29
JOURNAL OF CONSUMER PSYCHOLOGY, 7(2), 101-129 Copyright O 1998, Lawrence Erlbaum Associates, Inc. The Inclusion Effect and Category-Based Induction: Theory and Application to Brand Categories Christopher Joiner Department of Marketing and International Business Kansas State University Barbara Loken Department of Marketing and Logistics Management University of Minnesota This article investigates inferences between elements in brand categories using an argument strength research paradigm. In 3 laboratory studies of brand categories, we found that consumers often generalize possession of an attribute from a specific category (e.g., Sony televisions) to a more general category (e.g., all Sony products) more readily than they generalize the attribute from the specific category (e.g., Sony televisions) to another specific category (e.g., Sony bicycles)-a counternormative phenomenon called the inclusion effect. Consistent with the category induction model of Osherson, Smith, Wilkie, Lopez, and Shafir (1990), the similarity between the premise and conclusion categories (e.g., televisions and bicycles) was a strong predictor of argument strength judgments. We also found that the inclusion effect was attenuated when the specific conclusion category increased in its typicality to the general category (e.g., Sony cameras vs. Sony bicycles). Presenting the general conclusion argument immediately followed by the specific conclusion argument (as opposed to presenting each argument alone) seemed to accentuate the inclusion effect, despite our expectation that increased salience of the judgmental inconsistencies (that lead to the counternor- mative reasoning) would reduce the effect. Although several rival explanations of the inclusion effect (including a conversational norms explanation) were ruled out, a conclusion plausibility explanation also appeared to account for some of the results. Requests for reprints should be sent to ChristopherJoiner, Department of Marketing and International Business, Kansas State University, Manhattan, KS 66506. E-mail: [email protected]

The Inclusion Effect and Category-Based Induction: … Inclusion Effect and Category-Based Induction: ... Lauren sweatshirts to Ralph Lauren belts) ... (Cherniak, 1984; Lakoff,

  • Upload
    leanh

  • View
    218

  • Download
    0

Embed Size (px)

Citation preview

JOURNAL OF CONSUMER PSYCHOLOGY, 7(2), 101-129 Copyright O 1998, Lawrence Erlbaum Associates, Inc.

The Inclusion Effect and Category-Based Induction: Theory and Application to Brand Categories

Christopher Joiner Department of Marketing and International Business

Kansas State University

Barbara Loken Department of Marketing and Logistics Management

University of Minnesota

This article investigates inferences between elements in brand categories using an argument strength research paradigm. In 3 laboratory studies of brand categories, we found that consumers often generalize possession of an attribute from a specific category (e.g., Sony televisions) to a more general category (e.g., all Sony products) more readily than they generalize the attribute from the specific category (e.g., Sony televisions) to another specific category (e.g., Sony bicycles)-a counternormative phenomenon called the inclusion effect. Consistent with the category induction model of Osherson, Smith, Wilkie, Lopez, and Shafir (1990), the similarity between the premise and conclusion categories (e.g., televisions and bicycles) was a strong predictor of argument strength judgments. We also found that the inclusion effect was attenuated when the specific conclusion category increased in its typicality to the general category (e.g., Sony cameras vs. Sony bicycles). Presenting the general conclusion argument immediately followed by the specific conclusion argument (as opposed to presenting each argument alone) seemed to accentuate the inclusion effect, despite our expectation that increased salience of the judgmental inconsistencies (that lead to the counternor- mative reasoning) would reduce the effect. Although several rival explanations of the inclusion effect (including a conversational norms explanation) were ruled out, a conclusion plausibility explanation also appeared to account for some of the results.

Requests for reprints should be sent to Christopher Joiner, Department of Marketing and International Business, Kansas State University, Manhattan, KS 66506. E-mail: [email protected]

102 JOINER AND LOKEN

The increasing importance of viewing brands as categories is evident from the recent proliferation of academic research on brand extensions (Broniarczyk & Alba, 1994a; Dawar, 1996; Dawar & Anderson, 1994; Herr, Farquhar, & Fazio, 1996), brand dilution (Aaker & Keller, 1990; Loken & John, 1993; Romeo, 1990), and the structure of brand categories (Boush, 1993; Boush & Loken, 1991). Practitioners also often think of their companies' products in terms of brand categories; many companies are organized by brands, and recently several have been promoting within a single communication their brand names (e.g., Jell-0, Land O'Lakes, Healthy Choice, Johnson & Johnson) and the full range of products under a brand name (e.g., an array of desserts). As a result, consumers may be more inclined to think about an overall brand category when evaluating new products with the brand name or when considering new information relevant to a particular brand name.

Prior research has examined the types of attribute inferences consumers make about branded products. One stream of research (e.g., Broniarczyk & Alba, 1994a; Park, Milberg, & Lawson, 1991) has examined the effects of overall brand category attributes (e.g., Jell-0 products taste great) on new product introductions with the same brand name-that is, in the same-brand category. Brands are more extendable (i.e., consumers generalize brand attributes to new category members) if the new product shares a consistent brand concept (Park et al.) or if the brand's attributes are relevant to those attributes desired in the extension category (Broniarczyk & Alba, 1994a). Other research has examined the role of similarity, or fit, between existing brand category members and new products under the same brand name in generalizing attribute and attitude information. Consumer inferences about the attributes of, and affect toward, existing products of a brand category are more likely to generalize to a new product in the brand category when the extension is similar (vs. dissimilar) to the existing products (e.g., Aaker & Keller, 1990; Boush & Loken, 1991; Boush et al., 1987; Park et al.), when the brand's parent and extension categories are perceived to be closely related (Herr et al., 1996), or when advertising for new brand members is similar to advertising for existing ones (D. C. Smith & Park, 1992). Research also has examined the inferences that consumers make about the parent brand on the basis of attributes of a new brand extension. If the attributes of a new brand extension are inconsistent with those associated with the general brand category, the extension attributes can "dilute" or change perceptions of the brand category attributes (John, Loken, & Joiner, in press; Loken &John, 1993).

However, little is still known about the role of category membership in deter- mining inferences between new and existing brand category members. If two products are believed to belong to the same category, they may be perceived as sharing the category attributes (Barsalou, 1992). This tendency may be heightened for branded products when they are assigned a category by virtue of their brand name (Boush, 1993). If consumers believe that Gucci handbags are quality prod-

THE INCLUSION EFFECT 1 03

ucts, those consumers may infer that all Gucci products are quality products. They also may extend such induction to new brand extensions under the Gucci name (Gucci sheets or other new products may be perceived as high quality). On learning that an object possesses a new attribute, people may use an assigned category to guide how far they generalize the attribute to other known objects (Barsalou, 1992; E. E. Smith, Lopez, & Osherson, 1992). This phenomenon is named category-based induction.

Category-based inductions are more likely under certain conditions. For exam- ple, the extent to which consumers transfer an attribute from one existing product (e.g., Jell-0 gelatin) to a new product (e.g., Jell-0 cookies) may depend on how the new product (cookies) is related to the existing product (gelatin) or to the category overall (all Jell-0 products). When a marketing communication conveys that a particular branded product has a desirable attribute (e.g., Jell-0 gelatin tastes great), consumers may infer that the attribute describes some specific products (e.g., Jell-0 pudding) more than it does others (e.g., Jell-0 cookies). Our research examines how an attribute is transferred from one member of a category to other members of the category, as well as to the category overall.

