i LIBRARY Michigan State ; University PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. I DATE DUE DATE DUE DATE DUE I “103994 MSU is M Affirmative Action/Equal Opportunity Institmion GMMJ M BRX AN INVESTIGATION OF THE RELATIONSHIP BETWEEN BRAND INTERDEPENDENCE AND PROMOTION BEHAVIOR By William Kendall Meade II A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Marketing and Logistics July 1992 n‘t“ AV IMI BRAND D May do C oke and promote simultaneous mi- promote at 1 was an implicit i ”mm is obxious b Etgories “i111 low 1: Willi»? and com mm of research in Motion beimior. “11 Wow: (1) is iiii‘m 1990b), (2 5mm) between : i{mm‘SI‘J-1989). & C)//‘-C”~D /k./' Abstract AN INVESTIGATION OF THE RELATIONSHIP BETWEEN BRAND INTERDEPENDENCE AND PROMOTION BEHAVIOR By William Kendall Meade 11 Why do Coke and Pepsi never feature promote at the same time? Coke and 7Up promote simultaneously. Pepsi and 7Up promote simultaneously. Coke, 7Up, and Pepsi routinely promote at the same time as weaker bottlers. This dissertation identifies and measures an implicit coalition among strong soft drink bottlers. The study validates what, to many, is obvious but has never been tested enough to be called proven: that in categories with low brand loyalty, large national brands promote for a mixture of cooperative and competitive reasons. Three testable implications were derived from a stream of research in the promotions literature which relates brand interdependence to promotion behavior. Paraphrasing as questions, these implications ask whether the soft drink category: (1) is characterized by alternating promotions (Kinberg, Rao, and Shakun 1974; Lal 1990b), (2) exhibits competitive parity among strong bottlers and competitive asymmetry between strong and weak bottlers (Narasimhan 1988; Blattberg and Wisniewski 1989), and (3) exhibits equal promotional discounts for strong bottlers that are greater than the promotional discounts of weak bottlers (Narasimhan 11988; Raju, Srinivasan, and La] 1990). Single-source data for Grand Rapids, Michigan, provided by A C. Nielsen were combined with calendar marketing agreements (CMAs) provided by a strong bottler competing in the Grand Rapids market. The integration of these two data sources yielded ‘ use 1110‘ ’IPv“ ‘ a. . [358316111 heft. i4 "4 55:3de a database allowing three implications to be empirically tested. The test results are presented here, along with several suggestions for related research that will advance the understanding of single-source data in theoretical testing, refine the test of brand interdependence concepts, and explain promotional asymmetries. Copyright by William Kendall Meade II 1992 Dedication For Beth. Generosity mad- rcogmion as com tools would have t s: quality of the r recognition for be team. This st and foaming on '1; him of com Allen 3150 dm. 3013060!) expen SW aekno '33-‘11). and A M “39885 to Di'iSion Marker Sm General I flaring first-hat" .783] wOTid‘ 0ft M" Mike DL ”aimed Dex-e :Geaflh as we“ I7L , “q pTOV1ded ‘ Acknowledgments Generosity made this work possible. Dr. Robert W. Nason deserves special recognition as committee chair. Without Dr. Nason's help critical data sets and analytical tools would have been unavailable, deadlines would have been unset and/or unmet, and the quality of the manuscript would have sufi‘ered. Dr. Roger Calantone deserves special recognition for both encouraging and controlling creative assessment of the marketing literature. This study would have languished if not for Dr. Calantone's knack for finding and focusing on important issues. Dr. Bruce T. Allen and Dr. R. Dale Wilson shared the balance of committee responsibilities, ofi‘ering many helpful suggestions. Dr. John W. Allen also deserves thanks for providing encouragement and access to key grocery promotion experts. Special acknowledgment for help is due to Brooks Beverage Management, Inc. (BBMI), and A C. Nielsen. Mr. George P. Julius, Jr., President of BBMI, provided liberal access to personnel and soft drink promotion information. Mr. Dean Aukeman, Division Marketing Director, Mr. Robert W ave, Key Account Manager, and Mike Strahl, General Manager/VP. of 7Up Columbus, generously supported the research by relating first-hand experiences, providing source documents, and patiently explaining the ”real world” of soft drink promotion practice. Mr. Mike Dufl‘y, Marketing Manager, and Mr. J. Dennis Bender, Director of Marketing Analytical Development, both of A. C. Nielsen, provided invaluable assistance to this research as well. It is diflicult to communicate briefly the magnitude of their contribution. They provided single-source data, personal knowledge of promotion practice and seeing. and access armed afocus on 1 in: Bender as a “ ember almost as Dr. John letter In 00mg and I We! Questions al Mb are also due mg expertise 0 Fave Backie, o “1 b0th of Mich-,5 Pmdmg ”Perla Means. Mich} with mom] me c The authors Pan momma .fis .graduatim. Pre iced highicapac the resmd‘ler t] is famed an “Satchel had ‘ vii modeling, and access to promotion experts throughout A. C. Nielsen. In addition, they sustained a focus on cutting edge issues of practice. It would be fair to characterize J. Dennis Bender as a "shadow cabinet" committee member. Dennis encouraged the researcher almost as much as he discouraged the use of R2. Dr. John Totten, Vice President of Analytical and Technical Products, John Porter, Tom Goering, and Denise Masi, all of Nielsen, also provided their time generously to answer questions about sampling, coding of causal variables, and data structures. Special thanks are also due to Frank Slavic, Vice President of Custom Delivery Services, for lending expertise on time series issues and promotion evaluation. Faye Backie, Director of the Business Library, and Tom Davis, Systems Programmer III, both of Michigan State University, deserve thanks for the largely thankless job of providing superlative customer support. Dr. Frank Bacon, Adjunct Professor of Marketing, Michigan State University, and Joe E. Szalay, Consultant, supported this research indirectly by taking the trouble (early in the researchers career) to emphasize the indispensability of tenacity, resourcefulness, and persistence in primary market research. The author’s parents, Dr. John B. and Jean Meade supported the research with egg salad sandwiches, ”visits” for two grandsons as chapters neared completion, and a cash ”graduation" present which arrived a year early. Parents seem to know when their kids need high-capacity hard drives. Finally, Beth Meade supported the research by supporting the researcher throughout the PhD. She had babies as (she) planned, built a family life that renewed and sustained priorities, and said "you'll be fine" over and over. The researcher had doubts when hearing this; as usual, she was right. List of Tables ....... is. of Figures ..... l Intro A lmroductir B. Competiti< C Purpose at 11 Literature A Introducti 1. TheR 2. NCWI 3. Them B. Literature 1. Cncer 2- Price 3- Im'en' 4- Com; C ATlieore D' The BUSll E. The JOUn I. Pmmotio G. 3% HI. Procedur A IntroduCI Table of Contents List of Tables ......................................................................................................... xi List of Figures ....................................................................................................... xiv I. Introduction ...................................................................................... 1 A. Introduction ............................................................................................... l B. Competition and Promotion ....................................................................... 2 C. Purpose and Significance of the Study ........................................................ 3 II. Literature Review and Conceptual Development ........................................ 6 A Introduction ............................................................................................... 6 l. The Rapid Growth in Promotions ......................................................... 7 2. New Data Sources ................................................................................ 7 3. Theoretical Innovations ........................................................................ 11 B. Literature Review: Why Use Sales Promotions? ......................................... 11 l. Uncertainty about Response ................................................................. ll 2. Price Discrimination ............................................................................. 14 3. Inventory Carrying Cost ....................................................................... 22 4. Competition ......................................................................................... 25 C. A Theoretical Centerpiece .......................................................................... 32 D. The Business Press ..................................................................................... 34 E. The Journal Literature ................................................................................ 38 F. Promotion Depth ........................................................................................ 51 G. Summary .................................................................................................... 59 [11. Procedures and Method of Investigation ..................................................... 63 A. Introduction ............................................................................................... 63 B. Research Questions .................................................................................... 63 C. The Data .................................................................................................... 64 1. Single-Source Data ............................................................................... 64 2. Behavioral Data .................................................................................... 68 D. Research Hypotheses ................................................................................. 71 1. Research Question 1 ............................................................................. 72 2. Research Question 2 ............................................................................. 73 3. Research Question 3 ............................................................................. 74 E. Variable Identification and Operationalization ............................................ 75 l. Hypothesis 1 ........................................................................................ 76 a) Study Market: ................................................................................ 76 b) Competitors: ................................................................................... 77 c) G“ d) Prc e) SIT f) Av g) All h) 0r i) 0r 2. Hspot a) 0t 3. Hspot a) O; b) O; 4. Hspot S. Hypot 6. Hs'pot ll. Results... A Introduct B. Testing C .6096? pm ”.339 91.50!” ix c) Grocery Chains: .............................................................................. 78 d) Promotions: .................................................................................... 78 e) Strong Brands: ............................................................................... 83 t) Available Promotion Periods: .......................................................... 86 g) Altemation: .................................................................................... 86 h) Operational Hypothesis 1a .............................................................. 87 i) Operational Hypothesis lb .............................................................. 89 2. Hypothesis 2 ........................................................................................ 94 a) Operational Hypothesis 2 ................................................................ 96 3. Hypothesis 3 ........................................................................................ 98 a) Operational Hypothesis 3a .............................................................. 99 b) Operational Hypothesis 3b .............................................................. 100 4. Hypothesis 4 ........................................................................................ 101 5. Hypothesis 5 ........................................................................................ 105 6. Hypothesis 6 ........................................................................................ 107 Results ....................................................................................................... 108 Introduction ............................................................................................... 108 Testing Operational Hypothesis 1a ............................................................. 108 1. Operational Hypothesis la .................................................................... 109 Testing Operational Hypothesis 1b ............................................................. 114 1. Operational Hypothesis 1b .................................................................... 115 Testing the Second Research Hypothesis .................................................... 118 1. Operational Hypothesis 2 ...................................................................... 118 Testing Operational Hypotheses 3a and 3b ................................................. 122 Testing Research Hypothesis 4 ................................................................... 129 1. Regression Analysis .............................................................................. 133 2. Cross-Promotion Effects ...................................................................... 140 Testing Research Hypotheses 5 .................................................................. 142 Testing Hypothesis 6 .................................................................................. 147 Summary .................................................................................................... 147 Conclusions ....................................................................................... 149 Research Summary ..................................................................................... 149 1. Purpose ................................................................................................ 149 2. Questions ............................................................................................. 150 3. Method ................................................................................................ 151 4. Results 151 a) Question 1 ...................................................................................... 152 b) Question 2 ...................................................................................... 152 c) Question 3 ...................................................................................... 153 5. Summary .............................................................................................. 154 Theoretical Implications ............................................................................. 155 Methodological Implications ...................................................................... 156 Public Policy Implications ........................................................................... 156 Limitations ................................................................................................. 159 f future Resea r1 APPCUC A Calendar .‘xla B. Bundling ..... C Bidding ...... D. Rebates ....... E. Brands versr l Compliance 6. Forms and f H Promotional l. Promoriona' 1, Summary... lll Biblio .5 VII 3'" erzowmpow> X Future Research ......................................................................................... 160 Appendix A ....................................................................................... 162 Calendar Marketing Agreements ................................................................ 162 Bundling .................................................................................................... 162 Bidding ...................................................................................................... 164 Rebates ...................................................................................................... 164 Brands versus Bottlers ............................................................................... 167 Compliance Payments ................................................................................ 167 Forms and Functions of Exclusivity ............................................................ 169 Promotional Frequency .............................................................................. 171 Promotional Discount ................................................................................. 171 Summary .................................................................................................... 172 Bibliography ...................................................................................... 175 ., Contingency Tab' lnrentory Cost re Drivers of Promo Example Coding Research Questiq Prices for Z-Liter Relationships Ar Testing Operatid 2-2 2-3 3-1 3-2 3-3 3-4 3-6 3-7 3-8 3-9 3-10 3-1 1 3-12 3-13 3-14 3-15 4.1 4-2 4.3 4-4 4.5 List of Tables Inventory Cost versus Information ............................................................. 24 Contingency Table for the Calendar of Manufacturer Trade Deals .............. 49 Drivers of Promotional Discounts ............................................................... 52 Example Coding of Causal Data ................................................................. 66 Research Questions and Research Hypotheses ............................................ 75 Prices for 2-Liter Classic Coke ................................................................... 81 Relationships Among Causal Advertising Variables .................................... 83 Testing Operational Hypothesis 1a ............................................................. 88 Bottler Share of Units and Revenue in Chain 3 ........................................... 88 Agresti's Types of Models for Statistical Analysis ....................................... 91 Submodels of Equation 3-1 to Be Tested .................................................... 93 Selected lnforrnation on Grand Rapids Grocery Chains .............................. 97 Cross-Price Elasticities among Bottlers for 2-Liter Products ....................... 104 Summary of Cross-Price Elasticities among Bottlers for 2-Liter Products 104 Summary of Cross-Price and Promotion Effects ......................................... 105 Average Discounts for Each Promotion Type for 2-Liter Products ............. 106 Average Discount for Strong Bottlers ........................................................ 106 Average Discounts for Each Bottler Type and Bach Promotion Type ......... 107 Summary Results of Measures for Operational Hypothesis 1a ..................... 114 Association of Coke and Pepsi Promotions ................................................. 115 Association of Strong and Weak Bottler Promotions .................................. 115 Correlation among Bottler Promotions ....................................................... 117 Wilcoxon Signed-Rank Test Results ........................................................... 117 4-6 Pmmotionall o pronoun 01 lg Price Tiers L' 1,9 Promotion Sr L10 Summary lnl Lila Bottler-by-P tllb Bottler-by-P tlla Bottler-by-P Lllb Bottler-by-P #13 AVOVA Ce l-ll Treatmean llSa OLS Regres “Tb OLS Regres “Sc OLS Regref t-lSd OLSRegreg i16a OLS Rem. “Pb OLS Regre We OLS Regre W’d 0L5 Regre ll? 4-18 H9 #20 ill Su 5.] s2 4-6 4.7 4-8 4-9 4-10 4-11a 4-11b 4-12a 4-12b 4-13 4-14 4-15a 4-15b 4-15c 4-15d 446a 4-16b 4-16c 4-16d 4-17 4-18 4-19 4-20 4—21 A-2 xii Promotional Overlap between Grocery Chains ............................................ 119 Probability of Promotional Overlaps Occurring by Chance .......................... 1 19 Price Tiers Used to Code Promotional Overlaps ......................................... 121 Promotion Schedules - Three Grand Rapids Chains .................................... 122 Summary Information on Battler-by-Pramatian AN OVA ........................... 124 Bottler-by-Pramatian Analysis of Variance - Ounces Measure .................... 124 Battler-by-Pramatian Analysis of Variance - Lag Ounces Measure ............ 124 Battler-by-Pramatian Analysis of Variance - Cents Measure ...................... 124 Bottler-by-Pramation Analysis of Variance - Log Cents Measure ............... 124 ANOVA Cell Means .................................................................................. 125 Treatment Contrast p-values between Mean Sales Responses ..................... 128 OLS Regression Summary, Strong Battlers - Ounce Measure ..................... 134 OLS Regression Summary, Weak Battlers - Ounce Measure ...................... 135 OLS Regression Summary, Strong Battlers - Cent Measure ....................... 135 OLS Regression Summary, Weak Battlers - Cent Measure ......................... 13 5 OLS Regression Summary, Strong Battlers - Modified Ounce Measure ...... 139 OLS Regression Summary, Weak Battlers - Modified Ounce Measure ....... 139 OLS Regression Summary, Strong Battlers - Modified Cent Measure ........ 139 OLS Regression Summary, Weak Battlers - Modified Cent Measure .......... 140 Summary of Crass-Promotion Elasticities Between Strong and Weak Battlers, Ounce Sales ................................................................................. 141 Summary of Cross-Promotion Elasticities Between Strong and Weak Battlers, Cent Sales .................................................................................... 141 Average Discounts for 2-Liter Products ..................................................... 143 Average Discounts for 12-Ounce 12-Packs ................................................ 143 Summary of Research Questions, Research Hypotheses, and Operational Hypotheses ................................................................................................ 148 CMA Price/Promotion Schedule ................................................................. 163 Per-Case Rebates under Three Promotion Conditions ................................. 165 an. I e‘. I i-i A4 .l-S so Bottler Prc 1990 Fran Michigan .. Cases of 6 Promotion increment Structure A-3 A-4 A-6 xiii Battler Promotional Activity ...................................................................... 168 1990 Promotion Schedule for Soft Drinks, Small Independent Grocery in Michigan .................................................................................................... 171 Cases of 6-Pack Cans, Price, Cast and Gross Margin under Three Promotion Conditions ................................................................................ 172 Incremental Break-Even Sales Changes Implied by Promotional Price Structure .................................................................................................... 172 Alb iii All Types of Promo Kinds of Quanti An Explanation A Promotion Pr Pricing Strategi Dimensions of J Nielsen Single- Lili Factors for Grand Rapids 5 Example Data Example Data Grand Rapids ' Hypothesized H‘rPOthesized pr(titration ..... HlPOIhesized Pr(”muons .. . Hl'Pothesized Pr OmOilons . . . Hypolhesized omolions .. ASSOClall0n o M13001} 0 Package Sta ReSldUajs fOr List of Figures 2-1 Types of Promotions .................................................................................. 6 2-2 Kinds of Quantity Discounts Implementable with Coupons ......................... 16 2-3 An Explanation of Postpramation Sales Troughs ........................................ 21 2-4 A Promotion Prisoner's Dilemma ................................................................ 25 2-5 Pricing Strategies Induced by Brand Interdependence ................................. 45 26 Dimensions of Promotional Price Discrimination ........................................ 55 3-1 Nielsen Single-Source Data Structures ....................................................... 65 3-2 Lift Factors for In-Stare Promotions .......................................................... 70 3-3 Grand Rapids Soft Drink Market Size, Year 3 ............................................ 72 3-4 Example Data from the UPCs Table ........................................................... 77 3-5 Example Data from the Stores Table .......................................................... 78 3-6 Grand Rapids Weekly Market Share by Bottler .......................................... 84 3-7 Hypothesized Strong Battier Sales on and off Promotion ........................... 101 4-1a Hypothesized versus Actual Time on Promotion - Counting Any Promotion .................................................................................................. 110 4-1b Hypothesized versus Actual Time on Promotion - Counting Feature Promotions ................................................................................................ l 10 4-lc Hypothesized versus Actual Time on Promotion - Counting Secondary Promotions ................................................................................................ 111 4-1d Hypothesized versus Actual Time on Promotion - Counting Combined Promotions ................................................................................................ 113 4-2 Association of Coke and Pepsi Promotions ................................................. 115 4-3 Association of Strong and Weak Battier Promotions .................................. 116 4-4 Package Size Share of Volume - Three Grand Rapids Chains ..................... 120 4-5a Residuals far Ounce Measure AN OVA ...................................................... 125 xiv Residuals Residuals Residuals Observed Strong ar Grocery Market I Discoun‘ Dixoun Density Names I 4-5b 4-6a 4-6b 4-7 4-8 4-9 4-10 4-1 1 4-12 4-13 XV Residuals for Log Ounce Measure ANOVA ............................................... 126 Residuals for Cents Measure ANOVA ........................................................ 126 Residuals for Log Cents Measure ANOVA ................................................ 127 Observed Market Response to Strong Battier Promotions .......................... 129 Strong and Weak Battier Share of Volume ................................................. 133 Grocery Chain Market Share Fluctuations - Grand Rapids .......................... 137 Market Level Demand Fluctuation - Grand Rapids ..................................... 138 Discounts for 12-aunce Products - ANOVA Residuals ............................... 144 Discounts for Strong Battlers, All 12-aunce. Products ............................... 145 Density Plat of Strong Battier Discounts .................................................... 146 Names of Promotional Incentive Payments Used in Soft Drink Marketing .. 169 A central issue nelange of shon- 9901a! 1990b recitation price maid inventor from some coml P59 Explanatio sew am they all its has been i “Planations int Commune Work Firsn cc Rid NeSiin 199 I. Introduction A. In 'n A central issue in promotions research is why businesses prefer to run a complicated melange of short-term promotions rather than simply cut prices (Blattberg and Neslin 1990; Lal 1990b). Five answers have been ofi‘ered: uncertainty about customer reservation prices, uncertainty about search casts, price discrimination, minimization of channel inventory costs, and competition. The truth is probably that promotions result fi'om some combination of these factors, but it has not been possible as yet to synthesize these explanations into a general theory. Both the explanations and relevant data are so new that they are just beginning to be brought together in empirical tests. Consequently, there has been little empirical feedback (Box 1976; 1984) to help incorporate these mphnations into a coherent promotion theory. Competitive promotion explanations are likely to be a good starting point for empirical work. First, competition is the most commonly used explanation for promotion (Blattberg and Neslin 1990, p. 105). Second, competitive theories focus on the most intensely promoted mature product categories, where competition is likely to play a larger role than uncertainty, price discrimination, or channel costs (Lal 1990b). Third, data to test competitive models of promotion (UPC single-source data) have just begun to be used to test promotion theories (Raju, Srinivasan, and Lal 1990). One caveat about single-source data is in order. Authors working with these databases have found that coeficients in store-level demand models often have "wrong” signs. The response to this problem in the literature (and single-source practice) has been to use r17 I conventional statistical tr ilaberg and George l l maimed equations" (Si Siiisanalogous to a ti regression“ (SLR). SL' ancients (SLR explc attaining the variabi silos a Luce choice rr research on methods It Beause the Variant Wm it is logically mm and covariam dimmed but it is al mom methodo m“ “Conventional M011 PTOblems ( B . . {PM 2 conventional statistical techniques where they work and to extend them when they do not. Blattberg and George (1990, p. 21), for example, developed a technique called "seemingly unrelated equations” (SUE) to exploit cross-equation information and improve estimates. SUE is analogous to a traditional estimation technique called " seemingly unrelated regression" (SUR). SUE, however, is nontraditional because it exploits similarity of cocficients (SUR exploits similarity of residuals). Allenby (1989) developed a method for constraining the variability of crass-elasticities based on the assumption that consumers follow a Luce choice model. Finally, Blattberg and Neslin (1990, p. 371) have called for research on methods to constrain coefi'rcients. Because the variance in product sales is at the root of single-source estimation problems, it is logically possible that covariance may cause problems as well. The variance and covariance properties of single-source data are just beginning to be discovered, but it is already clear that these properties can invalidate conventional econometric methodology. Notwithstanding this hazard, the present study is fi'amed in terms of conventional techniques because there is no ex ante evidence indicating estimation problems or improved methodologies for store-level soft drink data. Warm The conventional economic wisdom is that competition disciplines firms to produce goods well and cheaply (Stigler 1968, p. 5). Competitive explanations of promotion, however, emphasize tacit or covert cooperation within outwardly competitive situations. For example, loyal Coke customers will not (within the relevant normal range of prices) buy Pepsi, and vice versa. Price promotions to loyal customers convert contribution margins to consumer surplus. Brand loyal market segments give Coke and Pepsi an incentive not to promote in every periOd in order to minimize margin losses to loyal customers. A segment of customers who will buy on a price basis if no national brands are rooted may g‘se Cc bolsand’or these 0 a lure oi the suite Elliot incentises l station [mint coalition linger Mil and l nadcpendence‘ l hganin 1914 and incur based or promoted brands W brands r distomers to lOt herdependeno “WWW ism mum 3 promoted may give Coke and Pepsi a joint interest in promoting to black the sales of store brands, and/or these customers may force Coke and Pepsi to compete or more than their fair share of the switcher market. In this partially competitive and partially cooperative situation, incentives for Coke and Pepsi ”mirror" one another and may result in an implicit coalition. Implicit coalitions are a recent development in tournament game theory (F ader and Hauser 1988) and provide qualitative support for what is called the ”brand interdependence" literature in this study. This literature consists of a stream of papers that began in 1974 and have developed explanations of promotion fi'equency and depth of discount based on competition. Brand interdependence explanations are based on which promoted brands win customers and which nonpromoted brands lose customers. When national brands win customers from local brands during promotions and do not lose customers to local brands when they are not promoted, an asymmetric pattern of brand interdependence exists. Because of this draw of sales from local brands, the brand interdependence literature argues that national brands like Coca-Cola and Pepsi have a joint incentive to alternate promotions to prevent competitive encroachment (Lal 1990a). C. i ' of th The purpose of this study is to subject the empirical implications of the brand interdependence literature to the strongest tests possible using the best data yet available. This study is potentially significant for the advancement of theory because the implications of a leading edge stream of research will be tested with data of unprecedented quality. Successful results would support what, to many, is obvious but has never been tested strongly enough to be called proven: that large national brands promote for a mixture of competitive and cooperative reasons. Conversely, unsuccessful results will grideaplace to star irsnininatiort invent: A second theoretic risoli drink bottlers. house they are in a reciandising than i iPC codes and nati ash iPC to its bot ”Wong bottlers fillers 0f Sliong a A third firearm. the discussion of ti Emmi“ in Panic [Embers and Wt fleas iKuma; ar 4 provide a place to start in testing competing explanations of promotion (price discrimination, inventory carrying costs, and so forth). A second theoretical contribution of this study is to examine the competitive dynamics of soft drink bottlers. Battlers are unique entities in consumer package goods marketing because they are in a position analogous to wholesalers, yet they play a much larger role in merchandising than is typical of wholesalers. Single-source data typically track sales by UPC codes and national brands within geographic markets. The present study, by relating each UPC to its bottler, tracks total bottler (ounce and dollar) sales within grocery chains. Measuring bottlers sales within grocery chains allows the interdependence between bottlers of strong and weak brands to be brought into sharp focus. A third theoretical contribution of this study is to draw several anomalous results into the discussion of the promotions literature in general and the brand interdependence literature in particular. These results are the asymmetry of competitive interactions (Blattberg and Wisniewski 1989), the asymmetry of complementary and substitute sales efi'ects (Kumar and Leone 1988; Walters 1991), and the stockup phenomenon (Litvack, Calantone, and Warshaw 1985). These results are "anomalous” in that while they have been empirically demonstrated beyond a reasonable doubt, they do not fit comfortably into microeconomic or marketing explanations of how markets work. Drawing these efi‘ects together is a first step in understanding some otherwise baming market responses to merchandising (such as the presence or absence of troughs alter promotions). Methodologically, the data obtained for this study are unique. Previous work has evaluated brand interdependence using scanner (but not single-source) data (Blattberg and Wisniewski 1989; Raju, Srinivasan, and Lal 1990; Lal 1990b; Walters 1991). This study is the first to group single-source data by the effective competitive actors and base promotion measurement on the promotion structure used by these competitors. this study is potenti ring category in supe h1991 soft drink sales .1idsenl992, p. 3). S ism of consumer pr inside influences by g because the industry is territories by the 1980 marathon ever sin. tith this increase. On PMS - has been till-limes 1987, Pa link are the most fre Watch recalled de Shoemaker 1991, p, g worked to Rise pr 0m mm °fpt0moti bediscussed in Chap1 This study is potentially significant for public policy as well. Soft drinks are the best selling category in supermarkets, accounting for 4 percent of sales (F eigner 1988, p. 187). In 1991 sofi drink sales were estimated at $9.75 billion, 94 percent in branded products (Nielsen 1992, p. 3). Soft drinks have been called "one of the great dream markets in the history of consumer products in the United States" because they have "responded to all outside influences by growing” (Tedlaw 1990, p. 26). Soft drinks are interesting not only because the industry is large and profitable. Granted the right of exclusive franchise territories by the 1980 Interbrand Competition Act the industry has been increasing in concentration ever since. Rampant price fixing (Galvin 1990, p. 27) has been coincident with this increase. One form of promotion in the soft drink industry - calendar marketing agreements - has been the subject of several news stories insinuating antitrust violations (60 Minutes 1987; Pasztor and Reibstein 1987; Consumer Reports 1988, 1991). Soft drink are the most frequently promoted products in grocery stores and have the most accurately recalled deal frequency and magnitude among consumers (Krishna, Currim, and Shoemaker 1991, p. 9). Instead of reducing prices to consumers, competition may have worked to raise promotion spending (Lodish 1986a, p. 36). The resulting patterns and fiequency of promotion raise interesting antitrust questions, the implications of which will be discussed in Chapter 5. Sales p: shsequen WI p it and retail. Trat i-D inc” Pifidur; Elicia; cm II. Literature Review and Conceptual Development A. Inr 'n Sales promotion emerged in the 19805 as a significant issue in marketing practice and subsequently developed into an active area of academic research (Blattberg and Neslin 1990, p. xi). Figure 2-1 shows the three main kinds of sales promotion: trade, consumer, and retailer. Figure 2-1 Types of Promotions Manufacturer > Trade Trade Promotions Consumer Retailer Promotions Promotions 9 4 Consumer Source: Blattberg“ Nellin(l990, p. 4). Trade promotions, ofi‘ered by manufacturers to wholesalers and retailers, are designed to increase product sales by providing trade moneys as compensation for supporting product promotions (Walters 1989, p. 253). Case allowances (that is, discounts hour the regular case price), advertising allowances, and display allowances are examples. Consrnner promotions are ofi‘ered by manufacturers and include coupons, refunds, and swath Pace muons manic resez minions throng mi sales and sham (Blanbe‘ l. The K Emmaxes of i is as $65 bum am $43 billior :34?) In 197: ifthe average I! '2‘ “WM 13 Mb aWroxir has and Ga iPTOmotion g, News 195 m Imprm'emen “Tip-3m OVer 3mm” him I“ of im'emo in“ ‘by‘Pmc 7 bonus packs. Price cuts, in-store coupons, and weekly specials are examples of retailer promotions. Academic research in this area has been stimulated by the relentless growth of promotions through the 19805, by the increasing availability of precise disaggregated data on retail sales and promotion, and by theoretical work in economics and consumer behavior (Blattberg and Neslin 1990, p. xi). 1. The Rapid Growth in Promotions Estimates of how much is spent annually on promotion in the United States range as high as $65 billion (Walters and MacKenzie 1988, p. 51). A more accurate figure may be about $43 billion, compared to around $30 billion spent on advertising (Lal 1990a, p. 247). In 1978 advertising accounted for 42 percent and trade promotion for 58 percent of the average marketing budget for consumer packaged goods manufacturers; by 1988 the respective figures were 31 percent and 69 percent. Sales promotions, for example, absorb approximately 25 percent of sales person and 30 percent of brand manager time at Proctor and Gamble (Buzzell, Quelch, and Salmon 1990, p. 141). Spending on all types of promotion grew at an annual rate of 12 percent through the 19803 (Blattberg and “fisniewski 1989, p. 291). 2.NDos Improvements in the price and performance of information systems and laser scanning equipment over the past decade have greatly enhanced electronic measurement of consumer behavior (Eskin 1987). Optical scanners enable retailers to reduce the labor costs of inventory tracking (Marketing News 1981) and to account for profitability on a product-by-product basis (Stern and El-Ansary 1988, p. 513). Scanners reduce costs by renting clerk cher p28) Wren inter sable to manuf: tree of retailer motions Scanner dara Resources. Inc decrrom'c track iii been lmom 2liorrnnrion. E ”Wham and rational broad ihi'iéie-Sonrce mm of Sn mi“ gTocer “iii Ciraerbi mamSeine: Tisearch 0n 8 reducing clerk checkout error and by increasing checkout speed (Totten and Block 1987, p. 28). When integrated with advertising and promotion information, scanner data are valuable to manufacturers as well, permitting them to calculate market response, the degree of retailer compliance with promotional agreements, and the total profitability of promotions. Scanner data in the United States are available from two companies. Information Resources, Inc. (IRI), was started in 1977 (Gage 1983) to capitalize on the potential of electronic tracking to improve market data collection (Rohm 1986). A. C. Nielsen long has been known for supplying advertising ratings as well as marketing research information. Both companies are developing "single-source” data supply capability (Abraham and Lodish 1990) by providing complete "causal" (in-store promotion) and national broadcast advertising information along with scanner data. To create complete single-source databases, IR] and Nielsen have established "instrumented markets” in a number of small to medium-sized cities. In these markets scanners are installed in all major grocery stores, there is a high level of cable television coverage, and there is split cable capability (Guadagni and Little 1983, p. 205), permitting control over which advertisements cable subscribers see. Instrumented markets allow relatively realistic field research on advertising and promotion to be conducted. Just as scanner data enable retailers to plan their assortrnents, integrated scanner and promotion data give manufacturers the potential for greater cost control and increased productivity in promotion spending. Research suggests that managing promotions is so complicated that managers focus on such intermediate issues as coupon redemption rates and advertising costs rather than total profitability (Neslin and Shoemaker 1983, p. 362). The reason may be that traditional accounting systems provide data for evaluating intermediate issues but not total promotion profitability. Information on the latter is very ricuh and exp6 :2: effects of mo can be used mrh as. allow manag total profits and t confidence. Consumer pro has been done by liomution (Km fixed costs for pr P-il. Because 0 more affordable. 