Marketing data driven techniques

Text Mining 669 Sentiment Analysis of Airline Tweets Jeffrey Breen of Cambridge Aviation Research (see pages 133–150 of Practical Text Mining, by Elder et. al, 2012, Academic Press) collected thousands of tweets that commented on airline service. Breen simply classified a tweet as positive or negative by counting the number of words associated with a positive sentiment and num- ber of words associated with negative sentiments (for a sample list of positive and negative words, see http://www.wjh.harvard.edu/~inquirer/Positiv.html and http://www.wjh.harvard.edu/~inquirer/Negativ.html. A tweet was classified as positive if the following was true: Number of Positive Words – Number of Negative Words >= 2 The tweet was classified as negative if the following was true: Number of Positive Words – Number of Negative Words <= -2 Breen found that that JetBlue (84-percent positive tweets) and Southwest (74- percent positive tweets) performed best. When Breen correlated each airline’s score with the national survey evaluation of airline service conducted by the American Consumer Satisfaction Index (ACSI) he found an amazing 0.90 correlation between the percentage of positive tweets for each airline and the airline’s ACSI score. Because tweets can easily be monitored in real time, an airline can track over time the percent- age of positive tweets and quickly see whether its service quality improves or declines. Using Twitter to Predict Movie Revenues Asu and Huberman of HP Labs used tweets to create accurate predictions of movie revenues in their article, “Predicting the Future with Social Media” submitted in 2010 (see the PDF at www.hpl.hp.com/research/scl/papers/socialmedia/ socialmedia.pdf). For 24 movies the authors used the number of tweets each day in the 7 days before release and the number of theaters in which the movie opened to predict each mov- ie’s opening weekend revenues. They simply ran a multiple regression (see Chapter 10, “Using Multiple Regression to Forecast Sales”) to predict opening weekend revenue from the aforementioned independent variables. Their regression yielded an adjusted R2 value of 97.3 percent. Adjusted R2 adjusts the R2 value discussed in Chapter 10 for the number of independent variables used in the regression. If a

670 Part XI: Internet and Social Marketing relatively unimportant independent variable is added to a regression, then R2 never decreases, but adjusted R2 decreases. Formally adjusted R2 may be computed as: 1 − (1 − R2)*(n − 1) / (n − k − 1), where n = number of observations and k = number of independent variables. The authors compared their predictions to predictions from the Hollywood Stock Exchange (www.hsx.com/). HSX is a prediction market in which buyers and sellers trade based on predictions for a movie’s total and opening weekend revenues. Using the HSX prediction and number of theaters as independent variables, a multiple regression yielded an adjusted R2 value of 96.5 percent of the variation in opening weekend revenues. Therefore, the author’s use of tweets to predict opening weekend movie revenues outperformed the highly regarded HSX predictions. In trying to predict a movie’s revenue during the second weekend, the authors applied sentiment analysis. They classified each tweet as positive, negative, or neu- tral. (See the paper for details of their methodology.) Then they defined the PNratio (Positive to Negative) as Number of Positive Tweets / Number of Negative Tweets. Using the tweet rates for the preceding seven days to predict second weekend revenues yielded an adjusted R2 of 0.84, whereas adding a PNratio as an independent variable increased the adjusted R2 to 0.94. As an example of how a large PNratio reflects favor- able word of mouth, consider the movie The Blind Side, for which Sandra Bullock won a Best Actress Oscar. In the week before its release, this movie had a PNratio of 5.02, but after 1 week the movie’s PNratio soared to 9.65. Amazingly, The Blind Side’s second week revenues of $40 million greatly exceeded the movie’s week 1 revenues of $34 million. Most movies experience a sharp drop-off in revenue dur- ing the second week, so The Blind Side illustrates the powerful effect that favorable word of mouth can have on future movie revenues. Using Twitter to Predict the Stock Market Bollen, Mao, and Zheng of Indiana University and the University of Manchester used the collective “mood” expressed by recent tweets to predict whether the Dow Jones Index would decrease or increase in their article, “Twitter Mood Predicts the stock market,” (Journal of Computational Science, Volume 2, 2011, pp. 1–8). They first used sentiment analysis to classify 9.8 million tweets as expressing a positive or negative mood about the economy. For Day t they defined PNt = ratio of positive to negative tweets and Dt = Dow Jones Index at end of day t − Dow Jones Index at end of day t − 1. The authors tried to predict Dt using PNt-1, PNt-2, PNt-2, Dt-1, Dt-2, and Dt-3. Predictions from a neural network (See Chapter 15, “Using Neural Networks to

Text Mining 671 Forecast Sales”) correctly predicted the direction of change in the Dow 84 percent of the time. This is truly amazing because the widely believed Efficient Market Hypothesis implies that the daily directional movement of a market index cannot be predicted with more than 50-percent accuracy. Using Tweets to Evaluate Super Bowl Ads In 2013 companies paid approximately $3 million dollars for a 30-second Super Bowl ad. Naturally, advertisers want to know if their ads are worthwhile. Professor Piyush Kumar of the University of Georgia analyzed more than one-million tweets that commented on Super Bowl 2011 ads. By performing sentiment analysis on the ads, Kumar found that the 2011 Bud Light ad (www.youtube.com/watch?v=I2AufnGmZ_U) was viewed as hilarious. Unfortunately for car manufacturers, only the Volkswagen ad http://www.youtube.com/watch?v= 9H0xPWAtaa8 received a lot of tweets. The lack of tweets concerning other auto ads indicates that these ads made little impact on Super Bowl viewers. Summary In this chapter you learned the following: ■ To glean impact from the text, the text must be given structure through a vector representation that implements a coding based on the words present in the text. ■ Binary coding simply records whether a word is present in a document. ■ Frequency coding counts the number of times a word is present in a document. ■ The term-frequency/inverse document frequency score adjusts frequency coding of a word to reduce the significance of a word that appears in many documents. ■ After text is coded many techniques, such as Naive Bayes, neural networks, logistic regression, multiple regression, discriminant analysis, principal com- ponents, and cluster analysis, can be used to gain useful insights. Exercises 1. Consider the following three snippets of text: ■ The rain in Spain falls mainly in the plain. ■ The Spanish World Cup team is awesome. ■ Spanish food is beyond awesome.

672 Part XI: Internet and Social Marketing After stemming and stopping these snippets, complete the following tasks: a. Create binary coding for each snippet. b. Create frequency coding for each snippet. c. Create tf-idf coding for each snippet. 2. Use your favorite search engine to find the definition of “Amazon mechani- cal Turk.” If you were conducting a text mining study, how would you use Amazon mechanical Turks? 3. Describe how text mining could be used to mechanically classify restaurant reviews as favorable or unfavorable. 4. Describe how text mining could be used to determine from a member of Congress’ tweets whether she is conservative or liberal. 5. Describe how text mining could be used to classify The New York Times stories as international news, political news, celebrity news, financial news, science and technology news, entertainment news, obituary, and sports news. 6. Alexander Hamilton, John Jay, and James Madison wrote The Federalist Papers, a series of 85 essays that provide reasons for ratifying the U.S. Constitution. The authorship of 73 of the papers is beyond dispute, but for the other 12 papers, the author is unknown. How can you use text mining in an attempt to determine the authorship of the 12 disputed Federalist Papers? 7. Suppose you are a brand manager for Lean Cuisine. How can use text mining of tweets on new products to predict the future success of new products? 8. Suppose that on the same day the Sofia Vergara Diet Pepsi ad with David Beckham aired on two different shows. How would you make a decision about future placement of the ad on the same two TV shows? 9. Two word phrases are known as bigrams. How can coding text with bigrams improve insights derived from text mining? What problems might arise in using bigrams?