A relevant inference phenomenon called the inclusion effect (or inclusion fallacy; see Osherson, Smith, Wilkie, Lopez, & Shafir, 1990; Shafir, Smith, & Osherson, 1990; Sloman, 1993; E. E. Smith et al., 1992) finds that consumers often generalize possession of an attribute from a specific category (e.g., Ralph Lauren sweatshirts have attribute X) to a more general category (e.g., all Ralph Lauren products have attribute X) more readily than they generalize possession of the same attribute from the same specific category to another specific category (e.g., Ralph Lauren belts). This may occur even though the latter category (Ralph Lauren belts) is a subset of the general category (all Ralph Lauren products). Such inductive inferences are considered counternormative because an argument is made to appear stronger despite the increased generality of its conclusion; typically, the probability of the conclusion must objectively decrease as its generality increases. The inclu- sion fallacy is a violation of the extension rule in probability theory, which states that if A is a subset of B, then the probability of A cannot exceed that of B. Data supporting the inclusion effect for natural, biological categories are reported by Osherson et al. (1990; see also Shafir et al., 1990, and E. E. Smith et al., 1992).

Surprisingly little is known about the nature and determinants of inductive inference processes that consumers use. Osherson et al. (1990) proposed a frame- work for studying category-based inductions by structuring them as sets of two arguments, with each set containing a premise category (assumed factual) and a conclusion category:

A: Ralph Lauren sweatshirts have attribute X Therefore, all Ralph Lauren products have attribute X

104 JOINER AND LOKEN

B: Ralph Lauren sweatshirts have attribute X Therefore, Ralph Lauren belts have attribute X

Each argument involves some attribute of the premise being transferred to the conclusion; therefore, the inferred attribute (X) remains fixed across premises and conclusions. In this paradigm, consumers are asked to judge which argument, A or B, is stronger. These argument strength judgments measure people's propensity to make generalizations or categorical inductions of the attribute from the premises to the conclusions. An argument is considered strong to the extent that belief in the premises leads one to believe in the conclusion. Premise information regarding products may be deduced from many sources (e.g., Consumer Reports, expert word of mouth, advertising from a trusted company, etc.). If the inferential link from the premises to conclusion is strong enough, some of the conviction associated with the premises is transmitted to the conclusion (Osherson, Smith, & Shafir, 1986). In this example, the inclusion effect is manifested if A (the general conclusion argument) is chosen or judged to be stronger than B (the specific conclusion argument).

Whereas consumer research on inferences has examined the relation between brands and attributes, research on similarity-based and category-based induction has received little attention (Broniarczyk & Alba, 1994b; Joiner & Loken, 1994). The present research focuses on documenting evidence for the inclusion effect in consumer contexts, exploring its theoretical underpinnings, and understanding the conditions under which it occurs. We report three empirical studies in this article. Studies 1 and 2 focused primarily on the aforementioned objectives. The third study evaluated the persistence of the inclusion effect phenomenon under a manipulation intended to increase the salience of the argument sets relative to each other.

THE INCLUSION EFFECT AND MODERATING INFLUENCES

Theoretical Bases

Osherson et al. (1990) argued that the cognitive bases for the inclusion effect derive from the similarity between the categories. Consumers may more readily generalize attributes from a specific to a general category (e.g., from Ralph Lauren sweatshirts to all Ralph Lauren products) than across specific categories (e.g., from Ralph Lauren sweatshirts to Ralph Lauren belts) when the former categories (sweatshirts and all products) are perceived to be more similar to each other than the latter categories (sweatshirts and belts). This explanation implies that when extending to new products, consumer inferences about the attribute rely on the perceived

THE INCLUSION EFFECT 105

similarity between the existing product that has the attribute and the new extension. If the two products are perceived as dissimilar, an inference about an attribute of one of the products may be generalized to the overall brand category more than to the other product. By the same token, a communication may inadvertently cause larger changes in consumers' beliefs about the overall brand category than in their beliefs about certain (e.g., dissimilar) individual products. Thus, by emphasizing the similarities or dissimilarities between two products, a firm could potentially influence the extent to which inferences are made between them.

A competing explanation for the inclusion effect derives from a prototype view of category structure (Rosch & Mervis, 1975). Much of our knowledge about categories of things is organized in terms of prototypical (typical) cases, and we frequently draw inferences based on such knowledge (Cherniak, 1984; Lakoff, 1987). Because this organization is heavily influenced by typical instances, con- sumers' beliefs about the attributes of a category are often more stable and contain lower variability than would be warranted if the attributes of all category members were considered (Rothbart & Lewis, 1988). The inclusion effect may occur because consumers are more apt to generalize attributes to the overall brand category (e.g., all Ralph Lauren products) when their mental representation of the category includes only the more typical (vs. the atypical) members of the category. Consid- eration of both typical and atypical members of the brand category during inferences may attenuate the inclusion effect. In contrast to the premise-conclusion similarity explanation, the focus here is on consumers' perceptions of the degree of typicality of one conclusion category to the other (i.e., the new extension to the overall brand category). Understanding consumers' perceptions about the specific product from which the attribute is generalized (e.g., Ralph Lauren sweatshirts) is relatively less important.

Meaningful Versus Blank Attributes

The extent to which consumers apply a prototype or category similarity heuristic may also depend on the meaningfulness of the stimuli to which they are exposed. When consumers are asked to reason about products in the abstract (e.g., using hypothetical category attributes), they may use more formal, reasoned, and analytic strategies associated with "scientific" thinking (see, e.g., Peters, Hammond, & Summers, 1974) and may exert more effort in searching beyond the most accessible information (as suggested by Alba & Hutchinson, 1987), perhaps even considering both typical and atypical members of the brand category. However, with more familiar stimuli, people may rely on everyday knowledge and reasoning. In fact, fallacies similar to the inclusion effect tend to occur primarily for concrete (i.e., meaningful) cases (Tversky & Kahneman, 1982). Alternatively, when individuals are exposed to meaningful attributes in an argument strength task, they may attempt

1 06 JOINER AND LOKEN

to simplify the task and use their prior beliefs about the conclusions, or the plausibility of the conclusions, rather than reason formally from premises to conclusions. In these instances, conclusion plausibility may be the determining factor in consumers' argument strength judgments. No clear guidelines emerge from past research about whether meaningful attributes will influence a consumer's processing style. Hence, we explored whether the inclusion effect is accentuated as the stimulus arguments become more meaningful and as "real" (vs. hypothetical) attributes increase an argument's meaning.

Task Effects

Finally, task structure may influence argument strength judgments. We examined two stimulus task effects. First, a forced-choice task was compared to a task in which the strength of each argument is measured separately. Second, the general argument was judged first, and the specific argument was judged immediately thereafter; this should increase the salience of the relation between the two argu- ments relative to when only one of the two arguments is judged. If participants have made an inference from a specific brand category member to all products in the brand, the salience of the judgment and its implications should increase the likelihood of the participants immediately making a similar judgment from the specific brand category member to an atypical category member. The inconsistency of inferring all category members but not an atypical category member should be salient and, therefore, should attenuate the inclusion effect.

STUDY 1

Study 1 examined the extent to which consumers make judgments about brand categories that are consistent with the inclusion effect and to what extent the meaningfulness of the attributes being inferred influences this effect.

Method

Design and measures. We compared two levels of attribute type (mean- ingful vs. blank) and three brand name replicates (Sony, Frito Lay, and Ralph Lauren). Both design factors were repeated measures, such that six (2 x 3) sets of arguments were tested. Thirty-eight participants completed one of two versions of a short survey consisting of a page of instructions explaining the argument strength judgment task, followed by the argument pairs (Arguments A and B), with each pair presented on a separate page. The six arguments were embedded within the

THE INCLUSION EFFECT 107

context of 14 argument strength judgments. In addition to the 6 focal argument strength judgments and argument "foils" (described in more detail subsequently), 6 other judgments were made about three unbranded products. These unbranded stimuli provided a neutral context and helped separate the target stimuli. Partici- pants also completed a series of questions designed to investigate rival explanations for the inclusion effect and provide typicality and pairwise similarity judgments. Two versions of the survey were developed so that an individual received an equal number of meaningful and blankarguments. Each version contained questions from each of the brand name replicates; the questions were systematically rotated so that a total of 19 participants answered each of the six sets of arguments.

