNSW Business School What to Expect (Contents in the Course) Exposed to a range of statistical tools and techniques, from classical statistical tools to emerging big data techniques. The emphasis is on how to apply and interpret the techniques to help answer marketing- related questions. Use the tools to address daily marketing problems. The tools are organized in that way. Work through a research project in a real context, from problems analysis to conclusions and communication. Learn widely used software for all the analyses, work-ready for a wide range of business, from local small business to multinational giants. Start to build an analytical way of thinking Equip growth (vs. fixed mindset) and intrinsic goals (See https://www.njlifehacks.com/intrinsic-goals-vs-extrinsic- goals/#:~:text=Intrinsic%20goals%3A%20These%20goals%20are,relationships%20and%20your%20personal%20growth. https://online.hbs.edu/blog/post/growth-mindset-vs-fixed-mindset) Dr. SunAh Kim UNSW Business School WOM from Previous Students “This course provided me with knowledge and skills I can apply” among surveyed students: – Over 96% students: “strongly agree” or “agree” “I was a former student in your marketing analytics course last semester and I just wanted to reach out to let you know how helpful and useful I have found your course since finishing university. I currently work at a marketing agency where I have had to use analytical techniques like price optimisation to help a client solve an issue.” Dr. SunAh Kim UNSW Business School WOM from Previous Students “Good news to share with you. I have just got an internship of developing a product recommendation system. The recommendation system is so like the recommendation system in our team project. Thanks for your course.” I also received several requests from students like this one: “I realise that I need some analytical tools taught in your course, but I can’t find the course materials. Can you please send me a copy……” Dr. SunAh Kim UNSW Business School Your voice matters! Incorporate Students Feedback Organize various analytical tools around daily marketing problems Create the group project using a real marketing context and real marketing problems Lectures and tutorials are recorded Let us know your feedback! Dr. SunAh Kim UNSW Business School Course Details Dr. SunAh Kim UNSW Business School Separate Dual Mode, Shared Content All students will attend online lectures. Tutorials are mainly face-to-face (F2F). The tutorial contents will be the same across different sections. – All interaction will take place in the classroom; there is no online channel for students to participate via the Internet. Dr. SunAh Kim UNSW Business School Time and Locations Lectures – Monday 11:00-13:00 – Zoom classroom: https://unsw.zoom.us/j/81909269609?pwd=lTM68CGd7za UGYxEJccOebiZgzJLhN.1 – Meeting ID: 819 0926 9609 – use passcode 30543054 if asked Tutorials – Please make sure you attend the enrolled tutorial section – Information on Moodle (“Course Outline and Tutor Info”) Dr. SunAh Kim UNSW Business School Tutors’ Contact Info Email, consultation time, consultation Zoom link: “Getting Started Hub” section on Moodle Email is the preferred contact method. We will respond within one business day. No appointment is needed if you wish to see us at consultation time. Outside the consultation time, the zoom meeting room is not monitored. If you require contact outside of the consultation times, please email the staff member with your question or to negotiate an alternate and mutually suitable consultation arrangement. Dr. SunAh Kim UNSW Business School Contact Information Preferred contact – Email: [email protected] – Will respond within one business day Office hour – Tuesday 13:00 – 13:30 Consultation Zoom link: https://unsw.zoom.us/j/82886235518?pwd=s14dANAHDv7GFhGnQ2I2bHauvar bZH.1 (use passcode 20002000 if asked) Dr. SunAh Kim UNSW Business School Detailed Course Schedule (Available on Moodle) Dr. SunAh Kim UNSW Business School Detailed Course Schedule (Available on Moodle) Dr. SunAh Kim UNSW Business School Textbook NO prescribed textbook Slides, reading materials and exercise datasets used in a particular week will be available on Moodle by Friday night of the previous week. – For example, Week 2 materials will be available by Week 1 Friday night. Dr. SunAh Kim UNSW Business School Software Microsoft EXCEL – Powerful, widely available and used – Most common way to store and process data in daily use – Makes the knowledge and skills you learn from this course work-ready for a wide range of business, from local small business to multinational giants. – Huge amount of online resources free to learn – Free MS Office for UNSW students: https://student.unsw.edu.au/notices/office Dr. SunAh Kim UNSW Business School Why Excel? Skills required from business job posts in NSW Dr. SunAh Kim Source: Labour Insight Jobs (Burning Glass Technologies) UNSW Business School Why Excel? Student WOM from the past MARK3054 course “Currently I am completing a Marketing internship at Nestle. I could not thank you enough for teaching us Microsoft Excel because it is a basic requirement for us, marketers, to know how to use it! I am so lucky to have taken your course because I didn't know a thing about Excel before this, and I could have died in the office haha! Once again, thank you so much for including excel in the syllabus; it is really useful for marketing students. ” Dr. SunAh Kim UNSW Business School Software Free Excel Add-in Software – Real Statistics Using Excel – http://www.real-statistics.com/. This website has rich Microsoft Excel resources, including Excel add-in software for statistical analyses, statistics instructions, examples, and discussion forums. – Works for both PC and Mac. Dr. SunAh Kim UNSW Business School Software R – Powerful (anything you know about stats), perhaps the most popular statistical software nowadays – Free and open source, updates of latest methods – Work-ready, suitable for virtually all kinds of business. – Huge amount of online resources free to learn – R download: https://www.r-project.org/ – RStudio (popular free integrated development environment for R) download: https://www.rstudio.com/ Dr. SunAh Kim UNSW Business School Software Expectation of R in this course – Be able to complete an analysis taught in the course with R, when the R script is provided to you. That is, you can operate R to run the code, read the results, and utilize the results for your analytical report. – Be able to make basic modifications to the scripts provided to you, so that you can use them to run similar analyses on other sets of data, e.g., for your team project. Dr. SunAh Kim UNSW Business School Software The combination of Excel and R is suitable to handle most small to moderate scale analytical problems, i.e., most of the analytical problems you will face in your work (unless you become a big data expert and handle large scale problems daily). A way I found very useful and popular: – Excel for handling and simple analyses – R for more sophisticated analyses – Excel for result demonstration and simple simulations Dr. SunAh Kim UNSW Business School Software Learning LinkedIn Learning – Has a number of online courses including many for Excel and R. – Free to you: https://www.myit.unsw.edu.au/services/staff/educational- technology/linkedin-learning YouTube.com – Has plenty of