Double Jeopardy

An executive summary for managers and executive readers can be found at the end of this issue

Comparing double jeopardy effects at the behavioral and attitudinal levels

Introduction

Subir Bandyopadhyay and Kunal Gupta

The authors Subir Bandyopadhyay is an Associate Professor of Marketing at the Indiana University Northwest, School of Business and Economics, Gary, Indiana, USA. Kunal Gupta is a Senior Consultant at Burke, Inc., Cincinnati, Ohio, USA.

Keywords Brands, Brand identity, Brand image

Abstract The marketing phenomenon known as the double jeopardy (DJ) effect has continued to intrigue marketing scholars and practitioners over the last four decades. It is often found that, vis-a`-vis the more popular brands, the less popular brands not only attract fewer customers but these customers buy these brands less frequently. The term “double jeopardy” is used to express this twin disadvantage faced by the less popular brands. Marketing researchers have shown that the DJ effect extends to many product categories (e.g. toothpaste or coffee), media (e.g. radio and television), and distribution channels (e.g. individual stores). Attitudinal measures are developed for both brand penetration and its frequency of use: two key elements used to measure the DJ effect. It is also empirically demonstrated, using attitudinal and behavioral data supplied by a large multinational company, how attitudinal measures unravel strengths and vulnerabilities of individual brands and how these insights can help managers in accurate brand positioning.

Electronic access The Emerald Research Register for this journal is available at www.emeraldinsight.com/researchregister The current issue and full text archive of this journal is available at www.emeraldinsight.com/1061-0421.htm

Journal of Product & Brand Management Volume 13 · Number 3 · 2004 · pp. 180-191 q Emerald Group Publishing Limited · ISSN 1061-0421 DOI 10.1108/10610420410538078

Brands with a small market share tend to suffer from lower levels of repeat purchase behavior than do high-market-share brands. This well-known phenomenon in competitive markets was termed “double jeopardy” by McPhee (1963) to highlight the plight of less popular comic strips. McPhee (1963) showed that, vis-a`-vis the more popular comic strips, the less popular ones were not only read by fewer people, but were also read less frequently by this smaller number of readers. The term “double jeopardy” was used to express this twin disadvantage faced by the less popular comic strips. In recent years, several researchers (e.g. Ehrenberg, 2002; Dall’Olmo Riley et al., 1997; Chaudhuri, 1995; Donthu, 1994; Ehrenberg et al., 1990; Fader and Schmittlein, 1993; Kahn et al., 1988) have contributed to DJ literature. Their studies have shown that the double jeopardy phenomenon extends to many branded packaged goods (toothpaste, coffee, etc.), the media (radio and TV programs), and distribution networks (e.g. individual stores) across various geographical markets. (See Ehrenberg and Goodhardt (1979) and Ehrenberg (1987) for more details on research findings.) The pioneering work in the field of double jeopardy (see, for example, Ehrenberg et al., 1990) has explained DJ as a simple statistical phenomenon, related to the size structure of the market. It has been empirically shown that all things being equal, the larger brands of a product category are bought more frequently vis-a`-vis the smaller brands of the same category. Ehrenberg et al. (1990) also argue that marketing inputs, e.g. price, promotion, advertising, distribution, and product formulation do not influence DJ directly, but influence the sales and the market share of the brand. Thereafter, it is the market shares of the various brands that determine these brands’ purchase frequency. One important characteristic of the research is that barring a few exceptions (see Barwise and Ehrenberg, 1985, 1987, 1988), most researchers have treated the DJ effect as a purely behavioral phenomenon. Even though the phenomenon of double jeopardy has been extensively observed and We thank Procter & Gamble of Canada for supplying the data used in this study. We also thank Soumita Banerjee for background research. The first author gratefully acknowledges the financial support of the Indiana University Northwest grant-in-aid program, Research and University Graduate School of Indiana University, Procter & Gamble, Canada, and the Social Sciences and Humanities Research Council of Canada Grant No. 410-95-0732.

180

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

studied from a behavioral perspective, it is surprising that the phenomenon has been hardly understood at the attitudinal level. Ehrenberg et al. (1990) alluded to attitudinal double jeopardy when they explained that because a small brand has a higher proportion of infrequent buyers, it receives fewer average positive attitudinal responses from its customers than a more popular brand receives from its more frequent customers. Their argument highlights a distinct advantage of measuring double jeopardy at the attitudinal level since attitude scores provide a valuable insight into how and which attributes influence behavior. In spite of the obvious advantages of attitudinal measures, few researchers have ventured to develop reliable measures of both penetration and frequency. In addition, a smaller number of studies (see, for example, Barwise and Ehrenberg, 1985, 1987, 1988) have indeed used a combination of attitudinal frequency measure and a behavioral penetration measure, few researchers have studied the double jeopardy effect using purely attitudinal measures of both penetration and frequency. Our research addresses this important issue in double jeopardy research. We will develop attitudinal measures of both penetration and frequency, and compare the DJ effects at the behavioral and attitudinal level. We will also identify a set of attributes (Barwise and Ehrenberg (1985) refer to these attributes as “descriptive factors”) that influences the attitudinal measures of penetration and frequency. This knowledge will help brand managers to identify the strengths and vulnerabilities of their brands, and thus orchestrate effective marketing and communication strategies. Such managerial insight is not readily available using a purely behavioral approach. The behavioral DJ approach may have other limitations as well. For example, Hofmeyr and Bennett (1994) argue that brand loyalty could not be equated solely with purchase propensity, since such a model cannot anticipate changes that should be measured to study the psychological shifts in attitude that precede behavioral shifts in brand purchases. They suggest a study of psychological factors that bind consumers to brands, instead of the hitherto conventional approach of looking only at brand size and behavioral loyalty. They develop a “conversion model” to measure brand commitment. This model divides the users of a brand into categories ranging from very strongly committed users, to convertible users, to non-users, in terms of how available these users are for conversion to the brand. The number of committed versus available users defines the psychological strength of a brand, and therefore the possibility of the brand defying the DJ effect. Besides these few studies, there are no other notable publications that have studied DJ at the attitudinal level, and this is where our research contributes useful information.