106$ warehouse < Web problerr 9 dificult and expensive to prepare (Kruger 1987, p. 147). Consequently, although costs and efi‘ects of most promotions are examined, the analysis does not yield estimates that can be used with any real confidence (Blattberg and Neslin 1990, p. 391). Single-source data allow managers to shift the focus of promotion evaluation from intermediate issues to total profits and to develop market response estimates which can be used with greater confidence. Consumer promotions through coupons are a good example. Evaluation traditionally has been done by integrating shipments or warehouse withdrawal data with promotion information (Kruger 1987 p. 141). This is a tremendous undertaking because of the large fixed costs for processing and storing massive quantities of information (Sanford 1989, p. 4). Because of economies of scale and scope, [RI and Nielsen make integrated data more afi‘ordable. In addition, integrated scanner databases provide better information than does warehouse withdrawal information because they solve three long-standing marketing research problems. First, single-source data precisely measure the start and finish dates of promotions (Wilson, Newman, and Hastack 1979, p. 41). Second, scanner data allow measurement of a more complete set of the "causal" merchandising and competitive activities that theoretically afi'ect customers at the time of purchase (Guadagni and Little 1983, p. 205). Third, these data largely solve the bias caused by using surrogates or measures remote from consumer purchasing to evaluate promotions. This third point merits some elaboration. A manufacturer's shipments, particularly during trade promotions, are subject to biases that vary across products, markets, and time (Kruger 1987, p. 149). One estimate is that $5 billion a year in grocery merchandise is bought in regions ofthe country where a manufacturer is ofi‘ering discounts and then is ”diverted" to higher priced regions (Buzzell, Quelch, and Salmon 1990, p. 143). Another bias is caused by ”forward" and/or ”bridge” buying. Retailers' purchases are based on . .l i ‘5»? min 1. Jul“ i9?” § Lat”. Inc: IO economic order quantity calculations, which include storage costs and/or capital carrying costs. Since price discounts by manufacturers can be used to offset these costs, retailers often buy more inventory than is needed to support a consumer promotion, known as ”forward buying.” Ifthey buy so many units that they do not need to purchase again until the next promotion, they are said to be ”bridge buying,” or buying "deal-to-deal." The units purchased beyond those needed to support consumer promotions then reduce or ”mortgage" normal sales. Thus, when shipments are used as a surrogate for sales, promotion analysis can be biased by diversion, forward buying, and/or bridge buying (Abraham and Lodish 1987a, p. 108). The gap between the agreement a retailer makes to price, display, or advertise the product and the retailer's execution of the agreement creates another surrogate variable because retailers do not "pass through" all the promotion support to which they agree. The contractual requirements for promotions, then, are surrogate measures for retailer performance. It is estimated that 65 percent of price reductions, 50 percent of in-store displays, and 80 percent of contracted size advertisements are passed through (Walters 1989, p. 263). Consequently, promotion evaluations based on integrated shipments and retailer accounting data are likely to be biased. To summarize, because scanners measure "actual consumer out-the-door takeaway" (Sanford 1989, p. 4), the biases inherent in traditional measurements are avoided. The more disaggregated the data, the better are the prospects for seeing the interrelationships between sales and causal variables (Ean 1985, p. 33). Because single-source databases are maintained by an independent organization, biases resulting fiom the gap between commitments and actual execution are also avoided. Reducing or removing these biases greatly increases the accuracy and reliability of estimates of promotion effects. Furthermore, economies of scale and scope reduce the cost of data preparation. Sound The man 1 ‘3'; Eisner search Lianne mien 1] promotion evaluation is coming within reach of many manufacturers as they come to rely more heavily on sales promotions. At the same time, the market for promotion evaluation is expanding along with academic research into this field. 3. Theoretical InnovaLtions The central issue for promotions research is why companies promote rather than permanently change their prices (Blattberg and Neslin 1990, p. 83). This question continues to challenge academicians and practitioners alike (Lal 1990b, p. 428). Five answers have been ofi‘ered: (1) uncertainty about reservation prices; (2) uncertainty about search costs; (3) difi‘erences in inventory carrying costs between levels in the marketing channel; (4) the opportunity to price discriminate; and (5) competition. The literature review in the following section discusses the two uncertainty explanations together; the remaining explanations then will be examined. B. Literature Review: Em Use Sales Promotions? 1. n ' ut Res onse Lazear (1986) based his explanation of promotions on the retailer‘s knowledge of demand for a product.‘ Suppose there is one buyer in a market and that the buyer is willing to pay price V for a product. The retailer does not know the buyer‘s reservation price V but has a prior distribution of buyer price sensitivity. {The retailer's problem is to set the price, R, to maximize expected profits over the number of chances s/he will have to sell the product. Ifone time period is available, and if the prior distribution of the market's ' Discussion in this section is based on Blattberg and Neslin (1990), pp. 83-86; 92- 95. amnion prices is not ore-half the area under is under the distribut" iadrnamic program the are under the prob period optimal strategvl Luear‘smodel is 5'1; operation for the bel m reduce price later Iitsons of competitiv. reference pOints (Nag mom“ reservation Second, W5 12 reservation prices is uniform, then the retailer's optimal strategy is to price the product at one-half the area under the distribution. This strategy yields a profit of one-quarter the area under the distribution. Lazear also models two time periods, shows that the solution is a dynamic programming problem, and indicates that the optimal prices are two-thirds of the area under the probability distribution for period 1, one-third for period 2. The two- period optimal strategy yields profits of one-third. Lazear‘s model is significant for three reasons. First, it provides a theoretical explanation for the behavior of retailers who price a product high when first available and then reduce price later in the product's life. This strategy is often recommended for reasons of competitive substitution (Dean 1970) or to establish consumer price-quality reference points (Nagle 1987). Lazear‘s point is that the retailer‘s uncertainty about customer reservation prices will prompt this strategy. Second, Lazear‘s approach shows that a two-price strategy produces higher profits than a one-price strategy (one-third versus one-fourth). Third, Lazear's model posits two empirical expectations. (1) In a two-period scenario, products with few buyers should have lower prices in the first period than do products with many buyers. The larger the population of buyers, the greater is the chance of finding buyers with reservation prices above the price in the first period. (2) Products that rapidly become obsolete should have lower prices in the first period and higher mark-downs in the second period than do products with a longer life. Another uncertainty used to explain promotion relates to search costs. Search costs determine the extent to which consumers will canvass markets for the lowest prices and product variety. Because consumers are heterogeneous, they have different costs of search, difl‘erent attitudes toward the costs of search, different abilities to process .53 nitration and‘ or difit picelei'els in equilibriu ndel based on “hetero isomers. informed an stopping by uninforme high fixed cost for shell shipping is more diftic are willing or unable arses of differential it Vanmwso p. 65 Muse COIISumel’s es Warmers, Varian m tall "indirect inOmia m“ “"95. Low till] have a larger m; 1 3 information, and/or different budget constraints. Thus, a market may support multiple price levels in equilibrium. Salop and Stiglitz (1977) derive a spatial two-price equilibrium model based on ”heterogeneity of consumer rationality" (ibid., p. 493) and two types of customers, informed and uninformed. They assume declining average costs and random shopping by uninformed customers. The first assumption is reasonable, since the retailer’s high fixed cost for shelf space falls as more units are sold. The assumption of random shopping is more difiicult to accept without an explanation of why uninformed customers are unwilling or unable to become informed (Blattberg and Neslin 1990, p. 94). The causes of difi‘erential information costs are not investigated by Salop and Stiglitz. Varian (1980 p. 651) criticized the Salop and Stiglitz model as being implausible because consumers eventually will learn which stores are low or high price. Rational consumers, Varian maintains, should be able to use what Salop and Stiglitz (1977 p. 509) call ”indirect information,” obtained fi'om past purchase experience and fi'om observable market shares. Low-price stores, which serve both informed and uninformed customers, will have a larger market share (Salop and Stiglitz 1977, p. 509). One factor not examined by Varian or Salop and Stiglitz is the consumer's opportunity cost of time (Becker 1965). Some retail stores might be able to charge persistently higher margins if they locate near consumers with high time costs. Shopping at more expensive stores might be rational for some customers because the time they save not searching for small savings is compensated for by time spent earning high wages (ibid., p. 503). This line of argument is also used to explain why prices are set on a take-it-or-leave-it basis (Scitovsky 1990, p. 139). Although the Salop-Stiglitz (197 7) model does not explain promotions, it does show that when customer heterogeneity is matched by retailer response, such as store location, V”? '_l ‘7 rents car mince itrses o for using v rm acetone sees the 121101; Pros man PTOi fQU (I) 14 profits can be captured. Customer uncertainty about where to find the lowest price introduces the potential for retailers to increase profits. The Salop-Stiglitz (1977) model focuses on spatial variation, whereas Varian (1980) emphasizes the time dimension; in doing so he shows that capturing profits from price uncertainty is a plausible explanation for using promotions. Varian (1980) and Salop and Stiglitz (1977) assume informed and uninformed customers. Although Salop and Stiglitz do not address promotions, Varian (1980, p. 652) sees them as a mechanism retailers use to randomize their pricing. This randomization prevents uninformed customers, who may happen to find a low price in a given store, from using indirect information to shop more eficiently in the future (Blattberg and Neslin 1990, p. 95). This mechanism then separates informed customers from the uninformed, who do not search for information. Furthermore, randomization need not occur all the time. If consumers cannot anticipate where or when an item will be promoted (ibid), retailers can use promotions to capture incremental profits from consumers. Lack of information is one promotion explanation. From the retailer's side, not knowmg customer reservation prices for goods that can be sold over multiple periods provides an incentive to price high when a product is introduced and lower as the product matures. Customer uncertainty, caused by search costs and consumer heterogeneity, provides another reason for promotions, namely, price discrimination. 2. Price Dim’ ‘n_ation In reviewing the economics literature on quantity discount structures, Dolan (1987) found three motivations for these structures: (1) to achieve perfect price discrimination, (2) to achieve partial price discrimination against heterogeneous customers, and (3) to thence the buyers o with sales promotic One form of quam lB'ntberg and Neslir pores are quantity di siege and a charge Mpan pricing asp to the fDiEd charge “11% COUpons avoir “he“ COUpons 2 Met the custorr or to a ml‘limum q mother fonn of qu is“ F1sure 2-2) - 15 influence the buyer's ordering pattern. All three of these motivations can be used to explain sales promotions. One form of quantity discount used in trade promotion is cumulative volume rebating (Blattberg and Neslin 1990, p. 317). Coupons can serve as quantity discounts. Two-part prices are quantity discounts (Dolan 1987, p. 3), that is, customers pay two prices (a fixed charge and a charge per unit) to consume a single product (Nagle 1987, p. 165). The two-part pricing aspects of coupons may be disguised by their face value and by the fact that the fixed charge only applies to a subset of customers. In other words, customers using coupons avoid the fixed charge and pay according to a uniform price schedule per unit, whereas customers who do not use coupons pay under a two-part tariff. When coupons are limited to a minimum quantity that is greater than the amount of product the customer would buy otherwise (Neslin, Henderson, and Quelch 1985, p. 153), or to a maximum quantity that is less than the amount the consumer would buy otherwise, another form of quantity discount, a two-block tarifi‘, can be implemented with coupons (see Figure 2-2). The advantage of a two-block tarrif is that different customers can be charged different marginal prices, whereas all customers pay the same marginal prices in a two-part tarifi‘ (Dolan 1987, p. 3). Coupons provide a means to implement three kinds of quantity discounts.Of the three motivations for quantity discounts listed by Dolan (ibid.), perfect price discrimination has not been investigated in the literature because it is unrealistic. To achieve perfect price discrimination, either homogeneous buyers or costless and complete control over the marketing mix would be required. The second motivation, partial price discrimination against heterogeneous customers, has been the main focus. There is some evidence, however, that the third motivation may be worth investigating. [mil 3.} m to Liidea, “I l6 Influencing buyer ordering patterns is an emerging topic in promotion research. In the soft drink industry, weekly promotion of major brands apparently prompts consumers not to stock up because they perceive that the next promotion will occur soon (Krishna, Currim, and Shoemaker 1991, p. 5). Thus, purchase patterns may neutralize a seller's ability to price discriminate against all but the most haphazard of shoppers (ibid.). Figure 2-2 Kinds of Quantity Discounts Implementable with Coupons Coat to Cost to Customer Customer Fixed Charge. F { Quanttty Quantity (3) Unttorrn Pnce Schedule (b) Two-Part Tami Cost to Customer | Block 2 : Price = p, I Bloctt r : Pflc. . D" ii Quantity 1c) Tweeatoctr Tantt Source: Dolan (1989, p. 3). Partial price discrimination takes place when market conditions allow firms to exploit heterogeneity by separating groups of customers with different demand elasticities. Airlines use time between reservation and flight to price discriminate between vacation travelers, who are price sensitive, and business travelers, who are less so (N agle 1987). The motivation usually cited for partial price discrimination is the desire to exploit the inelastic demand of small buyers for small quantifies (Dolan 1987, p. S), but the inelasticity may arise fiom many factors besides buyer size. Nara: research :sed a d coupon nor there i [\'C EDI FEB AGl Qfilfi. l7 Narasimhan (1984) evaluated demand elasticity differences between coupon users and nonusers fiom an econometric perspective. Diary panel data from National Panel Data research for 1,000 consumers across 20 product categories were analyzed. Narasirnhan used a demand equation to isolate any difl‘erence in price elasticity of demand between coupon users and nonusers: Ln(QTYi) = aan(PRICEi) + a2Ln(INCOMEi) + a3Ln(FAMSIZi) + a4Ln(EDUFEMi) + a5Ln(FEMEMPi) + a6Ln(AGEFEMi) +ei, where i varies across households and QTYi = annual quantity purchased; PRICE; = average price paid; IN COMEi = annual household income; FAMSIZi = number of household members; EDUFEMi = education level of female head of household; FEMEMP; = employment status of female head of household; AGEFEMi = age of the female head of household; and ei = a disturbance term satisfying OLS properties. Narasirnhan (ibid., p. 137) estimated these variables across 20 product categories for coupon users (those who made at least one purchase with a coupon) and nonusers. Four product categories that had statistically significant (or=.05) differences (paper towels, ready-to-eat cereals, dog food, and cat food) on nonprice variables were investigated but not omitted from the analysis. The data for all 20 product categories were then pooled, and a dummy price variable was added to the demand equation. The dummy variable is a7[5iLn(Pricei)], where Si is 0 for coupon nomrsers and 1 for coupon users. The results confirmed that coupon users are more price sensitive (that is price elastic) (Hozcr7<0) than nonusers in 16 of 20 product categories. The exceptions were paper towels, hair coloring, ready-to-eat cereals, and dog food. Narasirnhan (ibid., p. 138) hypothesized that, besides economic variables (household income and product price), tilts iaed tens hoes (Oh? I Trifle heel Ci Dbl 0dr l 8 family size and the opportunity cost of time would have a positive effect on coupon use. Since direct measures were not available, proxies (employment of husband and/or wife and the presence of children under age 18) were used. A t-test on mean quantity purchased ”revealed that users of coupons have a significantly higher average consumption than nonusers" (ibid.). The regression model developed was (ibid., p. 139): COUPONi = aoaNCOMEi) + (12(INCOMEi)2 + a3(EDUFEMi) + a4(FEMEMPi) + (15(QTYi) + or6(QTYi)2 + (17(DUMCHDi) + agoCCli + agOCCZi + 8i, where INCOME, EDUFEM, FEMEMP, and QTY are the same as above, i varies across households, and CUPUNT = the quantity bought using coupons; DUMCHD = 1 without children below 18, 0 otherwise; OCCI = I if husband is a professional, 0 otherwise; and OCC2 = 1 if husband is blue collar, 0 otherwise. Only partial support was found using the proxy variables (ibid., p. 146). From the signs of the coefficients and patterns of significance, Narasimhan (ibid., p. 142) deduced that the income effect on coupon use is nonlinear, that coupon use is likeliest among middle- income households, and that female education and presence of children positively affect coupon use. One variable, female employment, had the wrong sign. Narasimhan (ibid.)assumed that working women would be less likely to redeem coupons, but in fact they are more likely to be coupon users . Narasirnhan's article is the foundation of a stream of research on coupons and, more generally, sales promotions as price discrimination mechanisms. The data strongly supported the difl‘erence in price elasticities between coupon users and nonusers. Narasimhan, in the tradition of utility maximization, hypothesizes that rational mechanisms (opportunity costs) rather than preferences underlie the difl‘erence in demand elasticity between the two groups. ”That is, the difl'erence in demand elasticities is driven by m '\ “rid hen? er- 19 opportunity cost differences and not due to taste differences" (ibid., 132). Unfortunately, the proxy variables available did not rule out either preferences or opportunity costs. Narasimhan's approach precluded in-depth analysis of how coupons function as mechanisms of price discrimination. A coupon conceivably might discriminate because of a constellation of constraints that affect consumer price elasticity of demand. For example, a margarine coupon that requires a minimum purchase of two packages might, in the aggregate, increase profits for a manufacturer (or retailer) over a single low price by separating market segments. Price-sensitive customers will stock up, while nonprice sensitive customers will not take advantage of the offer. For example, large families might take advantage of the offer while single people might not because they cannot use two packages of margarine before they spoil. Against an economist's expectations, employed, married, high-income mothers may clip and use coupons from the newspaper because they can combine reading the paper with commuting or relaxing at work. There are many informational, economic, and time constraints that potentially could influence price sensitivity. Narasimhan's approach does not account for all rational, taste, or ideosyncratic variables. Instead, it validates the importance of connecting indicators of price sensitivity to promotion media. In subsequent research, Jeuland and Narasimhan (1985) generalized the price discrimination approach fiom coupons to sales promotions. The two consumer subpopulations hypothesized were customers' high or low inventory holding costs (ibid., p. 296). Because some people are constrained in the quantities they can store, retailers can stimulate demand among customers with low holding cost through periodic deals. In efi'ect, retailers maintain a price for each market segment by using consumer holding costs to separate market segments. A recent extension of the price discrimination approach has been made by Hoyt, Calantone, and di Benedetto (1990). This extension is discussed in 20 detail below with the literature on promotion depth, but in brief the authors determined suflicient conditions for promotions to price discriminate. They argue that at least two dimensions must be used to segment markets. That is, promotion will not increase profits unless something keeps nonprice-sensitive customers from purchasing at the same time as price-sensitive customers. In the above margarine example, the two dimensions of family size and price sensitivity were implicitly combined. Hoyt, Calantone, and di Benedetto go farther and show that efl‘ective price discrimination strategies can be highly counterintuitive. Although they did not empirically test their model, Jeuland and Narasimhan (1985, p. 304) predicted that " given a heterogeneous population, we should observe some consumers to stockpile when a deal is offered, and others less so - or even not at all." This hypothesis was lent support by an empirical study done by Litvack, Calantone, and Warshaw (1985), who investigated the efl‘ects of price increases and reductions on sales volume. They hypothesized that products can be characterized as stock up or nonstock up depending on response to nonpromoted price changes (ibid., p. 9). A before, during, and after experiment was conducted in stores to determine the efi‘ect of 20 percent price increases and decreases on 72 products. The researchers found that products characterized as stock up goods by retailers increased in standardized unit sales by 54.95 percent as a result ofa 20 percent price decrease, compared to an increase of 10.55 percent for nonstock ups (ibid., p. 19). Stock ups dropped 24.10 percent in standardized unit sales when price increased 20 percent, compared to a decrease of 7.6 percent for nonstock ups. At the least, the large unit volume response of stock up goods to the 20 percent price decrease indicates presence of the low holding cost consumers hypothesized by Jeuland and Narasimhan (1985). My 1 ‘ respects {Birth goods the inc sock 21 It is worth digressing for a moment to discuss the Litvack, Calantone, and Warshaw study (1985) because it may provide a structural explanation of some baffling customer responses to promotions. There has been very little theoretical work to date in this area (Blattberg and Neslin 1990, p. 360), and the distinction between stock up and nonstock up goods provides a building block toward a theoretical explanation. When considered with the incremental consumption efi‘ects of having an inventory of a product in the home, the stock up and nonstock up distinction provides an intuitively appealing explanation of why products do or do not have postpromotion sales troughs. Figure 2-3 shows hypothesized sales patterns during and after a promotion across these two dimensions. Figure 2-3 An Explanation of Postpromotron Sales Troughs I Type of Good Pattern of Con-twat Stock up Non-cock up 319 spike anal]. spike 1:an No Trough No Trough 819 Spike Small Spike Not: rant-mam 8.19 Trough 8.311 Trough Soft drinks and potato chips are examples of stock up goods with incremental consumption efl‘ects. When a customer encounters a price promotion on these products s/he will stock up, causing a large ”spike” in demand. Once in the home, the inventories of these products stimulate consumption, reducing the postpromotion "trough” that would occur. Cat food is an example of a stock up good that does not have incremental consumption effects. Even though a customer buys extra cat food when it is on sale, in-home inventory will not cause a cat owner to feed a cat more. In this case there should be a big spike during and a big trough after the promotion. A nonstock up good with incremental consumption efl‘ects might be fiesh baked goods. While the sales of these items might 22 increase fi'om promotion, their perishability prevents inventory fi'om affecting future purchases. Instant coffee or mustard are examples of nonstock up goods with no incremental consumption effects. When these items are promoted more will be purchased, but not too much more; each unit takes a while to consume, and the ”spikes" in demand for these products will not be big. Since units of coffee or mustard sold on promotion may ”mortgage” or ”cannibalize" firture full price sales, and household inventory will not stirrurlate additional consumption, some trough efl‘ect should be observable. The distinction between stock up and nonstock up goods for which Litvack, Calantone, and Warshaw (1985) argue provides a first step in theoretically explaining postpromotion sales troughs. To summarize the price discrimination literature, the ability of sales promotions to price discriminate has been strongly supported in the case of coupons (Narasimhan I984). The extension of the concept by Jeuland and Narasimhan (1985) to retailer dealing behavior is theoretically appealing, but much empirical work remains to be done. In particular, the observed difference in price elasticity between customers who are and are not promotion sensitive remains unexplained. Narasimhan (1984, p. 131, 138) argues for such microeconomic causes as the opportunity cost of time, but there are other possibilities. Such consumer behavior models as prospect theory, classical and operant conditioning, and price perception theory have only begun to be explored (Blattberg and Neslin 1990, p. 61). W The cost to retailers of storing and financing inventory is another explanation for the use of sales promotions. In 1981 Blattberg, Eppen, and Lieberman published a study of promotions based on data fiom the Chicago Tribune panel over the period 1958-1966. Steel i117: Lies foo: 23 Aluminum foil, facial tissue, liquid detergent, and waxed paper were the products studied. Stockpiling due to promotions occurred in all four categories, but for waxed paper and aluminum foil the response was almost twice that for facial tissue and liquid detergent. The researchers attributed the cause to inventory cost-shifting. Their work difl‘ers from the study by Jeuland and Narasimhan (1985) on price discrimination because the latter focused on heterogeneity in consumer holding costs. Blattberg, Eppen, and Lieberman (1981) hypothesized that retailers ofi’er promotions because their inventory holding costs are higher than the holding costs for some consumers. Retailers ofl’er promotional price reductions so that those consumers will stockpile; customers with high holding costs cannot do so and pay a higher price, which compensates the retailer for holding inventory. In effect, promotions shifl inventory to the lowest holding cost point in the marketing channel. Retailers are willing to lower prices and shifl inventory to the customer because they lower their costs even more. In contrast, Jeuland and Narasimhan (1985) assume that the retailer has the lowest holding cost in the channel, and thus they use consumers with high and low holding costs as a pricing segmentation variable. The fact that customers with high holding cost will buy limited quantities gives retailers some degree of monopoly power. Blattberg, Eppen, and Lieberman (1981) tested their model against a fi'amework that assumed promotions are used to induce product trial for a band; that is, manufacturers woo new consumers by lowering the level of price risk (ibid., p. 117). Blattberg, Eppen, and Lieberman refer to this as the "information explanation” of promotions. Because the panel data do not directly quantify holding costs (ibid., p. 123 ), the researchers evaluated holding cost and information indirectly. The three proxies were (1) whether consumers stockpile, (2) the sales volume of the most fiequently promoted products, and (3) the 24 packaging size of the most frequently promoted products. Table 2-1 contains the researchers' hypotheses and findings. Table 2-1 Inventory Cost versus Information Prediction of Prediction of Intonation Holding Costa Ruiz-ion m1 Model Finding Stockpiling No You You Volume of Frequently Low High High Promoted Products 8120 of Frequently Small Large Largo Promoted Products Source: Blattberg. Eppen, and Lieberman (1981 p. 128). Stockpiling was operationalized in two variables: increase in quantity purchased when on deal and mean number of days between purchases. The data on pooled stock keeping units indicated higher purchase quantities in eight of nine cases (ibid., p. 124). The timing analysis indicated a significant increase in time between purchases (ranging fi'om 23 percent to 36 percent) for all four product categories (ibid., p. 125). This test of the cost-shifting hypothesis is, however, weak. With respect to the information model, how does one distinguish between sales to new consumers induced to try a product and existing customers who stockpile? The researchers seem to assume that either the cost-shifting or the information model operates, but not both. Faced with a choice between models, they clearly prefer cost-shifting, but this choice seems arbitrary without fiuther justification. It is possible for both the cost-shifting and information mechanisms to work simultaneously. Subsequent research (Walters 1989, p. 265) on the relationship between sales volume and promotion fiequency, conducted on a broader sample of product categories, contradicts the findings of Blattberg, Eppen, and Lieberman. The cost-shifting hypothesis also does not account for the efi‘ects of price changes and inventory location (Blattberg 25 and Neslin 1990, p. 90). There is evidence that, for some product categories, presence in the consumer‘s pantry leads to incremental consumption (Williams 1983). Economic theory suggests that ”people will not buy less, and usually buy more, of a commodity when its price falls” (Stigler 1987, p. 22). Finally, for service industries like fast food, where finished products are not inventoried, promotions may increase market size rather than shift costs (Lodish 1986a, p. 40). 4. Competition Probably the most common explanation for why companies promote is the single- period prisoner's dilemma (Blattberg and Neslin 1990, p. 105). The prisoner's dilemma is a two-person noncooperative game with payofl‘s (see Figure 2-4) structured such that each player has an incentive to defect regardless of whether or not s/he thinks the other player will cooperate (Axelrod 1984, p. 9). The ”dilemma" is that players taken separately have the incentive to defect, while taken together they have the incentive to cooperate. The prisoners dilemma is a conflict between unilateral and group incentives (Fader and Hauser 1988, p. 554). This kind of conflict can take place at any level of the marketing channel 811d, consequently, can be used to explain all forms of promotions (trade, retailer, and consumer). Figure 2-4 A Promotion Prisoner’s Dilemma Whether a game is one-shot or infinitely repeated strongly affects human behavior in prisoner's dilemmas. In the context of a finite number of periods it is rational for each player. to ”defect.” On the next-to-last move, neither player will have an incentive to 26 cooperate, since both can anticipate a defection by the other player on the very last move (Axelrod 1984, p. 10). This line of reasoning implies that in a game of known length, any potential cooperation will unravel all the way back to mutual defection on the first move (ibid.). The finite-period prisoner’s dilemma leaves a rational actor no choice but to defect. Ifconsumer promotions are viewed as finite-horizon prisoner's dilemmas, then retailer interests are put in complete opposition. Each retailer tries to increase sales and profits by running promotions (Walters 1991, p. 17), but if all retailers promote simultaneously (other things equal), promotions will only reduce profits. In this case the retailer knows that promoting will only reduce margins, but if s/he does not promote and other retailers do, unit volume will be lost to customers switching to promoting retailers. When the time horizon of a prisoner's dilemma is infinite or unknown, potential for cooperation is created (Axelrod 1984, p. 11). Ifconsumer promotions are viewed in this context, then retailer interests are not put in total conflict. Because retailers will face each other for the foreseeable firture, choices made today influence later choices (ibid., p. 12). Axelrod calls this the "firture casting a shadow upon the present" (ibid.), and in effect a feedback loop is created. Each player takes the opponent's reaction into account in choosing what to do. While not having the power to end ”the game," retailers in the infinite horizon case have the power to minimize the damage promotions cause to one another. Although both finite and infinite-horizon prisoner's dilemmas can be used to explain why competition forces firms to promote, their implications are difl‘erent. The finite- horizon explanation, because of total opposition of interests, implies that promotions are a manifestation of perfect competition; instead of prices going down, promotion goes up "4 Pi g3 . it '0 27 (Lodish 1986a, p. 36). The infinite-horizon explanation suggests that although firms cannot avoid promoting, there is considerable latitude for them to avoid harming one another with myopic promotion behavior. From the finite-horizon perspective, promotions are an inescapable competitive mechanism that dissipates profits; from the infinite-horizon perspective, they provide a basis for tacit cooperation between rivals. The prisoner's dilemma has been used in social science to represent a wide variety of situations, including the arms race and economic rationality. It has even been called "the E. coli of social psychology" (Axelrod 1984, p. 28). Two branches of the game theory literature are applicable to promotions: the mathematical branch and the tournament branch. According to the literature, the former has not been of much prescriptive use for marketing problems (Lilien and Kotler 1983, p. 664) for two reasons. First, the mathematical complexity of modeling real situations with game theory is so monumental that it carmot presently, and may never, be accomplished (Dolan 1981, p. 227). Second, game-theoretical assumptions like randomized pricing are diametrically opposed to what managers actually do in practice (Blattberg and Neslin 1990, p. 103). There are some aspects of sofl drink promotions (avoiding wearout of a brand name or a package size by shifting promotions among brands and packages) in which mixed strategies may be used by bottlers (Dixit and Nalebufl‘ 1991, p. 23), but it appears that managers do not consciously employ randomization as a strategy in the game-theoretic sense. There are other problems in applying mathematical game theory to promotions in particular, and these fall into two general categories. The first relates to how well the economic effects of promotions are understood and/or measured. The second relates to such aspects of marketing as timing coordination and product positioning. Even if it were possible to express these factors mathematically they would firrther complicate an already massively complex mathematical problem. 28 In the promotions context, payofi‘s are not known, they change from one promotion to another, and they can be negative. Promotion profitability (the central piece of information needed to apply the mathematics of game theory) is difficult to determine and is rarely measured (Lodish 1986a, p. 163; Kruger 1987, p. 148). In commenting on state- of-the-art promotion evaluation systems, Kruger (1987) stated "it is not unusual to find that most trade promotions on a product lose money. In general, the better the evaluation method used, the more money is lost." The general ignorance of payoffs and/or the counterintuitive information provided by sophisticated promotion evaluation systems cause a problem in applying mathematical game theory. The prisoner's dilemma is not a game in which there is doubt about payofi‘s or whether one's move will help or hurt oneself. Kruger also says (ibid., p. 148) that long-run promotion effects (such as access to the distribution channel) exist. Although the mechanisms of these efl‘eCts have not been described in the literature, it seems plausible that they exist. The economic magnitude of these long-run effects would confound any mathematical formulation of promotions. Two other unknowns have not been measured: (1) the efi‘ect of one firm's price promotion on the profitability of another and (2) the efi‘ect of one firrn's nonprice promotion on the profitability of another. The economic mechanisms of promotions are not well enough understood quantitatively to allow the application of mathematical game theory if it were tractable. Another difficulty is that the logistics of sales promotions may prevent an equilibrium from being reached without explicit collusion. Retailer promotions require 4 to 6 weeks to schedule and cannot be cancelled quickly; a retailer who eliminates promotions without agreeing with rivals in advance (and thus risking antitrust prosecution) must endure 4 to 6 Cree diet. [Lire 29 weeks of competitors' promotions, even if they plan to follow a first-mover's strategy change (Blattberg and Neslin 1990, p. 107). The analogy between the prisoner's dilemma and price competition is intuitively appealing and may well be applicable to some price promotions; Urban and Star (1991, p. 181) cite the 1986 automobile industry interest rate price promotion. In this case, financing rates of 7.9 percent were introduced, and competitive one-upmanship subsequently whittled them to zero. This may resemble the prisoner's dilemma on the surface, but there are fundamental difl‘erences. The first difference is that prisoner's dilemmas are structured on profit results not price ofi‘ers. An argument for a price ofl‘er approach can be made, however. The prisoner's dilemma can operate on price ofl‘ers if (1) all players have the same cost structures and (2) market response is a fimction of price offers. The best available market response information, however, indicates that price promotions result in asymmetrical competition between brands (Blattberg and VVrsniewski 1989). This amounts to tacit verification of the idea that there may not be countermoves available to competitors for a firm's advertising, styling, display, and so forth (Shubik 1959, p. 349). ”Demand based investments" in branding, distribution, and advertising create mobility barriers (Wensley 1981, p. 180) that are not captured in the prisoner's dilemma. Because market response is not simply a function of price offers, it is unreasonable to assume that a price ofl‘er representation of the prisoner‘s dilemma is adequate. The analogy between game theory and promotions is clear and intuitively appealing, but it breaks down in the details of operationalization. In contrast to mathematical game theory, the tournament literature ofl‘ers prescriptive insights that often aptly express promotional practice. Axelrod (1984) and Fader and Hauser (1988) contribute to game theory explanations of promotion by identifying the major qualitative features of successful competitive strategies. in Ar nehdczc These adi‘i Cr Pill .i 51b 355635 COKE}: low“ 3O Axelrod's (1984, p. 29) work was pioneering because the existing literature did not reveal how to play the prisoner's dilemma well. The tournaments Axelrod conducted consisted of computer programs submitted by professional game theorists. The tournament was a ”round robin” so that each program played every other in a two- program game. Surprisingly, the simplest of all entries - TIT FOR TAT - won the tournament. TIT FOR TAT starts with a cooperative choice and thereafter does what the other player did on the previous move (ibid., p. 31). Axelrod highlighted four key success factors of TIT FOR TAT: What accounts for TIT FOR TAT's robust success is its combination of being nice, retaliatory, forgiving, and clear. Its niceness prevents it fi'om getting into unnecessary trouble. Its retaliation discourages the other side fiom persisting whenever defection is tried. Its forgiveness helps restore mutual cooperation. And its clarity makes it intelligible to the other player, thereby eliciting long-term cooperation (1984, p. 54). These four properties provide a mixed theoretical/empirical foundation for Axelrod's advice to players of prisoner's dilemma tournaments. The advice is as follows (1984, p.110) 1. Don't be envious. 2. Don't be the first to defect. 3. Reciprocate both cooperation and defection. 4. Don't be too clever. A subsequent tournament conducted using three-program games supported Axelrod's assessment of the four key factors (Fader and Hauser 1988, p. 568). This tournament consisted of a pricing game in which cooperators set price "high" and defectors set price "low” (ibid., p. 558). Fader and Hauser highlighted a fifth property of success, ”self- 31 awareness,” which also was present in Axelrod's tournament (1984, p. 30) but apparently was not used in the two-program context. Self-awareness allows players to consider the previous decisions and payoffs of all other players in the current game (F ader and Hauser 1988, p. 565). This makes it possible for two players in a three-person tournament to recognize and act on an implicit coalition against a third hostile player. Recognizing implicit coalitions proved to be the most important success factor in Fader and Hauser's tournament (ibid., p. 566). An implicit coalition exists if, given one player's defection, the profits of the remaining two players can be increased by their cooperation. This "happy medium” can raise profitability because it avoids the two extremes: defection of two players in response to one player's initial defection may be too severe a response, and the ”dove" strategy of cooperating may be suboptimal as well (ibid., p. 560). Several features of promotions - incomplete information about payoffs, time lags, geographic market differences, product position, and inability to communicate explicitly - make mathematical game theory a poor expression of reality. These same features make implicit coalitions more likely in the tournament context. National brands, for instance, may have incentives to form implicit coalitions. Store brands that are used by retailers to pressure national brands into promoting (Struse 1987, p. 150) can play the defector role. Having a difl‘erent defector in every chain's house brands conceivably could create an implicit coalition for the national brands that compete with one another in all markets. Furthermore, the customer loyalty to each national brand can fill the role of an outsider who stabilizes the coalition. Because Pepsi's loyal customers do not buy Coca-Cola when it is on sale, or vice versa, when Coca-Cola and Pepsi take advantageof an implicit coalition by alternating as high and low pricer, loyal customers pay price premiums to the high pricer. These price premiums are profits that cannot be switched fi'om one brand to ._.. f.) ‘1 1' If") 32 another, although they can easily be converted into consumers' surplus by myopic competition. The qualitative insights into competition that have emerged from the tournament work on game theory provide a common sense conceptual fiamework for understanding promotions. In implicit coalitions, cooperation is intermingled with competition, but without the collusion that would violate antitrust laws. Theoretically, promotion collusion should be rare, as the difficulty of detecting and countering promotions (Kotler, Armstrong, and Start 1991, p. 350) amplifies the destabilizing incentives of cartel membership (Martin 1988, p. 299). While mathematical game theory has not been prescriptively useful in marketing, the tournament literature has provided a conceptual fi'amework and five prescriptive insights. The tournament approach includes a plausible explanation for both cooperation and competition in promotions. C. ATh reti ent iece In 1974 a Management Science article by Kinberg, Rao, and Shakun, "A Mathematical Model for Price Promotions,” brought together the three essential issues characterizing current leading edge promotions research. These issues are: (l) recognition of the importance and sources of incremental unit sales, (2) the heterogeneity of customer response to promotions as a driver of dealing by premium brands, and (3) the importance of relative draw of customers to brands on promotion. Incremental sales are the fundamental objective of every promotion (Abraham and Lodish 1990, p. 50). A product on promotion may increase its sales by substituting for other products in the same category or by adding to category sales. The source of the increased sales is a critical issue, since it determines the incentives for manufacturers and retailers to promote and for retailers to pass through trade promotions (Blattberg and Xes‘ir. r l l l 1 1 l8} prod-.1: I “re a 33 Neslin 1990, p. 377). Kinberg, Rao, and Shakun list three sources of sales increases: (1) regular customers who buy larger quantities of a product, (2) customers of competitive products who switch to the promoted product, and (3) customers who are usually out of the market for a product but purchase it because it is on promotion. The heterogeneity of consumer demand is a second key assumption of Kinberg, Rao, and Shakun's promotion model. Two kinds of customers are assumed: quality conscious and price conscious. Product quality is assumed to be directly related to price. The assumption that customers difl‘er in quality and price consciousness implies that market response to a given stimulus will be asymmetrical. Because premium brands have higher quality levels, they can drop their prices temporarily and attract customers from private label brands. Because the latter lack the quality of a premium brand, they have no efl‘ective response to a national brand's promotional price. The resulting asymmetry between the national and private label brands anticipates current work on price-induced patterns of substitution (Blattberg and \lfrsniewski 1989, p. 294) and complementation (Walters 1991). The third aspect noted is the interdependence of premium brand sales. To say that all brands in a category are interdependent is not to say that there is competitive parity throughout the category. Premium brands have an advantage in that they can competitively draw - by price and nonprice promotions - customers from weaker brands. It is this advantage in "competitive draw” that determines the frequency and depth of promotions. The major findings of the article, which relate principally to competitive draw, are as follows. 