Index Symbols Advertising Pulsing Policies ages of subscribers, analyzing & (concatenate sign), 42 for Generating (ESPN magazine), 21–22, 3M example (inflection Awareness for New 25 points), 416 Products (Marketing airline miles A Science), 513 forecasting with neural The Accidental Influentials (Harvard Business Equation 4 model for networks, 258–259 Review), 653 measuring effectiveness Winter’s Method, 243–247 acquisition rate (customers), 366–367 of, 509–511 airlines, overbooking, 151–153 adaptive methods, 241 One Way TV Advertisements Albright, Chris, 99 adaptive/hybrid conjoint Work (Journal of algorithms, Evolutionary, 113 analysis, 280 ADBUDG curves Marketing Research), AllDifferent option (Solver), basics of, 484–485 505 276 fitting (Syntex Labs), Online Ad Auctions alternative hypothesis 486–489 additive model for estimating (American Economic contrasts (ANOVA), 602 trends/seasonality, Review), 533 linear regression, 182 228–231 Adelman, Sydney, 267 Planning Media Schedules in testing group means, 596 adjacent industries/categories, 431 the Presence of Dynamic Altman’s Z-statistic method, Adstock Model (advertising), 505–508 Advertising Quality 251 advertising. See also media selection models; (Marketing Science), ALVINN (Automated Land Pay per Click (PPC) advertising 513 Vehicle in a Neural Adstock Model, 505–508 pulsing vs. continuous Network), 250 spending, 511–514 American Consumer Some Optimization Problems Satisfaction Index (ACSI), in Advertising Media 669 (Journal of the Ameritech study (customer Operational Research value model), 336 Society), 517 Amour movie ratings, 396, Super Bowl ads, evaluating 398–400 with tweets, 671 Analysis of Variance, 599 Volkswagen ad, 671 Analysis Toolpak, 172–174 AdWords auctions (Google), ANOVA (Analysis of 533–536 Variance), One-way After Tax Profits equation, 345 contrasts, 601–603

674 Index forecasting one-way ANOVA, Bass diffusion model breakfast foods (MDS 599–601 basics of, 427–428 analysis), 566–570, one-way ANOVA example, deflating intentions data, 571–574 596–598 434–435 breast cancer test results overview, 595 estimating, 428–430 (Bayes Theorem), role of variance in, 598–599 forecasting new product 579–581 testing group means, sales with, 431–434 Breen, Jeffrey, 669 595–596 modifications of, 437 Brinton, Chris, 634 ANOVA, Two-way. See two- simulating sales of new Broadbent, Simon, 505 way ANOVA products with, 435–437 Bud Light ad, 671 Applied Multidimensional Bass version of The Tipping Building Models for Marketing Scaling (Holt Rinehart Point, 646–650 Decisions (Kluwer and Winston), 566 Bates Motel (revenue Publishing), 472 Applied Multivariate Statistical management) bundling, price. See price Analysis (Prentice-Hall), determining booking limits, bundling 548 150–151 Burkart, A.J., 517 array formulas for ESPN determining single profit- businesses, valuation of. See subscriber demographics, maximizing price, valuation (customer 78–79 146–147 value) array functions (Excel), 59–60 estimating demand curve, array LINEST function 144–146 (Excel), 388–390 segmenting customers with C attributes, product, 264–265 capacity constraints, C (local cluster coefficient) auctions (Google AdWords), 150 (networks), 631 533–535 segmenting customers with cable TV subscribers auto sales (regressions), two prices, 147–149 (customer value), 331, 186–191 Bayes Theorem, 579–581 333–334 autocorrelation (linear Before Tax Profits formula, cake sales (statistical regressions) 345 functions), 71–74 checking for (software sales), betweenness centrality 479 capacity constraints, non-independence of errors, (network nodes), segmenting customers 198–204 624–626 with (Bates Motel), 150 positive, 199 Bid Simulator feature car brands (perceptual maps), AVCO Financial (neural (AdWords), 536 570–571 networks), 252 bigrams, 672 car models (PCA), 556–557 bin ranges, 61 card sorting (similarity data), AVERAGE function (Excel), binary coding (text mining), 65 567 666–667 carpet cleaner (full profile AVERAGEIF/AVERAGEIFS binary dependent variables, functions (Excel), 72–74, conjoint analysis), 285–286, 288–289 265–271 599 BINOMDIST function (Excel), cars, driving with neural 657 networks, 251 B Binomial random variables, CART algorithm (decision 524, 656–658 trees), 408–410 bankruptcies, predicting with Blattberg, Robert, 365, 394 cash flows (health club neural networks, 251–252 The Blind Side (film revenues), business), 339–340, 345 Barnes and Noble (contrasts), 670 catsup market (neural 601–603 brand equity (discrete choice networks), 252 Bass, Frank, 517 analysis), 313–314 causal forecasting, 177

Index 675 ceilingAcq parameter, 366 cluster analysis. See also communalities (PCA), ceilingRet parameter, 366 clustering U.S. cities 555–556 cell phone sales (S curves), vs. decision trees, 408–409 competition (pricing), 136 418–420, 423–424 segmenting customers with, complementary products, centered moving averages, 271 pricing (Solver), 94–96 236–237 clustering U.S. cities computer hardware (price ceteris paribus, cluster analysis, 378–379 bundling), 109–110 190, 217 determining correct number Conditional Formatting Change Data Source option of clusters, 386 Formula option (Excel), (PivotTables), 8 finding optimal clusters with 464 ChapStick pricing (demand Solver, 380–382 Conditional Formatting icon curves), 96–99 identifying clusters, 379–380 sets (Excel), 43 charts, Excel. See Excel charts interpretation of clusters, conditional probability, Chase Bank (neural networks), 578–579 384–386 252 conjoint analysis overview, 378–379 check boxes for controlling setting up Solver model for, adaptive/hybrid, 280 chart data, 45–47 382–384 choice-based, 280 Chi Square Test, 319 standardizing demographic On the Creation of Acceptable Chiang, Mung, 634 attributes, 379–380 Conjoint Analysis chocolate preferences (discrete coefficients Experimental Designs choice analysis), coefficient of innovation, 428 (Decision Sciences), 305–309 logistic regression, 293 273 choice-based conjoint analysis, multiple linear regressions, developing conjoint 280 182 simulators, 277–279 churn rate, 328 regression, 185 full profile. See full profile class separation, 404 collaborative filtering conjoint analysis Classic PivotTable item-based, 398–400 generating product profiles Wizard, 28 item-based vs. user-based, with Solver, 272–276 classification 400–401 overview, 263 Classification and Regression product sets/attributes/levels, Netflix Prize Competition, Trees (Chapman and 263–265 401 Hall), 409 segmenting markets with, user-based, 393–397 error, 405, 410 271, 386–391 matrix (linear discriminant Collective Dynamics of ‘small- value-based pricing with, world’ networks (Nature), analysis), 589–590 271–272 631 proportional (linear connectors (The Tipping discriminant analysis), column graphs Point), 644 590 adding product images to, constraints on changing cells 32–34 classification algorithms (Solver model), 91 linear discriminant analysis. creating, 30 consumer surplus See linear discriminant Column Labels zone bundling products to extract, analysis (PivotTables), 6 108 Naive Bayes classifier, Column Sparklines, 51 defined, 107–108 581–586, 592 combination charts (Excel) contagion model (networks), overview, 577 adding labels/tables to charts, 641–646 click-through rate (PPC 34–36 continuous spending advertising), 533 adding product images to (advertising) closeness centrality (network column graphs, 32–34 defined, 505 nodes), 623–624 basics of, 29–32 vs. pulsing, 511–514