Our study aims to compare the DJ effect at the attitudinal (i.e. psychological) and the behavioral level for the same respondent for a given product category. By measuring DJ at these two levels, we try to study the shift in DJ patterns. Since marketing inputs such as product availability, pricing, advertising support, and promotion activities might influence the intended repeat purchase behavior, we expect more deviations from the double jeopardy pattern at the behavioral level as compared to the attitudinal level[1]. Studying individual brand level attitudinal performances offers several advantages. One such advantage is that it can shed more light on the possible reasons for the high usage rate of dominant brands. For example, if a dominant toothpaste brand such as Crest Original scores well on functional attributes (e.g. “cleans teeth well”, “leaves mouth fresh”, “removes stains” etc.), we have a better appreciation of why the brand has a high usage rate. Conversely, if a smaller toothpaste brand such as Macleans scores poorly on “valueoriented” attributes (e.g. “good value for money”, “has a low price” etc.), it gives a signal to its brand manager as to why the brand is not doing well. We will demonstrate this unique advantage of the attitudinal measures of DJ using the data for the toothpaste category. In addition, we will identify four factors or categories of attributes – functional, disease/health, quality and value – for all toothpaste brands. Then, we will compare the average factor scores for all brands to unravel the relative “strengths” and “vulnerabilities” for each brand. Also, we will discuss how this insight can help a brand manager to design an effective marketing strategy that builds on its strengths and guards against vulnerabilities. The rest of the paper is organized as follows. First, we develop attitudinal measures of the DJ effect. Specifically, we formalize the measures of attitudinal penetration and frequency, drawing on the literature on mere exposure effect and consumer memory. Next, we outline the measures of behavioral penetration and behavioral frequency. Subsequently, we outline the hypothesis developed, and the methodology for hypothesis testing. Then, we describe the data used for empirical testing, present the research findings, and discuss the implications of the results obtained. Finally, we conclude by outlining future areas of research, and the limitations of this study.

Attitudinal measures of the double jeopardy effect We will develop the attitudinal measures of frequency and penetration: two phenomena that together determine the extent of double jeopardy

181

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

(2) Low involvement everyday exposure to the brand or its communication casts doubt on the accuracy of the self-reported frequency scores of the respondents.

effect at this attitudinal level. To develop these measures, we will begin by highlighting current understanding of the exposure effect and its applicability to marketing studies, how the effect translates into (attribute) information storage in memory, and finally, how the stored information can be developed into recognition and frequency measures at the attitudinal level. Mere exposure effect Starting with the pioneering work of Maslow (1937), researchers have long observed that repeated, unreinforced exposure results in an increased positive attitude towards a stimulus. Although these early studies broached a conceptually novel idea, they often lacked the methodological rigor to establish unequivocal support for their results. However, through a more careful study, Zajonc (1968) demonstrated the “mere exposure effect” as the enhancement of the attitude of an individual towards a stimulus through mere repeated exposures. Zajonc (1968) defined “mere exposure” as a condition that makes the stimulus available to an individual’s perception. A typical exposure effect study was carried out in a laboratory setting, where the subject was exposed to neutral stimuli such as foreign words or ideographs (see, for example, Zajonc et al., 1972; Stang and O’Connell, 1974). The advantage of selecting neutral stimuli in all these studies has been the assurance of the absence of any preexisting favorable or unfavorable attitudes. Further, the laboratory settings have made possible accurate monitoring and control over the exposure and the frequency of exposure. This precision has enabled researchers to confidently support their hypothesis that a higher frequency of exposure does indeed result in a more positive attitude towards a stimulus. Beyond the laboratories, the relationship between exposure frequency and liking has also been established in naturalistic studies, with stimuli such as the names of public figures and common products. In these studies (see, for example, Bornstein, 1989), the exposure frequency has been estimated through either self-reported scores or an unobtrusive index such as the Thorndike and Lorge (1944) word frequency count. Mere exposure effect in marketing The use of such naturalistic studies is, however, likely to have shortcomings from a marketing perspective, especially for those studying frequently bought product categories. This can occur for two reasons: (1) Unobtrusive indices such as those used by Thorndike and Lorge (1944) are unavailable to give word frequency counts for brands of such product categories.

Thus, if a researcher were to ask a respondent about how many times he or she has been exposed to, for example, Pantene shampoo or its communication over the last month, the respondent has a high chance of erring while cumulating his or her personal (e.g. usage) and other passive (e.g. TV advertisement) exposures to the shampoo. On the other hand, the researcher will have more confidence in respondents’ answers to questions if they are familiar with a brand, or if they believe a certain attribute (property) to be associated with a brand. In a typical situation, a researcher might list all major attributes of the product category and all major brands of the same category. The researcher could then ask the subjects to respond to a “yes/no” type question if they know that one of the brands possesses one or more of these attributes. If a consumer highlights any brand-attribute pair, this can be an indication that the respondent recognizes the associated brand. Further, if the consumer confirms, for instance, five affirmations for Brand A, and three for Brand B, then a marketer would be interested in knowing if five versus three affirmations is any indication of Brand A being liked more than Brand B within the sample. In this paper, we explore the development of reliable measures of two important issues: brand recognition and brand liking, especially for lowinvolvement products. Information storage in memory First, it is important for us to understand how the information about the presence or absence of a brand-related attribute is stored in the memory. The issue can be resolved by understanding the debate between the proponents of the “strength hypothesis” and the “multiple-trace hypothesis”. In brief, if a customer believes that, for example, Pantene shampoo has five positive attributes, strength hypothesis suggests that these five would be stored together as one strong trace in the memory of the customer. On the other hand, multiple-trace hypothesis suggests that these attributes would be stored as five individual and different traces in memory. Extensive research (e.g. Hintzman and Block, 1971) has established the superiority of the multiple-trace theory, and researchers now use this theory to explain richer phenomena (Hintzman, 1984, 1986). Another understanding necessary for our research is in the case when, for instance, one of the five positive attributes of Brand A is highlighted multiple times by the respondent.