Stimuli. Real brand names were selected that were familiar to our population. Descriptions of the types of products made by (or "recently introduced by") the brand were provided and included examples of representative products in the category as well as all of the products used in the stimulus arguments. These descriptions were held constant across meaningful and blank attribute conditions and were included to control for participants' levels of familiarity with the products in the three brand replicates. In the blank attribute conditions, attributes were patterned after those of Osherson et al. (1990), such that the attributes were believable but not the ones about which most people would have prior beliefs (e.g., "contains Proseen 10"). In the meaningful attribute conditions, we used familiar attributes about which consumers were more likely to have prior beliefs. We also selected products for the specific argument conclusions for which participants would be somewhat unfamiliar. For some of the stimuli, it was decided to use hypothetical brand extension products to achieve this goal. It was explained to participants that the products in the arguments included a number of potential, or recently introduced, extensions of the brand name. The argument strength stimuli from Study 1 are presented in Table 1.

Measures. A single forced-choice question was used to measure the inci- dence of the inclusion effect (Osherson et al., 1990). For each stimulus pair, participants were asked to select the argument (A or B) that was more convincing "assuming that the facts above the line are true." The premises above the line in each argument were preceded by the term "Fact:" and the conclusions below the line were preceded by "Therefore:" to reinforce the nature of the task. Following the argument strength task, half of the participants completed similarity judgments for half of the 12 premise~onclusion pairs on a scale ranging from 0 (not at all similar) to 10 (extremely similar). The other half completed similarity judgments for the remaining stimulus pairs. Additionally, all participants judged the typicality of the conclusion in the specific conclusion stimulus (e.g., Ralph Lauren belts) as

1 08 JOINER AND LOKEN

TABLE 1 Argument Strength Stimuli (Study 1)

General Specific MeaningFcl Blank Brand Premise Conclusion Conclusion Attribute Attribute

Sony Stereo receivers A11 Sony electronic Clock radios Are dependable Contain products Cryson A

Frito Lay Mixed nuts All Frito Lay Frozen pizzas Are packaged Contain snack foods for freshness Proseen 10

Ralph Lauren Sweatshirts All Ralph Lauren Belts Are comfortable Have Ulric clothing stitching

members of the more general category (e.g., Ralph Lauren clothing) on a scale ranging from 0 (not at all typical) to 10 (extremely typical) for all stimuli.

Results

Support for the inclusion effect. Study 1 results are presented using 50% as the criterion against which the occurrence of the inclusion effect is measured. Although evidence of any degree of counternormative bias would be informative, we used the conservative 50% level to account for the possibility of "guessing." Using the binomial distribution, we constructed 95% confidence intervals around each of the observed inclusion effect proportions (i.e., those choosing the general, "A" argument) in our data. These observed inclusion effect proportions were then compared to the 50% criterion to determine if the observed effect was significant.

Results as a function of attribute type and product replicate condition are presented in Table 2. Overall, 72.8% of participants' judgments were consistent with the inclusion effect. This rate is significantly higher @ < .05) than the conservative 50% criterion. The inclusion effect rate for all of the individual conditions was significantly higher than 50% for all three conditions with mean- ingful attributes and for one of the three blank attribute conditions. In addition, only 5 of the 38 participants (13.2%) did not make at least one judgment consistent with the inclusion effect at the individual response level.

A 3 x 2 repeated measures analysis of variance (ANOVA) was also conducted on the proportion data (i.e., proportion of choices consistent with the inclusion effect). Because the dependent measure was dichotomous, logistic regression was also performed. The results for these two analyses were similar except for one variation: In the regression analysis, the brand replicate main effect was significant, Wald test = 16.78, p < .05, and in the ANOVA it was not, F(2, 36) < 1. However, follow-up analyses indicated that although the Sony replicate yielded lower rates of the inclusion effect (.66) than did the Ralph Lauren replicate (32) or the Frito

THE INCLUSION EFFECT 109

TABLE 2 Argument Strength Results (Studies 1 and 2)

Study I : Proportion of Forced Choices Consistent With Inclusion Effect

Brand Replicate Real Attribute Blank Attribute

Sony .737* ,579 Frito Lay .737* ,684 Ralph Lauren .842 * .789*

Study 2: Argument Strength Diference Scores

Moderate Typicality Low Typicality

Real Attributes Blank Attributes Real Attributes Blank Attributes

Sony -.02 1 -.396 3.479* 2.688* Gucci 2.208* -.208 3.646* 3.667* Kraft ,646 -.979 ,708 1.479* Jell-0 .I25 -.688 1.938* 1.104

Note. Study 1 n = 19 per cell. Study 2 n = 48 per cell. *For Study 1, entry is significantly different from 50%. ~ ' ( 1 ) > 4.25 at p < .0S. For Study 2, entry is

significantly different from zero, r(47) > 2.85 at p < .05.

Lay replicate (.71), no significant differences occurred between any of the repli- cates, all ts(18) < 1.40, ps > .15. Although meaningful attributes yielded more inclusion effects than did blank attributes (.77 vs. .68), the effect of attribute level, F(l, 18) = 2.03, and the attribute by brand interaction, F(2, 36) < 1, were both nonsignificant.

According to the Osherson et al. (1990) model, the difference between the premise+onclusion similarity of the two arguments (general and specific) is the driving force behind the inclusion effect. Consistent with this theorizing, partici- pants rated the similarity of the premise and conclusion in the general arguments (Ms = 8.32, 5.73, and 7.84 for Sony, Frito Lay, and Ralph Lauren replicates, respectively) as significantly greater than the similarity of the premise and conclu- sion in the specific arguments, Ms = 5.47,2.11, and 4.37, respectively; t(18) = 5.30, 6.33, and 4.97, respectively; ps < .001. The prototypicality theory was also supported in that the participants rated specific conclusion categories as low (M = 2.92 for Frito Lay) or moderate (Ms = 6.35 and 5.76 for Sony and Ralph Lauren, respectively) in typicality.

Conclusion plausibility. A rival hypothesis for our data pertains to the effects of the believability (plausibility) of the conclusions in the arguments in each stimulus pair, because the conclusion of one argument is different from the

1 10 JOINER AND LOKEN

conclusion of the second argument in each pair. We included an "argument strength foil" in Study 1, such that the "correct" choice (i.e., if participants were comprehending the argument strength task correctly) had an implausible conclusion (i.e., which prior beliefs would negate), and the "incorrect" choice had the more plausible conclusion.

Participants overwhelmingly chose the correct argument in a stimulus pair, even when it contained the less plausible conclusion. Across six brands, an average of only 11.2% (range = 5.3-15.8%) of the participants chose the incorrect but more plausible argument. We added two other types of argument strength judgments to our study to rule out plausibility effects. For some pairs of arguments, we held the conclusion constant (e.g., all Gucci coffee tables are high quality) but used different premise categories that varied in similarity to the conclusion (e.g., Gucci dining room tables are high quality vs. Gucci handbags are high quality). For other pairs of arguments, the conclusion was again held constant (e.g., all Sony electronic products are dependable), but we chose one of the premises to have the same brand name (e.g., Sony) as the conclusion and one to have a different brand name (e.g., Panasonic) as the conclusion. Clearly, argument strength (but not plausibility) should be higher for the same-brand attribute statements. Because a direct compari- son of the two choices would make the differences in the premises apparent, we asked our Study 1 participants instead to rate each of 14 arguments separately on a scale ranging from 0 (not at all convincing) to 10 (very convincing). Arguments within a pair were separated from each other by at least two other judgments and were on different pages of the survey. If participants were basing their judgments solely on the plausibility of the conclusion, the convincingness (argument strength) ratings of the two arguments in each pair should not differ. Seven pairs were tested, and all showed significant differences in the direction expected on the basis of argument strength (see Table 3). This result held even for one question that had a very implausible conclusion (i.e., Kraft ice cream bars have lots of cheese). Thus, the participants appeared to be making argument strength judgments, and our experimental instruction to assume that all premises were true seemed to have the desired effect.