676 Index contrasts (one-way ANOVA), Create PivotTable life cycles of, 347 601–603 dialog box, 4 Manage Marketing by the Conversion Rate (PPC advertising), 531–533 credit card transactions Customer Equity Test Copernican Principle (neural networks), 252 (Harvard Business basics of, 439–440 simulating remaining life of Credit Union (forecasting). Review), 365 products with, 440–441 See forecasting based on Managing Customers as corpus (text mining), 664 CORREL function (Excel), special events Investments (Pearson 174, 395 critical values formula Prentice-Hall), 328 correlations, sample, 544–548 correlations between variables (ANOVA), 603 optimizing acquisition/ finding with Data Analysis cross-sectional data, 179 retention spending on, Add-in, 172–174 crosstabs analysis (ESPN 368–373 “regression toward the mean” and, 174 magazine), 25–27 relationship between summarizing linear currentretentionrate spending and relationships with, 170–172 parameter, 366 acquisition/retention of, Cost per Click (PPC currentspendpercustomer 365–367 advertising), 532–534 parameter, 366 cost plus pricing, 271 count format (data), 298–299 curves for modeling resource- COUNT function (Excel), 71 COUNTBLANK function response relationships, D 484–489 (Excel), 71 COUNTIF function (Excel), customer lifetime value data 69–72 Customer Lifetime Value: controlling chart data with COUNTIFS functions (Excel), Foundations and Trends check boxes, 45–47 72–74, 406, 582 Counting Your Customers: in Marketing (Now Data Analysis Add-in, Who Are They and What Publishing), 335 172–174, 179–181 Will They Do Next? (Management Science), customer value multiplier, Data Analysis and Business 334 country clubs (two-part 328, 331 Modeling with Excel tariffs), 129 Courtyard by Marriott: estimating active customers, 2010, 42 Designing a Hotel Facility with Consumer- 334–335 Data Analysis dialog box, Based Marketing Models (Interfaces), 264 Friday Night Lights (FNL), 596 covariances of linear combinations of 333–334 data mining legends, 453 variables, 547–548 ideas to enhance customer Data Set Manager, 254–255 sample covariance, 543–544 value model, 335–336 Data Validation drop-down multiplier formula, 331 box, 328–329 overview, 327 rule for summarizing data sensitivity analysis, sets, 68 measuring with two- rule of thumb for way data tables, 330 summarizing data sets, template, 328–330 68, 70–71 varying customer margins, slicing/dicing sales data. See 331–333 PivotTables (Excel) customers Strategic Database Marketing customer-centric approach to (McGraw-Hill), 465 valuation. See valuation summarizing marketing data. (customer value) See Excel charts finding ideal points on summarizing with statistical perceptual maps, functions. See statistical 570–574 functions (Excel)

Index 677 summarizing with Subtotals clustering by attributes, price elasticity and, 317–318 pricing optimization, feature, 74–77 378–379 311–313 typical values for data sets, standardizing demographic random utility theory, 64–67 attributes (clustering), 303–305 testing for significance in, data points. See also outliers 379–380 314–315 (data points) dependent variables discriminant analysis forecasting sales from, in neural networks, 249 (regression), 251 discriminant score (linear 495–501 nonlinear relationships with discriminant analysis), Data Tables independent variables, 587 Disney World (bundling measuring sensitivity 192 products), 108 documents (text mining) analysis with, 330 relationships to independent defined, 664 term frequency/inverse recalculation tip, 351 variables, 161–162 document frequency Database Marketing (Springer), Dhoakia, Utpal, 353–357 score (tf-idf), 666–668 Dominick’s Finer Foods (shelf 335, 394 Diffusion of Innovations (Free space allocation), 492 Dreze, Xavier, 492 day of the week effect on sales Press), 415 dummy variables, 187–188, 268–270 (La Petit Bakery), 15–16 direct mail dynamic discrete choice, 315–316 Decision Calculus Modeling at optimizing campaigns with E Syntex Labs (Interfaces), Solver, 465–468 eating habits (PCA), 557 486 targeting with neural Economic Prediction Using decision trees networks, 251 Neural Networks: The Case of IBM Daily Stock vs. cluster analysis, 408–409 DIRECTV. See also cable TV Returns, 250 Efficient Market Hypothesis, CART algorithm, 409–411 subscribers (customer 671 eigenvalues, 555 constructing, 404–408 value) elasticity, price demand curves, 86–90 interpreting, 408 Base model, 431–434 discrete choice analysis, overview, 403–404 discount rate, per period, 328 317–318 property of uniform cross pruning, 409–410 discount sales, 138–141 elasticity, 318 deflating intentions data (Bass discrete choice analysis elevators, directing with model), 434–435 chocolate preferences, neural networks, 252 empirical generalization, 433 degree centrality (network 305–309 nodes), 621–622 Discrete Choice Methods with Deighton, John, 365 Simulation (Cambridge demand curves University Press), 304 for all products, finding with Discrete Choice Modeling SolverTable, 101–103 and Air Travel Demand estimating (revenue (Ashgate Publishing), management), 144–146 317–319 estimating linear and power, discrete choice theory, 280 85–90 dynamic discrete choice, forms of, 86–90 315–316 pricing using subjectively evaluating brand equity, estimated, 96–99 313–314 willingness to pay and incorporating price/brand (nonlinear pricing), equity into, 309–311 124–125 Independence of Irrelevant demographics Alternatives (IIA) analyzing effect on sales (La assumption, 316–317 Petit Bakery), 21–25 overview, 303