1. If all premium brands attempt to maximize profits, a noncooperative equilibrium results (Kinberg, Rao, and Shakun 1974, p. 951). 34 2. The smaller the difference in market share of premium brands, the greater is the chance of alternating promotions between or among the premium brands (ibid., p.955) 3. Solutions depend heavily on the relative market shares among the premium brands. If there are two premium brands with the same market share, the joint maximizing solution will be to alternate promotions. No premium brand will have an incentive to break this equilibrium (ibid., p. 957). 4. The sufficient condition for alternating promotions is that the incremental gains from private label brands are greater than the incremental loss of regular customers (ibid., p. 958). Why do businesses promote? Kinberg, Rao, and Shakun's answer is not strictly economic, game-theoretic, or behavioral. Their model clearly includes profit maximization (ibid., p. 951) yet does not exclude the possibility of nonrational behavior in the form of "lock ins" (ibid., p. 955). An important contribution is the qualitative implication they develop that interdependence of promotional response provides the basis for an implicit coalition of premium brands. In efl‘ect, they show that promotions have cooperative as well as competitive implications. Research along the lines of their mathematical modeling did not continue until the late 1980s, but earlier in the decade the business literature began to recognize links between competition and promotions. D. The Business Press The importance of promotions as a tool in interfirm rivalry long has been recognized in the business press. In the 1960s the use of promotions by an established brand (Formula 409 spray cleaner) was credited with the withdrawal of a new Procter & Gamble (P & G) product less than a year after its introduction. In 1967 P & G test marketed Cinch spray cleaner in Denver, Colorado. Formula 409, Inc, found out beforeth and quietly sabotaged the test market by stopping advertising, promotion, and restocking in Denver while P & G's test market ran (Solman and Friedman 1982, p. 26). When the national roll out of Cinch bl our in each reg with a half-gal rats later. P l peernpt a ner press reports Weapon. 35 out of Cinch began after very successfirl testing in Denver, Formula 409 preceded the roll out in each region with a $1.48 promotion: a l6-ounce bottle of Formula 409 bundled with a half-gallon refill (a six-month supply for the average household) (ibid., p. 27 ). Six years later, P & G used the same preemptive promotion strategy with Crest toothpaste to preempt a new product called Peak (Williams 1983). In these and other cases, business press reports clearly understood that promotions were being used as a competitive weapon. On October 25, 1987, the CBS program 60 Minutes (1987, p. 6) aired "Cola Payola," a documentary that covered a form of sales promotion called calendar marketing agreements (CMAs). These were portrayed as a means to avoid price competition (ibid., p. 9) and to limit small soda brands from encroaching on premium national brand sales (ibid., p. 8). This documentary was the first discussion of CMAs outside the soft drink trade (F elgner 1988, p. 213), and it clearly indicated that promotions were being used cooperatively. There is strong similarity between Kinberg, Rao, and Shakun's (1974) model of promotions and the defensive rivalry between Coca-Cola and Pepsi depicted by 60 Minutes. Observing the alternating Coca-Cola and Pepsi promotions, the Kinberg, Rao, and Shakun model would predict approximately equal relative market shares for the two and equal draw from private label brands for Coca-Cola and Pepsi. Unfortunately, no information about the relative market share and competitive draw parameters was revealed by ”Cola Payola.” The next coverage of CMAs quickly followed. In December fire Wall Street Journal ran a story, ”Cola Sellers May Have Bottled Up Their Competitors” (Pasztor and Reibstein 1987). It quoted the Coca-Cola Co. as saying its marketing programs ”encourage competition" by increasing the frequency of sofi drink promotions and by 'reducing irC‘rrarlorr promotior catered ir 1. Bi Cl an fr: 5; d R. C 3 \K 4 A a; pr fr 1: Six m. Sareszg WSpeqi ‘00er ‘- L o si b 2. Anorh the Septe 36 "reducing prices to consumers." The article reported testimony about CMAs in a civil suit in Charlotte, NC, where Coca-Cola was found guilty of conspiracy to control prices and promotions, and Pepsi settled out of court. Several promotion-related issues were covered in the newspaper article. 1. Bill-back payments totaling $900,000 in one year were made to 7-Eleven stores in Charlotte, NC. These payments were in exchange for a $1.59 promotional price and in-store display for 2-liter bottles. Coca-Cola compensated 7-Eleven with 108 free cases of 16-ounce cans of soda per store, for 26 weeks. Pepsi directly paid $400,000. Retailers passed through price discounts for Coca-Cola and Pepsi but not for RC Cola. Witnesses contended that bottlers coordinated promotion policy through retailers. A 1982 memo from a Coca-Cola executive "bluntly urged bottlers to refrain from aggressively promoting products when they knew special Pepsi promotional programs were under way and the memo noted that Pepsi bottlers 'will also refi'ain from this activity when Coca-Cola bottlers are featuring” (Pasztor and Reibstein 1987). , Six months after ”Cola Payola,” Consumer Reports (1988) ran a story entitled ”How to Save $2500 a Year in the Supermarket.” It considered promotions from the consumer’s perspective, covering advertising, trade promotions, retailer promotions, and even cooperative advertising. Two promotion-related issues were noted. 1. Under the bottlers' "calendar marketing agreements,” a store agrees to feature only one brand at any given time. Since Coca-Cola bottlers just happened to have signed such agreements for 26 weeks and Pepsi bottlers the other 26 weeks, lesser brands such as Royal Crown are effectively squeezed out. Sound like price-fixing? According to one supermarket owner who gave up his calendar marketing agreements, his soft drink prices decreased when all the companies had a chance to compete (Consumer Reports 1988). Another article on CMAs involved consumer expenditures for soft drinks, reported in the September issue of Supermarket Business (F elgner 1988). This article noted :crnnlalms frc lraddition C rein fact prc Not long l ”lamented Price fix past der liom - I there ha lire article 2 l, llOl A final . lage Story (Mi-5 cm W Weel. lament c term m2 Pacefim m Comps. rL . lec‘ 3 7 complaints fi'om small bottling companies about being locked out of retailer promotions. In addition, Coca-Cola was quoted as saying that, "when properly implemented, CMAs are in fact pro-competitive, benefiting both the trade and the consumer" (ibid., p. 188). Not long Thereafter, "The Price of Fixing Prices" in Beverage World (Galvin 1990) documented antitrust violations in the soft drink industry: Price fixing, it is fair to say, has been rampant in the sofi drink industry during the past decade. Since 1986, the Department of Justice has obtained guilty pleas from - or convictions of - more than 40 bottling companies and individuals. there has been only one jury acquittal (ibid., p. 27). The article also noted that 1. The Soft Drink Interbrand Competition Act of 1980 guaranteed exclusive franchise territories for bottlers (ibid., p. 27). 2. There had been ”intense consolidation” of the industry and subsequent ”all-out discounting war for most of the decade” (ibid., p. 27). 3. When bottlers met to fix prices, they agreed on prices only for cola products and not allied brands (ibid., p. 28). A final example is a Consumer Reports (1991) rating of colas that contained a one- page story, ”The Battle For Your Business" (ibid., p. 524). This article reiterated facets of CMAs covered in previous news stories: Coca-Cola and Pepsi dividing the year except for ”bad weeks in January and February when colder weather means sagging sales,” the payment of rebates to retailers by bottlers, the claims of small bottlers that CMAs are "a device that impairs the ability of smaller companies to stay competitive,” and the 1984 price-fixing case in North Carolina In summary, the business literature has indicated several ways promotions can be used in competition. In addition to sabotaging a competitor's test marketing and preempting the introduction of a rival product (Solman and Friedman 1982; Williams 1983), promorions preferred in lock out" Lilli] Not un R30. and exploiting flannel p lrnpliCit c Eioint or promotir consume Monte then finr Pfomoti PTOmoti A1101 38 promotions can be used by national brands to avoid price competition and "lock out" less preferred brands from promotions (60 Minutes; Pasztor and Reibstein 1987), as well as ”lock out” small competitors fiom promotions (60 Minutes; Felgner 1988). E. The Journal Literature Not until 1988 did the journals again explore along the lines pioneered by Kinberg, Rao, and Shakun (1974). An emerging theme is that promotions result from firms exploiting a mixture of joint and individual opportunities produced by brand loyalty and channel position. This approach fits well with the tournament literature's concept of implicit coalitions. Loyal customers and price switching segments exist, and there may be a joint opportunity - that is, an implicit coalition - for two preferred brands. When on promotion by itself, each brand captures profits from large volume sales to nonloyal consumers; when ofl‘ promotion, it captures higher margins at low volume sales from loyal customers. Brands with a significant base of loyal customers and/Or capacity limitations . then find it in their interest not to be promoted at the same time. The resulting alternating promotions can preemptively prevent or reduce the frequency of third-party brand promotions. Another party afl‘ecting implicit coalitions is the retailer. Because retailers have different economic interests than manufacturers, they merchandise store brands to force manufacturers of brands with loyal customers to promote and sell at a higher rate (Rao 1991, p. 141; Struse 1987, p. 150). Because retailers are primarily interested in category sales, however, they avoid merchandising multiple brands in each category (Walters and MacKenzie 1988, p. 55). Retailers appear to have the position in the channel and the economic incentive to (l) goad national brands into dealing and (2) police promotions and prevent their degeneration into pure price competition. The founc sersithiry. l hinted to b decision to < trorits from for profits g these two 3 Pt'r'lOdS alt: Negros r mod prir The hr PiOmotior brand sm' Whereas Whereas - Populatir mm 39 The foundation of implicit coalitions, however, is based on preferences and price sensitivity. Brand switchers are sensitive to the power of a bargain, that is, they can be induced to buy a product if the price is reduced enough. Regardless of whether the decision to cut price is made by the manufacturer or the retailer, the effect is to exchange profits from segments of the market that would have purchased the product at a high price for profits generated by scale economies. Because promotions are short-term price cuts, these two strategies need not be mutually exclusive. If promotion and nonpromotion periods alternate in some pattern that increases profits, mixing the high and low price strategies can make sense. This dual price strategy is similar to Lazear's (1986) two- period pricing model based on retailer uncertainty about demand. The first article in the current journal research investigating this approach to promotions is by Narasimhan (1988), who thinks that firms use promotional prices to woo brand switchers while minimizing the loss of profits from loyal customers (ibid., p. 428). Whereas Lazear assumes two time periods, Narasimhan assumes an infinite time horizon; whereas Lazear's retailer is uncertain about the distribution of reservation price in the population as a whole, Narasimhan's manufacturers know that there are two market segments: brand loyal and price switchers. Narasimhan (ibid., p. 429) explores two brands in a market with consumers who buy only from firm 1 (a1), who buy only fi'om firm 2 (a2), and who switch between the two firms ([3). These consumer segments are the whole market (a1+or2+B=1), that is, the market size is assumed to be constant. Narasimhan implicitly draws an interesting distinction between loyalty and preference. Ifthe switcher market segment prefers the weaker brand, the stronger brand will promote less ofien and offer a larger maximum discount. Ifthe switcher segment prefers the strong hurl. the strr inrir'asan ar Narasimhr represents hr relative draw lineal issue heterogene model addr wink the and the pr the model 1, 53 f1 “lib tl hrand 1nd v 40 brand, the stronger brand promotes more often and offers a smaller discount (Raju, Srinivasan, and Lal 1990, p. 278). Narasimhan's model holds incremental sales and consumption constant and, because it represents brand interdependence in a two-brand world, does not consider the issue of relative draw of weak brand customers by strong brands on promotion. Thus, of the three critical issues raised by Kinberg, Rao, and Shakun (1974) - incremental sales, heterogeneity of customer response, and sales draw fi'om weak brands - Narasimhan's model addresses one: preference heterogeneity. The essential contribution of the model is to link the frequency and depth of promotions to the size of loyal segments ((11 and a2) and the price sensitivity of the switcher segment (13). The two key qualitative findings of the model are: 1. Symmetry of draw (”pulling power") fiom the price switcher segment gives both firms the same average discount on promotion (ibid., p. 43 5), and 2. frequency of promotion for the premium brand firm (firm 1) is given by (a2+B)/(ot1+l3) (ibid., p. 441). The implication of this model is that price switchers drive the dynamics of promotion frequency and discount. Ifthey are a very small segment, then there is no incentive for either firm to discount or promote (ibid., p. 437). When switchers have a preference for the smaller brand at equal prices, the larger brand must promote less often and offer larger discounts (ibid., p. 438). In categories with many brands and intense rivalry, the brand with the least pulling power may not want to discount at all (ibid., p. 441). Price competition for switchers who are indifi‘erent between two brands helps the smaller share brand. The next article relevant to this research stream was an empirical study by Blattberg and Wisniewski (1989). This work investigated the effects of price deals on unit sales of hands and the and tuna). ll ofwsromerr tiers. based c preference d quilt}: tiers price tiers v Blatther annhalize [7 0f8 an 31568 out P0656 Sll’t equal) pr: Ware Ia Conditior Wine 1; Naragim to boy " CdS’tOmr some St 0th Waller 4 1 brands and their competitors in four product categories (flour, margarine, bathroom tissue, and tuna). The authors did not directly investigate incremental sales effects, heterogeneity of customer response, or competitive draw. Instead, they developed a theory of price tiers, based on an "idiosyncratic desire for quality" (ibid., p. 294), which gives a bimodal preference distribution (ibid., p. 305). Higher quality tiers have higher prices, and lower quality tiers have lower prices. The effect of price promotions in drawing sales between price tiers was the focus of this study. Blattberg and Wisniewski presented two key findings. First, brands on promotion cannibalize sales of brands in their own price tier (14 of 15 cases or=.05) and the tier below (7 of 8 cases). Second, lower tier brands on promotion rarely affect higher tiers (only 3 cases out of 45) in a scenario likely to find lower tiers affecting higher tiers (ibid., p. 303). These strongly asymmetrical effects within grocery categories indicate that (other things equal) premium brands on promotion make some incremental sales gains at the expense of private label brands. Kinberg, Rao, and Shakun (1974, p. 958) stated that a ”sufficient condition for alternating promotions to take place is that the incremental gains fiom private label brands are greater than the incremental loss fi'om regular customers. " Narasimhan (1988, p. 428) contends that firms fluctuate prices to induce brand switchers to buy their products while at the same time minimizing the loss of profits fi'om their loyal customers. Blattberg and Msniewski‘s discovery of asymmetric competitive draw lends some support to both the above brand interdependence arguments. Other types of promotion-induced asymmetries have been uncovered in recent work. Walters (1991) found asymmetries between both substitutes and complements: .. price promotions enabled Crearnette spaghetti, the market leader, to gain sales at the expense of private label spaghetti. In no instance did a low market share brand gain sales at the expense of a brand with a higher market share. Hence, price promotions appear to be an effective tool that brands having large *1 i inte‘ corr con are pro bra cor ltw 42 market shares in their categories can use to increase their sales temporarily at the expense of their competitors, but price promotions do not appear to build sales of low market share brands through cannibalization of high market share brands (ibid., p. 24). The pattern of result pertaining to complementary efi‘ects provides some interesting insights into price promotional efl‘ects. First, like substitution efi‘ects, complementary effects were not symmetrical. That is, price promotional activity on a brand (e.g., Crearnette spaghetti) resulted in increases in sales of a complementary brand (e. g. , Prego spaghetti sauce), but no increases in sales accrued to the brand (e. g., Crearnette spaghetti) when its complement was promoted (e.g., Prego spaghetti sauce). Second, price promotions on low share brands (e. g., private label spaghetti) significantly increased sales of high share complements (e.g., Prego Spaghetti sauce) as fi'equently as price promotions on high share brands (e.g. Prego spaghetti sauce) stimulated complementary sales of low share brands (e. g., private label spaghetti) (ibid., p. 24). These findrn' gs point out that asymmetries have intercategory as well as intracategory efi‘ects. Blattberg and Wisniewski (1989) confined their analysis to situations in which all products within a category were substitutes. In addition, the patterns of asymmetry observed by Walters appear to fit well the price-tier pattern hypothesized by Blattberg and Wisniewksi (1989). The next article in the analytical research stream is ”Effects of Brand Loyalty on Competitive Price Promotional Strategies,” by Raju, Srinivasan, and Lal (1990). The authors did not address incremental consumption efl‘ects because they assumed category volume would remain constant (ibid., p. 278), but they did examine consumer response heterogeneity and competitive draw as drivers of promotional fi'equency and depth. The specific kind of heterogeneity considered is customer loyalty in the context of three brands. Brand loyalty is defined as the price difi'erential needed to induce consumers who prefer a brand to switch to another (ibid., p. 276). A "strongest brand” is the brand whose customers must be ofl'ered the deepest discount before they change to a competing ofi‘ering. For instance, if Coke's customers will not change to Pepsi unless ofi‘ered a one dollar d of 50 C: "and Ir.) VJ Imph (Min Week SW 43 dollar discount per unit, and if Pepsi's customers change to Coke when offered a discount of 50 cents per unit, then, by Raju, Srinivasan, and Lal's definition Coke is the stronger brand. Two brands are premium price/quality brands, and the third brand is a private label. It is assumed that market segments are loyal to one brand and do not engage in variety- seeking behavior between or among brands (ibid., p. 278). The analytical fi'amework is a finitely repeated game. The authors choose finite games over infinitely repeated games to avoid solutions with a large number of equilibria — many of which are counterintuitive (ibid., p. 281). The two main contributions of this article are to extend Narasimhan's (1988) work from two-brand markets to markets of arbitrary but finite complexity (Raju, Srinivasan, and Lal 1990, p. 282) and to derive and empirically test implications of the promotion analytical framework. Four key implications are developed. 1. Brands in a product market with high brand loyalty across all brands will not be promoted (ibid., p. 296). 2. Promotion fi'equency and number of brands competing in a product market are directly related (ibid.). 3. When strong brands compete with weaker brands, the strong brand will promote less fi'equently (ibid., p. 297). 4. The discount ofl‘ered on promotion by a strong brand depends on the loyalty coefficients (ibid.). Implications 2 and 3 were empirically tested. The data used were weekly shelf prices and quantity sold of approximately 2,200 grocery items (UPCs), fi'om 27 categories, for 24 weeks. The authors judgrnentally assigned UPCs to 75 product markets by splitting 27 SAMI codes (ibid., p. 294). he pro; timber terreiar inns w: Conseq: t. 4: . .011? dons fm Worn: Wigner mmpelim 44 The findings for implication 2 were tested with a Kendall rank correlation test between the proportion of brands discounted by more than 5 cents in the product market and the number of brands and number of firms represented in the product market. The rank correlation between promotions and number of brands was .27 and for the number of firms was .26, which are both statistically significant at or=.05 (ibid., p. 295). Consequently, implication 2 was supported. Implication 3 was tested by treating a premium price/quality national brand as a "strong” brand and a store brand as a ”relatively weak brand. " The difference in mean number of promotions for 59 pairs of national and store brands was calculated. The authors controlled for package size and similarity (for example, compared a 6-pack to a 6-pack). In 31 of 59 cases (53 percent) the store brands offered discounts more ofien than the national brand, there were 11 ties (19 percent), and in 17 cases (29 percent) the national brand was promoted more frequently than the store brand. Implication 3, that strong brands promote less fiequently than weak brands, was supported by both the sign test and Wilcoxon test at or=.05 (ibid., p. 295). The finding that strong brands promote less ofien than weak is conspicuous because it is diametrically opposed to Kinberg, Rao, and Shakun's (1974) prediction as well as the predictions in two other papers by Lal (1990a, b). This reversal of promotion frequency is probably a product of the conceptual fi'arnework developed by Narasimhan (1988) and extended by Raju, Srinivasan, and Lal (1990). This framework does not consider implicit coalitions because it does not consider balanced market power. The essential motivation for promotion in Narasimhan's model is an imbalance of "strength" or customer willingness to pay a price premium. Given an imbalance, the brand managers for competing brands are assumed to find promotion schedules that maximize their respective profits in . 1M tthll L"lat is 45 profits. Figure 25 displays the equilibrium strategies in the form of cumulative density functions for the pricing strategies. Figure 2-5 Pricing Strategies Induced by Brand Interdependence our T 1.0 ”r" ‘L—J ”'71 tr airy/'4‘. r A " our.) 11'. 1.0 The +24% tr 0.5. + t. i " Source: Raju, Srinivasan, and La1(1990, p. 287). The strong brand density function is on top. In the figure, "r" stands for regular price and ”1w” for loyalty to the weaker brand (the price difference needed to make customers switch to the strong brand). Both the strong and weak brand density functions have jumps in them just below r. These jumps are caused by each brand's loyalty. The size of the jump is larger for the stronger brand because, by definition, it has more brand loyalty. This fiamework also may be applicable in cases where there is little if any brand loyalty as that is, where the jump below regular price does not exist. The complement of Raju, Srinivasan, and Lal's empirical finding is that 47.5 percent of the competitive brand pairs did not fit their model. The authors assumed that across all categories the loyalty of the stronger brand is large enough so that the weaker brand does not find it profitable to drop price as a means of attracting the strong brand's customers in .1; assu in a eqni ' ter inter pror althr Fun p. 2‘. bran 0i 5\ 46 (ibid., p. 288). This configuration of loyalty in the market leads to the strong brand promoting less often than the weak brand. The actual loyalty in many mature categories however, is not consistent with the assumption that strong brands have significantly more loyalty than weak brands. Blattberg and Wisniewski (1989, p. 303 ), for example, found that strong brands ran price promotions more often than weak brands. The next contribution in the brand interdependence literature was Lal's (1990a) article in Marketing Science, which broadened and extended the analytical framework. It assumed an infinite time horizon rather than one period and showed that price promotions in a triome (two national brands and one local brand [ibid., p. 250]) can be a Nash equilibrium in pure strategies. Up to this point work had relied on Varian's (1980) interpretation of mixed strategies as price promotions (Lal 1990a, p. 249). This interpretation has been criticized as an expressively weak and unnatural way to represent promotions in recent work (Rao 1991, p. 132). Managers do not price randomly, although they do try to steal sales fi'om one another (Blattberg and Neslin 1990, p. 103). Furthermore, because store brands are rarely promoted (Raju, Srinivasan, and Lal 1990, p. 296), they may not be managed in the same mixed-strategy game fashion as national brands (Rao 1991, p. 140). Lal (1990a) hypothesizes a number of consumers loyal to national brands (at), a number of switchers (B), and a reservation price for all consumers (r). Lal translates into mathematical terms an intuition about market segmentation and competition: Ideally, the national brands would like to realize the maximum rents fi'om the loyals and the switchers by charging a price as close as possible to [each segment's] reservation price. However, since the local brand has no loyal customers, it poses a threat to the national brands in the market for switchers by lowering its price. if the switching segment is profitable enough, the national firms jointly may find it desirable to charge the reservation price in one period followed by a promotion price such that only one of the national firms is always on promotion (1990a, p. 251). trait II“- (.4) anti 47 Brand loyal customers (the or segment) are the basis of the price discrimination aspect of this intuition, and price switchers (the B segment) are the basis of the competitive aspect. The essence of Lal's development focuses on the analytical properties of the market's demand structure. In Lal's fi'amework it consists of the market proportions loyal to national brands (or), switchers (B), the price for national brands (r), and the market's demand parameter (x). If an x exists such that Qaflr > x > 9aflr 24 4(3ar2 +2afl) 4(a+fl)2’ and if firms alternate promotions, then the profits fiom charging two prices are greater than charging a single price (Lal 1990a, p. 252). The promotional price will be: i: %%O(a+fl). 2-2 The demand parameter x will exist only if 822209, and the promotional price of the national brand must be low enough to attract price switchers from the local brand (ibid., p. 252). The proofs for these formulas are based on the standard financial assumption of maximizing net present value and hold for interest rates less than 60 percent per year (ibid., p. 255). On the basis of the values of the market parameters, Lal develops a pure Nash equilibrium explanation for promotions that assumes the national brands maximize profits. The promotional behavior that ensues is indistinguishable from collusion against the local brand. The incentive structure of the market is such that the national brands each credibly threaten the other not to ”defect" (that is, promote out of turn), since one rival can ‘punish' heat is never b hut sup profital llacKe ithid . nil be did he (J 48 "punish” the other and drive prices back to the collusive path (ibid., p. 253). Because the threat is credible (constitutes a perfect Nash equilibrium), game theory predicts that it will never be observed (ibid., p. 253). This punishment explanation is theoretically interesting but superfluous in the real world due to retailer incentives. Because retailers seek profitability for a product category or the store rather than a specific product (Walters and MacKenzie 1988, p. 52), they usually promote only one product in a category at a time (ibid., p. 55). Given this rate of promotion and the fact that retailers decide what products will be promoted, as a practical matter "defection” will be nonexistent, and punishment willbeunnecessary. In the Journal of Marketing Research, Lal (1990b) extended the pure strategy Nash equilibrium approach (Lal 1990a) by incorporating retailer incentives and store brands into the model. The result was additional qualitative insight into competition in the market he specified. 1. As the number of brands goes to infinity, the national brands will be increasingly better off by cooperating with each other than by cooperating with local brands (1990b, p. 434). 2. Compared to retailers without store brands, retailers with store brands should sell national brands at a higher price and should be able to buy the national brands on promotion for less (ibid., p. 43 5). 3. National brands should be promoted in only one store (or chain) in a market at a time (ibid., p. 436). 4. Manufacturer trade promotions of national brands should be negatively correlated with each other (ibid., p. 437). 5. There should be no trade deals for the local brand (ibid.). Lal cited Consumer Reports (1988) and ”Cola Payola" (60 Minutes 1987) in support of three propositions: (l) lesser brands are squeezed out, (2) Coca-Cola and Pepsi have about the same number of loyal consumers because each promotes 26 weeks out of the year. and promotio common For the l. promotir and con: ‘91 ll fibid, Cola ( COSts 49 year, and (3) the number of switchers is sufficiently large to keep national brands on promotion all the time (ibid., p. 436). In setting up a test, Lal noted that the two most common approaches are to test the theory's assumptions and implications (ibid., p. 43 7). For the latter, Lal used a chi-square to test for a negative correlation of manufacturer promotions (implication 4 above). The data consisted of weekly manufacturer trade deals and consumer panel information for lemon/lime liquid dishwashing detergent. The contingency tests were performed for Sunlight and Palmolive brands (see Table 2-2). Table 2-2 Contingency Table for the Calendar of Manufacturer Trade Deals (weeks) Pal-011w 1 Source: Lal (1990b, p. 439). Lal prefaced this analysis by stating: "Because we may not have the appropriate data set, the data analysis should be viewed as preliminary evidence rather than a rigorous test" (ibid., p. 43 7). Although Lal stated that his model was an "intuitive explanation" of the cola category, colas were not used because of "significant and sometimes insurmountable costs of collecting data for the more stringent test" (ibid.). The panel data apparently were used only to group the package sizes into the contingency table, not to develop a measure of loyalty (ibid., pp. 43 7-8). In conclusion, though the power of the empirical tests is not strong, the data seem to indicate that the prediction of the theory as related to implicitly colluding manufacturers cannot be rejected without firrther thought. Though the analysis afi‘ords merely anecdotal/supporting evidence, we are encouraged to collect such this; the n and 5 both ether di‘fet md S mesh flew in -s 50 data sets for additional product categories and/or geographic markets to conduct a more powerfirl test of the theory (Lal 1990b, p. 439). The final contribution considered is "Pricing and Promotions in Asymmetric Duopolies,” an article by Rao (1991) in Marketing Science. As does all work in this research stream, it assumes constant market size (ibid., p. 133). Rao extends the analytical/competitive framework by developing his model from the perspective of many consumer package goods manufacturers, who view promotions as part of the pricing decision (ibid., p. 132). The duopoly consists of a national brand and a private label, and it is asymmetric because customers are willing to pay a premium for the national brand but not the private label (ibid., p. 134). The two market segments are A, which is deal prone (ibid., p. 133), and B, which is willing to pay a price premium (ibid., p. 134). Brand 1 is the national brand, and brand 2 is the private label (ibid., p. 133). Rao distinguishes price from preference. Segment A chooses on the basis of price only, and segment B chooses based on price and preference (ibid., p. 134). He assumes that if both brands are the same price, half of segment A will buy the national brand, and the other halfthe private label, while all of segment B will buy the national brand. The price differential between private and national brands, which is maintained by retailers (Little and Shapiro 1980, p. 3207), is seen by Rao (1991, p. 138) as a market segmentation mechanism to increase profits. The centerpiece of his analysis is that the regular price, the frequency, and the depth of promotions, taken together, define the market segmentation. Because the firm's actions define the market, the interpretation of promotion shifts from the ”stealing customers" perspective to ”enforcing market position.” Rao concludes: My model suggests that we think of the pricing decision as defining the brand's franchise relative to its competition, then view promotion as a method of enforcing that position (ibid., p. 141). In Rao's analysis If it is too high, the runners and will brand lithe priva hen up. and the otters larger prom ensure the private promotion is used model Promotion M Several linkag Riggesred in the or should be 3e. 51 In Rao's analysis the price of the private brand has interesting dynamic consequences. If it is too high, the private label will compete with the national brand for nondeal-oriented customers and will cause (and have to endure) more frequent promotion by the national brand. If the private brand price is set lower, part of the nondeal-oriented segment is given up, and the national brand will be promoted less frequently. The national brand offers larger promotion depths (foregoing profits from nondeal-oriented customers) to ensure the private label does not promote (ibid., p. 141). In Lal‘s models (1990a, b) promotion is used to capture profits within a given preference structure; in Rao's (1991) model promotion is used by competitors to define market positions. F. Pr tin th Several linkages between market position and alternating promotions have been suggested in the literature, but no consensus has developed on how promotional prices are or should be set. Chevalier and Curhan (1976) reported that deals involving product categories with high sales volumes received greater price reductions than deals on product categories with low sales volumes. They argued that retailers try to maximize store/chain performance and that high sales categories have a greater effect on store/chain performance than do low sales volume categories (Walters 1989, p. 256). Walters found mixed support for this hypothesis in an investigation of two grocery chains (ibid., p. 265). Eight other hypotheses about promotional discounts were distilled fiom the literature and tested by Walters. These relationships and the empirical support found for them are listed in Table 2-3. The hypotheses shown in Table 2-3, with the exception of the last, put the retailer squarely in the middle of the deal magnitude decision. Walters reported that retailers must allocate their support of promotions because far more promotions are ofi‘ered than the retailer steer to pro .L '[__.1_]- :H .— In... . . Pim- COSZ‘ 52 retailers can participate in (ibid.). That price elasticity and category rank received mixed support may indicate that all retailers do not use the same information (decision rules) to set promotional discounts. Table 2-3 Drivers of Promotional Discounts Hypothesized Driver Chain 1 drain 2 1. Catcgory sales volu- 1s dinctly rclatod to deal It I minim“. 2. Rank in category is directly rclatod to dual magnitude. N I 3. Retailer pro-lotions acmnicd by census-r practicum N 1 1m dcal ngnltudc. 4. Pracucns custanizcd to the rctallcr's needs hm higher ! N dual magnitudc. 5. Trade doal ngnituds and retail deal magnitude are I 2 directly related. 6. Ad size is directly rclatcd to deal ngnltudc. I I 7. 812. of display roqucsccd is directly related to deal I I tudc. 8. title since last pro-Duos: is directly related to dcal Y I naqnltudc. ' 9. Prion elasticity of “and 1s dircctly rclaud to dcal I N “gum Source: Wales (1989). Current research indicates that retailers most often support medium price reductions accompanied by medium or large advertisements (ibid., p. 264). Practitioners have reported that store brands are used to force national brands to deal (Struse 1987, p. 150). This work raises the question of what effect retailers have on promotion depth. Walters reported: Retail managers participating in the study emphasized that they believe a promotional threshold exists among consumers and that ”at least a good size ad" (i.e., medium-sized ad) and price reductions of about 10 percent are necessary to interest consumers in the promotions (1989, p. 262). Promotional discounts also may be affected by the power that UPC scanning has provided retailers (Buzzell, Quelch, and Salmon 1990, p. 141). Before scanner data became available, the pricing problem was "how to set prices in a world of limited and costly information" (Avery 1980, p. s212), and the use of markup pricing was justified 53 because retailers relied on the manufacturer's "better data" to set prices (ibid., p. s213; Cannon and Bloom 1991, p. 169). UPC scanner and single-source data, however, have given retailers "better data" than manufacturers and tipped the balance of power in their favor (Kumar and Leone 1988). The pricing and promotion functions traditionally performed by manufacturers in US. channels, have begun to be taken ”hostage" by retailers. A brand is called a "hostage” when the trade controls more than 50 percent of a brand's direct marketing budget (Blattberg and Neslin 1990, p. 478). This forced functional shift of pricing fiom manufacturers to retailers may be manifested in promotions. Members at different levels in the marketing channel have different marginal revenue curves (Hawkins 1950, p. 181), yielding difi‘erent demand elasticities at each level in the channel (ibid., p. 183). This vertical structure of price elasticities (which become greater at each level down a channel) explains or is one reason retailers do not pass through all manufacturer price cuts (ibid.). Retailers see more elastic demand curves than national manufacturers, so less price stimulation is necessary to achieve a given response. Channel elasticity tiers also explain why chain stores are more sensitive than independent retailers to wholesale price changes (ibid., p. 184). Because chains internalize the wholesale level demand and cost curves, they can respond to changes in costs by maximizing total profit subject to consumer demand. Ifretailers can select price discounts, then they are in a sense monopsonists, and, they will select retail prices that maximize their profits but not necessarily those of manufacturers. Hawkins works out an example with three possible prices: ( l) a price that equates the retailer's but not the manufacturer‘s marginal costs and revenues, (2) a price that equates the manufacturer‘s marginals but not the retailer's, and (3) a price equating marginals of both manufacturer and retailer. Which price is selected depends on market 54 power. If retailers have the power to impose their price, promotional discounts will be smaller than if manufacturers impose their price. Among the factors affecting the depth of discount available to consumers, market power has not been incorporated into the literature. Another is kinked demand curves. While the literature on price dealing generally supports the concept of the downward- sloping demand curve (Calantone and others 1989, p. 4), the best data available indicate tlmt most grocery store demand curves are kinked (Totten and Block 1987, p. 42). The kinks probably occur because price increases beyond a "normal" range cause people to "cross-shop" or postpone purchase, and they later may buy from another store. This view of demand is consistent with the steep sales declines experienced by stock up goods (24.1 percent compared to 7.6 percent for nonstock up goods) from price increases (Litvack, Calantone, and Warshaw 1985, p. 18). It is also consistent with store loyalty measures showing that only 8 percent of shoppers are loyal to one store (Totten and Block 1987, p. 37). Kinked demand curves may affect price discounts in two ways. First, when it is kinked at high prices, a firm's demand curve indicates a price-insensitive market segment. For whatever reasons, some customers continue to buy at high prices, and these customers are a potential source of rents. The rents from price-insensitive customers are an opportunity cost of price discounts that can increase the break-even discount and volume required to make a promotional discount profitable. Second, a kink at the bottom of the demand curve indicates a floor price at which price has a diminishing effect. None of the literature on optimal price discounts takes demand curve kinks into account. A second approach to understanding deal magnitude selection is based on costs and price discrimination. Blattberg, Eppen, and Lieberman (1981, p. 126, 129) hypothesized 55 that depth of promotion depends on the relative inventory carrying costs of firms and consumers. This train of thought has been extended recently by Hoyt, Calantone, and di Benedetto (1990). This approach interprets discrepancies in consumer and retailer storage costs as a mechanism for price discrimination. The rationale is that consumers difl‘er along two dimensions, demand and inventory cost. The authors distinguish four classes of customers (ibid., p. S), as displayed in Figure 2-6. Figure 2-6 Dimensions of Promotional Price Discrimination Inventory Cost Less Price Sensitive lore Price Sensitive 319': M 31 Lou A, B, Hoyt, Calantone, and di Benedetto essentially investigate the necessary and sufficient conditions for dealing to extract more profit from a market than does a single price. The conditions consist of the ordering of reservation prices for the four market segments. The reservation prices are controlled by preferences, variable storage costs, fixed storage costs, and demand rate, so counterintuitive reservation price orderings are possible. For instance, if the reservation price ordering is A2>A1>BZ>B1, then there is no deal magnitude that will increase profits over a single price (ibid., p. 6). Common sense suggests that the ordering A1>A2>B1>B2 would maximize profits because it would allow sellers to segment the market between price-sensitive and nonprice-sensitive customers. The authors derive the conclusion, however, that under this market condition there is no deal price that results in some members of the less price-sensitive market segment (A) not inventorying when some members of the more price-sensitive segment stock up (ibid., p. 7). Under this ordering the ideal single price will be the average reservation price for either the price-sensitive (column B) or price-insensitive market segment (column A). 