tutorials for Excel and R as well, at various levels. – 1) use them as a systematic learning tool (e.g., an R course with a series of organized tutorials) – 2) search for a question you have with Excel or R. Many times, a solution to your question is just there, waiting for you to discover! Dr. SunAh Kim UNSW Business School Other Resources Marketing research handbook – Marketing Research: An Applied Orientation (6th Edition) by Malhotra. A global edition is available in Australia. Published in 2010, by Pearson Education, Inc. – This book can be used as a handbook of marketing research designs and classical analytical tools. Dr. SunAh Kim UNSW Business School Other Resources – Link on Moodle Implementing analytics in marketing strategies – Marketing Strategy by Robert W. Palmatier and Shrihari Sridhar. Published in 2017, by Palgrave. – This book provides a good sense as how analytics are utilised in marketing strategies. Multivariate statistics – Multivariate Data Analysis (7th Edition) by Hair et al. Published in 2010, by Pearson Education, Inc. – This book can provide you more details on multivariate statistics. Dr. SunAh Kim UNSW Business School Other Resources – Link on Moodle Big data analytics – Big Data: A Revolution that Will Transform How We Live, Work, and Think, by Viktor Mayer-Schönberger and Kenneth Cukier. Published in 2013, by Eamon Dolan / Houghton Mifflin Harcourt. This book is a good source to get a first understanding of big data. – Big Data in Practice, by Bernard Marr. Published in 2016, by John Wiley and Sons Ltd. This book provides 45 successful examples of companies using big data analytics to achieve extraordinary success. It is a good source to get a sense of how big data is used in business practice nowadays. Dr. SunAh Kim UNSW Business School Other Resources Big data analytics, e.g., – Machine Learning with R (by Brett Lantz) – Mastering Predictive Analytics with R (by Rui Miguel Forte) – Mastering Social Media Mining with R (by Sharan Kumar Ravindran and Vikram Garg) – These books are examples of technical books on how to use R to conduct big data analyses. They are for more advanced users. Dr. SunAh Kim UNSW Business School Other Resources Excel resources – http://www.real-statistics.com/. This website has rich Microsoft Excel resources, including Excel add-in software for statistical analyses, statistics instructions, examples, and discussion forums. – Marketing Analytics: Data-Driven Techniques with Microsoft Excel by Wayne L. Winston. Published in 2014, by John Wiley & Sons, Inc. An excellent resource that covers many analytical tools in marketing analytics, using Excel. You may use this book as a handbook and find out the solutions that you face (which may or may not be covered in this course). Dr. SunAh Kim UNSW Business School Other Resources R resources – The R book, by Michael J. Crawley. Published in 2012, by Wiley. One of the best-selling statistics book and R book. A very good introduction and handbook of R. – R for Marketing Research and Analytics, by Chris Chapman and Elea McDonnell Feit. Published in 2015 by Springer. This book shows you how to use R to address many analytical needs in marketing. Dr. SunAh Kim UNSW Business School Other Resources For those who would like to further develop R skills and become advanced users: – A list of R resources: https://www.pauljhurtado.com/R/ (you can also find other similar summaries online) – Many books on specific topics using R, e.g., R Data Visualization Cookbook (by Atmajitsinh Gohil), Machine Learning with R (by Brett Lantz), Mastering Predictive Analytics with R (by Rui Miguel Forte), Mastering Social Media Mining with R (by Sharan Kumar Ravindran and Vikram Garg), search www.amazon.com.au Dr. SunAh Kim UNSW Business School Other Resources Essay writing guide – Q Manual (link in the outline, or search “Q Manual” on Google) – A good guide for your essay writing. It also provides a referencing style guide. From time to time, you will be asked to do some additional readings. – In those cases, the reading materials will be made available on the course website on Moodle. – Readings marked as “optional” are not required for assessment Dr. SunAh Kim UNSW Business School PASS class – Strongly recommend this! Peer Assisted Study Sessions (PASS) are free, weekly, optional study sessions offered by the Business School. PASS sessions are facilitated by a student leader who has successfully completed the course in a previous term. You can come to: – Ask questions about specific problems or concepts that you encountered in tutorials and lectures – Work on a variety of problems with friendly and experienced leaders – Discuss general areas of concern for first year students, such as how to prepare for exams and manage time – Meet other students in an informal atmosphere Keep in mind that you can attend any class you like! Will update you on the details when the information is available shortly via Moodle. Dr. SunAh Kim UNSW Business School Assessment Items Dr. SunAh Kim UNSW Business School Assessment Details Dr. SunAh Kim UNSW Business School Team Project (30% in all; 25% team and 5% individual) An opportunity to take your knowledge of the techniques in the course and apply them to a real situation. FGP Customer Loyalty Program – Real business context – The nature of the program and merchant members has been disguised and the dataset has been re-generated to remove any private or sensitive information. – Brief and dataset available on Moodle Dr. SunAh Kim UNSW Business School Groups Forming Up to 5 people Come up with different roles and assign tasks to each role Complete the “Team Roles Assignment” sheet Send a copy to your tutor by your Week 3 tutorial Dr. SunAh Kim UNSW Business School Group Role Expectation All team members share the team responsibility and get credit from the team outcome. Free riders are penalized via peer evaluation. It is an analytical course. Everyone in the team is expected to go through all the analyses in the team project. (Different member may write up different sections.) – A good, integrated work usually shows a sign that analyses are built on each other, rather than several separated pieces. Dr. SunAh Kim UNSW Business School Research Plan + Report (19%) Two steps: research plan and final report You need to start to plan your project early. Each time after learning a new analytical tool in this course, you should revisit this plan and ask yourself: – can I use the new tool to solve the research questions (if you didn’t know how to solve them yet)? – or can I provide a better solution (if you already had a possible solution)? Dr. SunAh Kim UNSW Business School Research plan (not graded; feedback purpose) Steps: – Define one managerial problem that you plan to address in this project and the research questions associated with the problem – Specify information needed to answer the research questions, and – Provide descriptive analyses of the dataset – Propose expected outcomes from the analyses – In MS PowerPoint format for presentation – Submit slides on Moodle by 16:00 on 10 Oct (Week 5 Tue) – Present up to 10 minutes in Week 5 tutorial (5 min discussion) – Instructions available on Moodle Dr. SunAh Kim UNSW Business School Report (19%) Step 2: Final report – Analyse the data – Provide insights to managers – Write a concise and insightful report Due 16:00 on 14 Nov (Week 10 Mon) – Instructions available on Moodle – Read the instructions carefully for requirements and assessment criteria – In MS Word format for comments Dr. SunAh Kim UNSW Business School Presentation (10%) Present your findings during the tutorials in Week 10 – Up to 15 minutes – 5% individual presentation skills (AoL) – 5% overall team performance Slides due 16:00 on 14 Nov (Week 10 Tue) – A soft copy of your presentation slides. No slides accepted during presentation. – In MS PowerPoint format – Instructions available on Moodle Dr. SunAh Kim UNSW Business School Peer Evaluation of Teamwork (1%) Week 5: informal peer evaluation – Due by 16:00 on 11 Oct (Week 5 Fri) Week 10: formal peer evaluation – Due by 16:00 on 18 Nov (Week 11 Mon) Each student’s contribution score will be the average of the points received from their group members. – Adjustments to individual marks will occur where an individual student’s peer evaluation score falls below an acceptable level. Dr. SunAh Kim UNSW Business School Peer Evaluation Adjustment Policy (SoM) Marks will be adjusted where an individual student’s peer evaluation score is below an acceptable standard An average score received Downward (scale 1-5) adjustment Score = 1 50% 1 < Score < 2 20% 2 Regression. Similar.) Dr. SunAh Kim UNSW Business School Conduct Regression in R Model results all stored in a new data list called “fit” (you can have your own name) Linear model Equation to study Dataset used Summarize for the model model results Dr. SunAh Kim UNSW Business School Regression Result Dr. SunAh Kim UNSW Business School Regression Results Reading Regression worthwhile (ANOVA test) – Significance of the entire regression procedure Model fit: – R square: Percentage of the variance (changes) in the Y variable explained by the regression equation – Adjusted R square: R square adjusted to penalize the number of X variables (used to compare different models) Variable coefficient: effect size & significance (t test) – Impactful? Significant contribution? Dr. SunAh Kim UNSW Business School Regression Result Relationship: 𝑊𝑇𝑃 = 5.825 + 10.914×𝑊𝑒𝑖𝑔ℎ𝑡 + 𝜀 Dr. SunAh Kim UNSW Business School Regression Result (R) Dr. SunAh Kim UNSW Business School Back to Manager’s Question 𝑊𝑇𝑃 = 5.825 + 10.914×𝑊𝑒𝑖𝑔ℎ𝑡 + 𝜀 Holding everything else the same, if we pack coffee beans in a 0.5kg package, how much would consumers would like to pay? (Similarly, for 1kg and 2kg) How much do consumers value different package size regarding coffee beans? $10.91 for an additional kg of coffee bean Dr. SunAh Kim UNSW Business School Do the Same on Brand How much do consumers value our brand when buying coffee beans? Try it yourself as a homework Compare your results to mine (next slide) Dr. SunAh Kim UNSW Business School Value of the brand Anything wrong with this regression? 𝑊𝑇𝑃 = 17.496 + 2.592×𝐵𝑟𝑎𝑛𝑑 + 𝜀 Dr. SunAh Kim UNSW Business School Revisit Insights Needed Holding everything else the same, if we pack coffee beans in a 0.5kg package, how much would consumers would like to pay? (Similarly, for 1kg and 2kg, the other two typical package sizes.) – OR 𝑊𝑖𝑙𝑙𝑖𝑛𝑔𝑛𝑒𝑠𝑠 𝑡𝑜 𝑃𝑎𝑦 ~ 𝛽×𝑃𝑎𝑐𝑘𝑎𝑔𝑒 𝑠𝑖𝑧𝑒, controlling possible effect from other variables. Holding everything else the same, if the coffee beans are Vittoria’s product, how much more (or less) would consumers like to pay comparing to other brands? – OR 𝑊𝑖𝑙𝑙𝑖𝑛𝑔𝑛𝑒𝑠𝑠 𝑡𝑜 𝑃𝑎𝑦 ~ 𝛽×𝑉𝑖𝑡𝑡𝑜𝑟𝑖𝑎 𝑐𝑜𝑚𝑝𝑎𝑟𝑒𝑑 𝑡𝑜 𝑂𝑡ℎ𝑒𝑟, controlling possible effect from other variables. Dr. SunAh Kim UNSW Business School Multiple Regression 𝑌 = 𝛽! + 𝛽"𝑋" + 𝛽#𝑋# + ⋯ + 𝛽$ 𝑋$ + 𝜀 Find the best description of the relationship (same) – Minimises the sum of the squared errors over all the observations (same) In other words – Observations described by the relationship above, where 𝜀 is the error in the description. (same) – Pick a set of coefficients: 𝛽!, 𝛽", 𝛽#, … , 𝛽$ , so that the sum of the squared errors ∑ 𝜀 # is minimized. (a bit different) Dr. SunAh Kim UNSW Business School WTP ~ Weight + Brand 𝑊𝑇𝑃 = 3.713 + 11.048×𝑊𝑒𝑖𝑔ℎ𝑡 + 3.293×𝐵𝑟𝑎𝑛𝑑 + 𝜀 Dr. SunAh Kim UNSW Business School Multicollinearity Important assumption in multiple regression: – Independent variables (X’s) are uncorrelated – If this assumption is not obeyed, you have the problem of Multicollinearity Regression is pretty robust (i.e., it works fine when we violate some of the assumptions), but multicollinearity is an assumption you don’t want to violate… Dr. SunAh Kim UNSW Business School Multicollinearity How multicollinearity damage your regression? Consider the following example. – We know a relationship 𝑌 = 𝛽! + 𝛽𝑋 + 𝜀 – Now let us create: 𝑋" = 𝑋# = 𝑋 – Do we have a definitely relationship 𝑌~𝑋" + 𝑋#? Possible relationships: 𝛽 𝛽 – 𝑌 = 𝛽! + 𝛽𝑋" + 0𝑋# + 𝜀 𝑌 = 𝛽! + 𝑋" + 𝑋# + 𝜀 2 2 – 𝑌 = 𝛽! + 0𝑋" + 𝛽𝑋# + 𝜀 𝑌 = 𝛽! + 2𝛽𝑋" − 𝛽𝑋# + 𝜀 – …… – The model is not identifiable Dr. SunAh Kim UNSW Business School Multicollinearity Results of multicollinearity – Affects the significance of the coefficients – Reduces the efficiency of the estimates (identification problem) – Creates problems interpreting the coefficients Deal with multicollinearity – Check correlation before going to regression: might be a concern if correlation coefficient 𝑟 > 0.7 or 𝑟 < −0.7 – VIF test afterwards (rule of thumb): High multicollinearity if VIF(𝛽$ )>10 Dr. SunAh Kim UNSW Business School Compare Model Results Models Adj. R2 𝑊𝑇𝑃 = 5.825 + 10.914×𝑊𝑒𝑖𝑔ℎ𝑡 +𝜀 0.783 𝑊𝑇𝑃 = 17.496 + 2.592×𝐵𝑟𝑎𝑛𝑑 + 𝜀 0.026 𝑊𝑇𝑃 = 3.713 + 11.048×𝑊𝑒𝑖𝑔ℎ𝑡 + 3.293×𝐵𝑟𝑎𝑛𝑑 + 𝜀 0.828 The WTP model fits better by incorporating both weight and brand We have more confidence in the coefficients we get for both weight and brand Question. Can we still improve this model? Dr. SunAh Kim UNSW Business School Anyway to Improve? Does a consumer pay more for Vittoria just because of the label on the package? What does a brand name mean? – Better quality, better service – More socially recognised – A company that has social responsibility – …… How to show people buy Vittoria because their coffee bean is better? Dr. SunAh Kim UNSW Business School WTP ~ Weight + Brand + Weight * Brand Dr. SunAh Kim UNSW Business School Compare Model Results Models Adj. R2 𝑊𝑇𝑃 = 3.713 + 11.048×𝑊𝑒𝑖𝑔ℎ𝑡 + 3.293×𝐵𝑟𝑎𝑛𝑑 + 𝜀 0.828 𝑊𝑇𝑃 = 4.682 + 10.271×𝑊𝑒𝑖𝑔ℎ𝑡 + 1.650×𝐵𝑟𝑎𝑛𝑑 + 0.831 1.346×𝑊𝑒𝑖𝑔ℎ𝑡×𝐵𝑟𝑎𝑛𝑑 + 𝜀 What does the interaction term mean? – From weight perspective: if other brand (Brand=0), the effect of weight is 10.271, smaller than 11.048; if Vittoria (Brand=1), the effect of weight is 11.617 (10.271+1.346), larger than 11.048 – From brand perspective: the effect of brand Vittoria increases by weight. It is 2.996 (1.650+ 1.346) for 1kg package, smaller than 3.293; it is 4.342 (1.650+ 1.346*2) for 2kg package, larger than 3.293 Dr. SunAh Kim UNSW Business School Do We Want Many Interaction Terms? How many interaction terms if 3 IVs? 𝑌 = 𝛽! + 𝛽" 𝑋" + 𝛽# 𝑋# + 𝛽$ 𝑋$ +𝛾" 𝑋" 𝑋# + 𝛾# 𝑋" 𝑋$ + 𝛾$ 𝑋# 𝑋$ + 𝛾% 𝑋" 𝑋# 𝑋$ + 𝜀 More than 3 IVs? Shall we throw all the interaction terms in the regression? – N of interaction terms grows fast as we have more IVs – 3-way interaction is already very hard to interpret – Interactions increase multicollinearity concerns – Add interaction term if there is a good reason (e.g., theoretically there should be interaction, or the empirical results suggest potential interaction) Dr. SunAh Kim UNSW Business School Help Managers with Decision Making Result: 𝑊𝑇𝑃 = 4.682 + 10.271×𝑊𝑒𝑖𝑔ℎ𝑡 + 1.650× 𝐵𝑟𝑎𝑛𝑑 + 1.346×𝑊𝑒𝑖𝑔ℎ𝑡×𝐵𝑟𝑎𝑛𝑑 + 𝜀 WTP in different scenarios Weight Vittoria Other 0.5kg $12.14 $9.82 1kg $17.95 $14.95 1.5kg $23.76 $20.09 2kg $29.57 $25.22 How about 0.25kg or 2.5kg? Dr. SunAh Kim UNSW Business School Extrapolation Be extra careful when extrapolate out of the zone covered by the existing data. Weight WTP For example, in a simple relationship: 0kg $5.83?? 