Discussion

Study 1 clearly demonstrates that the inclusion effect (Osherson et al., 1990) occurs in consumers' reasoning about brands and products. Overall, 72.8% of judgments were consistent with this counternormative effect. Participants rated the product in the premise of the arguments as more similar to the general brand category than to the specific product in the conclusion of each of the argument pairs. These similarity ratings cannot account for differences between responses to the meaningful versus blank attribute arguments (because similarity ratings are the same for both), but

THE INCLUSION EFFECT 1 1 1

TABLE 3 Conclusion Plausibility Alternative Explanation Data:

Mean Argument Strength Ratings (Study 1)

Argument A Rating Argument B Rating - Brand A alarm clocks are dependable

Brand A clock radios.. . Sony televisions are dependable

All Sony electronic products.. . Frito Lay frozen egg rolls are packaged for freshness

Frito Lay frozen pizzas.. . Gucci dining room tables are high-quality products

Gucci coffee tables.. . Ralph Lauren sweatshirts contain Ulric stitching

All Ralph Lauren clothing.. . Jell-0 frozen entrees come in a variety of types

Jell-0 frozen vegetables.. . Kraft ice cream sandwiches have lots of cheese

Kraff ice cream bars.. .

6.47 Brand A CD players.. . 5.39* Brand A clock radios.. .

6.1 1 Panasonic juicers.. . 1.74* All Sony electronic products.. .

6.82 Frito Lay mixed nuts.. . 4.58" Frito Lay frozen pizzas.. .

7.7 1 Gucci handbags. .. 5.29% Gucci coffee tables.. .

4.61 Nike athletic shoes.. . .84* All Ralph Lauren clothing.. .

6.08 Pillsbury pie shells.. . 2.05* JeH-0 frozen vegetables.. .

4.2 1 McDonalds cheeseburgers.. . .34* Kraft ice cream bars.. .

Note. n = 38. *The difference between ratings for Arguments A and B, t(37) > 2.75, was significant at p < .01

they do suggest that the similarity between premise and conclusion categories may be playing a role.

In Study 1, our specific conclusions stimuli were either low or moderate in typicality, but we did not investigate how level of typicality (apart from prem- ise-conclusion similarity) influences the inclusion effect. If atypical products are not included in consumers' representations of a brand, participants should be more likely to evaluate a general conclusion argument (e.g., All Brand X products are high quality) as stronger than a specific conclusion argument (e.g., Brand X atypical product is high quality). In Study 2, we investigated the effects of both low and moderate levels of typicality on the incidence of the inclusion effect.

Although the attribute meaningfulness effects were in the expected direction in Study 1, the differences between meaningful and blank attributes were nonsignifi- cant. Two possibilities existed that may have weakened the effect. First, the meaningful attributes may not have been salient, and salient attributes may be more likely than nonsalient attributes to induce heuristic processes. Second, one of the attributes (dependable) in Study 1 was evaluative, more abstract, and probably generalizable to similar and dissimilar product categories; other attributes were concrete and descriptive (e.g., packaged for freshness) and may have been less generalizable (see, e.g., Johnson, 1984; Nakamoto, MacInnis, & Jung, 1993). As such, the effects due to concrete attributes may be idiosyncratic for a particular

1 1 2 JOINER AND LOKEN

subset of products. Therefore, in Study 2 we attempted to use attributes that, although abstract, were pretested to be salient for the subject population.

Additionally, in Study 2 we included an argument strength measure that does not rely on a forced-choice judgment. Although the forced-choice results of Study 1 provide evidence for the inclusion effect in consumer contexts and rule out a "guessing" explanation (otherwise, the judgments would have been split approxi- mately evenly between the general and specific arguments), they do not tell us whether the magnitude of the difference between the general and specific argument strengths is significant. If the differences between the two argument strengths are not significant, the results are less meaningful managerially. Therefore, in Study 2 an interval-scale strength measure, more directly measuring probabilistic infer- ences, was used to measure each argument individually. The forced-choice measure was also included in Study 2.

STUDY 2

The moderating role of the typicality of the specific conclusion in an argument pair was considered in Study 2. Additionally, we reexamined the role of attribute meaningfulness using attributes that were both salient and applicable to a wide range of products. As before, we predicted that the inclusion effect would be greater when arguments included real attributes. Finally, the interval scale measure of argument strength allowed participants to rate the two arguments as equally strong.

Method

Design and measures. The general argument strength task in Study 2 was the same as in Study 1. Study 2 included two levels of typicality of the product in the conclusion of the specific argument (moderate vs. low), two levels of attribute type (meaningful vs. blank), and four brand name replicates. The four replicates were split between two separate (between-subject) versions of the questionnaire so that two brand names (Jell-0 and Kraft) appeared in one version and the other two brand names (Sony and Gucci) appeared in the other. All other design factors were repeated measures, comprising eight (2 x 2 x 2) sets of arguments per version.

A total of 96 undergraduate marketing students completed argument strength judgments (n = 48 for the KraftIJell-0 version and n = 48 for the GucciISony version). Participants rated each of the arguments in the stimulus pair by answering: "In your opinion, how likely is it that the statement below the line in 'Argument A (B)' follows from the statement above the line?'Possible responses ranged from 0 (not at all likely) to 10 (extremely likely). An argument strength difference score (subtracting the rating for the specific argument from the rating for the general argument) was created for each stimulus pair and used in subsequent analyses. A

THE INCLUSION EFFECT 1 13

forced-choice task was also included, but results were generally consistent with the difference score measure and are not discussed further. These judgments were followed by similarity measures for either the Jell-O/Kraft or Gucci/Sony stimuli as well as typicality measures for all four brands (as in Study 1). The position of the general and specific arguments (A and B) within each stimulus pair was assigned randomly. Additionally, two sets of foils and a linguistic explanation question (described subsequently) were also included in the survey. The order of presentation of the 12 arguments was randomized in each version and presented to all partici- pants.

Pretests were conducted to develop appropriate attributes to be used in the stimulus arguments. A standard elicitation technique (Ajzen & Fishbein, 1980) was used to obtain attributes that were frequently mentioned as characteristics of the brand categories tested. From this list of salient attributes, attributes were selected for each of the meaningful attribute conditions. These attributes were fairly abstract and could be applied to nearly all product categories (high quality for three of the brands, expensive for the fourth). Blank attributes were similar to those used in Study 1. Recent research suggests that because members of a brand category are usually products that have survived in the marketplace for some time, they may be strongly associated with the brand and therefore at least moderately typical of it (D. C. Smith & Park, 1992). Therefore, we used hypothetical brand extension products to obtain specific conclusion product categories that would be low in typicality. This aspect of the stimuli was held constant across low- and moderate-typicality conditions by using new brand extensions in both conditions. Participants were told that the products in the arguments included a number of potential, or recently introduced, extensions of the brand. The stimuli used in Study 2 are presented in Table 4.

Results

Manipulation checks. As expected, the low-typicality products were all rated significantly less typical of the brand name than were the products used in the moderate typicality arguments, all ts(47) > 4.30, ps < .001 (see Table 5). Further, each of the moderate typicality conclusion categories was rated near the midpoint of the scale (between 5.04 and 7.02), and each of the low-typicality conclusion categories was rated near the low end (between 1.52 and 2.96).