56 An important implication of this research is that attempts to increase profits through price discrimination must split markets by more than one cost and/or value dimension at a time. Something must prevent less price-sensitive customers from stocking up when deals are targeted at more price-sensitive customers. Preference differences alone, though they may identify market segments, do not allow two-price strategies to increase profits over a single-price strategy. To increase profits, storage or other costs (such as the opportunity costs of time) must keep out of the market those customers who are willing to pay more until the price deal is over. This search for constraints to keep some customer groups out of the market while other customers are buying appears to be the opposite of price bundling. Bundling captures more revenue because wstomer groups have opposite orderings of reservation prices for two products. In contrast, price dealing captures more revenue because more price-sensitive and less price-sensitive customer groups - when total costs are taken into account - reverse their reservation price ordering for one product. The optimal prices and price discounts developed by Hoyt, Calantone, and di Benedetto ( 1990) are set just below the reservation prices of one of the cells in Figure 2-6 or the average of two cells, depending on the reservation price ordering. This fi'arnework gives subsequent researchers a means of putting analytical teeth into managerial pricing intuitions and the price discrimination literature's suggestion that discounts be set by extracting consumer surplus and setting marginal costs equal to marginal revenues (Dolan 1987, p. 20; Jeuland and Narasimhan 1985, p. 304). The marketing literature to date has explained deal magnitudes either by relating discounts to product, category, or retailer factors, or by relating discounts to price discrimination. In contrast to these general approaches the brand interdependence 57 explanation applies to a mature product category composed of a mix of strong and weak brands. Explaining why firms promote in this context is particularly dificult because the most intuitively appealing explanation - of converting product triers into loyal users - breaks down completely (Lal 1990b, p. 248). Why do manufacturers in mature product categories deal when they know that customers will not become loyal users? The brand interdependence explanation of promotional depth is based on asymmetry of competitive draw. Local brands or store brands on promotion do not draw sales from the name brands, while promoted name brands draw sales that would otherwise go to local brands (Blattberg and \Vrsniewski 1989). The brand interdependence explanation of promotion depth uses the idea of market position to predict depth of promotional discount. The brand interdependence literature conceptualizes "market position" or "strength" in terms of brand loyalty. Raju, Srinivasan, and Lal (1990, p. 276)define ”a brand's loyalty as the price differential needed to make consumers who prefer that brand switch to some competing brand. " Brands that command price premiums fi'om some proportion of the market are "stronger" than brands that cannot. The proportion of price-switchers who are attracted to strong brands when on promotion give national brands a straightforward tradeofl‘ between margin and volume. The profitability of this tradeoff provides the national brands with a means of selecting a (profit-maximizing) price discount while on promotion. One assumption of this approach is that national brand firms will not defect on one another by promoting simultaneously. As a practical matter, national brands do not defect because retailers try to maximize category and store profit. They merchandise only one brand in a category each week (Walters and MacKenzie 1988, p. 55), a policy which 58 effectively coordinates national brand promotions so that they generally do not overlap (Pasztor and Reibstein 1987). In the absence of the retailer's coordinating role, it could be argued that national brands see promotions as an infinite horizon prisoner's dilemma and avoid antagonizing one another with competition that will lower profits for all (Axelrod 1984; Fader and Hauser 1988). The coincidence of the joint incentives of the national brands with the profit goals of the retailer, though theoretically interesting, preclude the need for this argument. Depth of deal was left as an exogenous variable by Kinberg, Rao, and Shakun (1974). In laying out their argument, however, they assumed that national brands were promoted at the same price discount and drew competitors from the switcher population in proportion to nonpromoted market shares (ibid., p. 952). Although they did not incorporate it into their model, the authors apparently saw a connection between brand strength and promotional discount. This link subsequently has been used to ”endogenize" deal magnitude into brand interdependence models by Narasimhan (1988, p. 435), Lal (1990b, p. 249), and others (Rao 1991, p. 133). The brand interdependence approach is consistent with the findings of the implicit coalition literature (Axelrod 1984; Fader and Hauser 1988) in several respects. The asymmetry between brands with strong and weak market positions gives joint incentives to strong brands. Strong market positions that cause kinked demand curves are a source of ”competitionless" rents for each national brand, providing something like side payments for these brands at high prices. These rents may function like an outside player stabilizing a game (Fader and Hauser 1988, p. 553). A second joint incentive is limiting competitive encroachment (Lal 1990a). Ifnational brands can profitably absorb all the available promotional media, they preempt the local brand's share of voice (Lal 1990b, p. 432). 59 As the importance of retailers to promotions is recognized, one feature of the brand interdependence literature that increases in appeal is its ability to model retailer as well as manufacturer incentives (Lal 1990b, p. 429). Lal notes that although the retailer coordinates the alternating promotions strategy, prices at the retail level depend on the manufacturer's wholesale prices (ibid., p. 433). The brand interdependence view is of retailers promoting one brand at a time because they can accept one trade deal at a time rather than because they view multiple brand promotions as cannibalized profit (ibid.). In summarizing the literature, the mixed empirical support for the product, category, and retailer factors is not surprising when one keeps in mind that these variables are surrogate measures of the contribution margins, competitors, and volumes that drive profitability. The strong association of manufacturer-controlled factors (trade deal magnitude, ad size, and display size) with discounts (Walters 1989) also makes sense. The channel control literature indicates that manufacturers can influence the retailer's behavior as long as they are willing to pay (Bucklin 1973, p. 42). Understanding this profit-division problem, however, is a difi‘erent question than understanding which promotional depth is most profitable. _G_._§rmn_uy The tournament work of Axelrod (1984) and F ader and Hauser (1988) provides an intuitively appealing foundation for the brand interdependence explanation of promotions. The research stream beginning with Kinberg, Rao, and Shakun (1974) has progressively elaborated the connections between the fi'equency and depth of promotions and: (1) brand loyalty, (2) the asymmetric competitive draw of name brands, and (3) the frequency and depth of promotions. 6O Kinberg, Rao, and Shakun (1974) initiated the product-interdependence approach with a mathematical model that demonstrated the importance of relative market share to fiequency of promotions in a triopoly. Promotion depth was left as an exogenous variable. Narasimhan (1988) then made promotion depth an endogenous variable by connecting market share to both promotion fi'equency and depth of price discounts in a two-brand world. Raju, Srinivasan, and Lal (1990) extended the analytical mixed-strategy interpretation of promotions. They showed that if loyalty to one brand falls below a critical level it becomes optimal for other firms to ofl‘er price discounts. They also extended the number of brands modeled from two to any number in a product market. Finally, these authors tested and found support for two implications: (1) an increasing number of brands (and firms) in a category increases the likelihood of promotion, and (2) stronger brands will promote less than weaker brands. The depth of discount on promotion was developed in their model but not empirically tested. Lal (1990a) showed that the connection between market preference factors and promotion behavior can be characterized by a pure strategy Nash equilibrium. Of interest to Lal are fi'eedom to price discriminate against loyal customers and the prisoner's dilemma rivalry for price switchers. On the basis of discounted retums over an infinite horizon, Lal's analytical framework demonstrates what F ader and Hauser (198 8) would call an implicit coalition for brands with loyal customers to collude against brands without loyal customers. Lal (1990b) extended this fiamework to include store brands, calendar marketing agreements, and retailer pass through. Finally, Rao (1991) added to research in this field by conceptualizing customer preferences as drivers of promotion depth, fiequency, and ”regular" price. 61 The work on brand interdependence relies heavily on such analytically vague but intuitively appealing concepts as "premium brands" (Kinberg, Rao, and Shakun 1974, p. 950), ”brand switchers" (Narasimhan 1988, p. 428), "loyal consumers" (Raju, Srinivasan, and Lal 1990, p. 278), and "regular price" (ibid., 281). To date brand interdependence models have been constructed at the level of individual consumers and/or market segments (Kinberg, Rao, and Shakun 1974, p. 949; Narasimhan 1988, p. 438; Raju, Srinivasan, and Lal 1990, p. 279; La] 1990b, p. 430), and the implications have been drawn at the firm level. This approach has allowed significant progress to be made in explaining promotions, but the resulting explanations have not been strongly tested (Raju, Srinivasan, and Lal 1990, p. 298; Lal 1990b, p. 440). The key reason given for not rigorously testing these models is that data are lacking. Raju, Srinivasan, and Lal used chain level scanner data but causal data were unavailable (1990, p. 292); Lal had panel data available, but panels do not measure the essential constructs of his model (1990b, p. 430). The best panel data available contain purchases by household with associated causal factors (display, advertising, and promotion) but do not cover the variables needed for brand interdependence models, such as ”reservation prices" and market segments of ”loyals" and "switchers." In addition, Lal's data were fi'om the dishwashing category which may not be comparable to the soft drink category he developed his model to explain. Consequently, the brand interdependence researchers have used what data are available to test very general model implications at the market level. While not allowing the consumer-level theoretical constructs of the brand interdependence research stream, single-source data can support significantly improved tests by allowing tests of implications at the store, chain, and market level drawn from 62 brand interdependence models. The brand interdependence research stream has developed a collection of testable implications about promotions from a variety of analytical approaches. Lal alone has contributed to three distinct models, the most recent of which - because it is based on net present value proofs - is hardly distinguishable from microeconomics. Despite some diversity in the analytical representation used to characterize brand interdependence (that is, finitely repeated versus infinitely repeated games), there is extensive qualitative agreement about the linkages between brand preferences, product loyalty, competitive draw, and the incidence of promotions. This ”triangulation" of similar results from different models increases the chance that the inferred relationship is valid; triangulation makes the inference less fiagile and, consequently, more believable (Learner 1983, p. 38). III. Procedures and Method of Investigation A. Intr ' n This chapter presents the hypotheses, methodological procedures, and definitions of the variables necessary for understanding and interpreting the research results. First, the research questions are restated. Second, data for the study are reviewed. Third, hypotheses are developed. Fourth, operational hypotheses and techniques for their analysis are developed. B. R h estions Although the brand interdependence literature is rich in terms of theory, empirical testing has yielded conflicting results. This research seeks empirical evidence in three theoretical areas: alternating promotions; a strong brand versus weak brand pattern of promotional dominance; and the literature's predictions about deal magnitude for strong and weak brands. The three research questions are: 1. Is the soft drink category characterized by alternating promotions among premium national brands? 2. Is there competitive parity among strong brands and competitive asymmetry between strong and weak brands? 3. Are promotional discounts equal for strong brands, and are strong brand discounts larger than discounts for weak brands? 63 C. The Data Besides information from the promotions literature, two types of data were used in this study: (1) single-source data on the soft drink category and (2) behavioral data on how soft drink grocery promotions operate. 1. Single-Source Data Initially, management at A. C. Nielsen agreed to provide three years of store-level weekly carbonated beverage data for the top 50 cities in the United States - all 3,000 stores in the Nielsen database. When these data were requested, however, 24 packed- binary mainfiame tapes arrived at Nielsen in Chicago fi'om the corporate data processing center in Milwaukee. One week of Sun workstation and technician time would have been required to decompress and format these data, a cost so substantial that a written request for higher management approval would have been necessary. Because Nielsen's budget was tight, it was doubtfirl the request would have been approved. When Drs. Calantone and erson were consulted, they concurred that cutting the sample down to two cities would not impair the study. The resulting data set consists of three years of weekly data for 117 Michigan stores: 80 in Detroit and 37 in Grand Rapids. Nielsen organized the data into three 52-week years running from June to June. Detroit and Grand Rapids were selected because they inchrde the operations area of a cooperating soft drink bottler. A brief description is given here in order to document how the data shape the operationalization of the study. The datawere provided in four tables, as shown in Figure 31 Each table consists of variables in columns and data points in rows. Figure 3-1 contains three data points fiom each table for the third year of data. 65 Figure 3-1 Nielsen Single-Source Data Structures Weeks=52 Records leek Rumor tree): 30 90 2 07 07 90 7 07 14 9o 7 Store-Weeks=4l898 Records Storm State County City Ween!“ ACV Usability CustCount 1 21 050 nrrr 1 9100000 1 13045 1 21 050 bar 2 9100000 1 13045 1 21 050 arr 3 9100000 1 13045 UPCs=1,645 Records um UPC Code UPC Description Ounces nulti Pack 1 007020200005 A s I MB? C1" 12.000 1 2 007020210103 A s I RTBP C!" CH 12P 12.000 12 3 007020210105 A s I m! C! or 24 P 12.000 24 Ddump=l,099,66l Records Avg sell Sell Price Special ltereil'u- W W Unit Sales Price mltiple Obs 00 2 1 54 50 1 0 00 3 1 6 50 1 0 00 4 1 10 50 1 0 Not all variables in the single-source data were used in the study. For example, three identification codes for stores (Nielsen store code, Progressive Grocer store code, and an undefined "Masked ID” store code) were redundant to the store number and were not included. In total, eight redundant, inapplicable, and/or undefined variables were excluded in order to reduce disk space and processing time, and because they did not add information to the study. The most significant data issue for this research regards the "special obs" field of the ddump table. Store-level causal advertising, couponing, and display information that distinguish single-source data fiom scanner data are coded into this field. Nielsen's coding method assigns one bit to each tracked variable. This collapses the seventeen causal variables into one six-digit field by assigning powers of two to each bit and adding up the sum across all bits. To illustrate how the special obs field works, the causal variables, 66 their associated conversion values, and a worked out example for one value of the special obs are displayed in Table 3-1. Table 3-1 Example Coding of Causal Dat Binary to Code Causal c- line n-ad or ad store 8- b- illustration- best food famil tor- of ai in ai of This method is capable of capturing very subtle merchandising effects (multiple display types in a single store) for each UPC on a weekly basis. The richness of this coding is significant for this study because the range of expression captured in the single-source data highlights the oversimplification of promotion practice implicit in the brand interdependence research. To understand the actual complexity of soft drink promotions, the ddump table was queried for the number of distinct values in the special obs field. In Grand Rapids in year 3 there were 491 distinct values of special obs. Because each value of the special obs field corresponds to a unique combination of causal variables (that is, a distinct promotion), the single-source data indicate that 491 unique promotions were used in Grand Rapids in year 3. This compares to 641 promotions in Detroit during the same period. While this complexity is small relative to the 17 factorial (or 355.5 trillion) possible combinations of 67 causal variables, it is quite large in relation to the assumptions in the literature. Brand interdependence research has, without exception, assumed that promotions are either on or 08‘ and that they are identical for each competitor. A second discrepancy between the literature and the single-source data used here is the number of competitors. The literature has not singled out an operationalization of competitors; consequently, products, product lines, and firms have each been operationalized as competitors. For the most part, theoretical papers have , assumed " one product, one competitor, " while empirical papers have found approximately equal support for brands and firms as competitive units (Kinberg, Rao, and Shakun 1974; Raju, Srinivasan, and Lal 1990; Lal 1990b). This lack of unanimity is not an indication of theoretical weakness (Raju, Srinivasan, and Lal showed that the theory should hold for an "arbitrary but finite" number of competitors [1990, pp. 282, 295]) it is an issue because the single-source data are so rich. As can be seen fi'om the number of products in the UPCs table there are nearly as many products in the soft drink category in Grand Rapids and Detroit (1,645) as there were products (2,200) in the 29 categories studied by Raju, Srinivasan, and Lal (1990). Whereas the literature assumes a narrow range of potential behaviors in order to isolate competitive interplay, the single-source data are designed to capture a wide range of behaviors. The data are easily rich enough to allow "armchair” operationalization and testing of the brand interdependence literature. When occurring together such causal variables as "m-ad," "display-endaisle,” and ”coupon ad” could be assumed to indicate promotion, and brand families could be formed into competitors based on prior knowledge. There are two problems with an armchair strategy, however. First, the power of the test rests on assumption rather than observation. Second, the coding complexity is so great that without a concrete understanding of competition in stores and how single- source data represent competition, valid interpretation of the data is impossible. 68 To summarize, the single-source data are an almost overwhelmingly rich and complex source. The complexity of the 17 causal variables in the data challenges the assumption implicit in the brand interdependence literature that promotions are either on or off. The number of products in the category raises the question of whether brands or firms should be treated as competitors. The complexity mismatch between data and theory demands that the researcher understand practice in order validly to operationalize single-source data into brand interdependence constructs. A cooperating soft drink bottler provided this information. 2. Behavioral Data Brooks Beverage Management, Inc. (BBMI), a large bottler of nationally distributed brands (Squirt, Vemors, 7Up, DR. Pepper, RC Cola) agrwd to provide information on soft drink merchandising. A detailed discussion of data provided by BBMI is contained in Appendix A. Highlights are noted here to explain how promotion competition operates in the soft drink category and how single-source data capture and express this process. From the outset BBMI confirmed that alternating promotions are a ubiquitous and important facet of soft drink merchandising. Consistent with the trade literature, alternating promotions take the form of bottler-retailer contracts called calendar marketing agreements (CMAs). CMA costs at BBMI are growing at more than 65 percent per year. Four issues of single-source data interpretation and operationalization surfaced during interviews with BBMI: (1) retailer compliance, (2) promotion intensity, (3) product line merchandising, and (4) promotional bundling. Managers at BBMI stated: "We set up programs, and the retailers do whatever they want with them.” This compliance issue is important for two reasons. First, it highlights the innate heterogeneity of the retailing environment. Second, if retailers do not comply 69 with CMAs, the "special obs" field for a single promotion will vary across stores. Theoretically, because CMAs are negotiated with chains, every store in a chain with a CMA in force should have the same merchandising program (same value of the special obs field in ddump). Because all stores in a chain may not comply with promotions, because each store may comply idiosyncratically, and because the sensitivity of Nielsen‘s store audit detects compliance variation, bottler promotions cannot be "read" from the single- source data by looking up specific values of the special obs field. A second data issue is promotion intensity. Soft drink merchandising managers think of promotion as an ordinal variable with three levels: (1) no promotion, (2) secondary promotion, or (3) feature promotion. Feature promotions support the bottler’s products with all the promotional tools available to retailers. Secondary promotions involve smaller price discounts, less conspicuous displays, and less than the maximum of advertising. The price schedule in BBMI's CMA is set up in terms of these threepromotion levels. Merchandising control is a third issue that is important to operationalizing and interpreting scanner data. Bottlers are the effective competitive agents of the national brands they sell. Promotions are negotiated between retailers and bottlers and usually run by package across a bottler's product line. If, for example, 2-liter bottles are offered on promotion, all of a bottler‘s brands in that package size are included in the promotion. The fourth issue in understanding how competition will be manifested in single-source data is promotional bundling. Bottler-retailer promotions are in effect "bundles" of feature advertising, price discounts, and special displays. Bottlers quote prices to retailers contingent on retailer pass-through of these elements. Figure 3-2 is a taxonomy of possible soft drink promotions based on all possible combinations of retailer merchandising (Duffy 1991). BBMI was asked what the volume sales increases would be fiom each of these combinations. Numbers in the diagram indicate the bottler's judgment 70 about the efl‘ect of these promotional elements (individually and jointly) on unit sales as multiples of nonpromoted sales (Aukeman 1991). The bundled nature of promotions emerged while the BBMI marketing director was being interviewed to fill out Figure 3-2. When asked what the sales response would be in region ”B,” he replied: "Well, that should never happen [because the CMA does not allow it]." When asked about region "C," he replied: "That should never happen, but it does [because of partial retailer compliance]." Figure 3-2 Lifl Factors for In-Store Promotions No Promotion A \ , \ 'x 1 Source: Aukemln(l991). Bundling is an important theoretical issue because it appears to be the mechanism for coordinating alternating promotions. Retailer feature advertising (”featuring”) is the critical element in the bundle because - due to CMA clauses - Coke and Pepsi cannot be feature advertised simultaneously. Because deep price discounts are bundled with featuring, feature advertising effectively coordinates the alternation of deep discounts. Bundling is an important data issue for operationalizing brand interdependence because the bundling of promotional elements defines the form that alternating promotions take 71 and, consequently, the single-source data patterns that can be used to isolate alternating promotions. To summarize, the information provided by the bottler on its own and its competitor's promotion behavior indicates the following: 1. Aggregation is a significant source of diversity. Store compliance within and across chains and variation of promotion policy across chains reduces the chance that promotions can be easily identified in single-source data. 2. Alternating promotions should manifest themselves in alternation of bottlers who are on feature promotion. 3. At least in these markets, bottlers are the effective competitive actors on behalf of the national brands. 4. The exclusivity clauses in Coke and Pepsi CMAs and the practice of bundling featuring, display, and price discounts should prevent both Coke and Pepsi from being on a feature promotion at the same time. Thus, bottler information gives strong guidance on how the brand interdependence constructs can be operationalized to take best advantage of the single-source data. D. Research Hymtheses The following hypotheses translate the brand interdependence literature's panel-level theory into implications at the store, chain, and market level that can be addressed with single-source data. While the most common market-level variable discussed in the literature is market share, it will not be used in this study. Instead, implications are developed in terms of physical unit sales and dollar sales. Figure 3-3 displays an ounce measure of market size for the Grand Rapids area in year 3. The data consist of a simple sum of the ounces sold by bottler in all 37 stores in the Nielsen Grand Rapids sample. The data in Figure 3-3 nm fiom June 1990 to June 1991. While the brand interdependence literature assumes constant category size (Kinberg, Rao, and Shakun 1974; Narasimhan 1988; Raju, Srinivasan, and Lal 1990), the data for soft drinks are highly variable. For 72 example, week 2=July 4, 11=Labor Day, 24=Christmas, 28=Superbowl, 48=Memorial Day, and 52=the second week previous to July 4, 1991. This raises diflicult questions of how to operationalize market share - by weeks, quarters, or years. Millions of Ounces Figure 3-3 Grand Rapids Soft Drink Market Size, Year 3 ‘ ~. . ,. , ,": W3?" ‘q .é‘i‘ir t“,- ESE-3‘4"“. vth~,-a{.¢e.+wu ‘ .'. . ."’..-'..-.u.¢'. ‘, ‘ l4 71013l6192225283l34374043464952 Weeks .CANADADRY .CANFIELD ICOKE EJ7UPIRC/DR IFAYGO ElPEPSl .STOREBRAND .UNKNOWN It is important to keep in mind that the theoretically important variable is profit, not volume or market share (Totten and Block 1987, p. 23). In the brand interdependence literature, market share is a surrogate measure for profit (Kinberg, Rao, and Shakun 1974; Lal 1990b; Raju, Srinivasan, and Lal 1990). Because market share does not provide any information about profit that is not already in the single-source data, the implications suggested in the brand interdependence literature have been developed in terms of unit and dollar variables that can be directly observed in single-source data. 1. Resegch Quefiion 1 The first research question asks: I. Is the sofi drink category characterized by alternating promotions among premium national brands? 73 The literature clearly predicts that brands with approximately equal sales and brand loyalty will alternate promotions (Kinberg, Rao, and Shakun 1974, p. 955; Lal 1990b, p. 433). Alternating promotions for soft drinks have been cited in support of brand interdependence models (Lal 1990b, p. 43 7). BBMI stated that CMAs are in force every week at large chains. Consequently, it would be a genuine surprise and potentially a refutation of the brand interdependence approach if alternating promotions for soft drinks could not be observed at large chains. Two kinds of alternating promotions are hypothesized in the literature: within-chain and between-chains. Since each form of alternation can be tested, separate research hypotheses are stated for each. H1: Within chains, alternating promotions should take place during no less than 66 percent and no more than 94 percent of available promotion periods (Kinberg, Rao, and Shakun 1974, p. 958). Hz: A single national brand (such as Coke) should not be promoted in two different chains at the same time (Lal 1990b, p. 436). 2. Research Question 2 The second research question asks: 2. Is there competitive parity among strong brands and competitive asymmetry between strong and weak brands? The literature efl‘ectively asserts that competition is a symmetrical zero-sum game among strong brands and an asymmetrical zero-sum game between strong and weak brands (Lal 1990a, p. 251). To test this assertion, two effects - "strong brand versus strong brand” and ”strong brand versus weak brand” - can be examined separately. If the strong brands are generating equal quantities of sales, alternating promotions should be observable (Hl and H2). If, however, the strong brands are not generating equal sales, then the weaker brand should be promoting more often than the stronger (Raju, Srinivasan, and Lal 1990). The hypotheses for these brand parity efi‘ects are: 74 H3: Brands that alternate promotions will be approximately equal in sales (Kinberg, Rao, and Shakun 1974, p. 955; Narasimhan 1988, p. 441). H4: The brands that alternate promotions are the beneficiaries of an asymmetry in competitive draw from private label brands (Kinberg, Rao, and Shakun 1974, p. 952; Blattberg and Wisniewski 1989, p. 303). 3. Research Question 3 The third research question asks: 3. Are promotional discounts equal for strong brands, and are strong brand discounts larger than discounts for weak brands? The literature relies on the concept of Nash equilibrium to enforce alternating promotions (Lal 1990b). That is, firms choose strategies with which they are jointly comfortable (Moorthy 1985, p. 264). Accordingly, strong brands should be comfortable with the same promotional discount, and strong brands should have greater promotional discounts than weak brands (Narasimhan 1988, p. 435; Raju, Srinivasan, and Lal 1990, p. 286). The hypotheses for these Nash equilibrium efl‘ects are: H5: Brands that alternate promotions should have approximately equal discounts (Narasimhan 1988, p. 435; Raju, Srinivasan, and Lal 1990, p. 286). H6: Strong brands on promotion should have greater discounts than weak brands (Narasimhan 1988, p. 435; Raju, Srinivasan, and Lal 1990, p. 286). These hypotheses are the strongest theoretical test performed to date. For the first time, soft drink data will be used to test the brand interdependence theory that has been refined to explain soft drink promotions (Lal 1990b, p. 436). For the first time, single- source data and CMA data have been integrated. For the first time, it has been possible to group products by actual competitors. For the first time, interchain as well as intrachain data can be used to test brand interdependence hypotheses. Finally, for the first time, three central theoretical implications will be simultaneously tested: (1) alternating 75 promotions, (2) asymmetric competitive draw, and (3) promotional discounts. Table 3-2 summarizes the research questions and corresponding research hypotheses. Table 3-2 Research Questions and Research Hypotheses l. hhfldflmWMhmm‘pmmpMMM? H1: mmmmpmmmmammmmapmumme—u Mdmmmafihbflphdflflmlwapm H2: Aflfloflbr-l(nchaCoh)Meuhflo-wdhmmc&sdfie~head199th F436)» 2. hhnWeWmMMflWeMMm‘m-dwflhflfl H3: “MMWHRWMQHhmmhMMdShk-lfltpm Nor-mu l988.p.441). H4: mmmmmmmuma-thummmm mmm.musru—rs74,p.m; Mtg-IN 1989,p.303). 3. mWWQHMMMdmmbr-dmmhwm" he‘s? H5: MMderflepWMMesliyqdmmu-hhlmpflstm m-IUIMFZSS). H5: “swumflmmwumu—emm—hrmpmm w-IMIMJJ“). E. Vm'gble Identification and Qperationalization The variables used in this study, which differ by hypothesis, are defined below with each hypothesis. Because of the great complexity of the single-source data, considerable time was spent by the researcher in understanding the definitions of each field and the behavior of the data. As much information as possible was obtained about how Nielsen collects and codes single-source data. While " good statistical practitioners have always looked in detail at the data before producing summary statistics and tests of hypotheses" (Hoaglin, Mosteller, and Tukey 1983, p. 1), the researcher did not pretest the hypotheses. In the following operationalizations the results reflect a thorough data exploration. For example, while end-aisle displays are an important component of CMAs, and there is an end-aisle display variable in the single-source data, data exploration revealed that not one end aisle display occurred in year 3. None of the research hypotheses reflect preliminary "peeks" at the data. In addition, model specifications are developed based on existing models in the literature and the details of promotion practice. Anticipated changes to model specifications are noted with each model. 76 1. Hypothesis 1 H1: “Within chains, alternating promotions should take place during no less than 66 percent and no more than 94 percent of available promotion periods (Kinberg, Rao, and Shakun 1974, p. 958). To test H1, market, competitors, grocery chains, promotions, strong brands, available promotion periods, and alternation each must be operationalized. a Stud Market: The possible markets for this study were Detroit and Grand Rapids, Michigan. The latter was chosen for three reasons. First, for the primary year of analysis (year 3), the single-source data for Grand Rapids had a much smaller percentage of missing data MissimStae-Wecks _ 124 _ - - - MissingStore—Weeks _ 1,032 __ Possible Steamed: _ 1,924 " 6'4 percent nussrng) than Detrort (Possible Store-Weeks ’ 4,160 " 24.8 percent missing). Second, the ddump table for Grand Rapids was 48 percent smaller (423,679 records) than the ddump table for Detroit (675,982 records). Third, Brooks Beverage Management, Inc., personnel made available to the researcher were responsible for the Grand Rapids market, and the promotion practice information provided was drawn from Grand Rapids. The Grand Rapids market is operationalized as the data in the Nielsen sample fi'om that city. This market consists of 37 reporting stores in 19 grocery chains. The geographic area covered in this sample, however, includes store fiom both Kalamazoo and Lansing. Because location of stores is a closely guarded secret (stores are not supposed to know if they are in the Nielsen sample), it was not possible to narrow the study to stores within the immediate Grand Rapids area. 77 b) Competitors: Previous work in the literature has interpreted " competitors" as either individual products (Kinberg, Rao, and Shakun 1974) or both products and firms (Raju, Srinivasan, and Lal 1990). For soft drinks, these definitions represent two polar extremes because of the substantial role bottlers play midway between the products and the national firms (Aukeman 1991; Duffy 1991; Oliver 1986, p. 33). Because in each market national brands are merchandised to retailers by bottlers, because CMAs are agreements between bottlers and retailers, and because CMAs structure the merchandising of soft drinks, in this study bottlers are considered the effective competitive actors on behalf of national firms. There is one exception. Although retailers do not bottle their own private-label products, these are classified as a bottler under the name ”store brand. " Bottlers have been operationalized by adding one field to the UPC s table to indicate which bottler merchandises a given UPC (see Figure 3-4). The information was obtained by interviewing people at BBMI and at the Beverage Bureau, a trade association. Figure 3-4 Example Data fi'om the UPCs Table UPCs=1.645 Records UPCNue UPCCode UPCDescription Ounces lulti Pack Bottler 1 007020200005 A s I R 8133 C! 12.000 1 Pepsi 2 007020210103 A 6 I a man C! on 12P 12.000 12 Pepsi 3 007020210105 A s I R m C! at 24P 12.000 24 Pepsi Each store in Grand Rapids sold an average of 231 soft drink UPCs each week, representing eight bottlers: 7Up/RC/DR.,l Canfield, Canada Dry, Coke, Faygo, Pepsi, store brand, and unknown. The latter indicates primarily bottled water and other beverage 1 To be consistent with the labeling of bottlers by their respective brands, Brooks Beverage Management, Inc., will be abbreviated as "7Up/RC/Dr. " to highlight the multiple brands (7Up, RC Cola, and Dr. Pepper) sold by this bottler. 78 products, included in the Nielsen soft drink category. Neither BBMI nor the Beverage Bureau could identify bottlers of these products. 0) My)! Chains: Operationalizing grocery chains using single-source data required two steps. First, the stores were extracted fiom the store-weeks table to produce a table (Figure 3-1) with one row of data for each store (see Figure 3-5). Second, a chain identifier was extracted from the "ProductionlD" field in the stores table to create a new variable, "chain.” This variable then was used to group data for within-chain and between-chain tests. Figure 3-5 Example Data from the Stores Table Stores=117 Records Store)?“ Productionln 1 A9054 530 The within-chain hypotheses were tested using the data from the grocery chain with the largest total sales in the Nlelsen sample. The single-source data for this chain contained five stores totaling 65,363 records. Because the identity of this chain is proprietary, it is referred to as ”Chain 3.” It was selected for three reasons. First, it had no missing data in year 3. Second, as the largest chain in the city, it was the most likely place for alternating promotions to take place. Third, the cooperating bottler supplied partial promotional schedules for 1990 that could be used to validate the operationalization of promotions. d) Promotions: Promotions do no emerge directly fi'om single-source data, which capture raw data on price, displays, and advertising. Classification rules were necessary to recode single- source data into promotion types. To set up these rules, a systematic relationship had to be found between soft drink promotion behaviors and single-source measurements. 79 Managers in charge of soft drink promotions identified three promotion types by name: (1) feature promotions, (2) secondary promotions, and (3) no promotions. These correspond to the price tiers in the bottler's price list. In addition documentation provided by bottlers and exploratory analysis of the single-source data established that a bottler's products can be simultaneously on feature and secondary promotion. Altogether, then, there are four kinds of soft drink promotions. 1. Feature promotions These include advertising in the store flyer, some form of display in addition to normal in-aisle display, and the deepest discount available in a given week. Feature promotions usually take the form of all a bottler's products within a package size being promoted. 2. Secondary promotions These include advertising in the store flyer, some form of display in addition to normal in-aisle display, and a smaller discount than the week's feature promotion. Additional display often takes the form of a "shelf talker" sign on the normal in-aisle display. Usually, all a bottler's products within a package size are promoted. 3. Combined feature and secondary promotion Types 1 and 2 run simultaneously for a single competitor's products. Usually, secondary promotion includes a package that is different fiom the featured package. This second package is given a moderate discount and additional display. 4. None. Products are not displayed outside the soft drink aisle, and there is no additional advertising. The specifics of each promotion type vary from week to week, and this complexity is captured in single-source data. For example, feature promotions might consist of 2-liter products at $.77 per unit one week, 12-pack cans at $2.65 another, or 6-pack cans for $1.49. The variation of promotion conditions takes place across advertising and display dimensions as well. Special displays with thousands of product units are built occasionally, special coupon advertisements are run periodically, and special deep discounts are occasionally offered. This means that feature promotions will not be seen in the single-source data as a particular combination of single-source variables. 