𝑊𝑇𝑃 = 5.825 + 10.914×𝑊𝑒𝑖𝑔ℎ𝑡 + 𝜀 0.25kg $8.55 0.5kg $11.28 – A linear relationship seems to make sense between 0.5kg and 2kg (existing data) 1kg $16.74 2kg $27.65 – But it is not necessarily linear outside this 2.5kg $33.11 zone, particularly when weight < 0.5kg Dr. SunAh Kim UNSW Business School Extrapolation Example Dr. SunAh Kim UNSW Business School Conducting Regression Analysis Theory / belief Choose variables Run regression Significance of overall procedure Rethink Overall fit of equation variables Results meaningful? New ideas? Use equation Dr. SunAh Kim UNSW Business School Issues in Regression: Correlation vs Causality Ice cream causes polio? In 1950s, people believed so (https://theglyptodon.wordpress.com/2012/08/21/polio- caused-by-ice-cream/) – “Polio ~ Ice cream” – Actually, “polio ~ summer” and “ice cream ~ summer” TED: The danger of mixing up causality and correlation: http://www.youtube.com/watch?v=8B271L3NtAw Regression is about correlation, NOT causality – Theory and caution are needed to interpret regression results Dr. SunAh Kim UNSW Business School Issues in Regression: Insignificant Coefficients Remove insignificant coefficients and only keep the significant coefficients? Reasons to remove insignificant coefficients – Trying variables – If not sure the impact of an X is non-zero, one should not use X to estimate/predict Y (Avoid spurious correlations) Re-run regression after reducing coefficients, to get the final results Dr. SunAh Kim UNSW Business School Issues in Regression: Insignificant Coefficients Remove insignificant coefficients and only keep the significant coefficients? Reasons not to remove insignificant coefficients – Variable selection should be based on theory, not fishing – What is the magic of 0? Wouldn’t an insignificant 100 more likely be 100 than be 0? If it is indeed not different from 0, why would it hurt to include it? – If insignificant due to large variance, one should investigate the model approaches rather than throwing it away. Dr. SunAh Kim UNSW Business School Issues in Regression: Data Check Multicollinearity – Correlation between independent variables always exist – Check correlation before going to a regression – VIF measure, in this course, we use the following criterion: High multicollinearity if VIF(𝛽_𝑘)>10 Outlier – Understand why it is there – If caused by error, throw it away – If the data is correct, re-think the model Dr. SunAh Kim UNSW Business School Other Data re Consumers’ Preference/Value DVs IVs Behavioral data Individual/social characteristics / – Sales amount lifestyle – Choices – Gender, age, income – Number of purchase/visits – Marital status, interests in sth. – … – “Green is important to me” – … Intentional data – Willingness to pay Opinions – Likelihood to purchase/visit – Satisfaction – Likelihood to recommend – “Coffee is important to me” – … – … Dr. SunAh Kim UNSW Business School Dealing with Non-metric DV Example: choose to buy/consider or not Binary logistic regression: http://www.real-statistics.com/logistic-regression/ Example: choose one from a set of options, or rank order a set of options Multinomial and ordinal logistic regression [optional] http://www.real-statistics.com/multinomial-ordinal- logistic-regression/ Dr. SunAh Kim UNSW Business School Binary Logistic Regression Use the recorded video to go through binary logistic regression Binary logistic regression: replacing the dummy DV with a formula of probability, the IVs are the same as linear regression 𝑃 log & = 𝛽! + 𝛽" 𝑋" + 𝛽# 𝑋# + ⋯ + 𝛽' 𝑋' + 𝜀 1−𝑃 Dr. SunAh Kim UNSW Business School Binary Logistic Regression Prediction using binary logistic results 𝑒! 𝑃()! = 𝑒 ! + 𝑒*!+*","+*#,#+⋯+*$,$ 1 = 1 + 𝑒*!+*","+*#,#+⋯+*$,$ 𝑒 *!+*","+*#,#+⋯+*$,$ 𝑃()" = * +* , +* , +⋯+* , = 1 − 𝑃()! 1+𝑒 ! " " # # $ $ Dr. SunAh Kim UNSW Business School Example: Buy Coffee Beans or Not Question: if Vittoria launches a 1kg package coffee beans, priced at $14, how many customers will choose to buy it? Data – ID: Respondent’s ID number (1~500) – Weight: Weight of coffee bean package (0.5kg, 1kg, or 2kg) – Brand: Brand of the coffee bean (Vittoria or Other ) – Price: Price of the coffee bean (in $) – Choice: Previous purchase decisions (buy or not ) Dr. SunAh Kim UNSW Business School Binary Logistic using Real Stats Dr. SunAh Kim UNSW Business School Regression Results Goodness of fit: McFadden’s R2 (Pseudo R2) 𝑳𝑳𝟏 Log-likelihood of proposed model 𝑹𝟐𝑳 =𝟏− 𝑳𝑳𝟎 Log-likelihood of constant-only model The higher, the better Dr. SunAh Kim UNSW Business School Regression Results Predictive accuracy: 91.4% Model coefficients: Dr. SunAh Kim UNSW Business School Binary Logistic using R Model results all stored in a new data list called “fit” (you can have your own name) Generalized linear model Equation to study Summarize Dataset used Indicate this is a model results for the model binary logistic model Dr. SunAh Kim UNSW Business School Binary Logistic using R Use fitted function to calculate probability estimation Use the estimated probability to predict choices Calculate the success rate of prediction Get McFadden’s R2 Package “pscl” is used Dr. SunAh Kim UNSW Business School Regression Results Using R Dr. SunAh Kim UNSW Business School Prediction using Regression Results Probability of a customer not to buy given the brand is Vittoria, the package is 1kg, and the price is $14: 1 𝑃23()! = 1 + 𝑒 !.$55+"!.66×8&9:;9E& – P(not buy) = 0.377 or 37.7% – P(buy) = 1- 0.377 = 0.623 or 62.3% In other words, assuming homogeneous group, 62.3% of the customers will buy this product. Dr. SunAh Kim UNSW Business School Predict customers’ preference and choices Dr. SunAh Kim UNSW Business School Lastly Dr. SunAh Kim UNSW Business School Homework Exercise the examples in the lecture and tutorial this week Prepare for the next lecture and tutorial Team project – Complete group forming, hand in role assignment sheet by week 3 tutorial – Understand your data, applying descriptive statistics and regressions (using the full dataset) – Think about the research plan Dr. SunAh Kim UNSW Business School Next Week Lecture topic: – Use regression in Conjoint Analysis (CA) Tutorial topic: – Introduce Excel and R functions – Exercise regression and find out consumer preferences Dr. SunAh Kim UNSW Business School Marketing Analytics and Big Data (MARK3054) Topic 3: Conjoint Analysis Dr. SunAh Kim Senior Lecturer in Marketing, UNSW Business School [email protected] Dr. SunAh Kim UNSW Business School Design Products Based on Consumer Preferences Dr. SunAh Kim UNSW Business School Value of Good Design 80% of a product’s manufacturing costs are determined during the first 20% of its design (varies across product categories). Source: Mckinsey & Company Report Dr. SunAh Kim UNSW Business School Value of Product Superiority 203 products in B2B, how successful they are? – (1) whether it met or exceeded management’s criteria for success, – (2) the profitability level, – (3) market share at the end of three years, and – (4) whether it met company