Tests of the inclusion effect. Results as a function of individual attribute, conclusion typicality, and brand condition are presented in Table 2. A significant inclusion effect is indicated by an argument strength difference score that is significantly greater than zero. For two of the brand replicates (Gucci and Sony), participants' overall judgments were consistent with the inclusion effect, F(1,47)

1 1 4 JOINER AND LOKEN

TABLE 4 Argument Strength Stimuli (Studies 2 and 3)

General Meaningful ~ , ~ , , d premise Conclusion Specific Conclusion Attribute Blank Attribute

Sony Televisions All Sony Mod. Typ.: cameras Are high quality Have Cryson A products

Low Typ.: bicycles Gucci Handbags All Gucci Mod. Typ.: sheets Are expensive Have Ulric

products (Study 2) Low Typ.: coffee Are high quality

tables (Study 3) Kraft Macaroni All Kraft Mod. Typ.: pizza Are high quality Have Multivar 8

and cheese products Low Typ.: ice cream

bars Jell-0 Gelatin All Jell-0 Mod. Typ.: cake Are high quality Have Proseen 10

products mixes Low Typ.: frozen

vegetables

Note. Mod. Typ. = moderate-typicality condition stimulus; Low Typ. = low-typicality condition stimulus.

> 20, p < .001. For these replicates across typicality and attribute conditions, participants rated the general conclusion argument as significantly stronger than the specific conclusion argument. Seven of the eight low-typicality conditions and one of the eight moderate typicality conditions were significantly higher than zero, t(47) > 2.85, ps < .05. Four out of eight salient attribute conditions and four out of eight blank attribute conditions yielded significant inclusion effects, 447) > 3.20, p < .05. On an individual response level, only 7 individuals (7.3% of the sample) did not make at least one set of judgments consistent with the inclusion effect.

Because the data were collected from different sets of participants, separate 2 x 2 (Typicality x Attribute Meaningfulness) repeated measures ANOVAs were conducted on the difference score data for each of the four brand replicates. The main effect for typicality was significant for all four brand name replicates, Fs(1, 47) ranged from 8.15 to 38.29, ps < .01. Low-typicality conclusions accentuated the inclusion effect (Ms = 1.09, 1.52, 3.66, and 3.08 for Kraft, Jell-0, Gucci, and Sony, respectively) relative to moderate typicality conclusions (Ms = -.17, -.28, 1.00, and -.21, respectively). In fact, looking at means across attribute conditions, only the four low-typicality conditions, as well as the Gucci moderate-typicality condition, yielded significant inclusion effects, r(47) > 2.35, ps < .05. Results for the attribute main effect were consistent with expectations. Across typicality levels, the meaningful attribute conditions yielded a significantly higher level of the effect than did the blank conditions for Gucci (2.93 vs. 1.73), F(1,47) = 8.75, p < .01,

THE INCLUSION EFFECT 1 15

TABLE 5 Similarity and Typicality Means (Study 2)

A. Painvise Similarity Ratings

Moderate Typicality Low Typicality

Sony specific Sony general Difference Gucci specific Gucci general Difference Kraft specific Kraft general Difference Jell-0 specific Jell-0 general Difference

B. Typicality Ratings

Premise Conclusion-Moderate Conclusion-Law

Sony Gucci Kraft Jell-0

"r(47) > 2.15, p < .05, general versus specific similarity scores. 't(47) > 8.75, p < .01, moderate-typicality versus low-typicality similarity difference score.

*r(47) > 4.30, p < .05, moderate-conclusion versus low-conclusion typicality ratings.

and was marginally significant for Sony (1.73 vs. 1.15), F(1,47) = 2.87, p < .lo, and Jell-0 (1.03 vs. .21), F(1,47) = 2.83, p < . lo, consistent with predictions. The effect was in the predicted direction in the fourth brand condition (Kraft) but failed to reach significance (.68 vs. .25), F(1,47) = 1 . 3 3 , ~ > .lo. Finally, these two main effects were qualified by a significant typicality by attribute interaction for two of the brands: Kraft, F(1, 47) = 7.54, p < .01; Gucci, F(l, 47) = 12.25, p <. 01. Follow-up contrasts found that the attribute main effect (i.e., meaningful > blank) was significant for moderate typicality conditions, ts(47) > 4.40, ps < .01, but not for the low-typicality conditions, ts(47) < 1.10.

In sum, the overall pattern of results for the difference score data was quite consistent across brand replicates. Although the brand replicates did vary in the overall size of the effects, all of the brands showed a significant inclusion effect in at least some conditions. The effect was especially pronounced when the specific conclusion contained a category low in typicality. Arguments that contained a meaningful (salient and abstract) attribute were sometimes more likely to result in

1 1 6 JOINER AND LOKEN

judgments consistent with the effect than were arguments with blank attributes. hi^ effect wits stronger for moderate-typicality conclusion conditions than for low-typicality conditions (perhaps because under low-typicality conditions, both

and blank attributes yielded inclusion effects). In Study 2, to account for participants' argument strength judgments and the

inclusion effect, we also measured the relative ability of premise-conclusion similarity and conclusion typicality by using judgment type (rather than partici- pants) as our unit of analysis. Separate samples of participants provided prem- ise-conclusion similarity ratings (n = 45) and specific conclusion typicality judg- ments (n = 47). Because two similarity scores (for Argument A and Argument B) were obtained for each judgment, a difference score was computed between the average premise-conclusion similarity of the general and the specific argument categories. For typicality, one judgment was made for each argument (e.g., the typicality of Gucci sheets to all Gucci products). Correlation analyses revealed that both measures predicted argument strength judgments, p < .05. The greater the difference in the premise-conclusion similarity judgments between the general (A) and specific (B) arguments (r = .60), and the lower the typicality of the product in the specific argument's conclusion (r = -.64), the more likely individuals were to make judgments consistent with the inclusion effect (participants generalized more to the general conclusion).

For the meaningful arguments (n = 8) and from a separate sample (n = 30), we also measured plausibility of the conclusions by asking participants to judge the likelihood that each conclusion (A and B) was true on an 11-category likelihood scale ranging from 0 (extremely unlikely) to 10 (extremely likely). (For blank arguments, the measure is ambiguous.) We computed the difference score between the plausibility scores for the general and the specific argument's conclusions. In determining the relative ability of plausibility, typicality, and similarity to predict argument strength, we noted that the three measures were highly intercorrelated: 4 8 ) = -.76 for the typicality-similarity and typicality-plausibility pairs, and .79 for the similarity-plausibility pair; our degrees of freedom were small. Keeping in mind these limitations, the three measures among meaningful attribute stimuli were all significantly related to the argument strength difference score, r(8) = -30, .76, and .72, all ps c .05, for typicality, similarity, and plausibility, respectively.

Discussion

Study 2 suggests that the inclusion effect is a rather robust counternormative phenomenon. The results using an interval scale measure of argument strength in this study were consistent with those using a forced-choice measure in Study 1. Significantly, the two arguments in apair were rated as equally strong in only 14.2% of the responses. We found that the inclusion effect was more pronounced in situations where the specific argument conclusion contained a low-typicality

THE INCLUSION EFFECT 1 1 7

category. It appears that the presence of a low-typicality conclusion in an argument increases the likelihood that participants base their judgments on a representation of a brand that is heavily influenced by its most typical products. Consumers may perceive the general category (e.g., all Jell-0 products) in terms of an abstract prototype, in which the brand's typical products are well represented and the atypical products are not. As a result, the degree of overlap between a specific category (e.g., Jell-0 gelatin) and the general category is perceived as greater than between a specific category (e.g., Jell-0 gelatin) and another atypical, specific category (e.g., Jell-0 frozen vegetables). The inclusion effect also tended to be stronger for arguments that included meaningful attributes. Less meaningful, blank attributes may have produced more analytic reasoning and explicit consideration of all of a brand's products (both typical and atypical), attenuating the effect.