678 Index Empirical Generalizations generating product profiles HLOOKUP function, 99 and Marketing Science: A with, 272–276 IFERROR function, 42 Personal View (Marketing INDIRECT function, 273, Science), 433 maximizing lift of product categories, 454–456 275 ending credits (customers), INTERCEPT function, 371 maximizing quantity discounts with, 520–522 162–166, 170 entropy, 405 KURT function, 196 epidemic model (The Tipping MDS analysis of U.S. city LARGE function, 69 distances, 560–566 MATCH function, 41, 112, Point), 647–650 Equation 4 model nonlinear pricing strategies 396–397 and, 123–124, 126 matrix multiplication and (advertising), 509–511 error terms (linear optimizing direct mail transpose in, 545–546 campaigns with, MEDIAN function, 65 regressions) 465–468 MMULT function, 546 basics of, 178 MODE function, 65 nonconstant variance optimizing three-way lift OFFSET function, 560–561 with, 451–453 outlines, 77 (heteroscedasticity), PERCENTILE.EXC/ 197–198 running model multiple nonindependent, 198–199 times, 391 PERCENTRANK.EXC normally distributed, functions, 68–69 196–197 setting Mutation rate with, RAND() function, 349–350 errors/residuals, 167–168 384, 522 RANDBETWEEN function, ESPN magazine advertising 356–358 budget (Naive Bayes), Excel Report Filter, 11–14 586 Analysis ToolPak, 293 RSQ function, 162–166, ESPN magazine subscribers array functions, 59–60 170 (demographics) AVERAGE function, 65 SKEW function, 64 analyzing ages of, 21–22 AVERAGEIF/AVERAGEIFS Slicers feature, 11–14 analyzing gender of, 22–23 functions, 72–74, 599 SLOPE function, 170 array formulas for BINOMDIST function, SMALL function, 69 summarizing, 78–79 657–658 statistical functions. See constructing crosstabs calculating best-fitting trend statistical functions analysis of age lines with, 167 (Excel) /income, 25 Conditional Formatting STDEV function, 67 describing income Formula option, 464 STEYX function, 169 distribution of, 23–24 Conditional Formatting icon SUMIF/SUMIFS functions, describing location of, 24 sets, 43 69–74 overview, 21 CORREL function, 174, 395 TABLE feature, 39, 52–53 estimating Bass model, COUNT function, 71 TEXT function, 42 428–430 COUNTBLANK function, 71 TRANSPOSE function, evaluating user similarity, 394 COUNTIF/COUNTIFS 59–60 Evolutionary algorithms, 113 functions, 69–74, 406, TREND function, 207–208 Evolutionary engine, 92 582 Trendline feature, 96 Evolutionary Solver Excel 2010 Data Analysis VAR function, 67 basics of, 112–115 and Business Modeling VLOOKUP function, 15 finding optimal bundle (Microsoft Press), 65 VLOOKUP function/ prices with, 111–118 Excel Solver. See Excel formulas, 462 finding optimal linear Solver XNPV and XIRR functions, classification rules with, FREQUENCY function, 339–340, 345 587–591 61–62 GETPIVOTDATA function, 25–27, 52–54 GOAL SEEK command, 433

Index 679 Excel charts Solver Parameters dialog sales with no interactions combination charts. See box, 97–98 (two-way ANOVA), combination charts Two-Way SolverTable, 101 614 (Excel) An Explanation of Linear sales with Ratio to Moving controlling data with check Programming in Media Average Forecasting boxes, 45–47 Selection (Journal of Method, 235–238 dynamically updating labels, Marketing Research), software sales with 40–43 517 SCAN*PRO model, GETPIVOTDATA for end- eyeball approach, 339 475–479 of-week sales reports, E-ZPass example (conjoint stock markets with Twitter, 52–54 simulator), 278 670–671 overview, 29 Technological Forecasting PivotCharts, 36–38 for Decision-Making (McGraw-Hill), 425 sparklines for multiple data F series, 48–51 Farley, John, 429 forecasting based on special summarizing monthly sales- feasible solutions (Solver events force rankings, 43–45 model), 92 building basic model, updating automatically, Fidelity Corporation (neural 213–217 39–40 networks), 250 checking randomness of Excel Solver Field List (PivotTables), 5–8, forecast errors, activating, 90 17, 28 221–222 building special forecasting filtering, collaborative. See evaluating forecast accuracy, models with. See collaborative filtering 217–218 forecasting based on forecasting refining base model, special events accuracy of regression 218–221 defining Solver model, 90–91 forecasts, 183 Format Data Series, finding optimal clusters with, airline miles with neural 32–33 380–382 networks, 258–259 Format Trendline generating product profiles causal, 177 dialog box, 97 with (conjoint analysis), future months (Winter’s frequency coding (text 272–276 Method), 246 mining), 666–668 modeling trends/seasonality improving accuracy of FREQUENCY function with, 228–231 (movie revenues), (Excel), 61–62 optimizing acquisition/ 498–499 Friday Night Lights (FNL) TV retention spending, 367, movie revenues, show, 333–334 370, 372–373 495–498 Frontiers of Econometric pricing multiple products movie revenues with 3 weeks Behavior (Academic with SolverTable, of revenue, 499–501 Press), 317–318 99–103 movie revenues with Twitter, full profile conjoint pricing razors, 92–94 669–670 analysis pricing razors with new product sales with Bass determining product profiles, complementary model, 431–434 266–267 products, 94–96 one-way ANOVA, 599–601 overview, 265 Select a Solving Method sales for future quarters, ranking attributes/levels, drop-down menu, 237–238 270–271 91–92 sales from few data points, running regression with setting up model for cluster 495–501 dummy variables, analysis of U.S cities, sales with Bass model, 268–270 382–384 431–434 shortcomings of, 279