182

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

Does each exposure result in a separate trace for each encounter, or do multiple encounters strengthen the already existing memory trace of that attribute? Despite both schools of thought, the balance seems to be in favor of the recent work by Shiffrin et al. (1990) and by Murnane and Shiffrin (1991). These researchers demonstrated, using word-repetition exercises, that a repeated stimulus results in the same trace being multiplexed, rather than stored as a redundant trace[2]. Their conclusion seems to be somewhat in line with Petty and Cacioppo (1984) who argued that for lowinvolvement brands, argument number (i.e. the number of memory traces) is more important than argument importance[3].

subconsciously. Thus, it is very likely that for lowinvolvement product categories, frequency information may be used as a decision criterion employed by the customer to reduce cognitive load. That is to say, an average respondent likes Brand A more than Brand B if he or she believes Brand A has a higher number of net positive attributes than Brand B has. To summarize, we support the “multiple-trace” hypothesis that each brand attribute is stored as an individual trace in the memory of the respondent. We also agree with the argument that the more positive attributes the brand is believed to have, the greater the number of memory traces of the brand in the consumer’s memory is. Brand familiarity (we also call it attitudinal penetration) entails coming in contact with any such attribute, and for low-involvement product categories, brand liking (or attitudinal frequency) can be equated with the number of positive attributes a respondent believes the brand to have[4]. Thus, we define attitudinal penetration and attitudinal frequency in the following manner: . Attitudinal penetration – the ability to recognize any one memory trace (or an attribute) for a brand that has penetrated the respondent’s memory. . Attitudinal frequency – the number of affirmed memory traces (or attributes) for the same brand[5].

Brand recognition and brand liking In continuation with the multiple-trace hypothesis, how can a marketer assume that a respondent has recognized a certain brand? Is there a threshold number of attribute-brand affirmations, above which the researcher can say with confidence that the particular respondent recognizes the brand? An answer to this question has been provided in earlier works of Anderson and Bower (1972) and Hintzman and Block (1971). Both studies demonstrated that the recognition of an item entails contact with any one trace among the multiple traces that remained in the consumer’s memory. In line with these studies, we would thus use the affirmation of any one attribute-brand pairing as the necessary condition for brand recognition (or attitudinal penetration from the perspective of the double jeopardy phenomenon). For a low-involvement product category, brand liking has been linked to the concept of attribute frequency (Bettman and Park, 1980; Russo and Dosher, 1983). Petty and Cacioppo (1984) demonstrated that argument number can dominate argument importance if subject involvement is low; thus attribute frequency may be considered a peripheral route to persuasion. This hypothesis was further supported by Alba and Marmorstein (1987) who stated that in absence of the encoding of any other information, the net number of positive attributes towards a brand can influence judgment and choice. They suggested that according to the theory of automation (see Hasher and Zacks, 1979, 1984) frequency information: . acts continually; . cannot be improved by practice; . cannot be inhibited; . does not require conscious awareness; and . drains minimal cognitive resources. These characteristics make the frequency heuristic very different from other consumer decision heuristics as the frequency information may be acquired either with minimum effort or even

We argue that there is a positive relationship between any one memory trace for a brand and between the total number of traces for the same brand. The positive relationship is supported by past research on mere exposure effect, where more frequent exposure has been shown not only to develop effective responses towards the brands, but also to enhance brand image as being more trustworthy and reliable. We also agree with Harris et al. (1980) who review past studies and state that “frequency estimates are subsequent to and dependent on recognition decisions”. Thus, if we were to extend the argument of double jeopardy (McPhee, 1963; Ehrenberg et al., 1990) to the attitudinal level, we would hypothesize that there should be a positive and significant correlation between attitudinal penetration and attitudinal frequency.

Behavioral measures of the double jeopardy effect At the behavioral level, we use “number of purchasers in the last year” and “frequency of purchase in the last year” as measures of penetration and frequency, respectively. Similar behavioral measures are widely used in the DJ

183

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

(1) Pearson correlation coefficients; and (2) the w(1-b) model[6].

literature, for example, in the studies done by Fader and Schmittlein (1993) and Kahn et al. (1988). Table I summarizes the measures of penetration (bx) and frequency (wx) at the attitudinal and at the behavioral level.

Hypothesis We hypothesize that the presence of DJ at the attitudinal level will be stronger than at the behavioral level. The more the number of respondents believe that a given brand has a key attribute, the larger the average number of attributes checked by these respondents for the brand should be. This is in line with McPhee’s (1963) observation that a better-known brand is liked more. Similarly, low awareness about the key attribute of the brand should result in disproportionately lower scores for this brand on various attitudinal attributes. The logic for the hypothesized (high) strength of this attitudinal correlation can be understood by an argument made by Bower (1967). He stated that if, over a span of time, any single trace is successfully recalled, then what is retained each time is some increasing fraction of the components recalled on the previous occasion. Forgotten components (i.e. forgotten positive attributes) are likely to be recalled with a bias that is consistent with the remembered components, and the recalled pattern will change progressively over the recall series. Retention of any unique trace would thereby result in the respondents assigning positive (i.e. present) value to the forgotten information, thus progressively increasing their liking for the brand. Thus, for our product category, the correlation for the brands at the attitudinal level will progressively develop over time in a fashion consistent with that suggested by DJ. This should thus strengthen the attitudinal-level correlation for the observed brands. It is this relationship that may produce higher trial and repeat buying behavior (Rindfleisch and Inman, 1998), and is likely to undergo deviations at the behavioral level because of a gamut of marketing mix, and personal, and situational variables. The result is a weakening of the hypothesized correlation.