80 In addition, three data collection characteristics afl‘ect how soft drink promotions are captured in single-source data. First, selling price, not price discount, is recorded. Consequently, classifying a promotion when, as often happens, the only difference between a feature and a secondary promotion is deal discount requires that the latter be calculated. The literature assumes that promotions offer a discount from a "regular price" (Raju, Srinivasan, and Lal 1990, p. 287). To date, however, the—soft drink category has defied all attempts at developing a satisfactory definition of "regular price" (Bender 1991). Table 3-3 displays the prices charged for a 2-liter bottle of Classic Coke within a single Grand Rapids grocery chain. Rows represent prices, columns represent stores, and the cell counts represent the number of weeks (columns sum to 52) in which this price occurred for this product. One appealing definition of regular price is "the consumer‘s expectation of the current price based on the consumer's previous observation of current price" (Slavic 1988a, p. 2), but the single-source data do not measure consumer expectations. In Table 3-3 the $.89 price might be justified as "regular" because it is the most fi'equent. Yet, because soft drinks are stock up goods (Litvack, Calantone, and Warshaw 1985), an argument can be made that customers use the minimum price - rather than the most frequent price - as a reference (Totten 1991). Recent research has shown that customers can recall soft drink promotion fiequency better than any other category (Krishna, Currim, and Shoemaker 1991, p. 9). In particular, 11.1 percent ofcustomers can correctly recall both deal magnitude and fiequency of Coke promotions (ibid., p. 11). In the brand interdependence literature, the key "regular price" is the reservation price of the market segment loyal to premium brands (Raju, Srinivasan, and Lal 1990, p. 287). Again, this information is not available in single-source data. 8] Table 3-3 Prices for 2-Liter Classic Coke Chain 3 - Year 3 Store Store 109 110 3 1 1 4 2 32 5 2 Because consensus about how to operationalize "normal price" and, consequently, price discount has not yet developed, a practical compromise is needed. In this study discounts are calculated based on the highest price observed within each chain. The highest price is treated by bottlers as a reference price for negotiations with chains. This heuristic is the standard approach taken in the brand interdependence literature (Narasimhan 1988, p. 428; Raju, Srinivasan, and Lal 1990, p. 294). Display measurement is the second data collection characteristic affecting how soft drink promotions are captured. Although there are eight display variables in the single- source data, it was determined that Nielsen's coding scheme allows valid detection only of the presence or absence of displays for soft drinks. For example, because all the CMAs obtained stipulate either end-aisle or displays away from the soft drink aisle, end-aisle and store lobby displays would seem to be natural measures. Exploration of the data set revealed, however, not one end-aisle or lobby display in the data set for year 3. Correspondence with Nielsen produced the conclusion that ”we have not been splitting out display-end of aisle for '2 or 3' years" (Goering 1992). In addition, the display coding scheme includes heuristics that are inconsistent with straightforward interpretation of display variables; for example, "in stores which have a cross aisle in the middle, displays 82 that are on end caps in the middle of the store are actually coded as in-aisle displays" (ibid.) . It is important to note that the inability to extract measures perfectly consistent with CMAs is not a defect in single-source data. Rather, the enormous complexity of the grocery environment makes pragmatic compromise inescapable. The researcher's interpretation of display as being either "on" or "off" is consistent with how Nielsen uses the display measure (Bender 1991). Advertising measurement is the third data collection characteristic affecting how single- source data represent promotions. Again, the distinctions used by Nielsen to code the nine advertising variables do not capture CMA terms. In the case of advertising, however, the disparity between data coding and promotion practice is more likely attributable to the interaction of retailer merchandising variety and Nielsen's coding scheme. CMAs state advertising requirements in subjective ("highest form of advertising") rather than more objective ("illustrated ad covering 9 square inches of the first flyer page") terms. Since retailers vary their promotions and advertising placement weekly to avoid wearout, a soft drink feature promotion may have a large and prominently placed illustration in the store flyer one week, but an equivalent feature the next week may be smaller and less prominently placed. The bottler‘s stipulation for "highest form of advertising” in practice appears to mean "highest form of soft drink advertising in a given week. " Ifthe same brand were featured two weeks in a row, the advertising scores for the two feature promotions would be difl‘erent. While advertising is executed on a relative basis, the coding of advertisements into single-source data is more objective, using such criteria as prominence on the page and ad size. Exploratory analysis of soft drink promotions and the corresponding advertising variables brought this "relative advertising 83 versus objective coding" issue to light. Table 34 displays the correlations among the single-source advertising dummy variables. Table 3-4 Relationships among Causal Advertising Variables Chain 3 - Year 3 ADLINI PEARSON CORRELATION MATRIX NUMBER OF OBSERVATIONS: 303 Note: ThemhthisubleueouldeeednmforNielsclsadvctismgcamdvuiables As can be seen from this correlation matrix, counterintuitive combinations are possible. For instance, the correlation between ADMAJOR and ADB is higher (.714) than between ADMAJOR and ADA (.080). The most important combination from a bottler‘s perspective is ADILLUS and ADMAJOR. The .440 correlation can be interpreted to mean that 44 percent of the illustrated ads were large and prominently placed. As was the case for displays, the researcher concluded that single-source data indicate only presence or absence of advertising. On the basis of the above discussion, promotions are defined as follows: 1. A feature promotion takes place whenever there is any display, any advertising, and a price discount. 2. A secondary promotion takes place whenever there is any display, any advertising, and a smaller price discount than for the featured brand. Lfimflm Coke and Pepsi have been considered the "strong brands” in the soft drink category by observers of alternating promotions (Lal 1990b, p. 436; Williams 1983, p. 168; 60 84 Minutes 1987; Pasztor and Reibstein 1987; Consumer Reports 1988, 1991). Because, as discussed above bottlers are the effective competitive actors on behalf of national brands, in this study bottlers (as opposed to brands or UPCs) are operationalized as competitors. Because the brand interdependence literature has developed largely in terms of market share, bottler strength might be operationalized by bottler market share. Figure 3-6 plots weekly market share by bottler in Grand Rapids in year 3. Ifmarket share is used to determine strength, store brands are the strongest during stock up weeks. Yet, they are considered - almost by definition - weak brands by researchers (Raju, Srinivasan, and Lal 1990). A market share criterion for bottler strength clearly leads to an interesting view of the market. Figure 3-6 Grand Rapids Weekly Market Share by Bottler 100% , 80% g 60% 0 40% ' 20% .__ . , , l4 7101316192225283134374043464952 Weeks ICANADADRY ICANFIELD ICOKE D7UP/RC/DR IFAYGO DPEPSI ISTOREBRAND IUNKNOWN The literature explains tacit cooperation by implicit coalitions, but a market share operationalization of strength is not consistent with this explanation. The critical issue is whether some competitors recognize joint incentives to restrain behavior in the face of unrestrained behavior by others. Coke, Pepsi, and 7Up/RC/DR. as national brands have the necessary joint incentives to form an implicit coalition. These national brands compete 85 with one another in every market against a fragmented assortment of regional and store brands. From the implicit coalition perspective, the store brand (particularly in peak demand weeks) is a noncooperative player with whom the national brands resist competing on unfavorable terms. Because bottlers, not brands, are operationalized in this study as competitors, an implicit coalition at the level of the bottler is required to define the "strong" bottlers. An implicit coalition is encouraged by three factors: (1) competitors perceiving their self- interests to be similar (Nagle 1987, p. 87); (2) competitors recognizing that communal defection (price cutting) may be too severe a strategy (F ader and Hauser 1988, p. 566); and (3) an efl‘ective system for detecting price cutting (Nagle 1987, p. 90). Because bottlers of national soft drink brands compete with one another throughout a geographic market against a fi’agmented assortment of store and local brands, they are likely to perceive similar self-interests. Because the national brand bottlers recognize that at least some segments of the population prefer their products and are willing to pay premiums over store brand products, they recognize that price cutting to compete with fragmented competitors is suboptimal. Because national brand bottlers subscribe to syndicated data sources, negotiate prices carefully with chains, and distribute direct to stores ever week, they have ample information as to competitive price ofl‘ers. Consequently, "strong brands" in this study are operationalized as national brand bmers (Coke, Pepsi, 7Up/RC/DR.) because these bottlers appear to have an implicit coalition. Bottlers of store brands, Faygo, and Unknown have been characterized as "weak" brands because they do not have suflicient mutuality of interests or a sufficient number brand loyal customers to form an implicit coalition. 86 fl Available Promotion Periods: The brand interdependence literature implicitly has made several assumptions about promotions. In the preceding review of single-source data, the assumption that promotion is either "on" or ”ofi" was discussed. Another assumption is that only one competitor can promote at a time. At least in the soft drink category, this assumption is false. Secondary promotions, by definition, run parallel with feature promotions. Another intuitively appealing assumption is that only one competitor can be on feature or secondary promotion at a time. Exploration of soft drink promotions in Chain 3 and other chains showed this assumption to be false. Retailers sometimes run two secondary promotions in a week and even dual features once or twice a year. Promotions are not mutually exclusive. The brand interdependence variable "proportion of time on promotion" (Kinberg, Rao, and Shakun 1974) is operationalized in this study as the proportion of 52 available weeks that a bottler has products on promotion. The most salient alternative formulation would be to operationalize time on promotion as share of promotion. The considerable overlap of bottler promotions, however, would confound this measure. Alt tion: The last operationalization issue in testing hypothesis 1 is to define an alternating promotion. This issue has not been covered extensively in the literature. Kinberg, Rao, and Shakun (1974, p. 957) developed hypotheses about alternation in terms of the percentage of time bottlers were on promotion. Instead of using this measure in the first empirical test of alternating promotions, Lal (1990b, p. 439) used a chi-square to test for negative association between promotions of two firms in an implicit coalition. Thus, two difi‘erent operationalizations of alternation are implicit in the literature: percentage of time on promotion and negative association. Operational hypotheses for this study were developed using each approach. 87 h) Qfirational Hypothesis 1a The first research question for this study asks: "Is the soft drink category characterized by alternating promotions among premium national brands?" The first research hypothesis states: "Within chains, alternating promotions should take place during no less than 66 percent and no more than 94 percent of available promotion periods." To say that alternating promotions are taking place does not specify when or for how long an individual brand is on promotion. Operational Hypothesis la (OH la) links alternating promotions to the time strong bottlers are on promotion. OH la: Each premium brand should be on promotion between 33 percent and 4 7 percent of the time (Kinberg, Rao, and Shakun 1974, p. 958). The brand interdependence literature implies that strong bottlers will form an implicit coalition to extract rents from nonprice-sensitive market segments. In the extreme case, a duopoly of strong bottlers would alternate promotions during all the promotion time available (Kinberg, Rao, and Shakun 1974, p. 95 5). In the real world two forces reduce the time national brand bottlers are on promotion. First, retailers have almost the same standing as a manufacturer of strong brands (Williams 1983, p. 167; Kumar and Leone 1988, p. 178). This should lead retailers to promote their own brands in order to force national brand bottlers to deal (Lal 1990a; Struse 1987, p. 150) and in order to capture higher margins fi'om private label products. Second, antitrust prosecution is presently a significant threat in the soft drink industry (Galvin 1990, p. 27). If a national brand bottler were to monopolize all of a chain's soft drink promotion, this behavior would be easily observable by competitors. Consequently, it probably is in the interest of national brand bottlers not to promote all the time, or not to the complete exclusion of weaker competitors. At the same time, given the strength of Coke, Pepsi, and 7Up/RC/DR., it 88 would be surprising to find that time on promotion is evenly distributed among bottlers (Lal 1990b). OH la will be tested by tabulating within 95 percent confidence limits the percentage of time each bottler spends on promotion. If the confidence intervals around the actual fiequency of promotion for national brand bottlers do not include a point in the range from 33 to 47 percent, OH 1a will be rejected. Table 3-5 displays how the data will be arranged for the test. Table 3-5 Testing Operational Hypothesis 1a What actual frequency of promotion would be expected by chance? Two heuristics can help provide a reference point for promotion frequency. First, if promotions are divided evenly among the eight bottlers selling products through Chain 3, then the frequency of promotion for each bottler should be approximately 12.5 percent. Second, the strongest competing explanation for promotion fiequency within-chain is share of business. Using rank of product in category, Walters (1989, p. 265) found mixed support for this hypothesis. If promotions are divided among bottlers on this basis, then the profile of promotion fiequency observed in Chain 3 should be parallel to Table 3-6. Table 3-6 Bottler Share of Units and Revenue in Chain 3 Store- war/mink. Cote Payvo Msi brand m Share of Units 9.50 22.40 3.90 17.20 41.40 5.30 share of Revenue 12.30 23.90 3.30 19.10 35.00 5.50 89 Because the brand interdependence literature has not distinguished between types of promotions and has considered promoting to be mutually exclusive, the best way to measure time on promotion has not yet been established. Four candidates are: ( 1) percentage of time on any kind of promotion, (2) percentage of time on feature, (3) percentage of time on secondary, and (4) percentage of time simultaneously on feature and secondary promotions. Consequently, all four measures will be calculated, and one table analogous to Table 3-5 will be developed for each measure. i) gmtional Hymthesis lb OH 1b: The promotions of bottlers of strong brands are not independent of one another (Lal 1990b, p. 437). The second approach to operationalizing alternating promotions is to test for independence. If members of an implicit coalition alternate promotions, their promotions should not be independent. Lal (1990b, p. 439) used a chi-square to examine the independence of promotions for three sizes of Sunlight (made by Unilever) and Palmolive brands of dishwashing liquid. The chi-square, however, runs the risk of misinterpreting relationships when promotions depend on more than two competitors, package sizes, and/or timing variables. Dillon and Goldstein (1984, p. 314) list three weaknesses of analyzing data two dimensions at a time. 1. Interactions between more than two variables cannot be identified. 2. Marginal association (association when all levels of a third variable are pooled) can be dramatically difl‘erent from partial associations (association for two variables when each level of a third variable is taken into account). 3. All pairwise associations cannot be taken into account simultaneously. 90 Lal's chi-square analysis, for example, left out Procter and Gamble's dishwashing detergent and, consequently, pooled across P & G's promotions. If P & G concentrated its promotions on a single package size or ran them in opposition to either Unilever or Colgate-Palmolive products, the chi-square test could not accurately detect the independence of Sunlight and Palmolive. As an example of the difference in marginal and partial association, in Lal's data the 12-ounce package is an exception to his conclusion of negative association (ibid., p. 43 9). That size of Sunlight and Palmolive was on deal together (positively associated) 19.2 percent of the time, while the association for two other package sizes was 5.8 percent and 3.8 percent, respectively. If P & G concentrated its promotions in the lZ—ounce size, Lal's conclusion that Sunlight and Palmolive promotions are not independent might be incorrect. It could be that in the face of concentrated promotion of P & G's 12-ounce product (a partial association) the Sunlight and Palmolive promotions are positively associated, while over all packages (a marginal association) the promotions are negatively associated. Partial associations can be the opposite of marginal associations because of fiequency differences among outcomes or strong associations among particular combinations of variables. The reversal of marginal and partial associations is called Simpson's paradox (Agresti 1990, p. 138). Lal's approach left out a level of one variable (interpreting P & G as a level of a competitor dimension) and computed individual chi-squares for each of three package sizes rather than representing package sizes as levels of a package dimension. Consequently, Lal's test could not detect interactions between more than two variables, distinguish reversal of marginal from partial association, or simultaneously take all the pairwise associations into account. The analysis was, consequently, subject to all three weaknesses cited by Dillon and Goldstein (1984, p. 314). 91 In testing for independence of soft drink promotions there are potentially four dimensions that should be included: (1) bottler type (strong and weak); (2) promotion type (simultaneous feature and secondary, feature only, secondary only, and none); (3) package size (6-pack, 12-pack, or 2-liter); and (4) time (number of promotions in past 12 months). Loglinear models are one tool statisticians have developed to take multiple dimensions into account when testing for independence. Because loglinear models can analytically express and test multidimensional arrays of categorical data, they are a natural technique for assessing independence in this situation (Bishop, Fienberg, and Holland 1975, p. 31). Loglinear models are a member of the family of generalized linear models (GLMs). GLMs are specified by three components: a random component, which identifies the probability distribution of the response variable; a systematic component, which specifies a linear filnction of explanatory variables that is used as a predictor; and a link describing the functional relationship between the systematic component and the expected value of the random component (Agresti 1990, p. 80). The link function is what distinguishes loglinear models from more familiar multivariate methods (see Table 3-7). Table 3-7 Agresti's Types of Models for Statistical Analysis Poisson Source: Agenti(l990,p. 82). While a regression and ANOVA connect dependent and independent variables with an identity link filnction, loglinear models use cell counts in place of a dependent variable and link cell counts to independent variables through a log function. The nature and interpretation of dependent variables in loglinear models differ fi'om classical inference. 92 Researchers using loglinear models distinguish between testing for independence to uncover a relationship and testing for association that seeks to understand the nature and extent of a relationship (Dillon and Goldstein 1984, p. 310). Loglinear models are essentially a means of holding constant the effects of many dimensions in order to avoid confounding independence tests of hypothesized relationships. Loglinear models give researchers the capability to identify the order of interactions present in the association of multiple variables. Ifthere are no three-way interactions, for example, multidimensional data can safely be collapsed into two dimensions (Bishop, Fienberg, and Holland 1975, p. 39). Loglinear modeling of the model in equation 3-1 and its submodels will serve as the means for identifying the order of interactions taking place in soft drink promotions. logmijkl= Constant+B+P+p+T+B*P+P*p+p*T+B*p+P*T+B"T+ B*P*p + P*p*T + B’p*T + B*P*T + B*P*p*T, 3-1 where mijkl = cell count for the multidimensional table made up of bottlers i, packages j, promotions k, and time I, B= bottler, p= package. P= promotion, and T= time. Three types of independence can be identified with loglinear models: mutual, joint, and conditional (Agresti 1990, p. 142). If the sub model of equation 3-1 containing individual predictor variables (no interaction terms) explains as well as or better than models with interaction terms, then the individual predictor variables are mutually independent. If a sub model containing interaction terms for three of the predictor variables (but not the fourth) explains as well as or better than other submodels, the fourth variable is jointly independent. Ifa partial table of three variables is independent for a category of a fourth variable (as Sunlight and Palmolive were for 22-ounce and 32-ounce package sizes), then 93 the categories are said to be conditionally independent. prredictor variables are jointly independent, then they will be marginally independent and can be collapsed into fewer dimensions (ibid.). The likelihood ratio chi-square and the Pearson chi-square statistics for the submodels of equation 3-1 will be used to test OH 1b. Table 3-8 displays the nine submodels to be tested. If at least one of submodels one through seven is not significant at or=.05, then it will appear that promotions of strong bottlers are independent, and Hypothesis 1b will not be supported. Eight submodels are to be tested in order to isolate the order of interaction efl‘ects and the most useful explanatory dimensions. Table 3-8 Submodels of Equation 3-1 to Be Tested MM Note: mmmmmmummmmmdmmwmmofimmm (Agresti 1990,p. 144). FaemthPpTabhwidaewflim3-lbyushgmewmwrewmahiaadfialmdel (amodelwithalllowerordermainefl‘ectsandinteractiom). BMforBottlesztandsforpromotiontypemstandsfor Mdenandsfortime. The researcher hypothesizes that time and package type may not be necessary to explain alternating promotions (that is, these variables may be at least jointly independent of bottler and promotion type), so sub models excluding these variables will be tested. In effect, this hypothesis is a strong form of support for OH 1b, that is, regardless of timing and package, bottler promotions are not independent. 94 2. ngothesi_sg In trying to determine whether the Grand Rapids soft drink market is characterized by alternating promotions, the second hypothesis to be tested is: H2: A single national bottler (such as Coke) should not be promoted in two different chains at the same time (Lal 1990b, p. 436). Lal (Lal 1990b, p. 436) stated: "We would expect temporal price dispersion within a store and spatial price dispersion across chain stores such that no national brand is promoted at the same time at both stores. " Lal hypothesized that stores play a role analogous to brands but at one organization level above brands. As do brand managers of interdependent products who attempt to alternate promotions, store or chain managers with interdependent stores will alternate promotions. In Lal's framework, two strong stores with brand. loyal customers alternate promotions in response to one small store without loyal customers (ibid.). Lal did not extend his model to the complexity of a large market with many grocery chains. If five or six grocery chains were to participate in an implicit coalition, it is unclear how they would behave. Chains could take turns promoting, could promote the same bottler‘s products by taking turns in offering deep discounts, could promote the same bottler’s products in different packages, and so forth. While Lal extended the qualitative expectation of some kind of promotional alternation across space, the particular forms were not specified. Pragmatically, it seems unlikely that a bottler in a city with as many grocery chains as Grand Rapids (19) could avoid having promotions in multiple chains simultaneously. The Pepsi CMA makes the blanket suggestion that promotions run during odd weeks of the year and on four of six listed holidays. Ifchains were to conform to this suggestion, Pepsi promotions would run in all chains during the same weeks. Furthermore, because of Pepsi's attempts to obtain 26 weeks of promotion in odd-numbered weeks, in the presence 95 of a strong implicit coalition, Coke's promotions would be forced into even-numbered weeks. In the extreme case, this alignment of promotions could take place across chains throughout a market. It seems prudent, then, to look for very high as well as very low levels of interchain promotion coincidence, even though high levels have not been hypothesized in the brand interdependence literature. The practical need for bottlers to standardize promotions will itself generate some interchain promotional overlap. The Clayton Act as amended by the Robinson-Patman Act restricts a bottler's ability to price discriminate between grocery chains (Nagle 1987, p. 331). Because CMAs bundle pricing, advertising, and display, a bottler ofl‘ering all grocery chains the same price schedule is effectively offering the same promotion menu. If one is willing to assume that bottlers have similar price/promotion schedules, the concepts of "low" and "high" frequency can be grounded by the chance that different chains randomly choose the same promotion. The BBMI CMA, for example, contains three price tiers for four packages, or 12 price/promotion options. Three choices in the middle price tier (in-store display) are no longer available to retailers, however, because of poor pass-through. This leaves a total of nine promotions available to the chains in BBMI's market. If Coke, Pepsi, and BBMI each offer nine promotions, the chance of a bottler's promotion being selected is l in 27, or 3.7 percent. The chance that two chains will select the same promotion is then .0037’=.001369, or .1369 percent. In the Grand Rapids Nielsen sample there are 19 grocery chains. “0th 171 pairs (combinations) of 19 chains, there is a 23 percent chance (.001369‘171=.23) of observing two chains with the same promotion if all Grand Rapids chains promote in a given week. This implies a worst case frequency of one overlap every four or five weeks, or 11.96 overlaps per year. If many more than 12 overlaps occur in a year, this would lend support to Lal's qualitative hypothesis that promotional alternation at the 17mg level is taking place. At the same 96 time, however, this finding would falsify the hypothesis that alternation is taking place at the phaip level. The lower bound on promotion overlap depends on how many weeks chains promote soft drinks. If fewer chains promote, the chances of promotional overlap for a given bottler are decreased, Ifzero promotions are observed, support is lent to Lal's hypothesis that promotional alternation is taking place at the chain level. In light of this discussion, Lal's hypothesis that interchain promotion coincidence should not occur has been operationalized as: a) Mional Hypothesis 2 OH 2: lnterchain promotion coincidence is significantly greater than or significantly less than chance would predict, given the average number of weeks grocery chains promote soft drinks To test this operational hypothesis, the soft drink promotion schedules for the chains with complete data for year 3 will be searched for overlapping promotions. To make the analysis tractable, three of the nineteen chains in Grand Rapids will be analyzed. This sample makes the analysis tractable because it reduces the quantity of data to be analyzed (from 423,679 records in the ddump table to 123,619 records), because it reduces the number of weeks to be examined from 998 to 156, and because the three chains selected have no missing data points. Selected data for the chain stores in Grand Rapids are displayed in Table 3-9. Where the chains to be analyzed are indicated by three asterisks in the left-hand column. The chain in the top row of Table 3-9 was excluded fi'om the analysis because it contains anomalous data. Although Nielsen was not willing to identify the location of stores, by examining this chain's data and identification code and by talking with BBMI it was possible to identify the chain. These data represent an outside chain entering the :3: Nielse a year solidn 97' Nielsen Grand Rapids market area for the first time. In the bottler's words, for more than a year "it was a battle to the death with [Chain 3], and [Chain 3] won." It appears that soft drinks were used heavily as loss leaders to build market share. Table 3-9 Selected Information on Grand Rapids Grocery Chains Average Anna-ago A11 Chain Bacon-d: 1n ‘”“‘9' ‘CV Gusto-n: Count Chunodity number annp rear 3 Volume 34.200.000.00 34,951 34,200,000 "* 30.140.000.00 47,199 30,140,000 3 65,363 16,913,460.76 15,733 16,913,461 69* 16,750,000.00 20,600 16,750,000 2 32,565 14,700,000.00 13,030 14,700,000 13,460,040.22 15,227 13,460,040 13,100,000.00 15,077 13,100,000 12,106,333.33 16,659 12,706,333 so. 11,650,000.00 11,735 11,650,000 1 25,691 11,525,000 00 12,620 11,525,000 10,473,000.00 12,653 10,473,000 10,000,000.00 11,746 10,000,000 6,065,769.23 10,066 6,065,769 6,035,000.00 6,965 6.035,000 7,704,500.00 12,111 7,704,500 7,669,617.02 13,220 7,669,617 5,934,000.00 9,053 5,934,000 5,166,265.61 7,365 5,166,266 4,970,000.00 6,133 4,970,000 The expected overlap based on chance will be calculated by determining how many chains promoted in each week of year 3. A promotion overlap will be counted as two retailers promoting a strong bottler's products in the same package size on the same price tier. For example, 2-liter products promoted in Chain 1 at $.89 and in Chain 2 at $.88 would be counted as an overlap. Even though prices are not identical, if they difi‘er by less than the difi‘erence between price tiers in BBMI's CMA the promotion will be counted as overlapping. This interpretation of pricing allows some latitude for price zone difi‘erences within and between chains. 1 codes i count i1 6-pack brands imam t-test i oven: Oil' 98 This analysis will not be limited to looking for identical UPC codes across chains. UPC codes can be misleading indicators of promotions because they capture only some unit count information. For instance, Coke and Pepsi do not have any UPC codes to represent 6-pack cans, while 7Up/RC/DR. and the store brands do. In addition, CMAs state that all brands in a package size must be included in a promotion. Consequently, the data will interpret price per can as indicating an equivalent promotion in two chains. A two-tailed t-test at 0t=.05 then will be used to test OH 2. If there are significantly more promotional overlaps than predicted based on chance, or if there are significantly fewer promotions, OH 2 will be supported. 3. Hymthesis 3 The second research question of this study asks: ”Is there competitive parity among strong brands and competitive asymmetry between strong and weak brands?” In order to assess the competitive position of strong bottlers, two research hypotheses were developed. Hypothesis 3 tests whether strong bottlers have equal sales on and ofi‘ promotion, as implied in the literature (Kinberg, Rao, and Shakun 1974, p. 955; Narasimhan 1988, p. 44]). Approximately equal low sales when ofi‘ promotion and high sales when on is consistent with strong bottlers sharing the switching segment of the market. The second research hypothesis assessing competitive parity, H4, tests whether strong bottlers benefit from an asymmetry in competitive draw. If strong bottlers draw sales fi'om weak bottlers, and if weak bottlers are unable to reciprocate, the brand interdependence explanation of an implicit coalition based on competitive positions receives support. The brand interdependence theory predicts that where strong bottlers engage in alternating promotions, they should have approximately equal overall sales to loyal .—.~—_——... 4 _ gusto isolate the sal nonloy meawl are thr (ll 99 customers (Kinberg, Rao, and Shakun 1974). While sales to loyal customers are not isolated from overall sales in store-level data, it may be possible to obtain a rough idea of the sales to loyal customers for each bottler. .Selecting one bottler‘s sales in weeks when nonloyal customers buy another bottler‘s promoted products yields an approximate measure of sales to loyal customers. From the brand interdependence perspective there are three key market segments: (1) brand loyal customers who do not look for promotions and buy their preferred product regardless of promotions; (2) brand loyal customers who seek promotions, stock up when they find a promotion, and purchase nothing otherwise; and (3) ”cherry pickers,” who shop for promotions and seek top-tier brands at medium or bottom-tier prices. What constitutes brand loyalty is moot at present (Raju, Srinivasan, and Lal 1990). Perhaps it is more accurate to say that what constitutes brand disloyalty is debatable. While segment (2) may or may not be considered loyal, depending on perspective, segment (1) is by almost any measure considered loyal. Therefore, a coarse measure of sales to loyal customers is provided by the sales of a highly promoted product in the weeks it is off promotion. During these weeks only segment (1) customers are buying that product. To test Hypothesis 3, two operational hypotheses will be used. a) eratignal Hypothesis 36 OH 3a: Strong bottlers have approximately equal unit and dollar sales when they are ofl promotion (Kinberg, Rao, and Shakun 1974, p. 955; Narasimhan 1988, p. 441 ). OH 36 will be tested by extracting unit and dollar sales for strong bottler products in Chain 3 for year 3. Due to holiday peaks in demand, it is likely that strong bottler products ofi‘ promotion in "stock up weeks” will not be comparable to strong bottler products ofi‘ promotion during ”regular weeks. " To prevent comparing apples to oranges 100 the six holiday weeks in year 3 (week 2=July 4, 11=Labor Day, 24=Christmas, 28=SuperbowL 48=Mernorial Day, and 52=second week before July 4, 1991) will be excluded from the analysis. . OH 3a will be tested with a two-way ANOVA treatment contrast. The two factors to be included are strong bottlers (Coke, Pepsi, and 7Up/RC/DR.) and promotion type (none, secondary, feature, and feature and secondary). Ifthe average sales (in ounces and cents) for strong bottlers off promotion are not significantly (0r=.05) difi‘erent, then the hypothesis that strong bottlers have equal nonpromoted sales is supported. b) ngrational Hypothesis 3b OH 3b: Strong bottlers have approximately equal unit and dollar sales when they are on promotion (Kinberg, Rao, and Shakun 1974, p. 955; Narasimhan 1988, p. 441). As in OH 36, OH 3b will be tested by extracting unit and dollar sales for the products of the strong bottlers in Chain 3 for year 3. Due to holiday peaks in demand, the researcher expects that strong bottler sales ofi‘ promotion in "stock up weeks" will not be comparable to strong bottler sales 03 promotion during ”regular weeks. " To prevent comparing apples to oranges, the six holiday weeks (week 2=July 4, 11=Labor Day, 24=Christmas, 28=Superbowl, 48=Mernorial Day, and 52=second week before July 4, 1991) in year 3 will be excluded from the analysis. OH 3b will be tested with a two-way ANOVA treatment contrast. The two factors to be included are strong bottlers (Coke, Pepsi, and 7Up/RC/DR) and promotion type (none, secondary, feature, and feature and secondary). Because this study distinguishes among four types of promotion, OH 3b can be operationalized in a particularly strong form. In an extreme environment of constant alternating promotions, all strong bottlers would be predicted to experience the same market response for each type of promotion 101 (see Figure 3-7). If the market response (in ounces and cents) for at least one type of promotion is not the same across bottlers, then the hypothesis of equal sales on promotion will not be supported. ‘ Figure 3-7 Hypothesized Strong Bottler Sales on and off Promotion Feature + None Secondary Feature Secondary Coke ’asaa‘ 'ddod‘ rgggg‘ ’jjij‘ Pepsi bb.bb ee.ee hh.hh kk.ld( 7Up/RC/DR. cc.cc iii? 11.11 11.11 (3 Indicates Mean Sales Levels Not Significantly Different 4. Hypothes_i9»__4 Two hypotheses test the second research question of this study: Is there competitive parity among strong brands and competitive asymmetry between strong and weak brands? As noted above, Hypothesis 3 evaluates competitive parity by testing the implication that strong national bottlers should have equal sales to loyal customer segments (Kinberg, Rao, and Shakun 1974, p. 955; Narasimhan 1988, p. 441). Hypothesis 4 tests for asymmetric competitive draw between bottlers who alternate promotions and bottlers who promote infrequently if at all. H4: The brands that alternate promotions are the beneficiaries of an asymmetry in competitive draw from private label brands (Kinberg, Rao, and Shakun 1974, p. 952; Blattberg and Wisniewski 1989, p. 303). 102 Competitive draw can be tested by estimating the cross-price and cross-promotion elasticities of strong and weak bottlers, extending Blattberg and Wisniewski's (1989, p. 299) cross-price elasticity method. These authors estimated cross-price effects by developing a regression equation relating each UPC's sales to its own and its competitor's price promotions. This regression was then used to calculate the cross-price efi‘ects among brands. To keep the analysis tractable, only Chain 3 data for year 3 will be used. The following equation modifies Blattberg and Wisniewski's (1989) equation in order to use information on promotions provided by bottlers. Sit = CXMCit-B1Pit+BzFit+B3Sit+B4Bit+BsPit-36Fijt-B7Sijt-BsBijt+B9¢it). 3-2 where Sit = total unit sales of UPC i in week t (aggregated over 5 stores in Chain 3); cit = constant for UPC i in week t; Pit = price of UPC i in week t; Fit = dummy variable for feature promotion only for UPC i in week I; Sit = dummy variable for secondary promotion only for UPC i in week t; Bit = dummy variable for both feature and secondary promotion for UPC i in week t; Pit = price of competitive UPC i in week t; Fijt = dummy variable for COMPETITOR's feature promotion only for UPC i in week t; Siit = dummy variable for COMPETITOR's secondary promotion only for UPC i in week t; Bijt = dummy variable for COMPETITOR's both feature and secondary promotion for UPC i in week t; and (lit = dummy variable for stock up weeks (July 4, Labor Day, Superbowl, Memorial Day, and a second week before July 4, 1991, in Grand Rapids in week 52). Each UPC's competitors will consist of UPCs in the same package and multipack form that are promoted. Again, to avoid biasing the study, this specification was not pretested to evaluate how well it represents the data. Consequently, the equation is likely to need modification. Other studies provide some guidance here. For instance, in previous work with scanner data, interaction terms have been excluded because of insignificance (see 103 Kumar and Leone 1988, p. 182) and because good fits (R2 between .75 and .91) without interaction terms have been achieved (Blattberg and Wisniewski 1989, p. 302). For other issues there is less guidance, however. Given what customers know about soft drink promotions, shelf prices may not perform well. An industry expert told the researcher that he believes the reference price for sofi drinks is not the most frequently charged price, but rather the price on deal - which is one of the lowest prices (Totten 1991). Krishna, Currim, and Shoemaker (1991) have shown that peOple remember the depth and frequency of soft drink promotions better than any other category. Ifthis is the case, it may be that the degree to which a weekly price difi‘ers from its lowest (or highest) price may be the measure of price that triggers customer behavior. The operationalization of promotion in this study is new and may need modification. Given the strong information provided by Brooks Beverage Management, Inc., feature and secondary promotions should be significant predictors of sales. Due to the difficulty of operationalizing these variables using the single-source data, however, and/or due to measurement error in the data or error in the information provided by BBMI, these variables may not be significant. Regardless of the modifications that may be required in the analysis of the data, Equation 3-2 provides a theoretically and managerially grounded starting point for developing an equation to represent the data Alter an adequate specification for the regression equation has been found, it can be used to estimate the cross-price and promotion effects of soft drink products. Tables of cross-efi‘ects will be developed for price, feature, secondary, and simultaneous feature and secondary promotions. Sets of these tables will be developed on the basis of package and multipack combinations (by aggregating elasticity results across a bottler's products within a package form). Because competition is operationalized in terms of package-multipack combinations, the regression results will be nested within package-multipack and should be reported in this form. Table 3-10 provides an example of the reporting format to be 104 used. This example is for price elasticity effects for 2-liter products, but similar tables will be developed for all the packages possible. In addition to a table such as Table 3-10, three other tables will be provided for each combination: feature only effects, secondary effects, and simultaneous feature and secondary effects. Table 3-10 Cross-Price Elasticities among Bottlers for 2-Liter Products Amer Oxn- Sms or : Canada Can- Store rm- IDR. Dry field Coke raygo Pepsi Brand known No statistical test to gauge the significance of asymmetrical effects was used by Blattberg and Wisniewski ( 1989), but such a test is possible by extending their approach (see Table 3-11). If strong bottlers benefit from asymmetry of competitive draw, the average cross-effect that strong bottlers have on weak bottlers (the upper right-hand cell of Table 3-11) should be larger than the average effect that weak bottlers have on strong (lower left-hand cell of Table 3-11). A t-test can be used to determine if strong bottlers have a greater effect on weak bottlers than vice versa. Table 3-11 Summary of Cross-Price Elasticities among Bottlers for 2-Liter Products f Amer-mum” Whom: r Pnrasormalnom: Straw M r— Strong I Weak If the upper right-hand cell is significantly greater than the lower lefi-hand cell, then Hypothesis 4 is supported. In addition, tables for feature, secondary, and simultaneous feature and secondary promotions will be calculated for each package-multipack 105 combination. With the results of these four cross-effects, Hypothesis 4 can be considered supported to the extent that strong bottlers have greater effect on weak bottlers than vice versa (see Table 3-12). This is not a statistical test, but if strong bottlers have a greater efl'ect on weak bottlers across all variables, it would be convincing evidence for asymmetrical competitive draw and would support H4. Table 3-12 Summary of Cross-Price and Promotion Effects Strong Battlers have leak Battlers have > > "feet on leak Street on Strong Battlers Battlers Feature and Secondary Feature Secondary None 5. Hymthesis 5 Research question three asks: Are promotional discounts equal for strong brands, and are strong brand discounts larger than discounts for weak brands? The last two hypotheses were developed to address this question: H5: Brands that alternate promotions should have approximately equal discounts (Narasimhan 1988, p. 435; Raju, Srinivasan, and Lal 1990, p. 286). H6: Strong brands on promotion should have greater discounts than weak brands (Narasimhan 1988, p. 435; Raju, Srinivasan, and Lal 1990, p. 286). The operationalization of discounts used to isolate feature and secondary promotions for Hypothesis 1 will be used to test H5. In brief, this operationalization divides price for 6 UPC in a given week by the maximum price observed for that UPC within a chain for the year. This operationalization reflects the price structure of CMAs that set price tiers by discounting from a ”regular wholesale” price. Average discounts for each promotion type within Chain 3 during year 3 will be calculated and reported by bottler and package- multipack combination (see Table 3-13). 