680 Index G in SCAN*PRO model, 423 I setting bounds on changing Gangnam style video, 653 cells, 419, 474 ideal points on perceptual gender of subscribers, maps, 570–574 GRG Nonlinear engine, 92 analyzing (ESPN GRG Nonlinear option, 101 Idiot’s Bayes - not so stupid magazine), 22–23 group means, testing after all? (International A general theory of bibliometric Statistical Review), 592 and other cumulative (ANOVA), advantage processes 595–596 IFERROR function (Excel), 42 (Journal of American Groupon offers imitators (Bass model), 427 Society of Information analyzing with Monte Carlo impressions (online Sciences), 635 Generalized Second Price simulation, 357–359 advertising), 530 Auction (GSP), 534 lifetime customer value and, income of subscribers, genetic algorithms, 113 GETPIVOTDATA function 327 analyzing (ESPN (Excel) pizza parlor example, magazine), 23–25 for end-of-week sales Independence of Irrelevant 353–357 Alternatives (IIA) reports, 52–54 using one-way data tables to assumption, 316–317 pulling data from independent variables simulate, 357–359 dealing with insignificant PivotTables with, Grucca, Thomas, 252 25–27 Gini Index, 405, 410 (regressions), Gladwell, Malcolm, 641–642, 644 H 184–185 Global Seek command, 89 determining significant Goal Seek, 89, 433 Goldberg, David, 113 HAL Computer (linear (regressions), 183 Golliher, Sean, 637 Gompertz curves multiple regressions), logistic regression with vs. Pearl curves, 425 allocating supermarket shelf 178–186 multiple, 296–298 space with, 492 defined, 485 health club business valuation. in neural networks, 249 fitting S curves to data with, 422–424 See valuation (customer nonlinear relationships with Google AdWords auctions, 533–536 value) dependent variables, great salespeople (The Tipping Point), 645 Henderson, Bruce, 136 192 Greek Yogurt (decision trees), 404–408 heteroscedasticity, 197 relationships with dependent Green, Paul, 566 GRG Multistart engine histograms variables, 161–162 determining values in Bass model, 429 summarizing data with, INDIRECT function (Excel), finding values fitting sets of data, 423 59–64 273, 275 summarizing Monte Carlo inflection points (S curves), simulation results with, 415–416 359 Information Science (Oxford HLOOKUP function (Excel), University Press), 405 99 innovators (Bass model), 427 Hollywood Stock Exchange input cells (neural networks), (HSX), 670 249 Honda TV advertising, intentions data, deflating (Bass 518–527 model), 434–435 Housing Starts (Ratio interactions to Moving Average two-way ANOVA with, Forecasting), 238 614–616 Hughes, Arthur, 465 of variables (linear Hypothesis of No Linear regressions), 193–195 Regression, 182 intercept (constant term), 178

Index 681 INTERCEPT function (Excel), analyzing yearly trends, linear classification with 162–166, 170 19–20 more than two groups, Interfaces (1988), 492 day of the week effect on, 590–591 Interpretation by physicians of 15–16 model validation, 591 clinical laboratory results overview, 14–15 overview, 586–587 (New England Journal of labels linear forecasting models, 230 Medicine), 580 adding to charts, 34–36 linear media allocation, interval data, 560 dynamically updating in 517–520 Interviews with Real Traders charts, 40–43 linear pricing, 107 (Ward Systems Group), LARGE function (Excel), 69 linear regression. See simple 250 learning curves (pricing), linear regression inverse rankings (product 135–136 linear relationships, profiles), 268 least squares estimates summarizing with iPod sales (S curves), 420–422 (coefficients), 182 correlations, 170–172 IRR (internal rate of return), least-squares lines (Excel), 167 LINEST function (Excel), 339–340, 343 Issenberg, Sasha, 403 Lee, A.M., 517 388–390 item-based collaborative Legs value (staying power), links, measuring importance filtering, 398–401 496–497, 499–500 of (networks), 626–628 Lehmann, Donald, 331, 429 Little, John, 484 levels of product attributes, loading of variables on J 264–265 principal components, lifts 552 J.Crew (RFM analysis), calculating for multiple loans, determining with neural 460–468 two-way product networks, 252 combinations, 448–449 local cluster coefficient (C) K computing for two products, (networks), 630 445–449 location of subscribers, keywords (online advertising), computing three-way analyzing (ESPN 529 product, 449–453 magazine), 24 Klemz, Bruce, 252 optimizing store layouts Lodish, Leonard, 486, 492 Klout scores (networks), with, 454–456 Log Likelihoods, 309–311 636–637 linear combinations of log odds ratio, 289 kRet parameter, 366 variables, 547–548 Logistic curves Kumar, Piyush, 671 linear demand curves, 86–88 vs. Gompertz curves, 425 KURT function (Excel), 196 linear discriminant analysis basics of, 418–420 kurtosis, 196–197 classification matrix, logistic regression 589–590 dialog box, 294 L classification rules involving estimating model with nonlinearities/ maximum likelihood L value, definition/ interactions, 591 method, 290–293 computation of (network evaluating classification rule estimating probabilities with, nodes), 628–630 quality, 590 292–293 La Petit Bakery sales finding most important interpreting coefficients, 293 analyzing effect of attributes, 589 model, 289–290 promotions on, 20–21 finding optimal linear with multiple independent analyzing product classification rule, variables, 296–298 seasonality, 16–18 587–589 necessity of, 286–289

682 Index overview, 285–286 Marketing Decision Models Monte Carlo simulation performing with count data, (Prentice-Hall), 509 analyzing Groupon offers 298–299 Marketing Research with, 357–359 StatTools. See StatTools (Prenctice-Hall), 556 basics of, 347–348 logit regression model, 289 viral. See viral marketing media allocation simulation, logit transformation, 289 Markov Chains, 347–349 522–527 Lohan, Paris, 493 Martino, Joseph, 425 modeling customer values Lonsdale, Robert, 517 MATCH function (Excel), 41, using, 347–353 Loyalty Effect (Reichfeld and 112, 396–397 predicting market initiative Teal), 330 matrix multiplication/matrix successes using, Luenberger, David, 405 transposes (Excel), 353–357 545–546 summarizing results with mavens (The Tipping Point), histograms, 359 M 645 movie revenues maximum likelihood method, forecasting with 3 weeks of machine learning, 668 290–293, 307–310 revenue, 499–501 MAD (Mean Absolute MBA applicants (linear predicting, 495–498 Deviation), 259, 497–498 discriminant analysis), predicting with Twitter, magazine subscribers (logistic 590–591 669–670 regression), 286–299 McGrayne, Sharon, 579 movie reviews (text mining), Manage Marketing by the media selection models 668 Customer Equity Test linear media allocation, moving averages (Harvard Business 517–520 eliminating seasonality with, Review), 365 Monte Carlo media 225–228 Managing Customers as allocation simulation, Ratio to Moving Average Investments (Pearson 522–527 Forecasting Method, Prentice-Hall), 328 overview, 517 235–238 Mao’s Palace restaurant (linear quantity discounts and, multicollinearity (linear regression), 162–168 520–522 regressions), 204–207 MAPE (Mean Absolute Percent MEDIAN function (Excel), 65 multidimensional scaling Error), 247, 476–477, 497 Mellon Bank (neural (MDS) margins, varying customer, networks), 252 analysis of breakfast foods, 331–333 A Meta-Analysis of Applications 566–570 markdown pricing (revenue of Diffusion Models analysis of U.S. city management), 153–155 (Journal of Marketing distances, 560–566 market basket analysis Research), 429 finding ideal points on computing lift for two Milgram, Stanley, 628 perceptual maps, products, 445–449 misclassification rate (decision 570–574 computing three-way trees), 409 similarity data and, 559–560 product lifts, 449–453 mixed bundling (prices), 110 multinomial logit model, 304 data mining legends, 453 MMULT function (Excel), 546 multinomial logit version of optimizing store layouts with MODE function (Excel), 65 discrete choice, 319 lift, 454–456 A Model to Improve the Baseline multiple linear regression market values, using customer Estimation of Retail Sales, autocorrelation, 198–204 value to estimate, 344 472 basics of, 178–179 marketing Models and Managers: The heteroscedasticity, 197 Market Facts of Canada, 403 Concept of a Decision interactions of variables, 193 Marketing Analytics (Admiral Calculus (Management interpreting output of, Press), 523 Science), 484 182–186