Measuring the DJ effect We use two methods to compare the strength of the DJ effect at the attitudinal and behavioral levels:

Larger correlation coefficients between penetration and frequency indicate stronger DJ effects. We also use Ehrenberg’s w(1-b) model to compare the intensity of the double jeopardy effect at the two levels. Several researchers (see, for example, Ehrenberg et al. (1990) and Hofmeyr and Bennett (1994)) have used the w(1-b) model to measure the double jeopardy effect. To be specific, we observe the penetration (bx) and the frequency (wx) for all the brands in a product category at each of the two levels. The mean w(1-b) (or w0) is then calculated for all brands in the category put together. Thereafter, the value of w0/1-bx is computed for each of the brands. Finally, the fit between the observed value of frequency (wx) and the predicted value of wx for a given bx (w0/1-bx) is observed to check if smaller values of “b” also gave smaller values of “w”, thereby exhibiting the double jeopardy effect (Ehrenberg et al., 1990). At the behavioral level, we hypothesize that there are likely to be a lot of shifts between the intended and the actual purchase behavior. These shifts could be caused by one or more of the attitude behavior moderators, such as situational cues (Borgida and Campbell, 1982; Snyder and Kendzierski, 1982), personality factors (Bettman and Park, 1980), manner of attitude formation (Fazio and Zanna, 1981), temporal stability of the attitude (Schwartz, 1978) and the confidence with which the attitude is held (Sample and Warland, 1973). These effects are likely to influence the purchase probabilities of individual brands that should be expected from observance of pure DJ phenomenon at the attitudinal level. The result of these deviations would be a weaker fit between the observed purchase frequency (wx) and the frequency suggested by the DJ effect (w0/1-bx) at the behavioral level compared to that at the attitudinal level. Thus, we test the following hypothesis: H1. The presence of the double jeopardy effect would be stronger at the attitudinal level than at the behavioral level.

Data The analysis has been carried out for two product categories: toothpaste and laundry detergent; the data contain ten and eight popular brands respectively. The survey data were provided by a

Table I Definitions of penetration and frequency Level Penetration (bx)

Frequency (wx)

Attitudinal level Behavioral level

Average number of attitudinal attributes checked Frequency of purchase in the last year

Percentage of respondents checking any attribute Number of purchasers in the last year

184

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

multinational company specializing in packaged consumer goods. The data set contains the responses of 1,096 subjects (for toothpaste) and 1,431 subjects (for laundry detergent) residing in two major North American cities. The survey includes ten behavioral questions to measure the respondents’ usage patterns and their satisfaction towards the various brands, and 30 attitudinal questions to measure the respondents’ beliefs about these brands. For the toothpaste category, these 30 attitudinal questions are broadly categorized under four factors: (1) Functional (e.g. “cleans teeth well”, “leaves mouth fresh”, “removes stains”, etc.). (2) Disease/health (e.g. “reverses tooth decay”, “suitable for children”, “prevents root cavities”, etc.). (3) Quality (e.g. “recommended by the dentist”, “brand I can trust”, “good for teeth that are sensitive to heat/cold”, etc.). (4) Value (e.g. “good value for money”, “has a low price”, etc.). Although the behavioral measures in the data set are “recalled measures” rather than actual behavior, the potential lack of reliability in such measures is minimized because of the use of a freechoice as opposed to a forced-choice questionnaire format. This is in congruence with Barnard and Ehrenberg (1990) who argued that the DJ effect should occur for free-choice data: Double Jeopardy effects occur for the free-choice data but possibly not for the two forced-choice [scaling and ranking] techniques, because forcing a response should or could undermine the statistical selection effect basis for DJ that occurs in freechoice questioning (Barnard and Ehrenberg, 1990, p. 479).

Also, all our attributes are positively framed to circumvent the problem of low response rates for negative attributes (Barnard and Ehrenberg, 1990).

Results and discussion Correlation coefficients We compare the Pearson correlation coefficients between penetration and observed frequency and the predicted and the actual frequency (see Table II for the correlation coefficients). We expect that the correlation under both scenarios will be stronger at the attitudinal level compared to that at the behavioral level. Pearson correlation coefficients between observed penetration and observed frequency at the attitudinal and the behavioral levels are 0.952 and 0.773 respectively, for toothpastes; and 0.939 and 0.918 respectively, for laundry detergents. Both sets of correlation coefficients clearly support

Table II Correlation matrix Observed frequency Attitudinal Behavioral

Toothpaste Predicted frequency Penetration

Laundry detergent Predicted frequency Penetration

Attitudinal Behavioral Attitudinal Behavioral

0.939*

Attitudinal Behavioral Attitudinal Behavioral

0.970**

0.666 0.952* 0.773

0.950 0.939**

Notes: *significant at 1 per cent level; **significant at 10 per cent level

our hypothesis that the DJ effect at the attitudinal level is stronger compared to the behavioral level. The Pearson correlation coefficients between predicted and observed frequency at the attitudinal and the behavioral levels are 0.939 and 0.666 respectively, for toothpastes; and 0.97 and 0.95 respectively, for laundry detergents. These correlation coefficients are statistically different from one another at the 1 per cent significance level for toothpastes but only marginally significant for laundry detergents. Moreover, individually, these correlation coefficients are statistically different from zero. The magnitudes of the correlation coefficients support our hypothesis that the double jeopardy effect at the attitudinal level has less deviation than it does at the behavioral level. Also, the positive sign of the correlation coefficients indicates that the brands with a low penetration level generally have low levels of frequency. These results demonstrate the existence of the double jeopardy effect at both levels. Observed and predicted frequencies We compare the actual frequencies (wx) and the predicted frequencies (w0/1-bx) from the w(1-b) model. The values of wx and w0/1-bx are summarized in Tables III-VI for both measurement approaches and the two product categories. These values can be easily interpreted by considering the case for any one brand. Let us take Colgate Original as an example. Among the 1,096 respondents, about 63 per cent (see Table III) affirmed the presence of at least one positive attribute for the brand Colgate Original. This gives the brand an attitudinal penetration of 0.63. For this level of penetration, the respondents associated Colgate Original with 3.22 positive attributes on average. This is the level of the observed frequency for Colgate Original as defined in Table I. The corresponding frequency derived from the w(1-b) model (i.e. w0/1-bx) is 3.20. This figure is a close approximation of the actual frequency of 3.22 reported by the sample.