Linguistic effects. A rival explanation for the inclusion effect is based on conversational or linguistic norms (Dulany & Hilton, 1991; Grice, 1975). Accord- ing to this interpretation, participants may have assumed that the experimenter intended the general (all Sony products) and the specific conclusions (Sony bicycles) to be interpreted as independent. In other words, all Sony products may have been interpreted as all Sony products except Sony bicycles. For four of the brands used in Studies 1 and 2 (and for which the inclusion effect was high), we added two questions following the argument strength judgments. The first asked how strongly the participant agreed or disagreed (on a scale from 0, strongly disagree, to 10, strongly agree) that the writer meant that all Sony products included a range of products such as Sony televisions, Sony videocassette recorders, and Sony bicycles. As shown in Table 6, these means range from 7.11 to 8.16, generally supporting our proposed account and not the conversational norm account. A second question (see, e.g., Dulany & Hilton) asked participants to select one of three options stating what the writer (i.e., experimenter) meant by the phrase "all Sony products": (a) "all Sony products excluding Sony bicycles", (b) ". . . including Sony bicycles", or (c) "... neither including nor excluding Sony bicycles." As expected, the proportion of participants who reported that the writer (experimenter) intentionally excluded the atypical products from the general category phrases was relatively small (Table 6). In other words, few participants interpreted the general and specific categories as independent. These linguistic interpretations are thus less plausible than our proposed explanation, which implicates premise~onclusion similarity and conclusion typicality as moderators.

STUDY 3

In Study 3, we examined whether the stimulus task moderates the effect-in particular, whether presenting the arguments sequentially (i.e., rating the general argument immedi-

TABLE 6 Conversational Norms (Alternative Explanation) Data (Study 2)

Brand Inclusion Effect Proportion Experimenter Meaning Agree-Disagree Scale (0-10)

Sony .789 Excluding 21.1 % Including 63.2% Neither 15.8%

Gucci ,895 Excluding 00.0% Including 73.7% Neither 26.3%

Kraft .842 Excluding 10.5% Including 68.4% Neither 15.8%

Jell-0 ,737 Excluding 26.3% Including 57.9% Neither 15.8%

ately prior to the specific argument) increases the similarity of the responses to the two judgments, attenuating the inclusion effect. If judgments about the specific argument (regarding the atypical extension) immediately follow judgments about the general argument (regarding all products in the brand category), the inconsistencies inherent in the inclusion effect should become more salient than when the two arguments are presented alone. Therefore, we assumed that presenting the two arguments in this order would produce greater conformity between them. We expected this conformity to manifest itself in one of two ways. The participants judging both general and specific arguments could increase the strength of the specific argument relative to the group who rated only the specific argument to conform to the general argument--that is, individuals' responses to the general argument would frame, or provide a context for, judgments of the specific argument (see, e.g., Schwarz & Strack, 1991). Alternatively, the participants could decrease the strength of the general argument to conform more closely to judgments of the specific argument. This second possibility could occur if participants were considering the arguments simultaneously (rather than sequentially), were reading both arguments prior to rating them, or were simply going back and changing their answers to the general argument after rating the specific arguments.

Method

In Study 3, three groups of participants completed argument strength judgments. One group rated the strength of a general argument first before rating the strength of the specific argument. A second group rated the strength of a series of specific conclusion arguments. A third group rated the strength of a series of general conclusion arguments. We expected that participants in the first condition, who had made a judgment about a general conclusion argument first, might rate the specific

THE INCLUSION EFFECT 1 19

conclusion argument as stronger than would participants in the second condition, who judged only the specific conclusion argument. Alternatively, participants in the first condition might rate the general argument lower than those in the third group because the inconsistencies inherent in the low strength ascribed to the specific judgment were assumed to be more salient. If either of these expectations was borne out, we expected a low rate of the inclusion effect in the first condition.

Design and measures. The procedure used in Study 3 was similar to that used in Study 2. A total of 63 undergraduate marketing students completed argument strength judgments using the stimuli from Study 2 with minor modifica- tions. The study was a 4 x 2 x 2 x 3 design. The first three factors were the same as in Study 2 and were all manipulated within-subjects (four brands: Kraft, Jell-0, Gucci, or Sony; typicality: moderate or low; attribute: meaningful or blank). Participants rated each of the arguments on the same 11-point argument strength scale described in Study 2. These judgments were followed by the similarity and typicality measures described in Study 1. Three argument strength questionnaires were administered between-subject. The first group completed a task that was very similar to that in Study 2. Participants rated 16 general-specific argument pairs. Unlike in Study 2, however, the general arguments were always presented prior to the specific arguments in each pair, and participants did not complete a forced- choice argument strength question. The second group of participants rated only the 16 specific conclusion arguments: 4 (brands) x 2 (levels of typicality) x 2 (attrib- utes). The third group of participants rated only the eight general conclusion arguments: 4 (brands) x 2 (attributes). The order of presentation of the arguments was randomized with the restriction that the eight meaningful arguments preceded the eight blank arguments. The stimuli used in Study 3 are presented in Table 4.

Results

Manipulation checks. As expected, the low-typicality products were all rated significantly less typical of the brand name than the products used in the moderate typicality arguments, all ts(20) > 3.25, ps < .001 (see Table 7), with one exception. Kraft pizza was not rated as significantly more typical than Kraft ice cream bars, t(20) = 1.5 1, p = .146. Further, there was significant variability in the ratings of the moderate typicality conclusion categories (ranging from 3.14 to 8.05), as compared to the ratings in Study 2. Each of the low-typicality conclusion categories was rated near the low end of the scale (between 1.43 and 2.90).

Judgment conformity tests. Comparing the judgments made by partici- pants in the single and paired argument conditions suggest that the general pattern

120 JOINER AND LOKEN

TABLE 7 Similarity And Typicality Means (Study 3)

A. Pairwise Similarity Ratings

Moderate Typicality Law Typicalify

Sony specific Sony general Difference Gucci specific Gucci general Difference Kraft specific Kraft general Difference Jell-0 specific Jell-0 general Difference

-- -

B. Typicality Ratings

Premise Conclusion-Moderate Conclusion-Low

Sony Gucci Kraft Jell-0

"(20) > 4.70, p c .05, general versus specific similarity scores. 't(20) > 2.60, p < .05, moderate-typicality versus low-typicality similarity difference score.

*t(20) > 3.25, p c .05, moderate-conclusion versus low-conclusion typicality ratings.

of results was in a direction opposite to predictions (see Table 8). The average strength ratings for the specific arguments decreased (rather than increased, as expected) in 12 of the 16 conditions when presented in the context of the general argument than when presented alone. The decrease was statistically significant in 2 of these 12 conditions, ts(40) > 2.03, ps < .05. The average strength ratings of the general argument increased rather than decreased in 13 of the 16 conditions when it was in the context of a second specific argument. Six of these 13 conditions (all Sony and Kraft stimuli) showed a significant increase, all ts(40) > 2.75, ps < .05. In other words, presenting the general argument, followed by the specific argument, heightened the differences between the argument strength judgments rather than diminished them.