Index 683 multicollinearity, 204–207 networks. See also neural predicting stock markets nonlinear relationships networks with, 250 between variables, 192 defining/computing L value, regression and, 249–250 normally distributed error 628–630 NeuralTools (Excel add-in), terms, 196–197 Klout scores, 636–637 249 overview, 177 local cluster coefficient (C), New Way to Measure predicting potential 630 Consumers’ Judgments customer value with, measuring importance of (Harvard Business 335 links, 626–628 Review), 265 qualitative independent network contagion, New-Product Diffusion Models variables in, 186–191 641–646 (Springer), 433 R2 and, 168 A Networked Life (Cambridge nodes (networks) running with Data Analysis University Press), 401 betweenness centrality, Add-in, 179–181 Networks Illustrated 624–626 testing for nonlinearities/ (Edwiser Scholastic closeness centrality, interactions, 193–195 Press), 634 623–624 testing validity of regression nodes. See nodes (networks) degree centrality, 621–622 assumptions, 195–196 Power Law for, 634–636 entropy of (decision trees), validation of, 207–209 random, 631 405–407 multiplicative model for regular, 631–633 measuring importance of, estimating trends/ Rich Get Richer theory, 621 seasonality, 231–233 634–636 seeded, 645 multiplier formula (customer scale-free, 635 nonlinear pricing value), 331 six degrees of separation, defined, 107 Multistart option (Solver), 232 628 demand curves and Mutation rate (Evolutionary Strogatz and Watts network, willingness to pay, Solver), 114, 385 633–634 124–125 mxn matrix, 545–546 Neumann, James von, 348 optimizing nonstandard neural networks quantity discounts, analyzing scanner data with, 127–128 N 252 optimizing standard quantity credit card and loan discounts, 125–127 Naive Bayes classifier basics of, 581–586 applications, 252 optimizing two-part tariffs, Idiot’s Bayes - not so stupid direct mail targeting with, 129–131 after all? (International Statistical Review), 592 251 overview, 123–124 virtues of, 592 directing elevators with, profit maximizing with, named ranges (two product 252 125–131 lifts), 447 driving cars with, 251 nonlinear relationships Economic Prediction Using Nash equilibrium, 315–316 between variables, 192 negative autocorrelation, 200 Neural Networks: The nonmetric MDS, 560 negatively skewed histograms, Case of IBM Daily Stock nonstandard quantity Returns, 250 63 forecasting airline miles discounts (nonlinear neighbor nodes (networks), with, 258–259 pricing), 127–128 overview, 249 NPV (Net Present Value) of 630 predicting bankruptcies cash flows, Netflix Prize Competition with, 251–252 339–340, 343 predicting sales with, Null Hypothesis (collaborative filtering), 253–258 ANOVA, 596, 602 401 linear regressions, 182

684 Index O outlines, Excel, 77 pulling data with output cells (neural GETPIVOTDATA objective (target) cells (Solver networks), 249 function, 25–27 model), 90 overbooking models, pizza parlor OFFSET function (Excel), 151–153 customer acquisition/ 560–561 retention spending, 368–370 Old Spice video, 653 Groupon offers, 353–357 On the Creation of Acceptable P Planning Media Schedules in Conjoint Analysis the Presence of Dynamic Advertising Quality Experimental Designs Pareto 80–20 Principle, (Marketing Science), 513 (Decision Sciences), 273 10–11 Poisson random variables, 656–658 One Way TV Advertisements Pay per Click (PPC) positive autocorrelation, 199 Work (Journal of advertising positively skewed histograms, Marketing Research), 505 basics of, 529–530 63 power company (nonlinear One-way Analysis of Variance Bid Simulator feature pricing), 125–128 (ANOVA). See ANOVA (AdWords), 536 power curves, 484–486 power demand curves, 86–87, (Analysis of Variance), Google AdWords auctions, 88–90 One-way 533–535 Power Law for networks, one-way data tables profitability model for, 634–636 Power Pricing (Dolan), 96 measuring sensitivity 531–532 Practical Text Mining analysis with, 343–344 peak sales, time/value of, 430 (Academic Press), 669 Predicting the Future with One-Way SolverTable, 101 Pearl curves Social Media, 669 using to simulate Groupon basics of, 418–420 Predictions (Simon and offers, 357–359 vs. Gompertz curves, 425 Schuster), 415 Price, D. J., 635 Online Ad Auctions (American per period discount rate, 328 price bundling Economic Review), 533 per period retention rate, 328 extracting consumer surplus with, 107–108 ordinal data, 560 PERCENTILE.EXC/ finding optimal bundle orthogonal designs (product PERCENTRANK.EXC prices with Evolutionary Solver, 111–118 profiles), 266–267 functions (Excel), 68–69 mixed bundling, 110 Orthogonal Main-Effect perceptual maps (product overview, 107–108 pure bundling, 109–110 Plans for Asymmetrical comparisons), 570–574 pricing cost plus, 271 Factorial Experiments periods parameter (customer determining single profit- (Technometric), 266–267 margin values), 332 maximizing price (Bates Motel), 146–147 orthogonality (covariance), Peterson, Ann Furr, 252 549 PivotCharts (Excel), 36–38 Otis Elevator (neural PivotTables (Excel) networks), 252 analyzing demographics outliers (data points) effect on sales, 21–25 defined, 68 analyzing La Petit Bakery finding (movie revenues), sales. See La Petit 499 Bakery sales in regressions, 183–184 analyzing True Colors spotting omitted special Hardware sales. See forecasting factors, 220 True Color Hardware spotting omitted special sales forecasting factors with, Create PivotTable dialog 217–218 box, 4

Index 685 dropping prices over time, probit regression, 304–305 R 135–138 product profiles. See also R2 value, defining, 168–169 markdown (revenue conjoint analysis RAND() function (Excel), management), 153–155 defined, 263 349–350 multiple products with generating with Evolutionary RANDBETWEEN function SolverTable, 99–103 Solver, 272–276 (Excel), 356–358 optimization of (discrete ranking, 266–267 random forecast errors, choice analysis), products 311–313 New-Product Diffusion Models 221–222 random networks, 631 optimizing with Excel (Springer), 433 random utility theory, Solver, 90–96 product sets/attributes/levels 303–305 Power Pricing (Dolan), 96 (conjoint analysis), random variables price elasticity (demand 263–265 Profit per Click (PPC Binomial/Poisson, curves), 86–90 advertising), 532 656–658 price elasticity (discrete profitability model for PPC, 531–532 defined, 356 choice analysis), promotions, analyzing effect randomized block designs 317–318 on sales (La Petit Bakery), price reversals, 116–117 20–21 (two-way ANOVA), 608 price skimming, 136–138 property of uniform cross ranking product attributes/ (price–unit cost)*demand elasticity, 318 formula, 96, 102 proportional classification levels (conjoint analysis), razors, 92–94 (linear discriminant 270–271 razors with complementary analysis), 590 Rao, Vithala, 566 products, 94–96 pruning decision trees, Ratio to Moving Average reservation prices, 124 409–410 Forecasting Method using subjectively estimated pulsing (advertising) applying to monthly data, demand curves, 96–99 vs. continuous spending, value-based (conjoint 511–514 238 analysis), 271–272 defined, 505 calculating moving averages principle components analysis pure bundling (prices), (PCA) 109–110 /centered moving applications of, 556–557 averages, 236 basics of, 548–550 computing seasonal indexes, communalities, 555–556 237 determining number of PCs fitting trend lines to centered to retain, 554–555 moving averages, 237 finding first principal forecasting sales for future component, 550–552 finding PC3 through PC6, Q quarters, 237–238 554 overview, 235–236 finding second principal component, 552–554 qualitative independent Red Bus/Blue Bus Problem, overview, 541–542 prior probabilities, 579 variables in regression 316–317 probability, conditional, 578–579 analysis, 186–191 regression quantitative independent discriminant analysis version variables in regression of, 251 analysis, 186 logistic. See logistic quantity discounts regression defined, 123 model to predict sales, in media selection, 520–522 253–254 standard/nonstandard, multiple linear. See multiple 125–128 linear regression