106 Table 3-13 Average Discounts for Each Promotion Type for 2-Liter Products Can- Store Un- /DR. field Coke Brand known Feature and Feature None A pairwise t-test at 0t=.05 will be used to determine whether price discounts for the three strong bottlers are significantly different from one another for each type of promotion. Ifbottler discounts for each package-multipack combination and promotion type are not significantly different, H5 will be supported. Table 3-14 illustrates how pairwise comparisons will be tested across both promotion type and package-multipack combination. Separate tables will be developed for each package-multipack combination. It is unlikely, however, that all three strong bottlers would align their discounts for all promotions and package-multipack combinations. Because bottlers concentrate their promotions in a few packages, even if the brand interdependence theory is true, it is possible strong bottler discounts diverge for some promotion types and/or some packages. Some products are promoted occasionally (half-liter bottles), while others are heavily promoted (6—pack cans, 12-pack cans, and 2-1iter bottles). If the strong bottler discounts for the most fiequently feature-promoted package in Chain 3 are not significantly different, H5 will be considered to be supported. Table 3-14 Average Discount for Strong Bottlers Chain 3 - Year 3 107 6. Hypothesis—6 The second aspect of assessing whether promotional discounts conform to the pattern hypothesized in the brand interdependence literature is to determine whether strong bottlers have greater promotional discounts than weak bottlers. The price discount results developed in Table 3-14, where strong and weak bottlers have common package sizes, can be used to compare the average discounts of strong and weak bottlers (see Table 3-15). One table will be constructed for each package-multipack combination that strong and weak bottlers have in common. Ifthe strong bottler's average discount is significantly larger than the weak bottler's (as assessed by a t-test), H6 will be considered supported. Again, because bottlers concentrate their promotions in a few packages, even if the brand interdependence theory is true, it is possible that strong bottlers would have smaller discounts than weak bottlers for some promotion types and/or some packages. Ifthe strong bottler discounts for the most frequently featured package in Chain 3 are greater than weak bottler discounts, H6 will be considered supported. Table 3-15 Average Discounts for Each Bottler Type and Bach Promotion Type Chain 3 - Year 3 Feature None IV. Results A. In uin This chapter presents and discusses the results for each of the hypotheses stated in Chapter 3. Central to this investigation is an attempt to answer three questions relevant to the brand interdependence promotion explanation: 1. Is the soft drink category characterized by alternating promotions among premium national brands? 2. Is there competitive parity among strong brands and competitive asymmetry between strong and weak brands? 3. Are promotional discounts equal for strong brands, and are strong brand discounts larger than discounts for weaker brands? Each research question has been broken down into two hypotheses. In the cases of H1 and H3 these hypotheses have been firrther broken down into operational hypotheses addressing individual theoretical relationships. Results of the empirical tests of these operational hypotheses follow. A summary table of the results can be found beginning on page 148. B. Testing Operational Hyppthesis 1a H1: Within chains, alternating promotions should take place during no less than 66 percent and no more than 94 percent of available promotion periods (Kinberg, Rao, and Shakun 1974, p. 958). H1 is investigated by testing two operational hypotheses. Each applies an alternating promotion measure suggested in the literature. 108 109 1. Operational Hypothesis la 0H 1 a: Each premium brand should be on promotion between 33 percent and 4 7 percent of the time (Kinberg, Rao, and Shakun 1974, p. 958). For this hypothesis to be true, the confidence interval around time on promotion for 7Up/RC/DR, Coke, and Pepsi bottlers should overlap a range from 33 to 47 percent. Because the brand interdependence literature has not settled on an operationalization of ”time on promotion” that adequately copes with the complexity of practice, it is prudent to look at four measures of soft drink promotions. Figures 4-la through 4- l (1 illustrate the time bottlers were on promotion using four measures. The confidence interval for time the three premium bottlers spent on promotion overlaps the hypothesized 33 to 47 percent range in two of four cases. In each figure, the shaded range from 33 to 47 percent represents the time on promotion hypothesized by Kinberg, Rao, and Shakun (1974, p. 958). "UCL" and "LCL" represent the upper and lower confidence limits, respectively. ”Actual" is the observed percentage of time each bottler was on promotion. Dollar and unit market share are represented by upward and downward pointing triangles, respectively. The first measure of time on promotion was the time spent on any promotion (Figure 4-16). The confidence intervals for all three premium bottlers overlap the hypothesized frequency range, supporting OH la. The observed promotion frequencies for strong bottlers fall within the hypothesized range, supporting OH la. The second measure of time on promotion was the time that a bottler's product spent on feature promotion (Figure 4-lb). Using this measure the confidence intervals for all three premium bottlers again overlap the hypothesized frequency range and support OH la. Time on feature promotion is a narrower interpretation of time on promotion, and l l 0 this narrowness is reflected in lower time on promotion scores for all bottlers compared with the first measure. The frequency ranking of Pepsi and Coke bottler promotions reverses between the first and second measures, reflecting difi’erent promotion strategies. The Coke bottler relies more heavily on feature promotions and the Pepsi bottler on secondary promotions. Figure 4-la Hypothesized versus Actual Time on Promotion - Counting Any Promotion 100% 900A) .. ........... .. .......................... . .. . ....................... 800A) . . . ................ .. ............................. 70% .. . ........ .. .. . .......................... 60% , 50% _ g 40% ' 30% f ‘ 20% 10% ~ 0% M UCL- Sl.31% 55.87% 59.34% 12.17% 27.69% 25.29% ucrj 25.11% 3.75% 32.47% 0% 6.92% 5.48% Arm-l. 38.46% 42.31% 46.15% 5.77% 17.31% 15.38% SMSV 15% 30% 24% 4% 22% 5% 0: MS‘ 12% 29% 22% 5% 27% 4% Time on Promotion Figure 4- lb Hypothesized versus Actual Time on Promotion - Counting Feature Promotions 100% m ........ 80% 70%. . ............. . ..... 50% - 40% 30% Time on Promotion 111 The third measure of time on promotion is time on secondary promotion (Figure 4-10). This is a narrower measure than time on any promotion because it leaves out feature promotions. The focus, however, is secondary promotions, which are less sought alter because they have smaller lift factors (see Figure 3-2 regions H, G, and D) than feature promotions (Figure 3-2 region F). Secondary promotions are also not as exclusive as feature promotions. Secondary promotions ran simultaneously during 15 weeks, while feature promotions ran simultaneously during one week. Using this measure only, the confidence interval for Pepsi overlaps the hypothesized fi'equency, disconfirming OH la. Figure 4-1c Hypothesized versus Actual Time on Promotion - Counting Secondary Promotions 100% 90% 80% 70% 60% 50% 40% . 30% 20% 10% 0% um,- LCLI and. SMS V OrMS A Time on Promotion The secondary promotion measure seems unlikely to be a good measure of time on promotion because bottlers strongly prefer feature or combined feature and secondary promotions. Examining promotions from the vantage of secondary promotions is worthwhile, however, to ensure that a useful measure of promotion time is not overlooked. Secondary promotions may be a significant (though peripheral) means of competition. 112 The brand interdependence literature interprets alternating promotions as either a means of capturing rents fi'om loyal customers (Kinberg, Rao, and Shakun 1974; Narasimhan 1988) or a means of preventing fi'agmented bottlers fiom encroaching on strong bottler sales (Raju, Srinivasan, and Lal 1990; Lal 1990a; Lal 1990b). In the literature these competitive arguments assume that promotions are exclusive to a single competitor each week. This clearly is not the case in soft drink merchandising practice. Combining feature with secondary promotions reduces the number of competitor promotions, providing some degree of exclusivity to a bottler's own promotions. The majority of promotions, however, are not combinations. Consequently, secondary promotions might be a peripheral means of competition (or cooperation) for strong bottlers. Of the 11 secondary promotions in which the Coke bottler participated, 9 were simultaneous feature and secondary promotions, and 2 ran in opposition to 7Up/RC/DR. feature promotions. BBMI reported that it had "heard through the grapevine" that the Coke bottler had made clear its desire that Chain 3 not run 7U p/RC/DR. secondary promotions during Coke feature promotions (this occurred three times in year 3). Of the 20 secondary promotions in which the Pepsi bottler participated, 11 were simultaneous feature and secondary promotions, 7 ran in opposition to the 7U p/RC/DR. bottler's feature promotion, one ran in opposition to a Coke feature promotion, one ran in opposition to a store brand promotion, and the remaining promotion ran in opposition to an unknown bottler‘s feature promotion. Interestingly, Coke and Pepsi promoted at the same time twice in year 3. Pepsi ran one secondary promotion while Coke ran a feature promotion, and both Coke and Pepsi ran secondary promotions during a 7Up/RC/DR. feature promotion. Consequently, while it appears that strong bottlers may use secondary promotions to reduce the efi‘ectiveness of a rival's feature promotion, featuring is the principal means of implementing alternating promotions. 113 The fourth measure of time on promotion is time on both feature and secondary promotion (Figure 4-1d). This is the narrowest of the four measures of time on promotion and none of the confidence intervals for strong bottlers overlap the hypothesized frequency range. This measure disconfirrns OH la. Figure 4-ld Hypothesized versus Actual Time on Promotion - Counting Combined Promotions 100% 50% . ******* 20% i " 0% Time on Promotion UCL- 15 01% 25.29% 25.29% 0% LCLI 0.38% 5 48% 5.48% 0% Anni-II 7.69% 15 38% 15 36% 0% 0% 0% 5 MS V 15% 30% 24% 4% 02 MS A 12% 29% 22% 5% The apparent strengths and weaknesses of the four measures are noted below. 1. Any promotion. This measure captures both the featuring effect and the possible peripheral competitive effects of secondary promotions, but it does not distinguish between these effects. Bottlers may rely more heavily on feature promotions (as Coke does), on secondary promotions (as 7Up/RC/DR does), or on a mixture (as Pepsi does), and their fi'equency of promotion will not look different. This measure agrees with both brand interdependence theory and common sense observation (see 60 Minutes 1987). 2. Feature promotions. This measure removes secondary promotions and consequently, any peripheral role. The focus on featuring gives the measure clarity because bottlers desire as many feature promotions as possible. Secondary promotion activity is likely to be an important form of rivalry, however, (possibly a competitive game nested within the cooperative alternation game). To the extent secondary promotions are an important means of competing, a feature only measure may be unsatisfactory. Feature promotions agree to a limited extent with the brand interdependence theory and common sense observation, but featuring does not appear to be superior to the "any promotion” criterion. 114 3. Secondary promotions. This measure ignores any role for feature promotions, making it an unsatisfactory measure of promotion behavior. The secondary measure is theoretically valuable, however, because it highlights the distinct promotional strategies of strong bottlers. 4. Combined feature and secondary promotions. This is the most restrictive measure of promotions. The strength of this measure is that it is the closest approximation to promotional exclusivity of any of the available measures. Its weakness, however, is that it abstracts away from feature and secondary promotions. Featuring is a major avenue of competition - witnessed by the Coke bottler's preference for this form. Secondary and feature promotions are partially captured in the combined measure, but that excludes secondary promotions used as interference competition. For instance, a Coke secondary promotion during a 7Up/RC/DR. feature may have devastating effects on the 7Up/RC/DR. bottler. The ”any promotion” measure of time on promotion appears to be the best measure for the purposes of this study. Any promotion captures all competitive efi‘ects, agrees with common sense, and is consistent with an implicit coalition of strong bottlers competing with weak bottlers who promote infrequently. Table 4-1 provides summary results, which support the brand interdependence literature's hypothesis of alternation. Strong bottlers obtained 94 percent of the feature promotions and 73 percent of secondary promotions, for a total of 83 percent of all promotions. All three of these proportions fall within the 66 to 94 percent range of H1. Although all measures of promotions did not confirm H}, the most appealing measure (time on any promotion) confirmed OH la. Table 4-1 Summary Results of Measures for Operational Hypothesis la Confirm Disconrlrsr x x C. T ' 'nalH esislb The second avenue in the literature for detecting alternating promotions is testing to see if strong brand promotions are independent (Lal 1990b). 115 1. Operational Hypothesis 1b 0H 1 b: The promotions of bottlers of strong brands are not independent of each other (Lal 1990b, p. 437). As discussed in Chapter 3, loglinear modeling and chi-square analysis were to be the means of testing OH 1b. The promotions of strong bottlers were so extremely negatively associated that valid loglinear models and chi-squares could not be calculated. Tables 4-2 and 4-3 and Figures 4—2 and 4-3 illustrate the negative association of strong bottlers. Table 4-2 Association of Coke and Pepsi Promotions Figure 4-2 Association of Coke and Pepsi Promotions Table 4-3 Association of Strong and Weak Bottler Promotions leak Battlers Any Promotion Grand Total 11 49 14 52 116 Figure 4-3 Association of Strong and Weak Bottler Promotions Week laden The conventions for chi-square and loglinear modeling are that sparse cells (with fewer than five observations) make significance tests suspect. If more than one-fifth of the cells are sparse, then the statistical results are invalid (Wilkinson 1990, p. 506). Recent work has extended the viability of chi-square analysis to smaller samples and increasingly sparse tables (Agresti 1990, p. 246). Unfortunately, this work indicates that chi-squared approximations are at their worst when (as is the case in of Tables 4-2 and 4-3) tables are heterogeneous as well as sparse. Heterogeneity (the mix of firll and empty cells) is easy to see in Tables 4-2 and 4-3. For example, if soft drink promotions were independent and joint uniformly distributed between Coke and Pepsi, the expected frequency in Table 4-2 would be 3.25 (weeks) for each cell. There are nine cells in Table 4-2 representing positive association. Eight of these are empty, and the ninth cell contains one observation (representing 1.9 percent of the total). While the observed promotion pattern prevents the use of conventional statistical tests, it is important to recognize that this pattern is one observation short of being the strongest confirmation of OH 1b possible. It is, however, an analytical irony that the data are so negatively associated that significance tests cannot be used to assess the negative association. 1 1 7 A second approach to testing for negative association is to examine the correlation of bottler promotions. Table 4-4 shows these correlations. The strongest negative correlation is between Coke and Pepsi (-.624). All correlations among strong bottlers are negative. Table 4-4 Correlation among Bottler Promotions srvnrup can "P81 PMGO 8T0!!! amorous m . 000 can -0 . 32 1 . 000 m1 -0 . 133 -0 . 624 1 . 000 '5!“ -0.177 -0.061 0. 1 . m -0.166 -0.135 -0.252 -0.106 1. mm -0.216 -0.169 -0.100 -0.099 0.633 PEARSON CORRELATION MATRIX NUMBER OF OBSERVATIONS: 52 A third approach is to use the “Wilcoxon signed rank test to test for independence of bottler promotions. Table 4-5 shows the results of a Wilcoxon test for all bottlers in Chain 3 during year 3. Table 4-5 Wilcoxon Signed-Rank Test Results WIWDIM (mum manna-meanest) m can "PSI FA!“ m m m 1 1 m 1 9 0 ms: 23 0 moo noes 6 0 m z I (m or I!“ Ml)/m WC" or m ms) m cm "PSI moo 8m more! m 0.000 can 1. . ms: 0.946 .218 'MOO -3. 007 -4 . 18‘ m - . -3.422 m - . - . m—srnsn MIL!!!” mm W W!“ am can "PSI moo m M m 1 . 000 can . "PSI moo sm 118 These results indicate three broad patterns of promotion behavior. First, strong bottler (7Up, Coke, and Pepsi) promotions are not independent. Second, strong and weak bottler (F aygo, Store, and Unknown) promotions are independent, as all pairings of strong and weak bottlers are significant. Third, one weak bottler (Faygo) is marginally independent of the other two, who are not independent. Although the loglinear models and chi-squared tests of significance could not be used, it appears that OH lb is supported for these reasons. 1. The data pattern that invalidates significance testing is one observation away from being the strongest possible validation of the hypothesized negative association of strong bottler promotions. 2. The correlations among strong bottler promotions are negative, as predicted. 3. Strong bottler promotions as assessed by the Wilcoxon signed-rank test are not independent. D. Testing the Second Rpsearch Hymthesis The second research hypothesis tests whether bottlers run the same promotions across grocery chains. Lal (1990b, p. 436) hypothesized that a bottler would not promote in multiple chains at once. As discussed in Chapter 3, the large number of chains in Grand Rapids, the difficulty bottlers have in tacitly coordinating promotions across chains, and the bottlers' legal obligations to standardize promotions could cause more as well as less promotional overlap than would be expected by chance. Hypothesis 2 has, consequently, been modified to take the likelihood of positive association into account. 1. Qmatipnal Hymthesis 2 0H 2: Interchain promotion coincidence is significantly greater than or significantly less than chance would predict, given the average number of weeks grocery chains promote soft drinks. 119 The year 3 data for chains 1, 2, and 3 enable several operationalizations of promotional overlap. Table 4-6 displays promotional overlap measured in three ways: by counting the number of weeks any of a bottler's products were on promotion in two chains; second, by counting overlapping promotions for a single product size; and by counting the number of weeks a bottler ran the same promotion in two chains. Table 4-6 also shows the number of weeks each combination of grocery chains promoted soft drinks. Table 4-6 Promotional Overlap between Grocery Chains Battier, Package Size, 5 Price Silnltaneausly Promoting Soft Drinks 1 22 42 Inspection of Table 4-6 indicates that promotional overlap is as high. Bottler promotional overlaps between Chain 1 and Chain 2, for example, occurred in 15 of 22 weeks, or 68.2 percent of the time. Table 4-7 reports t-tests for all three measures of promotional overlap. Table 4-7 Probability of Promotional Overlaps Occumng by Chance Overlap of: overlap of: Overlap of: Bottler Battier 5 Battler, Package Sire Package Size, 6 9216. Chain 1 S 2 t-valne 3 . 1663 2 . 5912 -0 . 1063 4.2. 21 21 21 rain. 0.0047 0.0175 0.9146 Chain 1 S 3 tdvalue 1 . 6306 2 . 6473 2 . 1654 d.f. 27 27 27 15min. 0 . 0762 0 . 0134 0 . 0377 Chain 2 a 3 t-valne 9 . 9603 6 . 6342 3 . 6062 4.2. 41 41 41 15min. 0 . 0000 0 . 0000 0 . 0006 120 The probabilities used to calculate expected values were: [331 ]2 for the chance of a strong bottler‘s promotion overlapping in two stores; one in two for the chance of promoting in the same package size; and one in three for the chance of pricing in the same price tier. On average, 1.8 of the three strong bottlers were promoted in the three grocery chains each week. Although BBMI's CMA indicates three package sizes (12-ounce, 20-ounce, and 67.6-ounce) should be observed on promotion, the 20-ounce size was not promoted in these chains during year 3. While overlapping promotions for 12- and 67.6-ounce products are common, overlaps did not occur for medium-sized products (packages 16, 16.9, 20, and 33.8 ounces). The lack of promotion overlaps for medium-sized packages may be due to the fact that they account for a minority of sales and are more "fi'agmented" than any other category of package. For example, only 7Up/RC/DR. and Pepsi bottlers had products in 20-ounce packages. Medium-sized products accounted for 7 percent of promotions (177 of 2,560 promoted UPC-weeks) and 14 percent of the unit volume sold (see Figure 4-4). Figure 4-4 Package Size Share of Volume -Three Grand Rapids Chains 121 Expected values for promotional overlaps were calculated using two package sizes (12 and 67.6 ounces) and three price tiers (see Table 4-8). Price tiers are indispensable for distinguishing between feature and secondary promotions. Equivalent 12-packs were compared because UPC codes for 6-packs do not exist for Coke and Pepsi and because 12- packs were by far the most frequent can promotion. Table 4-8 Price Tiers Used to Code Promotional Overlaps Cans - 12- valent Wits .89- .99 Iodin- Disaonnt . 49- Discount Operational Hypothesis 2 states: Interchain promotion coincidence is significantly greater than or significantly less than chance would predict. Table 4-8 represents three measures of promotional coincidence probability. Because seven of the nine combinations indicate that promotion coincidence is significantly (p-values less than .05) greater than expected, OH 2 is not refuted by the data. Table 4-9 presents the schedule of promotions for the three sample chains for year 3, indicating the coincidence of promotions. To summarize, the first research question of this study asks Is the soft drink category characterized by alternating promotions among premium national brands? Operational Hypothesis la, that strong bottlers promote between 33 and 47 percent of the time, was derived fiom Kinberg, Rao, and Shakun (1974, p. 958). OH la was not refitted. Operational Hypothesis 1b, that strong brand promotions are not independent, was derived fi'om Lal's hypothesis of alternating promotions (1990b, p. 43 7). OH lb was not refirted. Operational Hypothesis 2, that interclmin promotion coincidence is significantly higher or lower than would be expected by chance, is a modification of Lal's prediction 122 that grocery chains alternate promotions to avoid overlap (1990b, p. 43 6). Lal's hypothesis was altered to incorporate the possibility that antitrust constraints and a large number of grocery chains may transform negative to positive promotion association. OH 2 was not refirted. Because OH la, 1b, and 2 were not refuted by the data, the answer to the first research question appears to be that the soft drink category (in the three chains) is characterized by alternating promotions. Table 4-9 Promotion Schedules -Three Grand Rapids Chains E. Tgsting Mational Hypgtheses 3a and 3b The second research question of this study asks: Is there competitive parity among strong brands and competitive asymmetry between strong and weak brands? To answer this question, two hypotheses were developed; one sought to verify the existence of competitive parity among strong bottlers (H3), and the other sought to confirm the asymmetry of price and promotional draw between strong and weak bottlers (H4). 123 Hypothesis 3 states: Brands that alternate promotions will have approximately equal sales (Kinberg, Rao, and Shakun 1974, p. 955; Narasimhan 1988, p. 441). This hypothesis was broken down into two operational hypotheses. 0H 30: Strong bottlers have approximately equal unit and dollar sales when they are of promotion (Kinberg, Rao, and Shakun 1974, p. 955; Narasimhan 1988, p. 441), and 011 3b: Strong bottlers have approximately equal unit and dollar sales when they are on promotion (Kinberg, Rao, and Shakun 1974, p. 955; Narasimhan 1988, p. 44 1). These hypotheses were tested with treatment contrasts in a two-way AN OVA. Table 4-10 provides information on the AN OVA dimensions. Six holiday weeks were removed from the data, leaving 46 weeks of data for each strong bottler. One alternative to removing these weeks would be to include a "holiday" dummy variable and expand the analysis to a three-way AN OVA As it would be irrational for bottlers to run secondary promotions on holidays, the secondary promotion cells in holiday weeks will be empty, preventing the AN OVA's estimation. Data were not adjusted for seasonality because only three years of data are available from Nielsen and it generally takes ten years of data to adjust reliably for seasonality (Lilien 1992). Results for the two-way ANOVA are presented in Tables 4-11a and 4-11b for ounces and in 4-12a and 4-12b for cents. ANOVAs were run using both linear and loglinear functional forms. One practitioner experienced in analyzing single-source data advised the use of the loglinear functional form based on the results of an unpublished Nielsen study of single-source data. Nielsen examined ”a halfdozen" products in each of 12 categories and concluded that loglinear was generally superior to other functional forms (Bender 1991). Linearity in natural units has been called an "absurd" assumption in the literature (Little and Lodish 1981, p. 24) and has been shown to exhibit the largest absolute magnitude of bias in estimating elasticities (Bolton 1989, p. 217). 124 Table 4-10 Summary Information on Bottler-by-Promotion ANOVA Feature and Bottler Coke 11 7 9 27 Table 4-1 la Bottler by Promotion Analysis of Variance - Ounces Measure Adjusted R2=.746 SOURCI BUM-OP-BQUARIS DP NIAN-SQUARI PHRAIIO P BOTTLER .4577902+14 2 .2288953+14 48.194 0.000 PROMOTION .106543B+15 3 .3551458+14 74.776 0.000 BOTTLIRPPROMOTION .2075363+14 6 .345894I+13 7.283 0.000 IRROR .5984313+16 126 .47494SI+12 Table 4-1 lb Bottler—by-Promotion Analysis of Variance - Log Ounces Measure Adjusted R2=.803 [ eouncs star-arms or tram-saunas r-mto P [ nor-nan 6.737 2 4.366 46.612 0.000 [ 211099011091 26.126 3 9.375 104.329 0.000 | Baum-9309902109: 1.694. 6 0.262 3.142 0.007 [ ms: 11.323 126 0 .090 Table 4-126 Bottler-by-Promotion Analysis of Variance - Cents Measure . 2 , Adjusted R =.708 [ norm stupor-saunas or tram-60mm r-mro 9 | [ 601-an .6509422+14 2 .4254712+14 36.016 0.000 | [ MIG! .220236s+15 3 .734119s+14 62.143 0. 000 J | mamrw .3409662+14 6 .566314s+13 4.611 0.000 | r m . 14664as+14 126 . 1161332+12 J Table 4-12b Bottler-by-Promotion Analysis of Variance - Log Cents Measure . 2 Adjusted R =.775 , [ saunas wu-or-sotms or nan-6mm r-mro 9 | F 60mm 5 . 709 2 2.655 39. 296 0 . 00g [ momma 19.116 3 6.373 67.727 0.000 | | th1m 1.372 6 0.229 3.146 0.007 | [ m 9.153 126 0.073 4 The overall AN OVAs explain a clear majority of the sales variation. Compared to linear forms, the loglinear form explains 5.7 percent more variation in ounces and 6.7 percent more variation in cents. In all cases bottler, promotion type, and their interaction are highly significant (p < 0.01). Table 4—13 contains the ANOVA cell means for both linear l 25 and loglinear functional forms. Figures 4-5a and 4-5b and 4-6a and 4-6b plot residuals against cell estimates for each ANOVA. Because the residuals do not manifest systematic patterns of deviation (trend or bias high/low) or heterogeneity of variance, it appears that the homogeneity of variance assumption is satisfied. Table 4-13 AN OVA Cell Means Feature + Battier Measure None Secondary Feature Secondary Ounces 1,443,858 4,638,340 2,610,688 4,978,201 Cents 2,773,848 6,818,815 4,596,567 7,535,141 Cake Lag(Ounces) 14.148 15.345 14.734 15.360 Lag(Cents) 14.803 15.728 15.310 15.781 Frequency 26 2 11 7 Ounces 1,078,728 2,026,295 2,260,149 3,749,898 Cents 2,121,290 3,404,379 3,678,352 5,931,987 Pepsi Lag(Ounces) 13.850 14.489 14.568 15.095 Lag(Cents) 14.532 15.026 15.079 15.563 Frequency 24 8 7 7 Ounces 626,185 1,548,920 1,938,441 1,734,102 Gents 1,370,214 2,621,922 3,808,602 3,272,475 70p/lC/BR. LagtOunces) 13.331 14.209 14.371 14.293 Lag(Oents) 14.113 14.750 15.066 14.940 Frequency 27 6 9 4 Figure 4-56 Residuals for Ounce Measure AN OVA Millions 2 I 0 e 1 ........................................... ...... ........ . ....................................................................... .... .................... :. C ' .6 ' : o t l ‘ - ' 9 0 'e ‘ e . I ‘ 6 O _1 ............................................................................................................................................................. O ,2 ......................................................................................................................................... e .................... 0 l 3 4 5 126 Figure 4-5b Residuals for Log Ounce Measure ANOVA Residual .OO' . o 0---” O- -1... o as. .- eti O -0. . rater :t'. 13 13.5 14 14.5 15 15.5 16 Estimate Figure 4-6a Residuals for Cents Measure AN OVA 3 Residual (Millions) _2 ..................................................................... _ ................................. 9 ........................................... 6 ............ -3 Estimate (Millions) 127 Figure 4-6b Residuals for Log Cents Measure AN OVA Residual 1.1... o e 1...... :1. . g . O . O -1 13 13.5 14 14.5 15 15.5 16 Estimate The significance of the bottler-by-promotion interaction indicates that strong bottlers followed difi‘erent strategies. Table 4-10 indicates that the Coke bottler ran intense promotions, that the 7Up/RC/DR bottler opted for less intense promotions, and that the Pepsi bottler used all promotion types uniformly. Because the overall ANOVAs seem to be well behaved, it is safe to compute and examine the treatment contrasts needed to test OH 36 and OH 3b. The null hypothesis for OH 36 is that "ofl‘ promotion” mean sales levels for all strong bottlers are equal. For example, the null hypothesis for OH 36 is that the Coke bottler's average sales ofi‘ promotion are equal to the Pepsi bottler‘s average sales off promotion. The strongest possible form of OH 3b is that mean sales levels are equal for all promotion types. If the strong form of OH 3b holds, every within-cell treatment contrast should not be significant (indicating mean sales levels of bottlers are not significantly difi‘erent). The brand interdependence hypothesis probably does not require perfect confirmation to receive support. Because implicit coalitions are based on mutually recognized interdependence, payofi‘s do not need to be symmetric to shape promotion behavior. The 128 theory to date has assumed symmetric interdependence as a simplifying assumption (Lal 1990b). The sufficient condition for alternating promotions stated by Kinberg, Rao, and Shakun (1974, p. 95 8) is that cooperation by strong bottlers increases their profitability more than it decreases the profitability of weak bottlers. A symmetrical division of the resulting revenue among strong bottlers is not a necessary condition. Table 4-14 displays the p—values for the treatment contrasts. Table 4-14 Treatment Contrast p-values between Mean Sales Responses Year 3 - Chain 3 109 Contrast Ounces Ounces . 0. 0.002 0.000 Figure 4-7 summarizes the results of the treatment contrasts. At least in Chain 3 strong bottlers do not have equal sales ofi‘ promotion. This finding disconfirrns OH 36. The average sales ofl‘ promotion for strong bottlers follow the ordering Coke > Pepsi > 7Up/RC/DR. The treatment contrasts in the ”Secondary," "Feature,” and ”Feature + Secondary" columns of Figure 4-7 relate to OH 3b. This hypothesis states that strong brands have equal unit and dollar sales on promotion. The numbers in each cell are promotion by bottler frequencies for Chain 3 in year 3. These data show some support for OH 3b. It is 129 possible that market response puts strong bottlers into a hierarchical implicit coalition. That is, the similarity of market response to Coke and Pepsi feature promotions provides an implicit coalition. Likewise, the similarity of market response to Pepsi and 7Up/RC/DR. feature promotions may produce another implicit coalition. Finally, because Coke is in a coalition with Pepsi, and Pepsi with 7Up/RC/DR., Coke may be in a coalition with 7Up/RC/DR even though Coke and 7Up/RC/DR sales on promotion are different. Such transitively coupled implicit coalitions are not covered in the literature at present, but this does not mean they are not possible. Figure 4-7 Observed Market Response to Strong Bottler Promotions Feature + None Secondary Feature S dsry 26 7 Coke 66.66 I jj.jj l 24 7 Pepsi bb.bb kk.kk 27 4 7Up/Rana. cc.cc l ( )memsaleslevelsarenotsigruficeulydiffauuforbothlogomcesmdlogccmson 3of'the4measures. ’- ’\ _\) men sales levels :6 not significantly different for log 01-1066. Overall, the data indicate some form of competitive parity in the similar market responses to bottler promotions. Thus, OH 3b receives some support, although the degree of confirmation of OH 3b cannot be assessed with precision. Bottler sales off promotion, however, do not confirm OH 36. F. Testy] 3 Research Hypothesis 4 The second research question of this study asks: Is there competitive parity between strong brands and competitive asymmetry between strong and weak brands? Hypothesis 3 addressed the issue of competitive parity among strong brands, and Hypothesis 4 130 addressed the issue of asymmetry by hypothesizing that brands that alternate promotions are beneficiaries of an asymmetry in competitive draw from private label brands. The original operationalization developed in Chapter 3 was designed to test cross- price, promotion, and competitive effects at the level of individual UPC 5. As knowledge of Nielsen's single-source data developed, however, it became apparent that a study of the UPC level of cross-promotion efl‘ects was not feasible. Error in store audits and errors integrating store causal effect with appropriate UPCs make single-source data unreliable for estimating cross-effects at the UPC level. Single-source data often indicate no promotion when a promotion is taking place. For example, in week 52 the single-source data indicate store brand UPCs were promoted in two stores of Chain 3 and not promoted in three other stores. According to the researcher's observation of Chain 3 and information from BBMI, the same soft drink promotions should run in all stores of Chain 3 each week. The three stores not indicating a promotion in week 52 sold as much or more of the store brand UPCs, as did the two stores indicating a promotion. In this case, only displays were indicated for stores 109-111. Somehow advertising in the store flier was not associated with store brand UPCs in these stores. Because the same store flier is used in all Chain 3 stores in a given week, and the store flier is used to indicate advertising, the promoted store brand UPCs should have been coded as advertised in all five Chain 3 stores. One solution to the problem of data reliability at the UPC level is to inspect and manually code promotions for the 1,600 UPCs in year 3. This approach, however, commits an error called the "regression fallacy" (Bawa and Shoemaker 1989, p. 74), which invalidates inferential statistics unless a holdout sample is used. A holdout sample would be inefi‘ective in this case because correcting by inspection leaves no data for a holdout sample. 131 A second possible solution is to analyze the data at a more aggregated level. Because the measurement error is biased against indicating promotions, when the data are aggregated, measurement error behaves as a filter and removes "background noise" of false promotions. That is, when data are aggregated to the bottler level across a chain, each store and each UPC functions as a repeated measure. When measures subject to Type H error are repeated, the chance of accurately detecting promotions is increased. Under these conditions, a positive indication in even one store can be a reliable promotion indicator. The benefits of using more aggregated data are: (l) the promotion detection criterion used in all previous analyses can be used in the present analysis; (2) the single- source data need not be changed in order to test hypotheses (This allows the use of inferential statistics). And (3) promotion measurement error is neutralized by aggregation. While aggregation to the bottler level solves the problem of accurate indication of promotion behavior, it also prevents modeling price effects. Because soft drink packages vary widely in per-ounce cost, and because aggregation pools pricing information across package sizes, price effects cannot be separated fiom other causal effects. Dropping price effects may enable a clearer analysis of promotional asymmetry, however, because the brand interdependence literature does not speak to how price, display, and advertising interact to produce asymmetries. Consequently, treating promotion effects as a gestalt may be a very good starting point for empirical testing. Another benefit of aggregating to the bottler level is that intrabottler competition cannot confound interbottler interactions. If, for example, the Coke bottler promotes 12-ounce cans, the competitive draw effects from Coke's 2-liter and 16—ounce packages must be addressed if the data are analyzed at the UPC level of aggregation. At the bottler 132 level of aggregation, however, these effects do not appear and, instead, the interbottler efl‘ects are brought into focus. Price effects were included in Equation 3-2 because Blattberg and Wisniewski's (1989) equation form was used to guide development of Equation 3-2. Aggregating to the bottler level and omitting price effects theoretically should not impair the test of the asymmetry hypothesis. Aggregation shifis the focus of the analysis fiom cross-price effects to cross-promotion effects (including price). The latter are precisely the constructs at issue in evaluating the brand interdependence explanation of promotion behavior. Because aggregating to the bottler level should eliminate measurement errors in promotion indicators, and because dropping price effects should not diminish the power of the test of the brand interdependence literature, the following analysis aggregates the single-source data to the, level of the bottler. 108(Sit) = cit + BlFit + B23it + 1338,, - B4Fit - BSsit - 36% + 37%, 4'1 where Sit = total unit sales of bottler i in week t (aggregated over 5 stores in Chain 3); Cit: constant forUPC i inweek t; Fit = dummy variable for feature promotion only for bottler i in week t; Sit = dummy variable for secondary promotion only for bottler i in week t; Bit = dummy variable for both feature and secondary promotion for bottler i in week t; Fit = dummy variable for COMPETITIVE bottler feature promotion only in week t; Sit = dummy variable for COMPETITIVE bottler secondary promotion only in week t; Bit = dummy variable for COMPETITIVE bottler feature and secondary promotion in week t; and (bit = dummy variable for stockup weeks (July 4, Labor Day, Superbowl, Memorial Day, and a second week before July 4, 1991, in Grand Rapids = week 52). Equation 4-1 is Equation 3-2 without price effects. This equation provides an analytical fi'amework for analyzing promotion asymmetry that is consistent with observation of: (1) Chain 3 promotion behavior, (2) the information provided by BBMI on how soft drink promotions work, and (3) the brand interdependence literature. Figure 4-8 133 displays the unit volume data for strong and weak bottlers that will be used to test for asymmetric promotion effects. Using the ounces measure, the correlation between strong and weak bottlers is -. 10 and using the cents measure, the correlation is -. 19. These negative correlations indicate that strong and weak bottlers are competitors. Figure 4-8 Strong and Weak Bottler Share of Volume 20 .2 0 irrrrrl'111'r'rrrrrrrrr'r'rrr'1'rrrrrfiur'rrrfirirrr 13 5 7 9 111315171921232527293133353739414345474951 Weeks {-30% *WedrOz. “MC” +WeskCents] 1. Reggsron Mysis Equation 4-1 was modified in three ways to develop a baseline model of cross- promotion effects. First, interactions between promotions and holidays were added for strong and weak bottlers. In exploring several specifications it became apparent that the holiday term (1370“) by itself was never significant, while the interactions between holidays and promotions were highly significant. The implication is that holidays affect only those bottlers on promotion. Second, because it was not possible to deseasonalize the data formally, two surrogate seasonality variables, the ”week number" and "cooling degree days," were added. Including surrogate seasonality variables like weekly dummies is a standard technique for controlling seasonality in the single-source industry (Bender 1991). Cooling degree days 134 for Grand Rapids are an intuitively appealing demand covariate if one believes that temperature drives thirst. Third, the term for weak brands' combined promotions (068“) was deleted because weak brands were never on combined feature and secondary promotions. These modifications produced a theoretically appealing baseline set of independent variables. An equation using the baseline independent variables is shown in Equation 4-2. 