Evidence of the inclusion effect. The argument strength ratings in the first group (see Table 9), in which both general and specific arguments were rated,

THE INCLUSION EFFECT 121

TABLE 8 Argument Strength Judgment Conformity (Study 3)

Specific Argument Strength Judgment General Argument Strength Judgment

Group I (Both Group 3 ("Speci$c'' Group I (Both Group 2 ("All" Arguments) Argument Only) Arguments) Argument Only)

Meaningful Attributes Moderate typicality Sony Gucci Kraft Jell-0 Low typicality Sony Gucci Kraft Jell-0

Blank Attributes Moderate typicality Sony Gucci Kraft Jell-0 Low typicality Sony Gucci Kraft Jell-0 - - - - --

*S~gn~ficant difference between means for Group 1 w~th corresponding mean In Group 2 or Group 3, t(40) > 2 0 2 , p < 05

showed that participants' judgments were consistent with the inclusion effect. Overall, across conditions, three of the four brands produced a significant inclusion effect, all Fs(l,20) > 9.50, ps < .05, whereas the fourth was marginally significant, Jell-0, F(1,20) = 3.90, p = .062. Across all the study cells, 10 of the 16 individual conditions produced significant effects, all ts(20) > 2.05, ps < .05. The attribute meaningfulness main effect (across the four brands) was nonsignificant, Ms = 1.29 and .76 for meaningful and blank, respectively; F(l, 20) = 2.21, p > .15, but went in the predicted direction (meaningful > blank) for each of the brands and ap- proached conventional levels of significance for the Kraft, F(l,20) = 3 . 4 6 , ~ = .077, and Gucci brands, F(1, 20) = 3.71, p = .069. The typicality main effect was significant across the four brands, Ms = 1.36 and .69 for low and moderate typicality, respectively; F(1, 20) = 4.48, p < .05, although the effect was signifi- cantly greater in the low-typicality condition than in the moderate-typicality

122 JOINER AND LOKEN

TABLE 9 Incidence of Inclusion Effect (Study 3)

Argument Strength Dtference Scores

Moderate Typicality Low Typicalily

Real Attributes Blank Attributes Real Attributes Blank Attributes

Sony .714* .67 1.91* 1.29* Gucci 1.29* 1.29* 2.33* .7 1 Kraft 1.43* .48 1.43* .67 Jell-0 .00 -.33 1.24* 1.29*

Note. n = 21 per cell. *Entry is significantly different from zero, t(20) > 2.05 at p < .05.

condition for the Jell-0 brand only, F(1,20) = 7.68, p = .012. The inclusion effect occurred more often for low typicality than for moderately-typical categories in the specific conclusion, consistent with the prototypicality explanation. No Typicality x Attribute interactions were significant at p < .05.

As in Study 2, the difference in the premise-conclusion similarity in the general and specific arguments was apparently a factor contributing to the incidence of the inclusion effect (see Table 7). Specifically, as the difference increased (i.e., the premise was seen as more similar to the general than to the specific conclusion category), the incidence of the effect increased.

Discussion

Surprisingly, emphasizing the relation between the two arguments decreased, rather than increased, participants' ratings of the specific argument's strength relative to a context in which the argument was presented alone. Even more important, presenting the arguments together increased, rather than decreased, the argument strength ratings of the general arguments-again, relative to a context where the argument was presented alone. As a result, we observed levels of the inclusion effect that were consistent with the previous two studies. These findings have at least two implications. First, the inclusion effect persists under conditions in which the salience of its illogic should have been higher. Second, the inclusion effect may be even greater under conditions in which both the general and specific arguments are presented together (vs. when only one of the two sets of arguments, general or specific, is shown).

Inducing participants to consider the implications of a target statement (premise) for the category as a whole did not increase their beliefs in the applicability of the target's implications for other specific (atypical) categories. These findings suggest "biased" representations or heuristics as causes of the inclusion effect. When

THE INCLUSION EFFECT 123

considering all of a brand's products in the context of a relatively atypical brand product, consumers may tend to use a representation of the brand that consists of one or more typical products and that excludes the brand's atypical products. For example, Study 3 presented a context that included a general brand category followed by an atypical branded product; it appeared to prompt consumers to create and use separate and contrasting brand representations rather than induce partici- pants to include the atypical products in their brand representation.

This would suggest that when consumers think about a brand category, the presence of an atypical branded product reinforces a brand representation that is dominated by the brand's most typical products. Rather than cause consumers to broaden their representation of the brand to include both typical and atypical products, this context may simply reinforce how different the atypical product is from the consumer's perceptions of the brand. In a sense, the consumer, although at some level acknowledging that the atypical product is a member of the brand category, does not seem to incorporate it into his or her thinking about the brand. When encountering the atypical product alone, its difference from the general brand category may be less dramatic and salient, accounting for the pattern of argument strength ratings observed in Study 3.

Significantly, the inclusion effect focuses only on the relative strength of the general and specific arguments. The inclusion effect is observed whenever a participant rates the general argument as stronger than the specific argument, regardless of the absolute level of strength of the arguments. However, it is useful to consider whether the effect is limited to only arguments of a certain absolute strength (e.g., only strong or only weak arguments). In Studies 2 and 3, an interval-scale measure of argument strength was used. This allowed us to examine the overall, or absolute, level of argument strength for our stimuli. Across the two studies the argument strength ratings for the specific arguments ranged from 2.27 to 6.54 on a 0 to 10 scale, and ratings for the general arguments ranged from 4.17 to 7.25 (see Table 8). There was no apparent relation between absolute levels of argument strength and the incidence of the effect.

The results of Study 3 show inclusion effect rates that are comparable to those of Studies 1 and 2. The findings show that participants persist in their counternormative reasoning even under conditions in which the inconsistencies should be more salient to them. The changed presentation of the arguments that was designed to induce participants to identify the inconsistency in rating the general argument as stronger than the specific argument did not reduce the incidence of the inclusion effect.

GENERAL DISCUSSION

Barsalou (1992) described the importance of category-based induction and why argument strength judgments can be critical: "On discovering that an entity exhibits

124 JOINER AND LOKEN

a new property, people may use a category to guide how far they project the property to other known entities" (p. 307). If the inferential link from premises to conclusions is strong enough, some of the conviction associated with the premises is transmitted to the conclusion (Osherson et al., 1986). When a property is true of a typical instance of a category, it is often assumed to be true of many other instances (Alba & Hutchinson, 1987; Rips, 1975). A particularly interesting and relevant question is, To how many other instances is the property extended or generalized? The inclusion effect is a counternormative phenomenon in which the property is more likely to be generalized to an entire general category than to a specific category (within the general category) that is low or even moderate in typicality. For example, our participants were more likely to generalize "high quality" to all Gucci products than to Gucci sheets on learning that Gucci handbags are high quality. We examined this phenomenon in three studies using different measures of argument strength as a means of determining the extent of people's generalizations.

Study 3 examined whether the incidence of the inclusion effect declined when the implications of the two argument strength judgments were salient to our participants (as compared to when they were not). Surprisingly, the inclusion effect was not reduced even when the two arguments (general and specific) were seen on the same page, with participants rating the strength of the general argument first. The context of the two judgments together led people to increase their strength ratings of the general argument, to decrease their ratings of the specific argument and, in fact, to provide judgments that tended to accentuate the inclusion effect. Making the judgments more salient may have heightened the distinction between people's "all brand products" representation (dominated by the brand's typical products) and the representations of other less typical products. This finding suggests that when atypical products appear within the context of the brand category, one could observe contrast effects such that the atypical product is viewed even more differently from the category than if it had been presented alone.

Although there are several rival explanations for the inclusion effect (e.g., conversational norms, conclusion validity, or plausibility), these were not generally supported in our three studies. Although we obtained mixed evidence for the conclusion plausibility explanation, even this account is equivocal and cannot fully account for the inclusion effects in our data. Nevertheless, these alternative expla- nations may guide future research that explores the multiple bases underlying the effect.