686 Index neural networks and, Rogers, Everett, 415 A Model to Improve the 249–250 Root Mean Squared Error Baseline Estimation of regression coefficients, 178, 185 (RMSE), 401 Retail Sales, 472 “regression toward the Row Labels zone modeling marketing mean” (correlations), 174 (PivotTables), 5 response to sales force running with dummy RSQ function (Excel), effort, 484–489 variables (product attributes), 268–270 162–166, 170 modeling trends and regressions rule of thumb for summarizing seasonality of, 225–233 accuracy of predictions from trend lines. See also data sets, 68, modeling trends in simple linear regression 70–71 (software), 479–480 regular networks, 631–633 Reichfeld, Frederic, 330 rxn matrix, 545 modeling with SCAN*PRO Report Filter (Excel), 11–14 Report Filter zone model, 472–475 (PivotTables), 6 optimizing allocation of sales reservation prices, 124 residuals/errors, 167–168 S effort, 489–492 resources S curves predicting with neural allocating scarce, 483–484 basics of, 415–417 networks, 253–258 curves for modeling fitting with seasonality, 420–422 setting quotas, 186 resource-response Gompertz curves and, simulating with Bass model, relationships, 484–489 422–424 retention rate Logistic curves and, 435–437 per period, 328 418–420 summarizing monthly sales- retentionrate parameter, 366–367 sales force rankings, 43–45 revenue management analyzing at Mao’s Palace True Color Hardware. See Bates Motel. See Bates Motel restaurant, 162–166 (revenue management) Copernican Principle to True Color Hardware markdown pricing, 153–155 predict future of, sales overbooking models, 439–441 using multiple linear 151–153 regressions to forecast. overview, 143–144 See multiple linear RFM (Recency/Frequency/ regressions Monetary Value) analysis computing R, F, and M, discount, 138–141 salvage values (customers), 460–462 forecasting from few data 371 direct mail campaign, 251 identifying profit-yielding points, 495–501 sample correlations, 544–548 combinations, 462–465 overview, 459–460 forecasting with Bass model, sample covariance, 543–544 success story, 465 Rich Get Richer theory 431–434 sample standard deviation(s), (networks), 634–636 RISKOptimizer, 523–526 forecasting with no 67, 543 interactions (two-way sample variance of data sets, ANOVA), 614 67 forecasting with Ratio scale-free networks, 635 to Moving Average SCAN*PRO model Forecasting Method, forecasting software sales 235–238 with, 475–479 forecasting with SCAN*PRO modeling Snickers sales model, 475–479 with, 472–475 great salespeople (The overview, 471–472 Tipping Point), 645 scanner data, analyzing with identifying relationship neural networks, 252 to marketing effort, seasonal indexes, 228, 231, 483–484 237 La Petit Bakery. See La Petit seasonality. See also Winter’s Bakery sales Method

Index 687 additive Solver model for defining R2 values, 168–169 spending estimating, 228–231 independent/dependent modeling relationship eliminating with moving variable relationships, between customer averages, 225–228 161–162 acquisition/retention fitting S curves with, SLOPE/INTERCEPT/RSQ and, 365–367 420–422 functions, 170 optimizing acquisition/ multiplicative Solver model Simplex LP engine (Solver), retention, 368–373 for estimating, 231–233 91–92 spentpercustomer parameter, of products, analyzing (La simulating 366 Petit Bakery), 16–18 conjoint simulators, 277–279 spread about typical data seeded nodes, 645 new product sales with Bass values, 64 segmenting customers (Bates model, 435–437 spread of typical value, 64 Motel) remaining product life with squared errors, minimizing, with capacity constraints, Copernican Principle, 499 150 440–441 standard deviation, sample, with two prices, 147–149 Single Factor option (Data 543 segmenting markets with Analysis dialog box), 596 standard error of regression conjoint analysis, 271, six degrees of separation (SER), 169 386–391 (networks), 628 standard quantity discounts sensitivity analysis SKEW function (Excel), 64 (nonlinear pricing) measuring with one-way data skimming, price, 136–138 defined, 123 tables, 343–344 Slicers feature (Excel), optimizing, 125–127 measuring with two-way 11–14 standardized values, 544–545 data tables, 330 slicing/dicing sales data. See standardizing demographic sentiment analysis (text PivotTables (Excel) attributes (clustering), mining) SLOPE function (Excel), 170 379–380 of airline tweets, 669 SMALL function (Excel), 69 states of the world (Bayes defined, 664 smoothing methods, 241 Theorem), 579 predicting movie revenues smoothing parameters/ statistical functions (Excel) with, 670 constants (Winter’s array formulas for sets, product, 264–265 Method), 242, 244–245 ESPN subscriber Sharda, Ramesh, 252 Snickers bars (SCAN*PRO demographics, 78–79 Shelf Management and Space model), 471–475 computing typical values Elasticity, 492 Solver, Evolutionary. See with, 64–67 Significance F values (multiple Evolutionary Solver COUNTIFS/SUMIFS/ linear regressions), 182, Solver, Excel. See Excel Solver AVERAGEIF/ 195 SolverTable AVERAGEIFS, 72–74 similarity data (MDS), 559– finding demand curves with, COUNTIF/SUMIF, 69–72 560, 567 101–103 LARGE/SMALL, 69 simple linear regression pricing multiple products overview, 64 accuracy of predictions from with, 99–103 PERCENTILE.EXC/ trend lines, 169–170 Some Optimization Problems PERCENTRANK.EXC, analyzing sales at Mao’s in Advertising Media 68–69 Palace restaurant, (Journal of the rule for summarizing data 162–166 Operational Research sets, 68 calculating best-fitting trend Society), 517 summarizing data with lines with Excel, 167 Sorger, Stephen, 523 subtotals, 74–77 computing errors/residuals, sparklines for multiple data summarizing variation with 167–168 series, 48–51 VAR/STDEV, 67