8it = 6it + BlFit + BZSit + B3Bit + B4Fit*¢it ' BSFit ' B6Sit - B7Fit*¢it + BgWeeknum + BgCDD, 4-2 Eight measures of unit volume (Sit) are available for each kind of bottler. These are: ounces, log ounces, ounce market share, log ounce market share, cents, log cents, cent market share, log cent market share. As a first step in analysis the baseline set of independent variables was regressed against each of the measures of unit volume. Tables 4-156 through 4-15d summarize the results for the baseline model. Table 4-156 OLS Regression Summary, Strong Bottlers - Ounce Measure 135 Table 4-15b OLS Regression Summary, Weak Bottlers - Ounce Measure Table 4-15c OLS Regression Summary, Strong Bottlers - Cent Measure Table 4-15d OLS Regression Summary, Weak Bottlers - Cent Measure 136 Four patterns emerge from these regressions. First, the ounce and cent measures for strong bottlers (Tables 4-15a and c) account for relatively little (one-third to one-half) of the variation. Second, share measures for strong bottlers account for much more (two- thirds) of the variation in sales. Third, ounce and cent measures for weak bottlers (Tables 4-15b and d) account for more than 70 percent of the variation. Fourth, share measures account for less variation for weak bottlers. Clearly, the same set of promotion variables does not account equally for the sales variation of strong and weak bottlers. The above R’s are comparable to previous studies. Kumar and Leone (1988, p. 184) explained an average of 42 percent of the sales variation, Blattberg and Msniewski (1989, p. 301) produced st between .7 5 and .95 for national brands (the analog of strong bottlers) and between .04 and .66 for generic brands (the analog of weak bottlers). An average of 65 percent of coefficients (not counting intercepts) in these previous studies were significant at or=.10 or greater. In the above regressions 63.3 percent of the coefiicients (not counting intercepts) are significant at the .10 level or better. Two independent variables, however, have counterintuitive signs. Weak bottler feature promotions decrease both strong and weak bottler sales, and cooling degree days (CDD) change sign when used in strong bottler market share calculations. Because there were feature promotions by weak bottlers in only three weeks, one of which fell on a stockup . holiday, there may have been insuficient information to estimate weak bottler feature efi‘ects. To compare, there were 11 weeks of secondary promotions for weak bottlers, while for strong bottlers there were 16 weeks of secondary, 29 weeks of feature, and 20 weeks of combined feature and secondary promotions. The change in R2 for strong bottlers between natural and market share units may indicate significant cross-shopping between grocery chains. Figure 4-9 displays the weekly change in market share between Chain 3 and the other Grand Rapids grocery 1 3 7 chains. Cross-shopping is likely because of both soft drink and consumer characteristics. Because the sofi drink category is fi'equently purchased, inventoriable, and fi'equently promoted, customers know a deep discount promotion when they see one (Kumar and Leone 1988, p. 179; Nagle 1987, p. 71). In addition, only 8 percent of grocery store customers are loyal to one store (Totten and Block 1987, p. 37). This indicates that customers shop multiple stores and can shift their purchasing to the outlets with good promotions. Colas are widely believed to be characterized by "brand loyal cherry picking" (ibid., p. 41). Ifconsumers know deep discounts when they see them, shop multiple stores during each sofi drink purchase cycle, and can shift their purchases to the best deals, the difference in R2 for Chain 3 natural and market share units seems reasonable. Figure 4-9 Grocery Chain Market Share Fluctuations - Grand Rapids 100%,}, 80% 60% § :1 5 40% 20% 0% ' 1"1' Ti'rrfi'.rv—r—r TV—r—r—rjlv 1"1"r"1"r"r'*I l4 7101316192225283134374043464952 Weeks ICIALIUI'HERS .CHAINI .CHAI'N2 WCHAINZII Another related limitation of the regressions is the low significance assigned to promotion variables. The strong appeal of these independent variables to both theoreticians and practitioners supports the expectation that these variables should unambiguously affect sales. Low significance combined with high appeal leads a researcher to ask whether other sources of variability might be confounding store level 138 significance measures. In practice, relatively liberal p-values of .15 to .20 are ofien used to assess the coefiicient significance for managerial purposes (Bender 1991). Even by such standards, however, the coefficients in Tables 4-15a through 4-15d are inadequate. One possible source of confounding is market level fluctuations in demand (see Figure 4-10). In an attempt to isolate the efi‘ects of causal variables within Chain 3, ounce and cent measures were indexed to remove cross-shopping and demand fluctuation efi‘ects. Figure 4-10 Market Level Demand Fluctuation - Grand Rapids Millions of Ounces 1‘1111717 "1"7"."1"I"T"I"1T'T":"1"!" ' 14 7101316192225283134374043464952 Weeks [BALLOTHERS ICHAIN] ICHAINZ ICl-IAIN3I To factor out these two effects, four indices were devised. Primary demand indices for cents and ounces were created by dividing each week's total demand (for Grand Rapids) by the average weekly demand (for Grand Rapids) for year 3. Similarly, cent and ounce market share indices for Chain 3 were created by dividing each week's market share for Chain 3 by the average weekly market share (for Chain 3) in year 3. Each dependent variable for each week was divided by the product of the appropriate primary demand and market har ' dex $“M‘flfifl‘é 15’ . Because coolin d da 3 and f tu s em (14 M .. 1) g egree y ea re promotions for weak brands had performed so poorly in the first set of regressions, these 139 variables were dropped, and a second set of regressions was estimated. The results of this analysis are contained in Tables 4-16a through 4-16d. Table 4-1 6a OLS Regression Summary, Strong Bottlers - Modified Ounce Measure 30¢ - . 840781 . 291232 . 380 Table 4-16b OLS Regression Summary, Weak Bottlers - Modified Ounce Measure Table 4- 1 6c OLS Regression Summary, Strong Battlers - Modified Cent Measure 140 Table 4-16d OLS Regression Summary, Weak Bottlers - Modified Cent Measure Removing the effects of cross-shopping and demand fluctuation decreases R25 for weak bottlers but raises R’s for strong bottlers. The number of coeflicients (not counting intercepts) significant at .10 or better increases to 82 percent. Two coeflicients were systematically weak in significance. Secondary promotions by strong bottlers have a marginally significant positive efi'ect on strong bottler sales and a less significant negative effect on weak bottler sales. The second irregular coefficient was the interaction between strong bottler feature promotions and holidays, which had a marginally significant effect on weak bottler sales (p=. 1228). Overall, the second set of regressions provides a satisfactory basis for estimating cross-promotion effects of strong and weak bottlers. 2. Cross-Promotion Effects Evaluating the cross-efi‘ects of promotions between strong and weak bottlers is complicated by two factors: promotion fi'equency and comparability. First, strong bottlers promote so often that even if they do not have an asymmetrical advantage on a per promotion basis, they may end up with an asymmetrical advantage by crowding out weak bottler promotions. Second, there is a " structural asymmetry” in promotion behavior because strong bottlers use difl‘erent forms of promotion than weak bottlers. For example, weak bottlers do not use combined feature and secondary promotions. This structural 141 asymmetry produces a comparability problem because it removes the possibility of simple paired comparisons between strong and weak bottlers for all promotion types. To address the comparability problem an index of cross-effects was constructed. The cross-elasticity for each promotion type was multiplied by its frequency of use in year 3. The elasticities for adjusted log ounces and adjusted log cents were used to construct the indices. Only elasticities significant at a=.10 or less were included in the computation. The intent in constructing this index was to capture the overall effect that strong and weak bottlers have on each other. Table 4-17 shows cross-efi‘ects for ounces and Table 4-18 for cents. Table 4-17 Summary of Cross-Promotion Elasticities Between Strong and Weak Bottlers, Ounce Sales Ounce Sales of Battlers are NA leak ~4. Table 4-18 Summary of Cross-Promotion Elasticities Between Strong and Weak Bottlers, Cent Sales Affect Cent Sales of Battier. Ibo are: m -21 . are leak -3. NA In ounce terms, strong bottler promotions have more than five times the impact on weak bottlers than vice versa. In cent terms the disparity is even greater - strong bottlers have a factor of seven advantage. Weak bottler secondary promotions have almost twice the effect on strong bottlers as strong bottlers have on weak (cross-elasticities of -.205 for weak and -.1 19 for strong) bottlers. The feature and combined promotions of strong bottlers have a greater efi‘ect on weak bottlers (cross-elasticities of -.358 and -.566 respectively) and are run frequently enough (29 feature promotions, and 20 combined promotions) to swamp the effect of weak bottler secondary promotions (ll of which ran 142 in year 3). Holiday secondary promotions are also more effective for weak bottlers (cross-elasticities of -.360 for weak and -.279 for strong bottlers), but there are so few stockup holidays in a year that this advantage does not accumulate to a significant competitive position. The property of this index that is most important for evaluating competitive interactions is its zero point. If strong bottler promotions had no effect on weak bottlers and vice versa, the index would be zero for each agent. If strong bottler promotions were somehow complements to weak bottler sales, the index would be positive. Because the cross-effects are both negative, strong bottler and weak bottler promotions are competitive. Because the above cross-promotion elasticity indices are of unknown distribution, a t-test cannot be used to assess the significance of the difl‘erence between strong and weak bottler index values. Because the index produces a single data point for the year, standard nonparametric techniques cannot be used. ‘ Ifa zero index value indicates no interaction, the 530 percent and 700 percent strong bottler index advantage strongly supports H4, that is brands that alternate promotions are beneficiaries of an asymmetry in competitive draw from private label brands. Strong bottler promotions have a large cumulative effect on weak bottlers, and weak bottler promotions exert little countervailing influence on strong bottlers. G. Testing Mob Hmthgm 5 The final research question of this study states ”Are promotional discounts equal for strong bottlers, and are strong bottler discounts larger than for weak bottlers?" Two hypotheses address this question. H5: Brands that alternate promotions should have approximately equal discounts (Narasimhan 1988, p. 435; Raju, Srinivasan, and Lal 1990, p. 286). 143 H6: Strong brands on promotion should have greater discounts than weak brands (Narasimhan 1988, p. 435; Raju, Srinivasan, and La] 1990, p. 286). Tables 4-19 and 4-20 display the average discounts for two package sizes within which strong bottler promotions overlap. If, as hypothesized, strong bottler discounts are not significantly different, a one-way AN OVA within these combinations should not be significant. The one-way ANOVAs, however, are either highly significant (p-value = .000 for 2-liter products) or marginally significant (p-value = .078 for 12-ounce products). For 12-0unce 12-pack products on feature promotion, however, discounts are not significantly different (p=.39027). Also, for strong bottler 12-ounce products ( 1 can or packs of 6, 12, 20 and 24) on feature promotion, discounts are not significantly difl‘erent (p=.718). The data show a mixture of strong and weak support for H5. Table 4-19 Average Discounts for 2-Liter Products F MIN/DR. Coke Pepsi J [4* Feature .394 .494 .143 | [ secondary .361 .407 .140 | Table 4-20 Average Discounts for lZ-Ounce 12-Packs I 7Up/ROI“. Coke P101 I I feature .430 .470 .460 | l seconaagx .245 .440 NA I It is possible that the dispersion of promotions among the 48 possible promotion combinations (six multipack combinations by two promotion types by four sizes) and a coincident dispersion of pricing strategies if bottlers do overlap in offerings make ANOVA an ineffective means of testing for true difl‘erences. For example, it is clear fi'om Table 4-19 that Pepsi did not strongly discount 2-liter products the three times they were on feature promotion. Instead, Pepsi's featuring emphasis focused on 20—ounce 8-pack products (promoted six times), for which discounts were not significantly different fi'om Coke and 7Up/RC/DR. discounts for featured 2-liters (p=.480). 144 Product and pricing difl‘erentiation for "taste goods" (like soft drinks) is a violation of Hotelling's principle of minimum differentiation (Moorthy 1985, p. 268), but it is not per se inconsistent with an implicit coalition explanation of promotions. If Pepsi has loyal customers who prefer a specific package (20-ounce bottles) and Coke's customers prefer another package (2-liters), it is conceivable that Coke and Pepsi could partition their promotional efforts so as to avoid direct competition for switchers. An explanation along these lines might explain why discounts are so similar for 12-ounce lZ-packs (because switchers prefer this package) at the same time they are difl‘erent for 20-ounce 8-packs. Another issue in testing Hypothesis 5 is whether AN OVA is an appropriate technique for detecting difl‘erences in discounts. A plot of the residuals against estimates for 12-ounce products shows some heterogeneity of variance (see Figure 4-11). Furthermore, a Tukey box plot (see Figure 4-12) reveals no significant difference between strong bottler discounts where ANOVA found a significant difl‘erence (p=.000). The v-shaped notches in Figure 4-12 indicate 95 percent confidence intervals around medians. Figure 4-12 makes clear that the variances in discounts are not equal for all strong bottlers. Figure 4-11 Discounts for lZ-ounce Products - ANOVA Residuals Residual 04 0.2 ........................... f. .............. , ' ....... I 0 t O . i .02 ........................... 3 .............. T. ............. . . 8 ...... .04 0.5 0.55 0.6 0.65 0.7 Estimate Chain3-Year3 145 Figure 4-12 Discounts for Strong Bottlers, All 12-ounce. Products 1 5 [ l T r 10 F q . . l] WSCOUNT .1 *0}— l“ 2U: O O I I I r~ N9 cote 9809‘ 7 ‘00 / Q d EOTTLER In addition, a density plot of strong bottler discounts shows them to be bimodal (see Figure 4-13). Bimodality could be caused by the price tiers that underlie promotional discounts. A superimposed normal distribution indicates the poor fit of strong bottler discounts to the normal distribution. The heterogeneity of variance and bimodality of strong bottler discounts make using ANOVA inappropriate (Wilkinson 1990, p. 484). Because the Kruskal-Wallace test is the distribution-free analog of AN OVA, the discount data for strong bottlers were reanalyzed using the Kruskal-Wallace test. The results indicated no significant difference of promotional discounts among strong bottlers for all 12-0unce products (p=.2643 ), 12-ounce products on feature promotion (p=.88684), and 12-ounce 12-packs on feature promotion (p=.2829). Two-liter products for strong bottlers had significantly different discounts (p=.000), although discounts for feature promotions of Pepsi 20-ounce 8-packs and 2-liter l-packs for Coke and 7Up/RC/DR. were not significantly difi'erent (p=.3 51). 146 Assessing the degree of support for the brand interdependence theory is complicated by the many sizes and quantity options available for bottlers to promote. On the one hand, the data provide moderate support for Hypothesis 5 because there are products (12-pack cans) for which the data strongly support the brand interdependence hypothesis of similar discounts. On the other hand, 2-liter products do not support this hypothesis without special pleading (that is, assuming that Pepsi is using 20-0unce products to do the same thing Coke and 7Up/RC/DR. are doing with 2-liter products). The brand interdependence literature does not deal with what might be called the microstructure of the market. Implicit coalition theory does not speak to the innumerable nuances of packaging, promotion, and product formulation that occupy much of a bottler's day and thoughts. Because the resolution of the brand interdependence theory is more coarse than the results at hand, it appears the most conservative conclusion is that there is a moderate degree of support for Hypothesis 5. Figure 4-13 Density Plot of Strong Bottler Discounts 04 40 F o 3 . » 30 o — 2o 8 E PROPORT lON PER BAR O to I 7”\ 3% Milk 025 045 065 085 105 DISCOUNT 147 H. Testing Hymthesis 6 Hypothesis 6 states: H6: Strong brands on promotion should have greater discounts than weak brands (Narasimhan 1988, p. 435; Raju, Srinivasan, and Lal 1990, p. 286). The data strongly support H6. Because of heterogeneity of variance and nonnormality of discount measures uncovered in the section above, the Kruskal-Wallace test was used instead of the t-test to test Hypothesis 6. This substitution is conservative because H6 hypothesizes a difference between the discounts of strong and weak bottlers, and the Kruskal-Wallace test is less sensitive than a t-test in detecting differences. Strong bottler discounts were found to be greater for all 2-liter products (p=.002), 2-liter products on secondary promotion (p=.002), and all 12-ounce products (p=.002). Discounts for 2-liter feature promotions and 12-ounce secondary promotions were not quite significantly different (p=.156 and p=.148, respectively). Soft drinks in cans were not feature promoted by weak bottlers and, consequently, could not be tested. 1. Sum Table 4—21 summarizes the results of the empirical tests. 148 Table 4-21 Summary of Research Questions, Research Hypotheses, and Operational Hypotheses l. hunummorywmwdumgpm-mmwmr H1: “C‘MWMWMJMMuhfipMflmmth-N predofevflnfleprunodonperlods(Khbcrg,Reo.-dSInk-l974.p.958) Ollie: Mpmhrdswhummflmdflmoflkh (Khberg,Rno.urlShd_l974,p.958). SUPPORTED. OHIII: mMofmnoIMMmmwofmmermFdfl) SUPPORTED. “2‘ Awwhfl(sfluC&)flkahMIMemmddlen-ehead 1990b, 3:? WMWhWMMMMHMMN Mflvuhnerqenbuofwednmflpn-ueoofldm SUPPORTED. 2. hfienmpefivepmymmghudsHMe-y-ufimm-Iflh-h? H3: “MMMHWWMWafisMM-IMIWQFM New“ imam) 0830: MWMeaqufldddohrnthymoflpv-mm Randy-19744;”; NWIMFMI) NOT SUPPORTED. 01135: MWMeWMqfl-m-dmmwhfieymmmm Randy-1974,5955; Narfl-IMFMI). SOME SUPPORT. H4: WMMWMmfier-Mhmmcdmmmw mmmumrmtpmzwuwmim.pmi SUPPORTED. 3. mmmwmmmummwmmhmmu brain? [15: mummwhewmquwmmimpoem muurmpm SUPPORTED FOR IZ-PACKS. INCONCLUSIVE FOR OTHER PRODUCTS. [16: “Mum“mmwuummwrmpaam MdUIMFM SUPPORTED. V. Conclusions A. Research SW The background to this study showed that the nature of promotion has been transformed in the past decade. New technologies like checkout coupons and UPC scanning, the evolution of promotion practices such as calendar marketing agreements (CMAs) and forward buying, and steady growth in promotion spending have moved sales promotion "out of the second-place shadow" of advertising (Lilien, Kotler, and Moorthy 1992, p. 324). Promoting has been explained from five theoretical perspectives: uncertainty about customer reservation prices, uncertainty about search costs, price discrimination, minimization of channel inventory costs, and competition. This study empirically tested the implications of an emerging stream of research, the "brand interdependence” literature, which explains promotions as a form of competition. The key mechanism hypothesized in this research stream is a link between brand preference and alternating promotions. Store-level single-source data for soft drinks supplied by A. C. Nielsen and CMAs provided by a cooperating bottler were integrated to test empirically the brand interdependence literature. 1. P_u_rpgse The purpose of this study was to subject empirical implications of the brand interdependence literature to the strongest empirical tests possible using the best data yet available. 149 150 2. Questions The brand interdependence explanation of promotion was tested by isolating three implications of the theory that were testable with the available data. The study cast each implication in the form of a question. 1. Is the. soft drink category characterized by alternating promotions among premium national brands? The brand interdependence explanation of promoting was developed, in large part, to explain anecdotal evidence of alternating promotions in the highly promoted and mature soft drink category (Lal 1990b). Consequently, if alternating promotions do not exist in the sofi drink category, the main empirical implication of this theory is false. This would provide evidence to reject the brand interdependence explanation of promotions. 2. Is there competitive parity among strong brands and competitive asymmetry between strong and weak brands? . The brand interdependence explanation of promotions relies on the notion of an implicit coalition among strong bottlers. Competitive parity among strong bottlers and asymmetric competition between strong and weak bottlers is the most likely foundation of an implicit coalition among strong bottlers. Ifthe market's response to top tier promotions is not symmetrical among strong bottlers and asymmetrical between strong and weak bottlers, the main hypothesized mechanism connecting brand preference to promotion behavior is severed. This would provide evidence to reject the brand interdependence explanation of promotions. 3. Are promotional discounts equal for strong brands, and are strong brand discounts larger than discounts for weak brands? A central premise of the brand interdependence literature is that strong bottlers "fluctuate” their prices in a way that allows them to share sales to customers who are 15 1 brand switchers (Narasimhan 1988, p. 428). This sharing of demand is based on the concept of a Nash equilibrium (Lal 1990b). A compelling way (because it is a Nash equilibrium) to share customers is for strong bottlers to use the same discounts (Narasimhan 1988, p. 43 5). If strong bottler discounts are not substantially similar, the main hypothesized mechanism enabling bottlers to coordinate promotions tacitly is severed. This would provide evidence to reject the brand interdependence explanation of promotions. 3. Method Single-source data are an intuitively appealing means of testing brand interdependence predictions. Yet, because single-source data capture merchandising detail at the UPC level and the implications of brand interdependence literature operate at the level of bottlers, some means of filtering competitor-level effects from UPC-level data was necessary. Information on soft drink promotion practice provided by a cooperating "strong” bottler completed the research We of this study. Consequently, two sources of data were used: store-level single-source data provided by A. C. Nielsen and information on promotional competition provided by a strong bottler in the Grand Rapids market. The fi'equency and discount strategies were extracted from the single-source data and then compared with predictions made by the brand interdependence literature. 4. Results Results of the study strongly indicate the existence of alternating promotions (question one), support the asymmetric draw aspect of question 2, and support the existence of deeper promotional discounts for strong bottlers (question 3). The only aspect of the study that did not receive any support was the equality of strong bottler sales off promotion. 152 :1) Question 1 Three operational hypotheses were used to test the first question. The proportion of time that strong bottlers were on any promotion (OH 1a) fell precisely into the interval hypothesized by Kinberg, Rao, and Shakun (1974, p. 958). In addition, the promotions of strong bottlers were not independent of one another (OH lb) as hypothesized by Lal (1990b, p. 43 7). Because the promotions of strong bottlers were extremely negatively associated, however, the chi-square and loglinear tests planned could not be used. This problem was addressed in three ways: first, by calculating correlation coefficients among strong bottler promotions (all were negative, supporting OH 1b); second, by performing a Wilcoxon test of independence (strong bottler promotions were not independent, p>.28 7, supporting OH lb); and third, by using logic to point out that the near perfect negative association of strong bottler promotions indicates these promotions are not independent. The third operational hypothesis (OH 2) confirmed that strong bottler promotions for three Grand Rapids grocery chains overlap significantly more often than would be predicted by chance (seven of nine measures of interchain promotional overlap were significant, 0r>.05), indicating the existence of alternating promotions in the Grand Rapids market. b 'n2 The second research question, addressing competitive parity among strong bottlers and competitive asymmetry between strong and weak bottlers, received mixed confirmation. As assessed by AN OVA treatment contrasts, strong bottler sales ofi‘ promotion are not equal (OH 3a), and strong bottler sales on promotion (OH 3b) are not equal for Coke and 7Up/RC/DR Strong bottlers do benefit from the hypothesized asymmetric promotional draw from weak bottlers (H4). 153 OH 3b received some important support, however, sales were not significantly different for Coke and Pepsi during feature and combined promotions, and sales for 7Up/RC/DR. and Pepsi were not significantly different during secondary and feature promotions. Market response to soft drink promotions provided some support for the brand interdependence approach and pointed toward necessary refinements. The results indicate that although strong bottlers do not service an equal quantity demanded from loyal customers (customers who buy their preferred bottler's products regardless of promotions by other bottlers), they do service an approximately equal quantity demanded fiom switchers (nonloyal customers). Implicit coalitions to date have been based on symmetrical payofi‘s for strong players. This is a simplifying assumption that provides an intuitively appealing suficient condition for an implicit coalition. The challenge raised by the current study is whether symmetric payoffs are a necessary condition for an implicit coalition. c) Question 3 The third research question addresses two aspects of promotional discounts: Are promotional discounts equal among strong bottlers (HS), and are strong bottler promotional discounts larger than weak bottler promotional discounts (H6)? Strong bottler discounts were not significantly different for 12-pack 12-ounce cans but were different for the only other size (2-liter) promoted by all three strong bottlers. Discounts were significantly different for 2-liter products because Pepsi's feature promotions did not deeply discount these products. Coke and 7Up/RC/DR discounts were not different for 2-liter feature (p=.255 using ANOVA and .124 using Kruskal-Wallis) or secondary (p=. 198 using ANOVA and .236 using Kruskal-Wallis) promotions. Strong bottlers ran 12-ounce promotions frequently (Coke ran 14, Pepsi ran 9, and 7Up/RC/DR ran 13 feature promotions in this size). Aside from 12-ounce products, howe‘ was a featu Other term the r is, tl equz Cok bon pro: bot bot the: t0 2 mt in r: app 154 however, strong bottlers appear to avoid feature promoting the same size package. Coke ran seven, 7Up/RC/DR. ran one, and Pepsi ran three, 2-liter feature promotions. Pepsi was alone in feature promoting 20-ounce products (6 times). A Kruskal-Wallis test of feature promotion discounts for Pepsi 20-ounce products and 2-liter products for the other strong bottlers found no significant difference (p=.351). Consequently, if viewed in terms of frequency of promotion rather than in terms of package sizes, H5 is confirmed for the most frequently and second most frequently promoted strong bottler products. That is, the most fi'equently feature promoted products (12-ounce cans) had approximately equal discounts, as did the second most frequently promoted products (2-liter bottles for Coke and 7Up/RC/DR., and 20-ounce bottles for Pepsi). 5. Sum The results of the study indicate the presence of an implicit coalition among strong bottlers and qualitatively support the brand interdependence explanation of soft drink promotions. While support for one operational hypothesis was absent (OH 3a: strong bottlers have equal sales ofl‘ promotion) and support for two other hypotheses was marginal (OH 3b: strong bottlers have the same sales on promotion and H5: strong bottlers have equal discounts), common sense refinements of the brand interdependence theory could bring these findings and the theory into accord. The refinements needed are to allow market response to promotions and size of loyal customer segnents to vary by strong bottler. The brand irnterdependence literature assumes strong bottlers are identical in power. It is this assumption, not the body of the brand interdependence theory, that appears not to hold in practice. LEE. This r1 promote perfect t] in the are Researcl interdepn t0 confin intewien Pepsi wt get mor A 50 require brands. Within level, g marker and pc' At SllOnE 155 B. Theoretical Implications This research has provided empirical support for the hypothesis that national brands promote for a mixture of cooperative and competitive reasons. The data were not a perfect fit with the theory, and refinement of the brand interdependence theory is needed in the area of necessary and sufficient conditions for implicit coalitions (see Future Research below). Nonetheless, the findings were broadly consistent with the brand interdependence theory. In addition, substantial qualitative and anecdotal evidence seems to confirm the presence of an implicit coalition among strong bottlers. During one interview a former Pepsi marketing manager (Duffy 1991) stated: ”When I worked at Pepsi we thought of the market as ours, theirs [Coke's], and up for gabs. We wanted to get more than our fair share of what was up for gabs. " A second implication of this study is that understanding competition for soft drinks requires attention to interactions among five aggegation levels. At the lowest level are brands. Brands aggegate to the second level of organization, bottlers. Bottlers compete within a third level of organization, grocery stores. Stores are aggregated into a fourth ‘ level, gocery chains. And gocery chains compete within a fifth level, the geogaphic market. Fluctuations at each level demand variation, market share shifts across chains, and poor compliance within a chain, can confound store-level analysis. A third theoretical implication is that the nature of the promotional asymmetry between strong and weak bottlers is different fi'om the asymmetry among price tiers previously reported (for flour, margarine, tissue, and tuna) by Blattberg and Msniewski (1989, p. 303). While lower tier products had little effect on upper tier brands in these four categories, weak bottlers (the analog to Blattberg and Msniewski’s lower tier products) had strong and significant efi‘ects on strong bottlers in the present study. The upshot is nha.atti smdy cor (liliern, B Then ofnucro order to promoti- would 3 make predict, predict drinkrr To 1 imPOrt Shge hisea diffi Phen Unde Pro /C7 1m iss - 156 that, at times, lower tier products can affect upper tier products. Overall, though, this study confirms the emerging consensus that promotional elasticities are asymmetric (Lilien, Kotler, and Moorthy 1992, p. 329). The most interesting theoretical implication of this study concerns the complementarity of microecononnic and game theory. This study has drawn heavily on both literatures in order to make sense of promotion practice. A microeconomic approach might predict that promotions should take place in proportion to market share. Results from this study would seem to agree, because the correlation between time on any promotion and bottler market share is .866 using cents and .755 using ounces in Chain 3. Price theory would not predict, however, that Coke and Pepsi promotions would not overlap. Game theory does predict alternation and, consequently, complements price theory in accounting for soft drink merchandising. C. Methodological Implications To make sense of single-source data in testing competitive strategies, it is critically important to start with and understand competition from the competitor's perspective. Single-source data so inundate a researcher with complex and detailed measurements that it is easy to let the measuring instrument, rather than the phenomena, drive a study. The difficult task is to find that combination of single-source measures best representing the phenomena of interest. The key methodological implication of the present study is that understanding the taxonomy (none, secondary, feature, combined) of soft drink promotions immeasurably increased the validity and interpretability of the theory tests. D. Public Polig Implications Because this study supported the existence of an implicit coalition, several public policy issues are raised. The efi‘ects that implicit coalitions and/or CMAs have on competition N: are th drink Currir antico POICIT Thr F irst, 1 is, con Pfom cllsto 1990 to b1 indi met 157 and consumers are by no means simple. In soft drink merchandising, the static nature of brands, product formulations, and business-to-business relationships forces bottlers to compete using a restricted set of variables centering on promotion. A kind of gaming rivalry results in which bottlers generate new kinds of promotions and test whether they mitigate competitive effects. CMAs may have evolved as a tactic because bottlers perceived that they alleviated competitive pressure. This produces a relatively benign process of evolution through tactic selection that is more diflicult to damn than a per se malignant process whereby national managers for Coke, Pepsi, and 7Up/RC/DR. covertly meet to agree on a collusion strategy. Nonetheless, bottlers appear to have sustainable market power. Because soft drinks are the top selling category in supermarkets (Felgner 1988, p. 187), because dominant soft drink brands are the most frequently promoted products in gocery stores (Krishna, Currim, and Shoemaker 1991, p. 9), and because the industry has been surrounded by anticompetitive practices like price fixing and exclusive territories (Galvin 1990, p. 27), potential harm to consumers and/or competition is a reasonable concern. There are three principle ways promotion tactics might mitigate competitive efl‘ects. First, the costs of competitive promotions may crowd out the benefits of competitiorn, that is, consumer's surplus (Lodish 1986a). Second, strong bottlers may use alternating promotions to price discriminate, capturing the value that would otherwise be lost to customers by employing myopic competitive tactics (Hoyt, Calantone, and di Benedetto 1990; Narasimhan 1988). Finally, alternating promotions may be useful to strong bottlers to block (or crowd out) weak bottler promotions (Lal 1990b). The theoretical literature on promotional price discrimination discussed in Chapter 2 indicates that alternating promotions cannot enable price discrimination unless some mechanism operates to prevent less price-sensitive customers fi'om stocking up (Hoyt, C alan custor DUI SC relatix costs questi the irc promr Th. censur- promc COHSU‘. Th. proble every highes gain a Prom Weak for i . feco 1980 the . disc 158 Calantone, and di Benedetto 1990). The most likely mecharnism is lethargy of brand loyal customers. Although this study did not assess which customers are lethargic in seeking out soft drink promotions, the econonnics of information (Becker 1965) would point to the relatively well-to-do consumers who because they make more, have higher opportunity costs of time which cause them to work more, and shop fewer grocery chains. The question requires empirical investigation to draw further conclusions. There is, however, the ironic possibility that strong bottlers price discriminate against the wealthy by promoting intensely. The issues of geatest practical importance are whether promotion costs crowd out consumer surplus and whether strong bottler promotions crowd out weak bottler promotions. Though underlining the importance of the effect(s) of promotion on consumer surplus, this study does not provide data to evaluate those effects. The last issue of crowding out of weak bottler promotions seems unlikely to be a problem. The largest gocery chain in the present study had alternating promotions in every week. Each week was put up for bid. Presumably the weeks are "sold" to the highest bidders by the chain. Consequently, while small bottlers may have the most to gain fi'om access to promotions, the strong bottlers probably have the most to lose. Promotions then are allocated by a market on an opportunity loss basis, and, consequently, weak bottlers are not being treated unfairly. The present study does not provide many answers. It does, however, lay a foundation for investigating four interesting public policy questions. First, antitrust cases have long recognized the riglnt of businesses to gain monopoly power through innovation (Petty 1989, p. 228). Should profits harvested from brand-loyal customers by promotions have the same legal protection as profits fiom innovation? Second, should promotional price discrimination tactics be viewed in a different light if they extract rents from a lethargic wealthy n antitrust r be illegal loyalty b: legally dc advertisi E. Lir Anc adveni Conse‘ 159 wealthy market segment rather than a poor price-conscious market segnent? Third, antitrust cases have considered ”fighting ship" and "fighting brand" preemptive tactics to be illegal (ibid., pp. 229-30). Should preemptive promotional tactics based on brand loyalty be considered illegal? Finally, independent recognition of mutual self-interest is legally defensible in the context of pricing (Nagle 1987, p. 86). Should this defense cover advertising, display, price discounting, and/or couponing? E. Limitations The generalizability of this study's results are limited. The method developed for coding promotions is dependent on information provided by the cooperating bottler and, ultimately, by conditions in the Grand Rapids market. In additiorn, the promotion coding scheme was developed to recover promotions from gocery chains that promote fiequently. The methods used in this study may or may not capture the competitive dynamics of bottler competition in smaller grocery chains. Another limitation of the study is the omission of broadcast advertising. While in-store advertising is included in Nielsen's single-source data, broadcast advertising is not. Consequently, to the extent that broadcast advertising affects demand, the results for question 2 (competitive parity among strong bottlers and asymmetry between strong and weak bottlers) may be biased. Results for questions 1 and 3 should not be affected, however. Not analyzing all Grand Rapids gocery chains simultaneously could possibly limit the findings of this study. Bottlers could conceivably partition sales by chains as well as by time (using CMAs). Ifsales are partitioned by chains across the Grand Rapids market, sales 03 promotion for strong bottlers could be equal at the same time ofi’ promotion sales in each chain are not equal. The brand interdependence literature has treated the 1 6O interdependence of bottlers and the interdependence of stores, but it has not treated this kind of irnteraction between jointly interdependent bottlers and stores. F. Future Reseggh The first issue for firture research is to isolate the necessary and sufficient conditions for an implicit coalition. Are symmetric payoffs for strong bottlers a necessary condition for an implicit coalition? The asymmetrical response between strong and weak bottlers might provide strong bottlers sufficient incentive for an implicit coalition even if they do not profit equally fi'om the coalition. A second issue for firture research is whether implicit coalitions ”nest” within price tiers at the same time they operate between price tiers. It is conceivable that two kinds of implicit coalitions (strong versus weak bottlers and Coke and Pepsi versus 7Up/RC/DR.) could operate simultaneously. A third question for firture research is whether Coke and 7Up/RC/DR. are in a coalition even though market responses to Coke and 7U p/RC/DR promotions are significantly difl‘erent. IfCoke and Pepsi are in a coalition, and Pepsi and 7Up/RC/DR. are in a coalition, could some form of competitive transitivity bind Coke and 7Up/RC/DR. into a coalition? A fourth issue for future research is an improved measure of alternating promotions. The two measures used here - time on promotion and negative association of promotions - do not finlly capture the alternation construct. Strong bottlers alternate promotions in order to share the switching segnent and to block the promotions of weak bottlers. Given the assertion that strong bottlers allow weak bottlers to promote only in "bad weeks in January or F ebruary" (Consumer Reports 1991, p. 524), counting the weeks each bottler is on promotion may not accurately estimate the access each bottler has to the market. 161 Another weakness of present measures is that they do not convey any information about sequence. Strong bottlers may block weak bottlers in subtle ways: by not letting a weak bottler promote at all (inviting antitrust action), by discounting to stockup switchers the week before weak bottlers promote, or by running secondary promotions alongside the weak bottler‘s feature promotion. Sequence information is indispensable for isolating all these blocking strategies. A fifth issue for firture research is a theoretical basis for coding and recoding single- source data. At least three ways apparently are possible: in absolute terms, in relative terms, and heuristically. Because it tracks data at the UPC level in many different store layouts, Nielsen's single-source data are coded in absolute terms (advertising data) and by using heuristics (display data). This coding scheme may not give the best information possible for the peculiar needs of soft drink bottlers, however. What matters to the bottler is display, price cut, and advertising relative to other bottlers. Given the quantity of data involved and present technologies for integrating causal data into single-source databases, Nielsen cannot be expected to code relative competitive information for each category. The research question is: Can single-source data coded on an absolute and/or heuristic basis be recoded to a relative basis? If so, can data coded on a relative basis be converted back to an absolute/heuristic basis? The consequences of aggegating single-source data provide a sixth area for future research. Blattberg and Neslin (1990, p. 183) state that if promotion schedules for all stores in a chain are the same, regession coefficients estimated fiom data aggregated to the chain level will be unbiased. Model coeficients or even the appropriate functional forms may vary by product, bottler, store, chairn, and/or market. Single-source data are presently enabling researchers to explore empirically these issues for the first time. E” Or markt markt COST. consi. Coca The l pron melc four row Wh use ver VI. Appendix A A. Cgmgr Mar_keting Agreements On June 6, 1991, the researcher visited the bottler's facility and met with the division marketing manager and key account manager. The subject of the meeting was calendar marketing agreements (CMAs) in the soft drink industry: how they work and what they cost. CMAs are agreements between retailers and bottlers on a merchandising bundle consisting of advertising, price cuts, and retailer support. CMAs for Pepsi of Michigan, Coca-Cola of Michigan, and the RC Cola/7Up bottler for Michigan were provided the researcher. CMAs figure prominently in tests of the brand interdependence literature because they appear to be the coordinating mechanism for alternating promotions. B. Bundling The RC Cola/7Up bottler’s pricing schedule provides some insight into how advertising, promotion, and price are packaged by CMAs (see Table A-l). CMAs are bundles of merchandising elements indexed by retailer display and advertising. Prices are quoted in four groups (the rows in Table A- 1) for package forrrns (the columns). The price in each row is set based on the retailer's level of merchandising support. The top row, "Regular Wholesale,” appears to be purely a reference price; according to the bottler, it is never used. The terms and condition for the other rows (merchandising conditions) are quoted verbatim from CMAs as follows. 1. Everyday Value Allowance Customer [the store] must reduce retail [price] to consumer to reflect promotional allowance. 162 4).. l 63 2. In-store Display Allowance Customer must display products separate and apart from the normal beverage department and reduce retail price to consumer to reflect promotional allowance. 3. Feature Ad Allowance Customer must feature advertise all available products in the highest form of account's normal means of advertising. Any form other than newspaper must be approved Customer must also reduce retail price to reflect promotional allowance. Table A-1 CMA Price/Promotion Schedule a-prr. e-plt. 12-plt. Package lG-ox . let. Z-Mte: Cane Cane Regular: Caee Cost 810 . 