sales and profit objectives. Which is a more powerful success driver? – Market attractiveness, or – Product superiority Robert G. Cooper, Winning at New Products (1993) Dr. SunAh Kim UNSW Business School Impact of Market Attractiveness on Product Success Success rate (%) 100 Mkt Share Mkt Share 36.5% 80 33.7% Mkt Share 60 31.7% 73.9 61.5 40 42.5 20 0 Low Moderate High Market Attractiveness Dr. SunAh Kim UNSW Business School Impact of Product Superiority on Product Success Mkt Share 53.5% Success rate (%) 100 98 80 Mkt Share 32.4% 60 58 40 Mkt Share 11.6% 20 18.4 0 Minimal Moderate Maximal Product Superiority Dr. SunAh Kim UNSW Business School New Product Development Process The engineers design a new product The marketers sell the new product Dr. SunAh Kim UNSW Business School Conjoint Analysis Helps Make New Offerings “More” Successful With a conjoint analysis, marketers can design and develop new products by thinking of products as bundles of attributes, then determining which combination of attributes is best suited to meet the preferences of customers. When to use it? – To identify product attribute trade-offs that customers are willing to make for a new product. – To predict the market share and impact of a proposed new product (i.e., bundle of attributes), proposed new product display, effects of logo etc. – To determine the amount that customers are willing to pay for a new product Dr. SunAh Kim UNSW Business School 88 Frozen Pizza Design Pizza features Type of crust (3 types) Topping (4 varieties) Type of cheese (3 types) Amount of cheese (2 levels) Price (3 levels) Feature levels Crust Topping Type of cheese Pan Pineapple Old English Thin Veggie Natural Swiss Thick Sausage Mozzarella Pepperoni Amount of cheese Price A little (50g). $9.99 A total of 216 (3x4x3x2x3) A lot (150g). $8.99 different pizzas can be $7.99 developed from these options Dr. SunAh Kim UNSW Business School Before we start, make a choice Dr. SunAh Kim UNSW Business School Make Your Choice Record your choice in sheet “My Conjoint” in “Pizza Conjoint Study.xlsx”. Choose 1 pizza that you like the most out of four. Dr. SunAh Kim UNSW Business School Decision Rules Compensatory – E.g., I will pay $9.99 for a sausage or pepperoni pizza, but $8.99 or lower for a pineapple or veggie pizza – Trade-off: a situational decision that involves diminishing or losing one quality, quantity or property of an option in return for gains in other aspects. E.g., here is a sausage for $9.99, and a pineapple for $8.99, which one shall I buy? Non-compensatory – Conjunctive: I must have thin crust and a lot of cheese. – Disjunctive: I can have either sausage topping or thin crust. – Lexicographic: I will first consider topping, it must be veggie or pineapple; then the type of cheese, … Dr. SunAh Kim UNSW Business School Direct Questions Ask respondents’ opinions about the features and feature levels, and how they would make purchase decisions regarding frozen pizza. Pros and Cons? – Easy to conduct – Explore unknown decision rules – Not systematic – Hard to accommodate tradeoff (a huge number of possible tradeoffs) – Hard to use the results Dr. SunAh Kim UNSW Business School Conjoint Analysis Participants make ratings/rankings/choices in a given set of products We estimate their value (utility) for each product feature and feature levels [Optional] A Conjoint research tool: www.conjoint.online – Helps with Conjoint design, data collection, analysis and report. – You may try for free. Dr. SunAh Kim UNSW Business School Dr. SunAh Kim UNSW Business School DV IV5 PRICE Independent Variables IV4: Amount of chees IV3: Type of chees IV2: Topping IV1: CRUST UNSW Business School Dr. SunAh Kim Complete Your Conjoint Ratings Record your ratings in sheet “My Conjoint” in “Pizza Conjoint Study.xlsx”. Dr. SunAh Kim UNSW Business School Conjoint Analysis Pros and Cons? – Systematic – Natural decision tasks – Various decision rules embedded in the observed decisions (e.g., ratings/rankings/choices) – Easy to use the results – Maybe not easy to analyze Dr. SunAh Kim UNSW Business School Conjoint Estimation Using Regression Dr. SunAh Kim UNSW Business School Conjoint Utility Computations % '! 𝑈 𝑃 = 𝛽! + & & 𝛽"& 𝑥"& + 𝜀 "#$ &#$ 𝑃: a particular product/concept of interest 𝑈 𝑃: the utility associated with product 𝑃 𝑚: number of features 𝑘! : number of levels on the 𝑖th feature (𝑖 = 1,2, … , 𝑚) 𝑥!" :the configuration of the 𝑗th level of the 𝑖th feature of product 𝑃 𝛽!" : utility associated with the 𝑗th level (𝑗 = 1,2, … , 𝑘! ) on the 𝑖th feature (partworth). Dr. SunAh Kim UNSW Business School Conjoint Utility Computations % '! 𝑈 𝑃 = 𝛽! + & & 𝛽"& 𝑥"& + 𝜀 "#$ &#$ = 𝛽! + 𝛽$$ 𝑥$$ +𝛽$( 𝑥$( + ⋯ + 𝛽$'" 𝑥$'" +𝛽($ 𝑥($ +𝛽(( 𝑥(( + ⋯ + 𝛽('# 𝑥('# … … +𝛽%$ 𝑥%$ +𝛽%( 𝑥%( + ⋯ + 𝛽%'$ 𝑥%'$ + 𝜀 Dr. SunAh Kim UNSW Business School Re Feature Levels Pizza features Type of crust (3 types) Topping (4 varieties) Type of cheese (3 types) Amount of cheese (2 levels) Price (3 levels) Feature levels Crust Topping Type of cheese Pan Pineapple Old English Thin Veggie Natural Swiss Thick Sausage Mozzarella Pepperoni Amount of cheese Price A little (50g). $9.99 A lot (150g). $8.99 $7.99 Dr. SunAh Kim UNSW Business School Dummy Variables Set “a little amount of cheese (50g)” as the base Amt_Lot = 1 if a product has a lot of cheese, otherwise 0 Amt_Lot A little cheese 0 A lot of cheese 1 Dr. SunAh Kim UNSW Business School Dummy Variables Set Pan Crust as the base Crust_Thin = 1 if a product has thin crust, otherwise 0 Crust_Thick = 1 if a product has thick crust, otherwise 0 Crust_Thin Crust_Thick Pan crust 0 0 Thin crust 1 0 Thick crust 0 1 Dr. SunAh Kim UNSW Business School Dummy Variables Set Pineapple as the base Top_Veg = 1 if a product has veggie topping, otherwise 0 Top_Saus = 1 if a product has sausage topping, otherwise 0 Top_Pep = 1 if a product has Pepperoni topping, otherwise 0 Top_Veg Top_Saus Top_Pep Pineapple 0 0 0 Veggie 1 0 0 Sausage 0 1 0 Pepperoni 0 0 1 Dr. SunAh Kim UNSW Business School Dummy Coding in Excel Dr. SunAh Kim UNSW Business School Dummy coding exercise An example of filling in the model with one data record The last level in each feature in the column is the baseline feature Crust Topping Type of Cheese Amount Price DV thin thick pan veg saus pepp pin swiss mozz Old 150 50 8.99 7.99 9.99 eng Dr. SunAh Kim UNSW Business School Dummy coding exercise An example of filling in the model with one data record Crust Topping Type of Cheese Amount Price DV thin thick pan veg saus pepp pin swiss mozz eng 150 50 8.99 7.99 9.99 0 0 0 Dr. SunAh Kim UNSW Business School Dummy coding exercise An example of filling in the model with one data record Crust Topping Type of Cheese Amount Price DV thin thick pan veg saus pepp pin swiss mozz eng 150 50 8.99 7.99 9.99 0 0 0 0 1 0 0 Dr. SunAh Kim UNSW Business School Dummy coding exercise An example of filling in the model with one data record Crust Topping Type of Cheese Amount Price DV thin thick pan veg saus pepp pin swiss mozz eng 150 50 8.99 7.99 9.99 0 0 0 0 1 0 0 0 1 0 Dr. SunAh Kim UNSW Business School Dummy coding exercise An example of filling in the model with one data record Crust Topping Type of Cheese Amount Price DV thin thick pan veg saus pepp pin swiss mozz eng 150 50 8.99 7.99 9.99 0 0 0 0 1 0 0 0 1 0 1 0 Dr. SunAh Kim UNSW Business School Dummy coding exercise and model 𝑈 𝑃 = 𝑎# + 𝑎$%&!'×𝐶𝑟𝑢𝑠𝑡_𝑇ℎ𝑖𝑛 + 𝑎$%&!