688 Index StatTools T three-way product lifts, interpreting logistic 449–453 regression output, 295 tables (Excel) logistic regression with adding to charts, 34–36 Thumbs up? Sentiment multiple independent summarizing weekly sales Classification using variables, 296–298 data with, 52–53 Machine Learning running logistic regression updating sales data with, 39 Techniques (Proceedings with, 293–295 of EMNNP), 668 Target store (market basket STDEV function (Excel), 67 analysis), 453 time series data, 179 steady state margin per The Tipping Point (Back Bay Technological Forecasting customer parameter for Decision-Making Books) (customer values), 332 (McGraw-Hill), 425 Bass version of, 646–650 stemming (text mining), 664 central thesis of, 646–647 STEYX function (Excel), 169 term frequency/inverse overview, 641 stock markets document frequency tokens (text mining), 664 predicting with neural score (tf-idf), 666–668 Train, Kenneth, 304–305 TRANSPOSE function (Excel), networks, 250 testing predicting with Twitter, logistic regression 59–60 hypotheses, 293 trend lines 670–671 for nonlinearities/ stopwords (text mining), 664 interactions (linear accuracy of predictions from, store layouts, optimizing with regressions), 193–195 169–170 for significance in discrete lifts, 454–456 choice analysis, calculating best-fitting with Strategic Database Marketing 314–315 Excel, 167 validity of regression (McGraw-Hill), 465 assumptions, 195–196 fitting quadratic demand Strogatz, Steven, 631 curve with, 96–98 Strogatz and Watts network, TEXT function (Excel), 42 text mining Format Trendline dialog 633–634 box, 97 Strogatz-Watts Small World definitions, 664 evaluating Super Bowl ads overlaying 12-month moving model, 644 average with, 227 structuring unstructured text with tweets, 671 movie reviews and, 668 trends. See also Winter’s (text mining), 664–668 overview, 663–664 Method Subtotals feature (Excel), Practical Text Mining additive Solver model for 74–77 (Academic Press), 669 estimating, 228–231 Sultan, Fareena, 429 predicting movie revenues Sum of Squared Errors, 599 fitting trend lines to centered SUMIF function (Excel), with Twitter, 669–670 moving averages, 237 predicting stock markets 69–72 multiplicative Solver model SUMIFS function (Excel), with Twitter, 670–671 for estimating, 231–233 sentiment analysis of airline 72–74 in software sales, modeling, Super Bowl ads, evaluating tweets, 669 479–480 structuring unstructured with tweets, 671 TREND function (Excel), supermarket shelf space text, 664–668 207–208 The Theory That Would Not (Gompertz curves), 492 True Color Hardware sales symmetric histograms, 63 Die (Yale University calculating revenue for each Syntex Labs (ADBUG curves), Press), 579 product, 9–10 calculating sales percentage 486–489, 492 per store, 4–8 overview, 3–4

Index 689 Pareto 80–20 Principle, user similarity, evaluating, 394 sample variance, 542–543 10–11 user-based collaborative Variance Inflation Factor Report Filter/Slicers features, filtering, 393–397, (VIF), 206 11–14 400–401 Verizon cell phones (price summarizing revenue by bundling), 111–113 month, 8–9 Victory Lab (Random House), Twitter V 403 evaluating Super Bowl ads viral marketing with, 671 validation of multiple linear overview, 653 predicting movie revenues regressions, 207–209 Watts model. See Watts with, 669–670 validity of regression model predicting stock markets assumptions, testing, VLOOKUP function/formulas with, 670–671 195–196 (Excel), 15, 462 two-part tariffs (nonlinear valuation (customer value) Volkswagen ad, 671 pricing) analysis of health club defined, 123 business, 340–343 optimizing, 129–131 computing cash flows, W two-way ANOVA 339–340 basics of, 607–608 estimating business market The Wall Street Journal (linear forecasting sales with no value using, 344 discriminant analysis), interactions, 614 Markov Chain model of, 587–591 with interactions, 614–616 347–353 Wanamaker, John, 505 with replication, 611–614 measuring sensitivity Ward Systems Group (neural without replication, 608–610 analysis with one-way networks), 250 two-way data tables tables, 343–344 Watts, Duncan, 631, 633–634, calculating lifts with, modeling with Monte Carlo 653 448–449 simulation, 347–353 Watts model measuring sensitivity Values zone (PivotTables), 6 basics of, 654–655 analysis with, 330 VAR function (Excel), 67 Binomial/Poisson random performing Monte Carlo variables. See also dependent variables, 656–658 simulation with, variables; independent building viral marketing 351–353 variables model, 658–660 two-way product lifts, Binomial random, 524 complex version of, 655–656 446–449 Binomial/Poisson random weak ties (networks), 634, 644 Two-Way SolverTable, 101 variables, 656–658 websites for downloading typical values for data sets, correlations between, NeuralTools Excel add-in, 64–67 170–174 249 nonlinear relationships Predicting the Future with between, 192 Social Media, 669 U variable cells, changing RISKOptimizer package, 523 (Solver model), 91 SolverTable, 99 Ulam, Stanislaw, 348 Varian, Hal, 533 websites for further updating charts automatically, variance information 39–40 of linear combinations of AdWords, 536 U.S. city distances (MDS variables, 547–548 Bud Light ad, 671 analysis), 560–566 role in ANOVA, 598–599 computer driven cars, 250

690 Index customer value concept for Old Spice video, 653 Within Groups Sum of banks, 335 positive/negative tweet Squares, 599 Economic Prediction Using words, 669 Wittink, Dirk, 472 Neural Networks: The Sean Golliher, 637 Working Capital equation, 345 Case of IBM Daily Stock Shelf Management and Space Returns, 250 Elasticity, 492 Gangnam style video, 653 SSRN direct mailing article, X Hollywood Stock Exchange 251 Xbox/PlayStation/Wii (discrete (HSX) predictions, 670 Volkswagen ad, 671 analysis), 309–316, 318 White, Halbert, 250 impression advertising, 530 Wiley Publishing (one-way XNPV/XIRR functions (Excel), Interviews with Real Traders, 339–340, 345 ANOVA), 596–601 250 Wilson, Rick, 252 X-Y Scatter Chart, 42 Klout scores, 635 logistic regression for Wingdings 3 font, 43 predicting churn rate, Win-Loss Sparklines, 51 Y 335 Winter’s Method Market Facts of Canada, 403 yearly trends, analyzing (La A Model to Improve the estimating smoothing Petit Bakery), 19–20 Baseline Estimation of constants, 244–245 Retail Sales, 472 yield management. See forecasting future months, revenue management 246 neural networks for initializing, 243–244 predicting market share, Mean of Absolute Percentage Z 252 Error (MAPE), 247 New Way to Measure overview, 241 z scores (clusters), 384–385 Consumers’ Judgments, parameter definitions for, Zipf’s Law, 638 265 241–242 zones (PivotTables), 5–6