185

0.918

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

Table III Penetration and frequencies at the attitudinal level (toothpastes) Brand

Penetration bx

Observed frequency wx

Predicted frequency wo/12bx

Deviation wx-w0/12bx

0.76 0.66 0.63 0.45 0.44 0.36 0.39 0.42 0.42 0.35

4.87 4.30 3.22 2.79 2.20 1.93 1.67 1.41 2.00 1.57

4.88 3.46 3.20 2.15 2.13 1.86 1.93 2.04 2.05 1.83

20.01 +0.84 +0.02 +0.64 +0.07 +0.07 20.26 20.63 20.05 20.26

1. Crest Original 2. Crest Tartar 3. Colgate Original 4. Colgate Tartar 5. Aquafresh Original 6. Aquafresh Tartar 7. Macleans 8. Sensodyne 9. Close Up 10. Arm & Hammer

Table IV Penetration and frequencies at the behavioral level (toothpastes) Brand

Penetration bx

Observed frequency wx

Predicted frequency wo/12bx

Deviation wx-w0/12bx

0.26 0.54 0.18 0.11 0.15 0.06 0.12 0.06 0.11 0.02

3.96 3.99 3.29 3.39 3.26 3.14 3.27 2.71 2.91 2.00

3.54 5.70 3.20 2.95 3.08 2.78 2.96 2.78 2.94 2.68

+0.42 21.71 +0.09 +0.44 +0.18 +0.36 +0.31 20.07 20.03 20.68

1. Crest Original 2. Crest Tartar 3. Colgate Original 4. Colgate Tartar 5. Aquafresh Original 6. Aquafresh Tartar 7. Macleans 8. Sensodyne 9. Close Up 10. Arm & Hammer

Table V Penetration and frequencies at the attitudinal level (laundry detergents) Brand 1. Ultra Cheer 2. Liquid Tide 3. Ivory Snow 4. Tide 5. Wisk 6. ABC 7. Arctic 8. Sunlight

Penetration bx

Observed frequency wx

Predicted frequency wo/12 bx

Deviation wx-w0/12bx

0.56 0.44 0.61 0.78 0.49 0.57 0.43 0.64

4.94 3.29 4.11 10.46 3.28 4.20 1.84 6.43

4.25 3.28 4.74 8.28 3.66 4.29 3.24 5.12

+0.69 +0.01 20.53 +2.18 20.38 20.09 21.60 +1.31

Table VI Penetration and frequencies at the behavioral level (laundry detergents) Brand 1. Ultra Cheer 2. Liquid Tide 3. Ivory Snow 4. Tide 5. Wisk 6. ABC 7. Arctic 8. Sunlight

Penetration bx

Observed frequency wx

Predicted frequency wo/12 bx

Deviation wx-w0/12bx

0.22 0.15 0.10 0.51 0.18 0.25 0.05 0.43

1.01 1.08 0.71 1.91 1.20 1.12 0.57 1.63

1.05 0.96 0.91 1.68 1.01 1.10 0.86 1.43

20.04 +0.12 20.20 +0.23 +0.19 +0.02 20.29 +0.20

At the behavioral level, about 18 per cent of the sample had bought Colgate Original within the past year (see Table IV for more details). Thus, the

level of penetration is 0.18. Given this penetration level, the w(1-b) model suggests that each of those respondents who bought Colgate Original in the

186

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

last year should have bought it about 3.20 times during the year. This value is marginally smaller than the actual purchase frequency of 3.29 reported by those respondents who bought the brand. Hence, for both attitudinal and behavioral levels, Colgate Original marginally outperforms the expected frequency level. Table IV results also show that the predicted frequencies (w0/1-bx) of most of the larger brands (e.g. Crest Original, Colgate Original, and Colgate Tartar) are lower than their observed frequencies (wx). This result is in line with the findings of Fader and Schmittlein (1993) who show, using behavioral measures, that larger brands can have higher repeat purchase rates than those predicted. Regarding Crest Tartar, the lower-than-expected frequency may be attributable to a scaling factor. It is obvious that the penetration rate for Crest Tartar (bx ¼ 0.54) is disproportionately higher than that of a comparable brand such as Colgate Tartar (bx ¼ 0.11). As a result, its predicted frequency (wo/1-bx ¼ 5.70) is also disproportionately higher than that of Colgate Tartar (wo/1-bx ¼ 2.95), as well as its own observed frequency (wx ¼ 3.99). Results for laundry detergents, as shown in Tables V and VI, also show that most major brands (e.g. Ultra Cheer, Liquid Tide and Tide) outperform their expected usage frequency level at the attitudinal level. The only exception is Ivory Snow; it underperforms both at the attitudinal (4.11 vs 4.74) and at the behavioral levels (0.71 vs 0.91). Of all the smaller brands, only Sunlight has been able to defy the DJ effect both at the attitudinal (6.43 vs 5.12) and the behavioral levels (1.63 vs. 1.43). Interestingly, Ultra Cheer outperforms expected frequency at the attitudinal level (4.94 vs 4.25) but falls short at the behavioral level (1.01 vs. 1.05). Thus, it is evident that Ultra Cheer has failed to translate the positive attitude of consumers to increased usage frequency. Thus, brand managers of Ultra Cheer should re-evaluate the distribution, pricing, and promotion strategies to achieve a desired level of brand usage commensurate with consumer attitude towards the brand.

repositioning strategy to leverage a positive image (or rectify a negative image).

Research implications Based on our analysis, we suggest that the double jeopardy effect be measured both at the behavioral and attitudinal levels. While behavioral measures (e.g. brand usage rate and brand penetration) are simple and easily available with most sales data, attitudinal measurements unravel important brand strengths (e.g. positive perceptions) and brand vulnerabilities (e.g. negative perceptions). A better understanding of the positive (or negative) consumer perceptions about the brand will help the brand manager to devise a suitable

Niche brands It is interesting to find that a niche brand[7] such as Aquafresh Tartar has clearly exhibited a higher purchase frequency than is suggested by the w(1-b) model (3.14 vs 2.78 as shown in Table IV) thereby suggesting a high usage rate among a small number of loyal customers. Similar figures are also found for Colgate Tartar (3.39 vs 2.95 as shown in Table IV). These results are in line with the findings of an earlier study by Kahn et al. (1988) who showed that niche brands often tend to defy the DJ pattern. As regards attitudinal penetration, niche brands score lower than regular brands because the select customers of a niche brand form only a small portion of the overall customer base. Table III results support this characteristic of niche brands. All tartar variants of Crest (0.66 vs 0.76), Colgate (0.45 vs 0.63) and Aquafresh (0.36 vs 0.44) show smaller attitudinal penetration than do the original formulae. Results for the behavioral penetration follow the expected pattern except for Crest Tartar, which has an unusually high penetration of 0.54 as compared to 0.26 for Crest Original (see Table IV for details). Our analysis has been able to shed more light on the possible reasons for the high attitudinal performance scores of individual brands. Table VII shows the mean scores of the four categories of attributes (i.e. functional, disease/health, quality, and value) on which 30 attitudinal questions were asked. It is evident that all the “tartar” brands perform better than their corresponding “original” formulae under the “functional” category. On the other three factors, the performance of the tartar variants is comparable with the original variants despite the fact that the tartar brands, with the exception of Crest Tartar, have much lower purchase penetrations (as shown in column bx in Table IV). Exploring in still greater detail, in Table VIII we highlight the possible causes for the high scores in these two categories. The tartar variants performed exceptionally well on a few attributes (e.g. “removes stains from teeth”, “leaves mouth Table VII Attribute category scores for the six major toothpaste brands Crest Original Tartar Disease/health Functional Quality Value