() ×𝐶𝑟𝑢𝑠𝑡_𝑇ℎ𝑖𝑐𝑘 + 𝑎*+,- ×𝑇𝑜𝑝_𝑉𝑒𝑔 + 𝑎*./01 ×𝑇𝑜𝑝_𝑆𝑎𝑢𝑠 + 𝑎*2,3 ×𝑇𝑜𝑝_𝑃𝑒𝑝 + 𝑎4.5!11 ×𝐶ℎ𝑠_𝑆𝑤𝑖𝑠𝑠 + 𝑎46788 ×𝐶ℎ𝑠_𝑀𝑜𝑧𝑧 + 𝑎9:7; ×𝐴𝑚𝑡_𝐿𝑜𝑡 + 𝑎??×𝑃𝑟𝑖𝑐𝑒_899 + 𝑎> ×0 + 𝛽;_?>> ×1 + 𝜀 Dr. SunAh Kim UNSW Business School Dr. SunAh Kim UNSW Business School Partworth Which crust is this participant’s favorite? Which cheese is this participant’s favorite? Which feature is the most important to this participant? – The range of a feature – Do NOT forget the baseline feature level! Dr. SunAh Kim UNSW Business School Partworth Range: 91.250 Range: 0 Range: 16.250 Range: 5.625 Range: 6.875 Which crust is this participant’s favorite? Which cheese is this participant’s favorite? Which feature is the most important to this participant? – The range of a feature – Do NOT forget the baseline feature level! Dr. SunAh Kim UNSW Business School Apply Partworth in Prediction Dr. SunAh Kim UNSW Business School Partworth Estimation Homework: – Try to estimate for other participants in the Conjoint data set – Try to estimate your own rating and check if the partworth matches your preferences – Try choice prediction using the estimated partworth in the Conjoint dataset Dr. SunAh Kim UNSW Business School Discussion In the dataset, the predictions for R18 and Z77 were wrong. How did it get wrong? How to model possible combinations such as sausage topping goes best with Swiss cheese? How to model continuous variables, e.g., continuous price? Dr. SunAh Kim UNSW Business School Conjoint Outputs The basic outputs of conjoint analysis are: – A numerical assessment of the relative importance that customers attach to features of a product category – Estimation of the value (utility) provided to customers by a certain offering (with specific features and feature levels) – Identification of product designs that maximize market share or other indices (e.g., maximize profit), by conducting simulations. Dr. SunAh Kim UNSW Business School [Optional] Choice based Conjoint Dr. SunAh Kim UNSW Business School [Optional] Choice based Conjoint With no buy option (allow you to estimate WTP) Dr. SunAh Kim UNSW Business School [Optional] Choice based Conjoint Partworth estimation: logistic regression Multinomial logistic regression: http://www.real- statistics.com/multinomial-ordinal-logistic-regression/ Dr. SunAh Kim UNSW Business School Lastly Dr. SunAh Kim UNSW Business School Homework Exercise the examples in the lecture and tutorial this week Review examples and exercise from the next lecture and tutorials Team project – Finalise team organisation and submit team role assignment if you haven’t done so. – Start to prepare for the research plan Dr. SunAh Kim UNSW Business School Next Week Lecture topic: – Customer value assessment Tutorial topic: – Exercise Conjoint Analysis estimation using regression – Exercise customer value calculation Dr. SunAh Kim UNSW Business School Quiz 1 on Week 5 Thursday Open on Week 5 Thursday (10 October), from 00 am to 23:59, Sydney time. One attempt only. 8-10 multiple choice questions, 30mins, 10% course scores Lecture contents from Weeks 1-3 (including the corresponding exercise in Week 4 tutorial), except optional contents. In some questions, you will be asked to use Real-stats add-in or R to perform analysis on the given dataset. Access the quiz on Moodle, in the “Online Quizzes” section Dr. SunAh Kim UNSW Business School If you have not yet don’t, please Review the “binary logistics” regression video from Week 2 moodle Review the conjoint video link in Week 3 Try to replicate the analysis results table that we see from each lecture by yourself after each class and get familiar with reading/interpreting the results! Dr. SunAh Kim UNSW Business School Marketing Analytics and Big Data (MARK3054) Topic 4: Customer Value Assessment Dr. SunAh Kim Senior Lecturer in Marketing, UNSW Business School [email protected] Dr. SunAh Kim UNSW Business School Motivations Dr. SunAh Kim UNSW Business School Value of Customers Transaction view: – It costs us $100 to acquire a customer. – The profit of selling a product is $40. – Deal or no deal? Profit base profit acquisition cost Dr. SunAh Kim UNSW Business School Value of Customers Long-term relationship view: relationship value cost savings price premium demand increase Profit base profit acquisition cost Time Dr. SunAh Kim UNSW Business School Customer Profit Patterns Over Time Profit per Customer (in dollars) Industry by Year of Relationship 1 2 3 4 5 Credit Card Issuance & Servicing -21 42 44 49 55 Industrial Laundry 144 166 192 222 256 Industrial Distribution 45 99 121 141 168 Auto Servicing 25 35 70 88 88 Source: Based on data from Reicheld and Sasser Dr. SunAh Kim UNSW Business School Customer Loyalty and Profits Reducing customer defections by 5% (e.g. from 15% to 10%) means: (Reichheld 1996, HBR) Dr. SunAh Kim UNSW Business School Acquisition, Retention and Growth Dr. SunAh Kim UNSW Business School CLV: Customer Lifetime Value Economic Value: (Discounted) revenue flow less cost-to-serve Total Lifetime Value of Customer Relationship Value: Reference Referral Learning Innovation, etc. Dr. SunAh Kim UNSW Business School Economic CLV ç Increases Expected revenue cash flow (minus) ç Lowers Expected cost to serve cash flow Expected profit cash flow Loyalty ç Lowers Risk adjustment (discount) Discounted cash flow (present value) Dr. SunAh Kim UNSW Business School Simple Idea of Customer Value to Firms Dr. SunAh Kim UNSW Business School Risk Adjustment Adjustment 1: customers may leave (-> retention rate) Adjustment 2: other discounts, e.g., financial interest rate, business risks, political risks (-> discount rate) Discount rate – Major sources: financial interest rate and business risks – Mainly associated with the nature of business, important for the estimation of firm value (in finance) – Mostly as given for the analysis of customers Dr. SunAh Kim UNSW Business School CLV Calculation (if the customer will be with us forever) 𝑪𝑳𝑽 = 𝑹𝟏 − 𝑪𝟏 + 𝑹𝟐 − 𝑪𝟐 ×𝒅 + … + 𝑹𝒏 − 𝑪𝒏 ×𝒅𝒏$𝟏 − 𝑨 𝑹𝒊 : revenue from the customer in period 𝒊 𝑪𝒊 : cost to serve the customer in period 𝒊 𝒅 = 𝟏/ 𝟏 + 𝒓 : discount factor ($1 in the next period is worth $d today), where 𝒓 is the discount rate 𝑨: acquisition cost Dr. SunAh Kim UNSW Business School CLV Calculation Example Period Revenue Cost Discount Discounted Value 1 $25 $5 1 ($25-$5)*1 2 $25 $5 1/(1+15%) ($25-$5)/(1+15%) 3 $25 $5 1/(1+15%)2 ($25-$5)/(1+15%)2 … … … … … i $25 $5 1/(1+15%)i-1 ($25-$5)/(1+15%)i-1 Dr. SunAh Kim UNSW Business School CLV Calculation (considering customer leave) 𝑪𝑳𝑽 = 𝒑𝟏 × 𝑹𝟏 − 𝑪𝟏 + 𝒑𝟐 × 𝑹𝟐 − 𝑪𝟐 ×𝒅 + … + 𝒑𝒏 × 𝑹𝒏 − 𝑪𝒏 ×𝒅𝒏$𝟏 − 𝑨 𝒑𝒊 : likelihood of retaining the customer in period 𝒊 𝑹𝒊 : revenue from the customer in period 𝒊 𝑪𝒊 : cost to serve the customer in period 𝒊 𝒅 = 𝟏/ 𝟏 + 𝒓 : discount factor, where 𝒓 is the discount rate 𝑨: acquisition cost Dr. SunAh Kim UNSW Business School CLV Calculation Example Period Retention Revenue Cost Discount Discounted Value 1 100% $25 $5 1 ($25-$5)*1 2 80% $25 $5 1/(1+15%) 80%*($25-$5)/(1+15%) 3 80%2 $25 $5 1/(1+15%)2 80%2*($25-$5)/(1+15%)2 … … … … … i 80%i-1 $25 $5 1/(1+15%)i-1 80%i-1*($25-$5)/(1+15%)i-1 Dr. SunAh Kim UNSW Business School CLV Calculation 𝑪𝑳𝑽 = 𝒑𝟏 × 𝑹𝟏 − 𝑪𝟏 + 𝒑𝟐 × 𝑹𝟐 − 𝑪𝟐 ×𝒅 + … + 𝒑𝒏 × 𝑹𝒏 − 𝑪𝒏 ×𝒅𝒏$𝟏 − 𝑨 To simplify, let 𝑹𝒊 = 𝑹, 𝑪𝒊 = 𝑪, 𝒑𝒊 = 𝒑𝒊"𝟏 (𝒑𝟏 = 𝟏, 𝒑𝟐 = 𝒑, … 𝒑𝒏 = 𝒑𝒏"𝟏 ) 𝑪𝑳𝑽 = 𝑹 − 𝑪 + 𝒑× 𝑹 − 𝑪 ×𝒅 + … + 𝒑𝒏$𝟏 × 𝑹 − 𝑪 ×𝒅𝒏$𝟏 − 𝑨 Dr. SunAh Kim UNSW Business School CLV Calculation 𝑪𝑳𝑽 = 𝑹 − 𝑪 + 𝒑× 𝑹 − 𝑪 ×𝒅 + … + 𝒑𝒏"𝟏 × 𝑹 − 𝑪 ×𝒅𝒏"𝟏 − 𝑨 = 𝑹 − 𝑪 𝟏 + 𝒑𝒅 + 𝒑𝒅 𝟐 + ⋯ + 𝒑𝒅 𝒏"𝟏 −𝑨 𝟏 − 𝒑𝒅 𝒏 = 𝑹−𝑪 −𝑨 𝟏 − 𝒑𝒅 𝟏 = 𝑹−𝑪 −𝑨 (when 𝒏 → +∞) 𝟏 − 𝒑𝒅 𝟏+𝒓 𝟏 = 𝑹−𝑪 −𝑨 (given 𝒅 = ) 𝟏+𝒓−𝒑 𝟏&𝒓 Dr. SunAh Kim UNSW Business School CLV Calculation (Simplified) 𝟏+𝒓 𝑪𝑳𝑽 = 𝑹−𝑪 −𝑨 𝟏−𝒑+𝒓 𝑹: revenue from the customer in each period C: cost to serve the customer in each period 𝑨: acquisition cost 𝒓: discount rate 𝒑: retention rate (if a customer is active in period 𝒊, the likelihood of the customer being active in period 𝒊 + 𝟏 is 𝒑) Dr. SunAh Kim UNSW Business School CLV Calculation Consider the example: – It costs $15 to acquire a customer – Costs $5 to support and service that customer – The gross revenue stream from the customer is $25 in each period – The discount rate is 15% – The retention rate is 80% Please calculate the CLV of the customer Dr. SunAh Kim UNSW Business School CLV Calculation 𝟏+𝒓 𝑪𝑳𝑽 = 𝑹−𝑪 −𝑨 𝟏−𝒑+𝒓 𝟏 + 𝟏𝟓% = 𝟐𝟓 − 𝟓 − 𝟏𝟓 𝟏 − 𝟖𝟎% + 𝟏𝟓% = 𝟓𝟎. 