50 $12 . 25 $9 . 55 59 . 55 wholesale Unit Cost 83 . 50 $1 . 53 82 . 39 84 . 7B Pmtlon 9 9 310 I 291 9 460 9 550 Allowance $2 . 85 $2 . 7O 8 . 60 $3 . 15 My Case Cost $7 . 65 59 . 55 $8 . 95 $6 . 40 Value out: Coat 82 . 55 81 . 19 $2 . 230 83 . 20 Suggested Retail $3.19 $1.49 $2.79 $3.99 nan-gm 20.14 19.94 19.04 19.84 Pmtion I I 430 In-etore Allowance NA 83 . 18 “A Display Case Cost $6 . 37 Unit Cost $1 . 59 Any Suggested 2 week- Retall 81 . 99 Karma 204 Ma: 9 I 301 9 235 I 495 9 558 feature Allowance $4 . 54 84 . 39 $3 . 60 $3 . 20 Ad Can Cost 85.96 87.86 $5.95 $6.35 Unit Cost $1 . 98 5 . 9O 81 . 49 $3 . 18 Any suggested 2 Weeks “tall $1.99 8.99 81.49 33.19 Source: Brooks Beverage Management. lnc(l991). When asked why the "In-store Display” row is empty except for 6-pack cans, the bottler answered that retailer pass-through for the other products is poor: "We set up progams, and retailers do whatever they want with them.” This is consistent with findings by Walters (1989, p. 263) that displays have the lowest pass-through of all forms of gocery promotion, with the observation in the literature that retailers appear to have substantial power over manufacturers (Kumar and Leone 1988), and with the finding that bill-back ; Salmon 1' C Bid: Strong bottlersr and retai year." 1 Food rel advertis- the RC I $13,333 65.4 pe bottler’: Becaus Compe l'Olun Sales I 164 bill-back promotions give manufacturers little if any actual control (Buzzell, Quelch, and Salmon 1990, pp. 141-2). C. Bidding Strong bottlers bid for promotions. Request-for-bid letters are sent by retailers to bottlers on a quarterly basis. Bottlers are asked to bid on selected weeks in the quarter, and retailers accept bids based on "dollars available, cost of goods, brands, and time of year.” The "dollars available" are trade promotion moneys ostensibly for advertising. Food retailers in general cover almost 100 percent of their advertising with cooperative advertising payments (Blattberg and Neslin 1990, p. 348). The advertising copayments by the RC Cola/7up bottler for feature advertising range fiom $0 to $30,000 and averaged $13,333.33 for six features in the first quarter of 1991. These "ad payments” grew by 65.4 percent in one large Michigan chain between 1990 and 1991. Ad payments for the bottler's total market grew by an average of 36.6 percent between 1990 and 1991. Because these payments are gowing far faster than advertising costs, they are not soley compensation for advertising costs. D. Rebates An inspection of the "Feature Ad” row in Table A-1 shows that all the packages have the same wholesale and retail price. Retailers make their margin on feature promotions in the form of rebates (or bill-back allowances) fi'om the bottler after the promotion is completed. Table A-2 shows the rebates the cooperating bottler pays by package type and volume condition. ”Incremental cases" are the number of cases above the previous year's sales plus a 7 percent growth factor. 165 Table A-2 Per-Case Rebates under Three Promotion Conditions Everyday In-store feature-Ad a Value Display Display B-Pack lG-oz . Returnable 0 O 3 . 20 Cans O 0 8 . 30 Z-Litor Bottles 0 0 8 . 30 Incruaental Cases 0 0 3.50 Source: Brooks Beverage Management, Inc (1991). This per-case rebate could be interpreted as "payola.” The 60 Minutes documentary presented payment contracts from bottler to retailer showing $144,000 per quarter, but the program did not present the wholesale or retail price information needed to determine whether these payments were payola or bill—back allowances. If the Coca-Cola and Pepsi bottlers covered in the 60 Minutes special conduct business in the same way as indicated in the CMAs obtained by the researcher, the payments interpreted by the documentary as payola were in fact the retailer's goss margins. In the gocery trade, 46 percent of promotions combine discounts from regular price with per-case rebates fi'om manufacturers (Walters 1989, p. 260). The usual explanation for the use of bill-backs or rebates by the manufacturer is that they allow control over retailer compliance with promotional agreements (Blattberg and Neslin 1990, p. 318). This explanation, however, ignores the reciprocal interdependence between bottlers and retail chains. If a bottler unilaterally reduces a bill-back allowance (because of poor compliance) to a retailer the retailer has an easy means of retaliation, that is, not allowing the bottler to promote again. The bottler personnel interviewed claimed that they are controlled by retailers and that CMAs bottlers to husband their limited irnfluence over retailers and focus this influence on a few irntensely profitable promotions. Rebates could be viewed as a means the stronger partner to finance a weaker partner's promotional costs. Bill-backs decrease capital requirements (Buzzell, Quelch, and Salmon 1990.11 143 the bottler t It is wor course of tl competitior manufactur intense corr the zone lei level. Fron function be In the C feature ads Strongest p it s6561115 re Because bt Consumer Yet, rCtznili Iimited tolt PTOmotiOD A third manufacn A fou cha"I'lel fr"(luentl backs b 3 166 1990, p. 143), and are paid quarterly. This resembles a functional shift of financing from the bottler to the retail chain (Stern and El-Ansary 1982, p. 13). It is worth noticing another fimctional shift explanation of rebates uncovered in the course of this study. The editor of a gocery trade magazine stated that he thought competition at the price zone level was responsible for a shift of the pricing firnction from manufacturers to retailers. For example, a chain operating with two price zones, one with intense competition and the other with little competition, requires products to be priced at the zone level. This creates tremendous pressure for pricing to take place at the retail level. From this perspective, rebates may simply be a means to shift and share the pricing function between the wholesale and retail levels in the marketing channel. In the CMAs obtained for this study, rebates are only used when retailers agree to feature ads and feature displays, that is, rebates are part of a bundle producing the bottler's strongest promotion. Ifthe primary finnction of rebates is to provide a compliance check, it seems reasonable they could be used to monitor less intense forms of promotion. Because bottlers deliver direct to stores they can check compliance directly. Consequently, bottlers theoretically should not need to use bill-backs for this finnction. Yet, retailers have a limited tolerance for compliance checks. To make the most of this limited tolerance, bill-backs may be used by bottlers only for their most importarnt promotions. A third reason for bill-backs may be Federal Trade Commission regulations that require manufacturers not pay retailers urnless they perform (Totten and Block 1987, p. 12). A fourth reason for rebates may be a need for flexibility resulting from the shift in channel power from mamrfacturers to retailers. Terms of rebate agreements change fiequently as bottlers bid for feature promotions. When payment is made through bill- backs, bottlers can change their bids up to the moment a promotion is run. Paying on a bill-back basi Bottlers may because of ti chain in the r to all other 1 chains) are 1 offer arrives terms. If sc be tacit pric E. Brand All stror pellsi's CM available ir Will be feat States; "Ct accOuln’t'sl complexitj According 167 bill-back basis gives retailers and bottlers "eleventh-hour" flexibility in dividing margins. Bottlers may be able to charge different prices to different customers in the short run because of this flexibility. Once an ageement is reached between a bottler and a gocery chain in the eleventh-hour, under the antitrust laws the bottler must offer the same terms to all other gocery chains. Because the other chains (particularly small or unsophisticated chains) are likely to have frozen their promotion schedules by the time an eleventh-hour offer arrives, transactions costs preven them from accepting the bottler's eleventh-hour terms. If some chains are persistently less flexible and/or unsophisticated, bill-backs may be tacit price discrimination mecharnisms. E. Br v B lrs All strong bottler CMAs apply across all brands and packages carried by a bottler. Pepsi's CMA product line requirement is that "feature pricing should include all brands available in that package. " Coca-Cola's CMA does not explicitly state that all products will be featured, but it implies that this will be the case. The RC Cola/7Up bottler's CMA states: "Customer must feature advertise all available products in the highest form of account's normal means of advertising. " These cross-brand requirements collapse the complexity of arranging promotions irnto a more manageable menu of offerings. According to the bottler interviewed, three package forms (2-liter bottles, 6-pack cans, and 12-pack cans) are the principal focus of soft drink promotions. F. Compliinnce Paments Street money, a lump-sum payment made by manufacturers to retailers to run promotions, is ofl‘ered because wholesalers do not pass through trade deals (Blattberg and Neslin 1990, p. 319). Ad payments (see Table A—3), given the constant costs of the 168 retailer's store flier and the bidding for feature weeks, resemble street money because both are essentially direct payments to obtain retailer support. Table A-3 Bottler Promotional Activity Ad Paymt Time Frame Activity 820,000 1/14/91 2—n1m feature 0 1/21/91 l2-Pack Can new“ 8.20/08. Bill-Back on Cases 801d. Unit Cost I 52.88 0 2/12/91 Grand Opening 3 Detroit Stores lZ-Pack Cans e 82.675 Unit $10,000 2/10/91 lZ-Pack Can Feature 9 82.77 85,000 2-Liter Secondary E 5.98 515,000 3/4/91 Choice of Package Feature $30,000 3/25/91 112-Pack or 2-Liter Raster 400,000 Total Source: Rooks Beverage Manager“ lnc(l991). This analogy between street money and ad payments is incomplete for two reasons: Ad payments are made to chain buyers, not individual stores, and Ad payments are rrnade even though promotional pass-through is not a problem. On the one hand, many of the chains using CMAs do their own distribution and would capture the incentives of a trade promotion. On the other hand, soft drink companies deliver directly to stores, so they do not run trade promotions. Although ad payments and street money are difl'erent in substance, they are similar in function. Street money was not discussed in the 60 Minutes documerntary, but it is another promotion practice that could be confinsed with payola. A dealer-loader performs a similar function to street money and ad payments but is offered at the store level rather than at the chain level. A dealer-loader, which is an inducement to a store manager to comply with a promotion, often takes the form of tacit gifts (such as microwave ovens, children's toys, or television sets) that are used initially in an in-store display but are not retrieved by the soft drink company after the promotion. An industry expert related that ”the ultimate dealer-loader" was given by Pepsi - Michael 169 Jackson concert tickets (Duffy 1991). Figure A-l shows the three forms of compliance payments used in soft drink promotions. Figure A-l Names of Promotional Incentive Payments Used in Soft Drink Marketing Bottler "Ad-Copayment" "Dealer-Loader" "Street Money" Y Chain Buyer i Chain Store Manager Independen't] The incentives provided by ad payments, dealer-loaders, and street money are irnportarnt to recognize for the purposes of this study because they have substantial effect in channels. These payments are not, however, captured in single-source data so their roles as opportunity costs cannot be assessed from these data alone. G. Forms and Functiorg of Exclusivity The Pepsi and Coca-Cola CMAs require that the other bottler's products not be featured at the same time. Pepsi stipulates that ”no products distributed by the Coca-Cola Bottling Company of Michigan are to be featured on Pepsi ad weeks.” Coca-Cola states: ”All ad weeks to be exclusive of Pepsi Cola Products. " The RC Cola/7up bottler does not constrain the advertising of other soft drink products. These restrictions seem to be common sense; a bottler paying for a retailer's support should be able to expect at a minimum that a direct competitor not receive that support at the same time. During the 170 June 6 interview, bottler personnel reported that at one time Coke and Pepsi CMAs excluded RC Cola as well. RC Cola then sued Coke and Pepsi and succeeded in having references to RC Cola deleted fi'om Coke and Pepsi CMAs. Because price and display are bundled with feature advertising, the result is price and display exclusivity in any given week. The limited number of weeks available causes a ”crowding out" phenomenon. Bottlers of products that have larger profit contribution from promotions have more incentive to pay gocery retailers than do bottlers of products with smaller profits. These "stronger" bottlers then outbid ”weaker" bottlers for the promotions, and the weaker bottlers end up with few, if any, feature promotions. In addition to feature promotion crowding out, secondary promotions can crowd out smaller bottlers. When Coca-Cola is featured, Pepsi or the other bottlers can have smaller in-store displays and smaller price discounts, called ”secondaries," and these are negotiated by bottlers in the same way as feature promotions. Table A-3 showed a $5,000 payment for a secondary promotion for RC Cola/7Up 2-liter bottles on 2/18/91. Secondary promotions may be used offensively or defensively. RC Cola/7Up may have paid more for its secondary to prevent a secondary promotion by Coca-Cola or Pepsi. If market segnents have ”trigger prices” at which dramatic shifts occur between brands, a secondary promotion can provide a bottler with some control over prices and brand shifting. In some instances secondary promotions of one strong brand while another is on a feature promotion might be construed as tacit cooperation to keep smaller bottlers fi'om obtaining time on promotion. Observation of several stores in the cooperating bottler‘s market during spring and summer 1991 confirmed the presence of combined feature and secondary promotions by strong bottlers. The clear two reta C01 qu: ch: 171 H. Promotional Frequency Both the Coca-Cola and Pepsi CMAs include weekly ”year-at-a-glance" calendars. The Pepsi CMA stipulates a minimum of 26 weeks of feature activity, and the calendar clearly marks Pepsi's preference for the first and third weeks of each month. One of the two Coca—Cola CMAs obtained included a completed promotional calendar for a small retailer in 1990 (see Table A-4). Coca-Cola was promoted 20 weeks, Pepsi 18 weeks, RC Cola/7Up 12 weeks, Perrier 1 week (September 10) and Spartan brands 1 week (September 17). Table A-4 Promotion Schedule for Soft Drinks, Small Independent Grocery in Michigan eh. i; E; 29 ED E; as e E 12 9 he“ [no e si Tap/RC he si mam e si mm si e 30 Y 4 O IO 3 30 - l3 0 7 B . 0 I? 4 e film e ”up/RC i e Inc 1 Coke rier ppartan si t. Bit... :2... ii... a... E;. .. a... t... E; a... a... in. El L1 MN I ; WM E81 COR. Dole Lit Cote Coke L. L81 Source Beverage Manama. Inc(l991). i I . Promptiopal Discomt The goss margin per case of 6-pack cans for retailers and bottlers is displayed in Table A-5, along with case costs to retailers and prices to end consumers. The break-even sales quantities for the price changes implied can be calculated using Nagle's break-even sales change formula (1987, p. 30). Cas _______al_ro 0;]. dis im b< 172 Table A—5 Cases of 6-Pack Cans, Price, Cost and Gross Margin under Three Promotion Conditions Everyday Value In-store Display Feature-Ad a Display rat-tune s Price $11.16 $7.96 $5.95 Retailer' s Cost $8 . 95 56 . 37 85 . 95 Retailer's at 52.21 81.59 5.00 Retailer's 0114 19.804 19.974 5.044 Rebate . oo .00 .30 Bottler' s Price $8 . 95 86 . 37 35 . 95 Iottler' s Cost 34.37 4 $4.37 94.37 Bottler's 34 $4.58 $2.00 $1.58 Dottler's m4 51.174 31.404 26.554 % B.E. sales change = Cf: 83:32:13,“ 9 100. A-l Table A-6 provides insight into channel incentives. Retailers should be willing to display products as long as displays increase sales. The bottler's price structure, however, implies that displays must increase sales by more than 30 percent in order to be profitable. The disparity between the retailer‘s required sales increase of 501.2 percent and the bottler's required 27 percent increase for feature-and-display promotions seems to be consistent with Hawkins's (1950) theory that channel members at different levels face different demand curves. Table A-6 Incremental Break-Even Sales Changes Implied by Promotional Price Structure I May Day Value 111-Store Diglay Feature-Ad a DisplaJLI I Retailer n/a 100 .004 501 .024] I lottler u/a 129.044 126.584 I J. Sum The trade literature on soft drink promotions has identified the existence of alternating promotions and implies contracts are underly the relationships between bottlers of national brands and gocery chains (Pasztor and Reibstein 1987; 60 Minutes 1987). The trade literature has also been acutely sensitive to the weaker bottler's access to the distribution channel (60 substantia11 cooperatin- What h. market acr promotior willing to C 0ca~C 01 loss invol because t‘ promotio constrain This r- these agr meChanis competit hYPOthes mechanis The Colman and me power Pencile betw cOl’rlp . l 73 channel (60 Minutes 198 7; Consumer Reports 1991). The above review of CMAs is substantially confirms the trade literature, although the information provided by the cooperating bottler goes substantially beyond what has been published to date. What has been learned fiom bottlers suggests why the alternating promotion and market access issues are linked in the trade literature. First, retailers select products for promotion based on opportunity costs. This shifts promotion time slots to those bottlers willing to pay the most for them. Second, because customers respond more strongly to Coca-Cola and Pepsi on promotion than to other (weaker) brands, there is an opportunity- loss involved for retailers in promoting weaker brands. Weaker brands do not promote because they cannot bid with the brands that receive strong consumer responses to promotions. The access of weaker soft drink brands to promotions is effectively constrained by customer preferences and scale econonnies. This review of CMAs from the leading soft drink companies has shed light on how these ageements, by bundling merchandising elements together, can act as a coordination mecharnism that prevents the unraveling of soft drink merchandising into price competition. This adds to previous work (Lal 1990b; Pasztor and Reibstein 1987) which hypothesized that retailers somehow coordinate promotions but did not specify the mecharnisrns. The presence of CMAs does not mean that prices, displays, and advertisements will be constant. Rivalry with retailers and other bottlers periodically alters the details of pricing and merchandising - the 65 percent per year growth in "ad payments” indicates that the power of retailers is gowing, and during the plant visit the managers at the bottler penciled price changes on the current CMA. Instead, CMAs structure the relationship between marketing mix elements for each bottler as well as promotion intensities between competitors. At least in Michigarn, bottlers are the efi‘ective competitive actors on behalf 174 of the national brands, competing as much with gocery chains for margins as with other bottlers for promotion time slots. First, CMAs allow high, medium, and low levels of advertising, discounting, and display to overlap. Second, the exclusivity clauses in CMAs keep high-intensity promotions separated in time which may prevent escalation that could degenerate into price competition. Abraham Agesti, Allenby Aukem Avery, Axelro Bass, ] BatSel Bawa Beck VII Bibliography Abraham, Magid M ., and Leonard M. Lodish (1990), "Getting the Most out of Advertising and Promotion," Harvard Business Review, 68, No. 3, pp. 50-58. (1987a), "Promoter: An Automated Promotion Evaluation System," Marketing Science, 6, No. 2, pp. 101-123. (1987b), Reply to "Steps Toward Mastering Trade Promotions,” by Michael W. Kruger, Marketing Science, 6, No. 2,.pp. 147-9. Agesti, Alan (1990), Categorical Data Analysis, New York: John wiley & Sons. Allenby, Greg M. (1989), "A Unified Approach to Identifying, Estimating, and Testing Demand Structures with Aggregate Scanner Data," Marketing Science, 8, No. 3,.pp. 265-80. Aukeman, Dean (1991), Division Marketing Director, Brooks Beverage Management, Inc., personal interview. Avery, Robert B. (1980), ”Comments on 'A Theory for Pricing Nonfeatured Products in Supermarkets," Journal of Business - Special Issue, 53, No. 3, pt. 2, pp. 8211-8213. Axelrod, Robert (1984), The Evolution of Cooperation, New York: Basic Books. Bass, Frank M., M. Givon, M. Kalwarni, D. Reibstein, and G. P. Wright (1984), ”An Investigation into the Order of the brand Choice Process,” Marketing Science, 3, No. 4, (Fall), pp. 267-87. Batsell, Richard R., and John C. Polking (1985), ”A New Class of Market Share Models, " Marketing Science, 4, No. 2, pp. 177-98. Bawa, Kapil, Jane T. Landwehr and Aradhna Krishna (1989), ”Consumer Response to Retailer’s Marketing Environments: An Analysis of Coffee Purchase Data,” Journal of Retailing, 65, No. 4, pp. 471-95. Bawa, Kapil and Robert W. Shoemaker (1989), ”Analyzing Incremental Sales fi'om a Direct Mail Coupon Promotion," Journal of Marketing, 53, No. 2, pp. 66-78. Becker, Gary S. (1965), ”A Theory of the Allocation of Time,” Economic Journal, 65, No. 299, pp. 493-517. 175 Bender. J . pe Best. ROE C BishOp, ‘. A. Blattberg F 7 r Blattber l Blattber Blattbe: Blattbe Bolton BOX. ( BTOOk Bucklj 176 Bender, J. Dennis (1991), Director of Marketing Analytical Development, A. C. Nielsen, personal conversation. Best, Roger J. (1976), "The Predictive Aspects of a Joint-Space Theory of Stochastic Change," Journal of Marketing Research, 13, (May), pp. 198-204. Bishop, Yvonne M., Stephen E. F ienberg, and Paul W. Holland (1975), Discrete Multivariate Analysis: Mary and Practice, Cambridge: MIT Press. Blattberg, Robert C ., and Edward I. George (1990), ”Shrinkage Estimation of Price and Promotional Elasticities: Seemingly Unrelated Equations,” Technical Report No. 79, Graduate School of Business, University of Chigago, (July 1989, March revision). Blattberg, Robert C., and Scott A. Neslin (1990), Sales Promotions: Concepts, Methods, and Strategies, Englewood Cliffs, NJ: Prentice-Hall. Blattberg, Robert C., and Kenneth J. Wisniewski (1989), ”Price-Induced Patterns of Competition," Marketing Science, 8 N4, pp. 291-309. Blattberg, Robert C., and Alan Levin (1987a), ”Modelling the Effectiveness and Profitability of Trade Promotions," Marketing Science, 6, No. 2, pp. 124-46. Blattberg, Robert C. (1987b), Reply to ”Steps Toward Mastering Trade Promotions, " by Michael W. Kruger, Mm'keting Science, 6, No. 2, pp. 147-49. Gary D. Eppen, and Joshua Lieberman (1981), "A Theoretical and Empirical Evaluation of Price Deals for Consumer Nondurables," Journal of Marketing, 45, (Wmter), pp. 116-29. Bolton, Ruth (1989), ”Robustness of Retail Level Price Elasticity Estimates,” Journal of Retailing, 65, No. 2, pp. 193-219. Box, George E. P. (1984), "The Importance of Practice in the Development of Statistics," T echnometrics, 26, No. 1, (February), pp. 1-8. , (1976), ”Science and Statistics,” Journal of the American Staistical Association, 71, No. 356, (December), pp. 791-98. Brooks Beverage Management, Inc. (1991), unpublished Calendar Marketing Agreements for bottlers of Coke, Pepsi, and Brooks Beverage products. Bucklin, Louis P. ( 1973), ”A Theory of Channel Control,” Journal of Marketing, 37, (January), pp. 39-47. 177 Bucklin, Randolph E., and James M. Lattin (1991), "A Model of Product Category Competition Among Grocery Retailers, " Journal of Retailing, forthcoming (March 1991 working paper). Butler, Ben (1991), President, The Hendry Corporation, personal conversation. Buzzell, Robert D., John A. Quelch, and Walter J. Salmon (1990), ”The Costly Bargain of Trade Promotion," Harvard Business Review, 68, No. 2, pp. 141—49. Calantone, Roger J., Cornelia Droge, David S. Litvack, and C. Anthony Di Benedetto (1989), "Flanking in a Price War," Interfaces, 19, No. 2, pp. 1-12. Campbell, Donald T., and Donald W. F iske (1959), ”Convergent and Discriminant Validation by the Multitrait-Multimethod Matrix,” Psychological Bulletin, 56, No. 2, (March), pp. 81-105. Reprinted in Methodology and Epistemology for Social Science: Selected Papers of Donald T. Campbell (1988), E. Samuel Overrnan, ed. Chicago: University of Chicago Press. Campbell, Donald T., and Julian C. Stanley (1963), Experimental and Quasi- Erperimental Designs for Research, Boston: Houghton Mifilin. Cannon, Joseph P., and Paul N. Bloom (1991), "Are Slotting Allowances Legal Under the Antitrust Laws?" Journal of Public Policy and Marketing, 10, No. 1, (Spring), 167-86. Carpenter, Gregory S., Lee G. Cooper, Dominique M. Hanssens, and David F. Midgley (1988), ”Modeling Asymmetric Competition,” Marketing Science, 7, No. 4, pp. 393-412. Chambers, John M., William S. Cleveland, Beat Kleiner, and Paul A Tukey (1983), Grapical Methods for Data Analysis, Boston: Duxbury Press. Chevalier, Michel, and Ronald C. Curhan (1976), ”Retail Promotions as a Function of Trade Promotions: A Descriptive Analysis,” Sloan Management Review, pp. 19-32. Cochran, Mlliam G. (1952), "The x2 Test of Goodness of Fit,” The Annals of Mathematical Statistics, ed. H. C. Carver and S. S. Wilks, 23, Baltimore: Waverly Press, pp. 3 15-45. Consumer Reports (1991), "Who's Got the Right One?" 56, No. 8, pp. 518-24. , (1988), ”How to Save $2,500 a Year in the Supermarket,” 53, No. 3, pp. 153-63. Cook “10‘ Ana Cooper. LE Ela C ox, Reav Ric Dean Joe‘: Mr PP Deaton. [- Sr Dickson, S Dillon, V1 A Dixit, At \ Dolan, I 1 Doyle, 1 WE, : Esk‘m’ ‘ 178 Cook, Thomas D., and Donald T. Campbell (1979), Quasi-Experimentation: Design & Analysis Issues for Field Settings, Boston: Houghton Mifilin. Cooper, Lee G. (1988), "Competitive Maps: The Structure Underlying Asymmetric Cross Elasticities," Management Science, 34, No. 6, (June), pp. 707-23. Cox, Reavis (1964), "Introduction," Theory in Marketing: Second Series, Homewood: Richard D. Irwin . Dean, Joel (1970), "Techniques for Pricing New Producs and Services," Handbook of Modern Marketing, Victor P. Buell and Carl Heyel, eds. New York: McGraw-Hill, pp. 51-61. Deaton, Angus ( 1987), "Estimation of Own- and Cross-Price Elasticities from Household Survey Data," Journal of Econometrics, 36, No. 1-2, pp. 7-30. Dickson, Peter R., and Alan G. Sawyer (1990), "The Price Knowledge and Search of Supermarket Shoppers," Journal of Marketing, 54, (July), pp. 42-53. Dillon, William R, and Mattew Goldstein (1984), Multivariate Analysis: Methods and Applications, New York: John Wiley. Dixit, Avinash K., and Barry J. Nalebufl‘ ( 1991), Thinking Strategically, New York. W. W. Norton & company. Dolan, Robert J. (1987), "Quantity Discounts: Managerial Issues and Research Opportunities," Marketing Science, 6, No. 1, pp. 1-22. (1981), "Models of Competition: A Review of Theory and Empirical Evidence,” Review of Marketing, Ben M. Enis and Kenneth J. Roering, eds. Chicago: American Marketing Association, pp. 224-34. Doyle, Peter, and John Saunders (1985), "The Lead Efl‘ect of Marketing Decisions," Journal of Marketing Research, 22, No. 1, (February), pp. 54-65. Duffy, Mike (1991), Marketing Manager, A. C. Nielsen, personal conversation. Eskin, Gerald J. (1988), "Single Sorce Data: the US Experience, " Journal of the Market Resewch Society, 31, No. 4, pp. 501-07. (1987), "Applications of Electronic Single-Source Measurement Systems," European Research, 15, No. 1, pp. 12-20. (1985), "Tracking Advertising and Promotion Performance with Single-Source Data," Journal of Advertising Research, 25, No. 1, pp. 31, 33-39. Fader, Pi Felgner, Gage, T Galvin, Goefing Guadag Han. R Harms Hardy Haust Haw! 179 Fader, Peter S., and John R. Hauser (1988), "Implicit Coalitions in a Generalized Prisoner's Dilemma," Journal of Conflict Resolution, 32, No. 3, (September), 553-82. , and Leonard M. Lodish (1990), "A Cross-Category Analysis of Category Structure and Promotional Activity for Grocery Products, " Journal of Marketing, 54, (October), pp. 52-65. , and Leigh McAlister (1989), "A Promotion-Induced Choice Restriction Model of Consumer Choice," Wharton School of Economics (University of Pennsylvania), Marketing Department Working paper 88-104 (February revision). Farris, Paul W., and Robert D. Buzzell ( 1979), "Why Advertising and Promotional Costs Vary: Some Cross-Sectional Analyses," Journal of Marketing, 43 (Fall), pp. 112-22. Felgner, Brent (1988), "Consumer Expenditures Study: Soft Drinks," Supermarket Business, 43, No. 9, pp. 187-213. Gage, Theodore J. (1983), "IRI Develops New Systems for More Exposure," Advertising Age, 54, No. 8, pp. M29-M30. Galvin, Andrew (1990), "The Price of Fixing Prices," Beverage World, 109, No. 1457, pp. 24-34. ‘ Goering, Tom (1992), Account Manager, A. C. Nielsen, personal correspondence. Guadagni, Peter M., and John D. C. Little (1983), "A Logit Model of Brand Choice Calibrated on Scanner Data," Marketing Science, 2, No. 3, pp. 203-38. Hall, Robert E., Jack Johnson, and David M. Lilien (1990), Micro TSP User's Manual Version 7, Irvine: Quantitative Micro Software. Hannsens, Dominique M., Leonard J. Parsons, and Randall L. Schultz (1990), Market Response Models: Econometric and Time Series Analysis, Boston: Kluwer. Hardy, Kenneth G. (1986), "Key Success Factors for Manufacturer's Sales Promotions in Package Goods," Journal of Marketing, 50, (July), pp. 13-23. Hauser, John R, and Steven M. Shugan (1983), "Defensive Marketing Strategies," Marketing Science, 2, No. 4, (Fall), pp. 319-60. Hawkins, E. R. (1950), "Vertical Price Relationships," Theory in Marketing: Selected Essays, Reavis Cox and Wroe Alderson, eds. Homewood: Richard D. Irwin. 180 Hee, Judith Ann (1981), "Determining Manufacturer's Couponing Strategies," M.A., Sloan School of Management, Massachusetts Institute of Technology. Helmer, Richard M., and Johny K. Johansson (1977), "An Exposition of the Box-Jenkins Transfer Function Analysis with an Application to the Advertising-Sales Relationship," Journal of Marketing Research, 14, No. 2 (May), pp. 227-39. Hoaglin, David C., Frederick Mosteller, and John W. Tukey (1983), Understanding Robust and Exploratory Data Analysis, New York: John Wiley. Hoyt, William H., Roger J. Calantone, and C. Anthony di Benedetto (1990), "Price Dealing for Consumer Non-Durables with Price Discrimination and Heterogeneous Inventorying Costs, " working paper, (September revision). Hoyer, Wayne D. (1984), "An Examination of Consumer Decision Making for a Common Repeat Purchase Product," Journal of Consumer Research, 11, (December), pp. 822-29. Jenkins, Gwilym M. (1979), Practical Experiences with Modelling and Forecasting Time Series, Jersey, Channel Islands, UK: Jenkins and Partners. Jeuland, Abel P., and Chakravarthi Narasimhan (1985), "Dealing - Temporary Price Cuts - by Seller as a Buyer Discrimination Mechanism," Journal of Business, 58, No. 3, pp. 295-308. Jones, J. Morgan, and Jane T. Landwehr (1988), "Removing Heterogeneity Bias for Logit Model Estimation,” Mw'keting Science, 7, No. 1 Winter), pp. 41-59. and Fred S. Zufryden (1982), "An Approach for Assessing Demographic and Price Influences on Brand Purchase Behavior," Journal of Marketing, 46, No. 1 (\Vrnter), pp. 36-46. Kahn, Barbara E., Manohar U. Kalwani, and Donald G. Morrison (1988), "Niching Versus Change-of-Pace Brands: Using Purchase Frequency and Penetration Rates to Infer Brand Positionings," Journal of Marketing Research, 25, (November), pp. 384-90. Kalwani, Manohar U., and Donald G. Morrison (1977), "A Parsimonious Description of the Hendry System," Management Science, 23, No. 5, pp. 467-77. Katona, George (1953), "Rational Behavior and Economic Behavior," Psychological Review, 60, No. 5, pp. 307-18. Reprinted in Jagdish N. Sheth and Dennis E. Garret, eds. (1986), Marketing Theory: Classic and Contemporary Readings, Cincinnati: South-Western, pp. 461-75. 181 Kennedy, Peter (1992), A Guide to Econometrics, 3rd ed, Mass: MIT Press. (1985), A Guide To Econometrics 2nd ed., Mass: MIT Press. Kinberg, Yoram, Ambar G. Rao, and Melvin F. Shakun (1974), "A Mathematical Model for Price Promotions," Management Science, 20, No. 6, pp. 948-59. Klein, Robert L. (1981), "Using Supermarket Scanner Panels to Measure the Efl‘ectiveness of Coupon Promotions," Marketing: Measurement and Analysis, Proceedings of the Third ORSA/TTMS Special Interest Conference on Market Measurement and Analysis, ed. John W. Keon, pp. 118-24. Providence, RI: The Institute of Management Sciences. Kotler, Philip, Gary Armstrong, and Richard G. Starr, Jr. (1991), Principles of Marketing Instructor '5 Manual, Englewood Clifi‘s, NJ: Prentice-Hall. Krishna, Aradhna, lrnran S. Currim, and Robert W. Shoemaker (1991), "Consumer Perceptions of Promotional Activity,” Journal of Marketing, 55 (April), pp. 4- 16. Kruger, Michael W. (1987), "Commentary: Steps Toward Mastering Trade Promotions," Marketing Science, 6, No. 2, pp. 147-49. Kuehn, Alfed A, and Albert C. Rholofl‘ ( 1967), "Consumer Response to Promotions," in Promotional Decisions Using Mathematical Models, ed. Patrick J. Robinson, Boston: Allyn & Bacon, pp. 45-148. Kumar, V., and Robert P. Leone (1988), "Measuring the Efl‘ect of Retail Store Promotions on Brand and Store Substitution," Journal of Marketing Research, 25, No. 3, pp. 178-85. Lal, Rajiv (1990a), "Price Promotions: Limiting Competitive Encroachment," Marketing Science, 9, No. 3, pp. 247-62. (1990b), "Manufacturer Trade Deals and Retail Price Promotions," Journal of Marketing Research, 27, (November), pp. 428-44. Lattin, James M. (1991), Professor of Marketiing, Stanford University, personal conversation. , and Randolph E. Bucklin (1989), "Reference Efl‘ects of Price and Promotion on Brand Choice Behavior," Journal of Marketing Research, 26, (August), No. 3, pp. 299-3 10. 182 Lattin, James M., and Leigh McAlister (1985), "Using a Variety-Seeking Model to Identify Substitute and Complementary Relationships Among Competing Products," Journal of Marketing Research, 22, (August), pp. 330-39. Lazear, Edward P. (1986), "Retail Pricing and Clearance Sales, " American Economic Review, 76, No. 1, pp. 14-32. Learner, Edward E. (1983), "Let's Take the Con out of Econometrics," American Economic Review, 73, No. 1, (March), pp. 31-43. Leone, Robert P. (1987), "Forecasting the Effect of an Environmental Change on Market Performance: An Intervention Time-Series Approach," International Journal of Forecasting, 3, No. 3-4, pp. 463-78. Lilien, David M. (1992), author of MicroTSP, personal conversation. Lilien, Gary L., Philip Kotler, and K. Sridhar Moorthy (1992), Marketing Models, Englewood Cliffs, N. J: Prentice-Hall. Lilien, Gary L., and Philip Kotler, (1983), Marketing Decision Mala'ng: A Model-Building Approach, New York: Harper & Row. Little, John D. C., and Leonard M. Lodish (1981), "Commentary on: 'Judgment Based Marketing Decision Models,” Journal of Marketing, 45, No. 3, pp. 24-49. Little, John D. C., and Jeremy F. Shapiro (1980), ”A Theory for Pricing Nonfeatured Products in Supermarkets,” Journal of Business, 53, No. 3, pt. 2, pp. sl99-s209. Little, John D. C. (1966), "A Model of Adaptive Control of Promotional Spending," Operations Research, 14, No. 6, pp. 1075-97. Litvack, David S. Roger J. Calantone, and Paul R. Warshaw (1985), "An Examination of Short-Term Retail Grocery Price Effects," Journal of Retailing, 61, No. 3, pp. 9-25. Lodish, Leonard M. (1986a), The Advertising and Promotion Challenge: Vaugely Right or Precisely Wrong?, New York: Oxford University Press. and David J. Reibstein (1986b), "New Gold Mines and Minefields in Market Research," Harvard Business Review, 64, No. 1, pp. 168-82. Mahajan, Vijay, Stuart 1. Bretschneider and John W. Bradford (1980), "Feedback Approaches to Modeling Structural Shifts in Market Response,” Journal of Marketing, 44, (Wmter), pp. 71-80. 183 Makridakis, Spyros and Steven C. Wheelwright (1989), Forecasting Methods for Management, 5th ed. New York: John Wiley and Sons. Marketing News (1981), 15, No. 6, (Section 2), p. 20. Martin, Stephen (1988), Industrial Economics: Economic Analysis and Public Policy, New York: Macmillan Publishing. McAlister, Leigh (1985), "The Impact of Price Promotions on a Brand's Market Share, Sales Pattern and Profitability,” Working Paper # 1622-85, Sloan School of Management, Massachusetts Institute of Technology, (February). Meyer, Robert J ., and J 0210 Assuncao (1990), "The Optimality of Consumer Stockpiling Strategies," Marketing Science, 9, No. 1, (Winter), pp. 18-41. Monroe, Kent B. (1990), Pricing: Making Profitable Decisions, 2nd ed., New York: McGraw-Hill. Moorthy, K. Sridhar (1985), ”Using Game Theory to Model Competition," Journal of Marketing Research, 22, (August), pp. 262-82. Moriarty, Mark M. (1985), "Retail Promotional Efl‘ects on Intra- and Interbrand Sales Performance," Journal of Retailing, 61, No. 3, pp. 27-47. Mosteller, Frederick, and John W. Tukey (1977), Data Analysis and Regression: A Second Course in Statistics, Reading: Addison-Wesley. Nagle, Thomas s. (1987), The Strategy and Tactics ofPricing: A Guide to Profitable Decision Making, Englewood Clifl‘s, NJ: Prentice-Hall. Narasimhan, Chakravarthi (1988), "Competitive Promotional Strategies," Journal of Business, 61 N4, pp. 427-49. (1984), "A Price Discrimination Theory of Coupons,” Marketing Science, 3, No. 2, pp. 128-47. Neslin, Scott A (1990), "A Market Response Model for Coupon Promotions," Marketing Science, 9 N2, pp. 125-45. Caroline Henderson and John Quelch (1985), "Consumer Promotions and the Acceleration of Product Purchases," Marketing Science, 4, No. 2, pp. 147-65. Neslin, Scott A,and Robert W. Shoemaker (1989), "An Alternative Explanation for Lower Repeat Rates After Promotion Purchases," Journal of Marketing Research, 26, (May), pp. 205-13. 184 (1983), "A Model for Evaluating the Profitability of Coupon Promotions," Marketing Science, 2, No. 4, pp. 361-88. Nielsen, A. C. (1992), "Nielsen Highlights - Category Performance, Total US. - Supermarkets $2 Million and Over, 52-Weeks Thru [sic] lst Quarter 1991," Nielsen Business Information Series, (August). Oliver, Thomas (1986), The Real Coke, The Real Story, New York: Penguin Books. Pasztor, Andy, and Larry Reibstein (1987), "Cola Sellers May Have Bottled Up Their Competitors," Wall Street Journal, (December 9), p. 6. Pedrick, James H., and Fred S. Zufryden (1991), "Evaluating the Impact of Advertising Media Plans: A Model of Consumer Purchase Dynamics Using Single-Source Data," Marketing Science, 10, No. 2, pp. 111-30. Petty, Ross D. (1989), "Predatory Promotion: A Theory of Antitrust Liability Whose Time Has Come?," American Business Law Journal, 27, (Summer), pp. 215-49. Platt, John R. (1964), "Strong Inference,” Science, 146, No. 3642, pp. 347-53. Quandt, Richard E. (1964), "Estimating the Effectiveness of Advertising: Some Pitfalls in Econometric Methods, " Journal of Marketing Research, 1, No. 2, (May), pp. 51-60. - Quelch, John A., Scott A. Neslin, and Lois B. Olson (1987), "Opportunities and Risks of Durable Goods Promotion," Sloan Management Review, 27, No. 2, pp. 27-38. Raju, Jagmohan S., V. Srinivasan, and Rajiv Lal (1990), "The Efl‘ects of Brand Loyalty on Competitive Price Promotional Strategies," Management Science, 36, No. 3, pp. 276-304. Rao, Ram C. (1991), "Pricing and Promotions in Asymmetric Duopolies," Marketing Science, 10, No. 2, pp. 131-44. Rao, Vithala R., and L. Joseph Thomas (1973 ), "Dynamic Models for Sales Promotion Policies," Operational Research Quarterly, 24, No. 3, pp. 403-17. Rasmusen, Eric (1989), Games and Information: An Introduction to Game Theory, New York: Basil Blackwell. Rinne, Heikki, and Michael Geurts (1988), "A forecasting model to evaluate the profitability of price promotions," European Journal of Operational Research, 33, No. 3, pp. 279-89. 185 Rohm, Wendy Goldman (1986), "A Capital Idea," Infosystems, 33, No. 11, (November), pp. 48-52. . Ruback, Richard S., and Martin B. Zimmerman (1984), ”Unionization and Profitability: Evidence from the Capital Market," Journal of Political Economy, 92, No. 6, pp. 1134-157. Salop, Steven, and Joseph Stiglitz (1982), "The Theory of Sales: A Simple Model of Equilibrium Price Dispersion with Identical Agents," American Economic Review, 72, No. 5, pp. 1121-130. (1977), "Bargains and Ripoffs: A Model of Monopolisticlly Competitive Price Dispersion," Review of Economic Studies, 54, No. 135, pp. 493-510. Sanford, Thomas E. (1989), "Better Tools for Marketing and Sales," Nielsen Researcher, 2, No. 1, pp. 2-7. Scitovsky, Tibor (1990), "The Benefits of Asymmetric Markets," Journal of Economic Perspectives, 4, No. 1, pp. 135-48. Shubik, M. (1959), Strategy and Market Structure, New York: Wiley. 60 Minutes (1987), transcript, 20, No. 6, (October 25), pp. 6-10. Slavic, Frank (1991), Vice President, Custom Delivery Services, A. C. Nielsen, personal conversation. (1988b), ”Chapter 4: Baseline Definition," SCAN *PRO MONITOR Manual, Nielsen Marketing Research. (1988a), "Baseline Methodology," SCAN *PRO, Nielsen Marketing Research. Sobel, Joel (1984), "The Timing of Sales," Review of Economic Studies, 41, pp. 353-68. Solrnan, Paul, and Thomas Friedman (1982), Life and Death on the Corporate Battlefield, New York: Simon and Schuster. Stern, Louis W., and Adel I. El-Ansary (1988), Marketing Channels, 3rd ed., Englewood Clifi‘s, N.J.: Prentice-Hall. (1982), Marketing Channels, 2nd ed., Englewood Clifl‘s, N.J.: Prentice-Hall. Stigler, George J. (1987), The Theory of Price, 4th ed., New York: MacMillan Publishing. (1968), The Organization of Industry, Chicago: University of Chicago Press. 186 Struse, Rudolph W. III (1987), "Approaches to Promotion Evaluation: A Practitioner's Viewpoint," Marketing Science, 6, No. 2, pp. 150-51. Tedlow, Richard S. (1990), New and Improved: The Story of Mass Marketing in America, New York: Basic Books. Totten, John C. (1991), Vice President, A. C. Nielsen, personal conversation. and Martin P. Block (1987), Analyzing Sales Promotion: Text and Cases, Chicago: Commerce Communications Inc. Urban, Glen L., and Steven H. Star (1991), Advanced Marketing Strateg, Englewood Clifl‘s, N.J.: Prentice-Hall. Varian, Hal R. (1980), "A Model of Sales," American Economic Review, 70, No. 4, pp. 651-59. Walters, Rockney G. (1991), "Assessing the Impact of Retail Price Promotions on Product Substitution, Complementary Purchase, and Interstore Sales Displacement," Journal of Marketing, 55, (April), pp. 17-28. (1989), "An Empirical Investigation into Retailer Response to Manufacturer Trade Promotions," Journal of Retailing, 65, No. 2, pp. 253-72. and Scott B. MacKenzie (1988), "A Structural Equations Analysis of the Impact of Price Promotions on Store Performance," Journal of Marketing Research, 25, (February), pp. 51-63. Wensley, Robin (1981), "Strategic Marketing: Betas, Boxes, or Basics," Journal of Marketing, 45, (Summer), pp. 173-82. Wichem, Dean W., and Richard H. Jones (1977), "Assessing the Impact of Market Disturbances Using Intervention Analysis," Management Science, 24, No. 3, (November), pp. 329-37. Wildt, A. R. (1974), "Multifirrn Analysis of Competitive Decision Variables," Journal of Marketing Research, 40, (February), pp. 50-62. Williams, Monci Jo (1983), "The No-Win Game of Price Promotion," Fortune, July 1 1, pp. 93-102. Wilkinson, Leland (1990), SYS TA T: The System for Statistics, Evanston, IL: SYSTAT, Inc. 187 Wilson, R. Dale, Larry M. Newman, and Manoj Hastak (1979), "On the Validity of Research Methods in Consumer Dealing Activity: An Analysis of Timing Issues," in Educators' Conference Proceedings, eds. Neil Beckwith, Michael Houston, Robert Mittelstaedt, Kent B. Monroe, and Scott Ward, pp. 41-46. Chicago: American Marketing Association. Wind, Yoram, and Vijay Mahajan (1981), "Market Share: Concepts, Findings and Directions for Future Research, " in Review of Marketing, Ben Ennis and Kenneth Roering, eds. Chicago: American Marketing Associaton, pp. 41-6. "‘llllllllllllllllES