187

8.7 11.5 6.6 2.2

8.6 13.0 6.0 10.6

Attribute category Colgate Original Tartar 6.3 9.3 4.1 6.8

5.7 9.8 3.7 6.0

Aquafresh Original Tartar 4.6 8.5 2.6 4.6

3.5 8.9 2.6 4.4

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

Table VIII Attribute scores for the six major toothpaste brands Attribute Colgate Original Tartar

Crest Removes stains from teeth Leaves mouth refreshed Is a brand I can trust Is gentle on teeth Prevents gum disease

Original

Tartar

3.8 8.8 5.2 3.1 3.8

16.7 13.6 5.7 4.1 4.0

1.8 6.8 2.8 1.9 2.2

refreshed”, etc.); these high attribute scores result in a high overall rating for the entire attitudinal category. Thus, the observance of DJ defiance by niche brands becomes much clearer when one links the high usage frequency of these brands to the attitudinal-level beliefs of customers. However, despite demonstrating a strong link between consumer behavior and attitudes through the example of niche brands, we are still confounded by the unusual usage figures for Crest Tartar. This niche brand shows a lower usage frequency than is predicted by the w(1-b) model (3.99 vs 5.70 as shown in Table IV). However, it is interesting to note that for the purely attitudinal measures, Crest Tartar performs much better than the predicted frequency (4.30 vs 3.46 as shown in Table III). Brand managers for Crest Tartar would do well to increase the usage frequency of their brand commensurate with its attitudinal strength.

Future research and limitations We have developed measures of attitudinal penetration and attitudinal frequency based on a psychological theory of memory and cognition. We have also demonstrated, for two product categories, that the DJ effect is more predominant when we use attitudinal level measures, compared to behavioral measures. In the process, we also showed how attitudinal measures provide interesting insights with regard to strengths and weaknesses of individual brands (e.g. niche brands). However, our study has several limitations. In the survey, behavioral measures for both penetration and frequency were based on the respondents’ recall of their purchase history, and hence may not be completely accurate. For example, a respondent may accurately recall the attributes they recognize about different brands, but not as accurately the number of purchases of that brand in a year. However, as stated earlier, free recall measures used in the survey questionnaire overcome the possible demerits of this shortcoming (Barnard and Ehrenberg, 1990). Future studies should try to collect purchase data

11.9 9.7 3.4 2.9 2.4

Aquafresh Original Tartar 1.3 4.8 1.8 1.3 1.2

9.0 7.9 2.4 2.3 1.6

from the panel of respondents during or immediately after the purchase using scanners and then collect attitudinal data from the same panel through a survey. The survey used a two-point (yes/no) scale to measure if respondents felt that a given brand had each of the 30 attributes. Unfortunately, this type of scale cannot accurately capture the degree of confidence in the response. In future, a sevenpoint Likert-type scale should be used to provide richer data for better understanding of the attitude-behavior relationship. In conclusion, our research may be seen as a starting point in exploring DJ effects at the attitudinal level. Further support for our work across other product categories and different markets should make marketers think about the possible causes for the deviations from the DJ effect at the behavioral level vis-a`-vis the attitudinal level. Understanding these deviations could lead to valuable insights into consumer attitude and behavior, and various marketing input variables that influence consumer attitude leading to desirable consumer behavior.

Notes

188

1 Specifically, the price of a given brand and the price, advertising, and promotions of competing brands will influence deviation from double jeopardy at the behavioral level. In general, we expect the pricing cues to cause more deviations from DJ at the behavioral level as compared to advertising cues which can affect both behavior as well as attitude. 2 Furthermore, it must be mentioned that in keeping with earlier research (Ratcliff et al., 1990), we do not expect a positive-list strength effect for our “yes/no”-type responses. That is, we do not expect strengthening one trace to harm the memory of other traces, because of the “yes/no” nature of the questions. In addition, conditions that might generate the positive-list strength effect (Murnane and Shiffrin, 1991) are absent for lowinvolvement brand communication. 3 The coexistence of strength hypothesis (for a repeated attribute) and multiple-trace hypotheses (for distinct attributes) in no way contradicts the research of Hintzman and Block (1971). These authors had explicitly stated that it is not mandatory for the two hypotheses to be mutually exclusive.

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

4 An alternative approach to measure brand liking is to use actual attitudinal measures of brand liking instead of attitudinal frequency. However, given our data limitations, we were unable to explore this measure. We propose this as a possible future extension of our work. 5 Again, past theoretical support is given by studies such as those undertaken by Hintzman and Block (1971) and Anderson and Bower (1972), who define frequency as the number of memory traces contacted. 6 The w(12 b) model outlines the relationship between the penetration bx and frequency of usage wx for brand X with the corresponding values of by and wy of brand Y. The full form of the model is wx(12bx) ¼ wy(12by)¼ wo where wo (a constant) is the estimated average value of w(12 b) across all brands in the category. 7 A niche brand caters to a distinct segment of the market (or market niche) with a distinct set of needs. Customers in a market niche are willing to pay a premium for a brand specially designed for them.