𝟕𝟏 Dr. SunAh Kim UNSW Business School Excel Model … … … … … … Dr. SunAh Kim UNSW Business School Excel Model Advantages See what you get Simulation and visualization (smart worksheet) Sensitivity analysis (next slide) Accommodate loyalty effects flexibly, e.g., – Growing margins (higher revenue and/or lower cost) – Different discount rate from year 5 – Different retention rate from year 5 Dr. SunAh Kim UNSW Business School Excel Model: Sensitivity Analysis How will the CLV result if the retention rate and the discount rate is different? Dr. SunAh Kim UNSW Business School Individual or Segment Level Estimation Dr. SunAh Kim UNSW Business School So Far… What have we achieved? – Overall estimation of CLV (aggregate level) What info do we use? – Average retention rate – Expected average revenue and cost in the future – Average acquisition cost – Discount rate/factor Dr. SunAh Kim UNSW Business School Individual or Segment Level CLV Estimation Using the same approach, we can estimate the CLV for an individual or a segment, if the following info is known: – Individual / segment retention rate – Expected individual / segment revenue and cost in the future – Individual / segment acquisition cost Dr. SunAh Kim UNSW Business School Issues in Predicting for Each Individual Precise prediction for each individual is difficult We can estimate whether a customer will buy and how much s/he will buy if we understand their preferences, but: – Limited observations for each individual – Limited data for observed behavior, e.g., we observe which product a customer buy, but don’t know what potential options are in the customer’s choice set. – Data from research studies (e.g., Conjoint) are good, but it is not possible to conduct studies on all the potential customers. Dr. SunAh Kim UNSW Business School Individual Level Estimation However, it is usually possible to tell: – The purchase likelihood of a customer like this person – The purchase amount of a customer like this person Assumption: – We can find a set of variables to represent the latent traits that determine a customer’s purchase decisions – It implies that, customers with the same values for this set of variables will behave the same (or at least similarly) Dr. SunAh Kim UNSW Business School Segment Level Estimation When we define a group of similar customers as a segment, we can estimate: – The purchase likelihood of the customers in this segment – The purchase amount of the customers in this segment Assumption: – Customers within a segment are similar and customers from different segments are different, in term of their purchase decisions Dr. SunAh Kim UNSW Business School Useful Variables for Individual Level CLV Est. Assumption: a customer’s historical purchase behavior is a good predictor of their future purchase behavior, and the historical purchase behavior can be summarized as: R Recency Time/purchase occasions since the last purchase Number of purchase occasions since the first F Frequency purchase (in a certain time period) Monetary Amount spent since the first purchase (in a M Value certain time period) Dr. SunAh Kim UNSW Business School Individual Level Estimation Using Regression If we are interested in the purchase amount of a customer: – IVs: recency, frequency and monetary value in year N – DV: purchase amount in year N+1 Linear regression Amount = 𝛽! + 𝛽" recency + 𝛽# frequency + 𝛽$ monetary + 𝜀 After calibration, if we know RFM in year M, we can predict the purchase amount in year M+1 Dr. SunAh Kim UNSW Business School Individual Level Estimation Using Regression If we are interested in the purchase likelihood of a customer: – IVs: recency, frequency and monetary value in year N – DV: buy or not in year N+1 Binary logistic regression 𝐵𝑢𝑦 = 𝛽! + 𝛽" recency + 𝛽# frequency + 𝛽$ monetary + 𝜀 After calibration, if we know RFM in year M, we can predict the purchase likelihood in year M+1 Dr. SunAh Kim UNSW Business School Enhance the Regression By incorporating more info, e.g., 𝐵𝑢𝑦 = 𝛽! + 𝛽" recency + 𝛽# frequency + 𝛽$ monetary +𝛽% gender + 𝛽& age + ⋯ + 𝜀 [Optional] By allowing nonlinear forms, e.g., 𝐵𝑢𝑦 = 𝛽! + 𝛽" recency + 𝛽# frequency +𝛽$ monetary + 𝛽% monetary # + 𝛽& monetary $ + ⋯ + 𝜀 𝐵𝑢𝑦 = 𝛽! + 𝛽" recency + 𝛽# monetary Frequency freq1 freq2 freq3 < 50 0 0 0 +𝛽$ freq1 + 𝛽% freq2 + 𝛽& freq3 50 ~100 1 0 0 +⋯+ 𝜀 101~200 0 1 0 >200 0 0 1 Dr. SunAh Kim UNSW Business School Discussion Dr. SunAh Kim UNSW Business School Discussion Now that we know how to calculate CLV, how does it help us with marketing decisions? Can we use CLV for the team project? (Discussion slides on team project available under week 7 section.) How to get recency, frequency, and monetary value in our dataset? Dr. SunAh Kim UNSW Business School Lastly Dr. SunAh Kim UNSW Business School Homework Exercise the examples in the lecture and tutorial this week Prepare for the next lecture Quiz 1 is scheduled on Thursday (10 Oct) Team project – Prepare for the research plan – Submit slides on Moodle, by 12:00 pm (noon) on 8 Oct (Week 5 Tue) Dr. SunAh Kim UNSW Business School Next Week No live lecture (public holiday) Pre-recorded lecture on Segmentation techniques will be posted online – Cluster analysis Tutorial topic: – Research plan presentation Dr. SunAh Kim UNSW Business School Marketing Analytics and Big Data (MARK3054) Topic 5: Customer Heterogeneity and Segmentation Dr. SunAh Kim Senior Lecturer in Marketing, UNSW Business School [email protected] Dr. SunAh Kim UNSW Business School Motivations Dr. SunAh Kim UNSW Business School Market Segmentation Consumers are different. Market segmentation is the subdividing of a market into distinct subsets of consumers. – Members are different across segments but similar within. Dr. SunAh Kim UNSW Business School STP Framework A decision process – Whose needs are similar within-groups and different between-groups (Segmentation) – Who can be reached profitably (Targeting) – With a focused marketing program (Positioning) Dr. SunAh Kim UNSW Business School The Many Uses of Segmentation Short term applications: Sales force allocation/planning Channel assignment Communication program Pricing Today’s competitors and my current relative advantages or disadvantages Dr. SunAh Kim UNSW Business School The Many Uses of Segmentation Longer term applications : Planning for segment development/growth New and evolving market segments to serve Not in-kind competition/threats (satisfying customer needs in different ways) Dr. SunAh Kim UNSW Business School Needs-Based Segmentation Distinguish Between Bases and Descriptors Bases—characteristics that tell us why segments differ (e.g., needs, preferences, decision processes). Descriptors—characteristics that help us find and reach segments. (Business markets) (Consumer markets) Industry Gender/Age/Income Size Education Location Profession Organizational structure Lifestyles Media habits Dr. SunAh Kim UNSW Business School Segmentation (for Carpet Fibers) Perceptions/Ratings for one respondent: Customer Values Strength (Importance)......A........ D............. Distance between B.. C.. A,B,C,D: segments C and D Location of.............. segment centers. Typical members: A: schools....... B: light commercial C: indoor/outdoor carpeting D: health clubs Water Resistance (Importance) Dr. SunAh Kim UNSW Business School Inter-dependence among Needs When needs are multi-dimensional, such a 2-D map is not enough to study the inter-dependence among the needs Cluster analysis is a widely used tool to model the inter- dependence among needs, e.g., – Hierarchical cluster analysis – K-means cluster analysis and Partitioning Around Medoids (PAM) cluster analysis Dr. SunAh Kim UNSW Business School A New PDA in 2001: ConneCtor Combines the features of – Mobile phone – PDA Dr. SunAh Kim UNSW Business School Telecom Products Dr. SunAh Kim UNSW Business School New PDA: Managers’ Question Managerial problem: – To whom shall we sell this new PDA? Research question – What segments exist in the market? (Features of each segment?) – Who are the consumers in each segment? – Which segment(s) is most attractive? Decision – Select segments and allocate resources. Dr. SunAh Kim UNSW Business School A Research Study 15 bases variables Dr. SunAh Kim UNSW Business School A Research Study 17 descriptors variables Dr. SunAh Kim UNSW Business School Cluster Analysis Cluster analysis is a convenient method commonly used in many disciplines to categorize entities (individuals, objects, and so on) into groups that are homogenous along a range of observed characteristics (variables) In marketing, clustering is used to partition data such that the resultant groups are internally homogenous (cohesive), but externally heterogeneous (separated) Dr. SunAh Kim UNSW Business School A Simple Example of Grouping Dr. SunAh Kim UNSW Business School Text Analytics and Sentiment Mining Using SAS Another Example: How to Group? Dr. SunAh Kim UNSW Business School Text Analytics and Sentiment Mining Using SAS Another Example: Best Grouping? Dr. SunAh Kim UNSW Business School Text Analytics and Sentiment Mining Using SAS Another Example: OR … Dr. SunAh Kim UNSW Business School Text Analytics and Sentiment Mining Using SAS Another Example: OR … Dr. SunAh Kim UNSW Business School Text Analytics and Sentiment Mining Using SAS Conducting Cluster Analysis Formulate the problem Select a distance measure (we use Euclidean distance) Initial understanding of the clusters Decide on the N of clusters Adjust N of clusters if Find out clusters given N necessary Interpret and profile clusters Dr. SunAh Kim UNSW Business School Euclidean Distance (2-dimension) Dr. SunAh Kim UNSW Business School Euclidean Distance (n-dimension) If 𝒑 = 𝑝! , 𝑝" , … , 𝑝# and 𝒒 = 𝑞! , 𝑞" , … , 𝑞# are two points in Euclidean n-space, the distance (𝑑) between 𝒑 and 𝒒 is given by: 𝑑 𝒑, 𝒒 = 𝑑 𝒒, 𝒑 Dr. SunAh Kim UNSW Business School Hierarchical Cluster Analysis A method seeks to build a hierarchy or tree-like structure of clusters. Dr. SunAh Kim UNSW Business School Hierarchical Cluster Analysis 3.. 6 Importance.. of Quality 2 9 5. 4.... 8 1 7 Importance of Price Dr. SunAh Kim UNSW Business School New PDA Hierarchical Cluster Analysis Dr. SunAh Kim UNSW Business School Why Standardization? Consider distance calculation of two variables: – Innovator: rating 1 vs 7 Huge difference in terms of needs for innovation In distance calculation: (7-1)2 = 36 – Price for the device: $201 vs $207 Small difference in terms of WTP for device In distance calculation: (207-201)2 = 36 Standardization: 𝑋 − 𝑚𝑒𝑎𝑛 𝑋 𝑠𝑡𝑑 𝑋 Dr. SunAh Kim UNSW Business School New PDA Dendrogram Dr. SunAh Kim UNSW Business School New PDA Dendrogram (4 Groups) Dr. SunAh Kim UNSW Business School Optimal N of Clusters Many different indices have been developed to suggest how many clusters might be optimal Dr. SunAh Kim UNSW Business School Optimal N of Clusters Dr. SunAh Kim UNSW Business School Optimal N of Clusters Dr. SunAh Kim UNSW Business School Decide N of Clusters Software makes suggestions – Software makes suggestions from statistical view – Many indices may suggest different optimal numbers, each number makes sense from a certain viewpoint You make the decision – Are the cluster profiles meaningful? – Is the cluster division useful for my marketing decisions? E.g., do the profiles help identify my customer segments; are the segments large enough to be useful (check cluster size)? Dr. SunAh Kim UNSW Business School Conducting Cluster Analysis Formulate the problem Select a distance measure (we use Euclidean distance) Initial understanding of the clusters Decide on the N of clusters Adjust N of clusters if Find out clusters given N necessary Interpret and profile clusters Dr. SunAh Kim UNSW Business School K-means Cluster Analysis A method aims to partition 𝒏 observations into 𝒌 clusters in which each observation belongs to the cluster with the nearest mean. It is more precise than hierarchical cluster analysis, because the method allows the re-assignment of cases, to achieve the optimal classification (i.e., similar within groups and different across groups) Dr. SunAh Kim UNSW Business School k-Means Clustering Algorithm Training Data 1. Select inputs. 2. Select k cluster centers. 3. Assign cases to closest center. 4. Update cluster centers. 5. Reassign cases. 6. Repeat steps 4 and 5 until convergence. Dr. SunAh Kim SAS Visual Stats for Professors (Chapter 2) UNSW Business School k-Means Clustering Algorithm Training Data 1. Select inputs. 2. Select k cluster centers. 3. Assign cases to closest center. 4. Update cluster centers. 5. Reassign cases. 6. Repeat steps 4 and 5 until convergence. Dr. SunAh Kim SAS Visual Stats for Professors (Chapter 2) UNSW Business School k-Means Clustering Algorithm Training Data 1. Select inputs. 2. Select k cluster centers. 3. Assign cases to closest center. 4. Update cluster centers. 5. Reassign cases. 6. Repeat steps 4 and 5 until convergence. Dr. SunAh Kim SAS Visual Stats for Professors (Chapter 2) UNSW Business School k-Means Clustering Algorithm Training Data 1. Select inputs. 2. Select k cluster centers. 3. Assign cases to closest center. 4. Update cluster centers. 5. Reassign cases. 6. Repeat steps 4 and 5 until convergence. Dr. SunAh Kim SAS Visual Stats for Professors (Chapter 2) UNSW Business School k-Means Clustering Algorithm Training Data 1. Select inputs. 2. Select k cluster centers. 3. Assign cases to closest center. 4. Update cluster centers. 5. Reassign cases. 6. Repeat steps 4 and 5 until convergence. Dr. SunAh Kim SAS Visual Stats for Professors (Chapter 2) UNSW Business School k-Means Clustering Algorithm Training Data 1. Select inputs.

Marketing Analytics and Big Data (MARK3054) UNSW Business School PDF

Document Details

Tags

Related

Summary

Full Transcript

Upgrade to continue