References Alba, J.W. and Marmorstein, H. (1987), “Frequency information as a dimension of consumer knowledge”, Journal of Consumer Research, Vol. 14, June, pp. 14-25. Anderson, J.R. and Bower, G.H. (1972), “Recognition and retrieval processes in free recall”, Psychological Review, Vol. 79, pp. 97-123. Barnard, N.R. and Ehrenberg, A.S.C. (1990), “Robust measures of consumer brand beliefs”, Journal of Marketing Research, Vol. 27, pp. 477-84. Barwise, T.P. and Ehrenberg, A.S.C. (1985), “Consumer beliefs and brand usage”, Journal of the Market Research Society, Vol. 1 No. 1, pp. 37-49. Barwise, T.P. and Ehrenberg, A.S.C. (1987), “The liking and viewing of regular TV programs”, Journal of Consumer Research, Vol. 14, June, pp. 63-70. Barwise, T.P. and Ehrenberg, A.S.C. (1988), Television and its Audience, Sage Publications, Beverly Hills, CA and London. Bettman, J. and Park, W. (1980), “Effects of prior knowledge and phase of the choice process on consumer decision process: a protocol analysis”, Journal of Consumer Research, Vol. 7, December, pp. 234-48. Borgida, E. and Campbell, B. (1982), “Belief relevance and attitude-behavior consistency: the moderating role of personal experience”, Journal of Personality and Social Psychology, Vol. 42, pp. 239-47. Bornstein, R.F. (1989), “Exposure and affect: overview and metaanalysis of research, 1968-1987”, Psychological Bulletin, Vol. 106 No. 2, pp. 265-89. Bower, G.H. (1967), “A multi-component theory of the memory trace”, in Spence, K.W. and Taylor Spence, J. (Eds), Psychology of Learning and Motivation, Vol. 1, Academic Press, New York, NY. Chaudhuri, A. (1995), “Brand equity or double jeopardy?”, Journal of Product & Brand Management, Vol. 4 No. 1, pp. 26-32. Dall’Olmo Riley, F., Ehrenberg, A.S.C., Castleberry, S.B., Barwise, T.P. and Barnard, N.R. (1997), “The variability of attitudinal repeat-rates”, International Journal of Research in Marketing, Vol. 14, pp. 437-50. Donthu, N. (1994), “Double jeopardy in television program choice”, Journal of the Academy of Marketing Science, Vol. 22 No. 2, pp. 180-5.

Ehrenberg, A.S.C. (1987), “New brands and the existing market”, CMaC working paper, London Business School, London. Ehrenberg, A.S.C. (2002), “Double jeopardy revisited, again”, Marketing Research, Vol. 14 No. 1, pp. 40-2. Ehrenberg, A.S.C. and Goodhardt, G.J. (1979), Understanding Buyer Behavior, J. Walter Thompson and MRCA, New York, NY. Ehrenberg, A.S.C., Goodhardt, G.J. and Barwise, P.B. (1990), “Double jeopardy revisited”, Journal of Marketing, Vol. 54, July, pp. 82-91. Fader, P.S. and Schmittlein, D.C. (1993), “Excess behavioral loyalty for high share brands: deviations from the Dirichlet model for repeat purchasing”, Journal of Marketing Research, Vol. 30, November, pp. 478-93. Fazio, R. and Zanna, M. (1981), “Direct experience and attitudebehavior consistency”, in Berkowitz, L. (Ed.), Advances in Experimental Social Psychology, Vol. 14, Academic Press, New York, NY, pp. 162-202. Harris, G., Begg, I. and Mitterer, J. (1980), “On the relation between frequency estimates and recognition memory”, Memory and Cognition, Vol. 8 No. 1, pp. 99-104. Hasher, L. and Zacks, R. (1979), “Automatic and effortful processes in memory”, Journal of Experimental Psychology: General, Vol. 108, September, pp. 356-88. Hasher, L. and Zacks, R. (1984), “Automatic processing of fundamental information”, American Psychologist, Vol. 39, pp. 1372-88. Hintzman, D.L. (1984), “MINERVA 2: a simulation model of human memory”, Behavior Research Methods, Instruments and Computers, Vol. 16, pp. 96-101. Hintzman, D.L. (1986), “‘Schema abstraction’ in a multiple trace memory model”, Psychological Review, Vol. 93 No. 4, pp. 411-28. Hintzman, D.L. and Block, R.A. (1971), “Repetition and memory: evidence for a multiple-trace hypothesis”, Journal of Experimental Psychology, Vol. 88 No. 3, pp. 297-306. Hofmeyr, J. and Bennett, R. (1994), “Double jeopardy and consumer commitment”, Canadian Journal of Marketing Research, Vol. 13, pp. 12-20. Kahn, B.E., Kalwani, M.U. and Morrison, D.G. (1988), “Niching versus change-of-pace brands: using purchase frequencies and penetration rates to infer brand positionings”, Journal of Marketing Research, Vol. 25, November, pp. 384-90. McPhee, W.N. (1963), Formal Theories of Mass Behavior, Free Press, New York, NY. Maslow, A.H. (1937), “The influence of familiarization on preference”, Journal of Experimental Psychology, Vol. 21, pp. 162-80. Murnane, K. and Shiffrin, R.M. (1991), “Word repetitions in sentence recognition”, Memory and Cognition, Vol. 19 No. 2, pp. 119-30. Petty, R. and Cacioppo, J. (1984), “The effects of involvement on responses to argument quantity and quality: central and peripheral routes to persuasion”, Journal of Personality and Social Psychology, Vol. 46, January, pp. 69-81. Ratcliff, R., Clark, S.E. and Shiffrin, R.M. (1990), “The list strength effect: I. Data and discussion”, Journal of Experimental Psychology: Learning, Memory and Cognition, Vol. 16, pp. 163-78. Rindfleisch, A. and Inman, J.J. (1998), “Explaining the familiarityliking relationship: mere exposure, information availability, or social desirability?”, Marketing Letters, Vol. 9 No. 1, pp. 1-15. Russo, J. and Dosher, B. (1983), “Strategies for multi-attribute binary choice”, Journal of Experimental Psychology:

189

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

Learning, Memory and Cognition, Vol. 9, October, pp. 676-96. Sample, J. and Warland, R. (1973), “Attitude and prediction of behavior”, Social Forces, Vol. 51, pp. 292-304. Schwartz, S. (1978), “Temporal instability as a moderator of the attitude-behavior relationship”, Journal of Personality and Social Psychology, Vol. 36, pp. 715-24. Shiffrin, R.M., Ratcliff, R. and Clark, S. (1990), “The list-strength effect: II. Theoretical mechanisms”, Journal of Experimental Psychology: Learning, Memory and Cognition, Vol. 16, pp. 179-95. Snyder, M. and Kendzierski, D. (1982), “Acting on one’s attitude: procedures for linking attitude and behavior”, Journal of Experimental Psychology, Vol. 18, pp. 165-83. Stang, D.J. and O’Connell, E.J. (1974), “The computer as experimenter in social psychology research”, Behavior Research Methods and Instrumentation, Vol. 6, pp. 223-31. Thorndike, E.L. and Lorge, I. (1944), The Teacher’s Word Book of 30,000 Words, Teachers College, Columbia University, New York, NY. Zajonc, R.B. (1968), “Attitudinal effects of mere exposure”, Journal of Personality and Social Psychology Monographs, Vol. 9, pp. 1-27. Zajonc, R.B., Shaver, P., Tavris, C. and Van Kreveld, D. (1972), “Exposure, satiation, and stimulus discriminability”, Journal of Personality and Social Psychology, Vol. 21, pp. 270-80.

behaviour emerging from statistical analysis or whether the effect reflects psychological factors at the level of the individual consumer. What Bandyopadhyay and Gupta do is to look at consumer behaviour and consumer attitudes to assess whether the “double jeopardy” effect derives from the manner in which consumers absorb, assess and store information about brands. In making this assessment the authors recognise that debate continues as to the way in which individuals store memories and information. Clearly, this debate is important to marketers since it helps to determine the manner in which we present our message. Nevertheless, research does show that the normal assumptions about promotional actions are robust – repetition does increase awareness and stimulates positive attitudes to the brand, consumers store repeated exposures to a given brand attribute as well as information about different attributes associated with a particular brand. The question that the authors ask here is, accepting the limitations of current research in consumer psychology, whether the “double jeopardy” effect shows more strongly at the attitudinal level that at the behavioural level. If this is the case, then marketers can be confident that the effect derives from the consumer’s memory and psyche rather than being a statistical quirk. At the same time we can also begin to examine ways in which our strategies might change in response to the “double jeopardy” effect.

Further reading Hawkins, S.A. and Hoch, S.J. (1992), “Low involvement learning: memory without evaluation”, Journal of Consumer Research, Vol. 19, September, pp. 212-25.

Executive summary This executive summary has been provided to allow managers and executives a rapid appreciation of the content of this article. Those with a particular interest in the topic covered may then read the article in toto to take advantage of the more comprehensive description of the research undertaken and its results to get the full benefit of the material present.

Double jeopardy – it is all in the consumer’s mind The “double jeopardy” effect (where brands with a greater market share not only have more customers but these customers are more loyal to the brand) presents a challenge to marketers. Once we have got over our initial depression about the effect, we need to consider how to develop strategies that respond to our brand’s market position, market share and performance. At the theoretical level it is also important that we begin to understand whether the “double jeopardy” effect is a function of aggregate

Bigger brands get more attention from the consumer Brand managers have often thought and talked about “share of mind” when they consider the individual consumer. This piece of “cod” psychology helps practitioners to appreciate that, while we tend to look at markets and market segmentation in the aggregate this must be constructed from a series of individual consumer behaviours and attitudes. Bigger brands, by securing a larger share of mind, get more attention than smaller brands. Thus the consumer of the smaller brand is more likely to recall positive attitudes associated with the larger brand. But the consumer of a market leading brand, where the same situation exists, is less likely to recall positive attributes associated with the smaller brand. The small brand user is therefore more likely to switch to a larger brand than vice versa. The result of this psychological situation is, when consumer actions are aggregated, that larger brands not only have more consumers but those consumers are less likely to switch. The double jeopardy effect results from larger brands having a greater number of stored “traces” of positive attributes and, at the same time, a greater range of positive attributes. The strength of bigger brands

190

Comparing double jeopardy effects

Journal of Product & Brand Management

Subir Bandyopadhyay and Kunal Gupta

Volume 13 · Number 3 · 2004 · 180-191

lies in the ubiquity of the brand message rather than in market share per se. Strategies to gain share of mind If it was that case that “share of mind” derived from the number of different attributes associated with a given brand then it is unlikely that the “double jeopardy” effect would be so prevalent. In most consumer goods markets there are a number of brands each having a set of positive attributes that may be stored by consumers. What the “double jeopardy” effect suggests is that the mind stores the same information as many times as that information is received. In making a brand choice we draw on the information about the different brands and, since we have more pieces of data about one brand compared to another, we are more likely to recall a positive attribute about that brand. The challenge for the marketer is to place the brand in the situation where the consumer’s mind is more likely to recall a positive attribute for our brand compared to one for a competing brand. This leads us to develop strategies based on repetition of the message across time and the full range of communications media. And this message needs to focus on a small number of positive attributes – it seems not to matter as much about the range of positives but about the number of times any positive message reaches the consumer. This situation reinforces the difficulty faced by smaller brands with smaller budgets – repeated

messages require heavy media spend which further underlines the advantages enjoyed by bigger brands. To succeed therefore, marketers of smaller brands need to concentrate their investment. Two routes open up – focusing on a given region or looking at a particular niche market. In the latter case there is plenty of evidence (support by the authors here) that identifying a niche market protects smaller brands from larger brands. The niche only appeals to a limited group of consumers so it is possible that the smaller brand can achieve a leading position within the minds of this deliberately limited selection of consumers. Regional focus does not often arise in thinking about brand strategies especially in a world where things are seen as “global” and some media (e.g. the Internet) do not have recognised boundaries. Nevertheless, most brand messages are received on a very local basis through broadcast and printed media and through point-of-sale promotion, packaging and merchandising. Perhaps a concentrated focus on a given geographical market could provide a strong basis for a brand to achieve a stronger position without having to find similar levels of promotional spend as the big international brands. (A pre´cis of the article “Comparing double jeopardy effects at the behavioral and attitudinal levels”. Supplied by Marketing Consultants for Emerald.)

191