Consistent with the category induction models of Osherson et al. (1990), the similarity between the premise and conclusion categories was a strong predictor of argument strength judgments. The stronger the similarity between the premise and the general conclusion categories (vs. the specific conclusion categories), the greater the incidence of inclusion effects. However, the incidence of inclusion effects was mitigated when the specific conclusion category was more typical of the general category. Thus, people were less cautious about generalizing from one

THE INCLUSION EFFECT 125

specific category to another when the latter was moderately (vs. not at all) typical of the more general category. Perhaps consumers perceive the general category (e.g., all Jell-0 products) as an abstract prototype, in which the brand's typical products are well represented and the atypical products are not. Hence, the per- ceived overlap between a specific category (e.g., Jell-0 gelatin) and the general category may be greater than that between a specific category (e.g., Jell-0 gelatin) and another specific category that is less typical (e.g., Jell-0 frozen vegetables). Consistent with this interpretation, the results of Study 2 (and to a lesser extent Study 3) show more attenuation of the inclusion effect when a specific conclusion category was moderately typical of the general category (e.g., Jell-0 cake mixes) than when it was not at all typical (e.g., Jell-0 frozen vegetables).

Although similarity and typicality are strongly related, participants' comparison points vary for the two types of judgments. The typicality measure (e.g., how typical are Ralph Lauren belts as a Ralph Lauren clothing product) does not draw in the premises of the arguments. The similarity measure, of course, does (e.g., how similar are Ralph Lauren sweatshirts and Ralph Lauren belts vs. how similar are Ralph Lauren sweatshirts and all Ralph Lauren clothing products). Of importance, we did not examine whether similarity between the specific and the general conclusion category influences the inclusion effect. Rather, we focused on how the typicality of the specific conclusion as a member of the general conclusion category influences the incidence of the inclusion effect.

For the stimuli used in these studies, the premise categories (e.g., Gucci hand- bags, Sony televisions) were highly typical of the general conclusion category (e.g., all Gucci or all Sony products). Hence, the perceived similarity between the (prototypical) premise and specific conclusion categories may have been close to the perceived typicality of the specific to the general category (see Barsalou, 1985). Although the typicality of the specific conclusion could have been measured using pairwise similarity ratings of the conclusions in the two arguments, we chose to measure typicality (not similarity) of the specific argument's conclusion and premise~onclusion similarities in the two arguments. Further research might examine conditions under which the two constructs are uncorrelated (e.g., using atypical premise categories). However, consumer inferences based on atypical instances (e.g., in premises) may not be very common (see Rips, 1975).

Meaningfulness of stimuli. We received mixed support for the hypothesis that the incidence of the inclusion effect would be higher when the stimulus attributes describing the categories were more meaningful than when they were blank. The data in Study 1 were directionally consistent with the effect but were not statistically significant. In Studies 2 and 3, when all attributes tested were abstract and transferable and when the type of meaningful (salient) and blank attributes were held constant across brand names, we found that meaningful, salient

attributes led to higher rates of the inclusion effect than did blank attributes for three of the four brand categories in Study 2, p < .lo, and for two of the four brand categories in Study 3,p < .lo. In no instances did we find significantly higher effects for blank attributes.

Although both premise-conclusion similarity and conclusion typicality received empirical support as moderators of the inclusion effect, neither one can directly account for these attribute differences because the similarity and typicality meas- ures were independent of the attribute context. We speculate, however, that mean- ingful (vs. blank) attributes may be more likely to trigger people's prior category knowledge and the use of intuitive reasoning strategies. Perhaps our participants interpreted a general category (e.g., all Jell-0 products) in narrower terms (e.g., a prototype, a specific typical exemplar, or a set of similarity judgments) when the arguments had meaningful attributes. In contrast, hypothetical attributes may have encouraged logical, rule-based processing. Alternatively, conclusion plausibility may be a contributing factor underlying the inclusion effect for meaningful attrib- utes, a possibility that received some support in Study 2. Nevertheless, theequivocal nature of these attribute results should temper conclusions.

CONCLUSION

Marketing communications provide important informational inputs to consumers' preferences, decision making, and judgment processes. An important element of understanding these processes is understanding how consumers use categorical inductions to derive conclusions about branded products based on statements they hear from a variety of sources. The results suggest several implications for market- ing communication decisions. If our inclusion effect results are generalizable, a company may wish to focus its communication efforts on a particular, typical product with the knowledge that the effort is likely to be generalized to the overall brand category. If positive attributes are generalized to the broad brand category, the brand's consumer-based equity (Keller, 1993) should increase, enhancing its ability to leverage this equity with future brand extensions. This process may provide some insight into the means by which brand-specific associations are created (e.g., Broniarczyk & Alba, 1994a).

Our results also suggest that communications about typical products are less likely to be generalized to products atypical of the brand. In fact, presenting atypical products in the context of the brand category may even heighten (rather than bridge) the differences between typical and atypical products. Therefore, although typical products may benefit from inferences from communications for other products in the brand category, atypical products may need to have specific and separate communications directly targeted toward them. Future research might be directed

THE INCLUSION EFFECT 127

toward further understanding the cognitive bases for the inclusion effects, as well as toward understanding other categorical induction phenomena (see Kardes, 1993; Osherson et al., 1990) in consumer judgment processes.

ACKNOWLEDGMENTS

This research was funded in part by grants from the McKnight Fund at the Carlson School of Management (CSOM) and from the CSOM marketing department.

We thank Paul Herr, three Journal of Consumer Psychology reviewers, and Jim Ward for thoughtful comments on an earlier draft of this article.

REFERENCES

Aaker, D. A,, & Keller, K. L. (1990). Consumer evaluations of brand extensions. Journal ofMarketing, 54, 27-41.

Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood Cliffs, NJ: Prentice Hall.

Alba, J. W., & Hutchinson, J. W. (1987). Dimensions of consumer expertise. Journal c?fConsumer Research, 13, 41 1-454.

Barsalou, L. W. (1985). Ideals, central tendency, and frequency of instantiation as determinants of graded stmcture in categories. Journal of Experimental Psychology: Learning, Memory, und Cognition, 1 I, 629-654.

Barsalou, L. W. (1992). Cognitive psychology: An overview for cognitive scientists. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Boush, D. M. (1993). Brands as categories. In D. A. Aaker & A. L. Biel (Eds.), Brand equity and advertising (pp. 299-312). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Boush, D. M., & Loken, B. (1991). A process-tracing study of brand extension evaluation. Journal o$

Marketing Research, 28, 16-28. Boush, D. M., Shipp, S., Loken, B., Gencturk, E., Crockett, S., Kennedy, E., Minshall, B., Misurell, D.,

Rochford, L., & Strobel, J. (1987). Affect generalization to similar and dissimilar brand extensions. Psychology and Marketing, 4, 225-237.

Broniarczyk, S. M., & Alba, J. W. (1994a). The importance of the brand in brand extension. Journal of'

Marketing Research, 31. 214-228. Broniarczyk, S. M., & Alba, J. W. (1994b). The role of consumers' intuitions in inference making.

Journal of Consumer Research, 21, 393-407. Cherniak, C. (1984). Prototypicality and deductive reasoning. Journal r?fVerbal Learning and Verbal

Behavior, 23, 625442. Dawar, N. (1996). Extensions of broad brands: The role of retrieval in evaluations of fit. Journal (if

Consumer Psychology, 5, 189-207. Dawar, N., & Anderson, P. F. (1994). The effects of order and direction on multiple brand extensions.

Journal of Business Research, 30, 119-129. Dulany, D. E., & Hilton, D. J. (1991). Conversational implicature, conscious representation, and the

conjunction fallacy. Social Cognilion, 9, 85-1 10.

128 JOINER AND LOKEN

Grice, H. P. (1975). Logic and conversation. In P. Cole & J. L. Morgan (Eds.), Syntax and semantics: 3. Speech acts (pp. 41-58). New York: Academic.

Herr, P. M., Farquhar, P. H., & Fazio, R. H. (1996). Impact of dominance and relatedness on brand extensions. Journal of Consumer Psychology, 5, 135-159.

John, D. R., Loken, B., & Joiner, C. (1998). The negative impact of extensions: Can you dilute flagship products? Journal cfMarketing, 62, 19-32.

Johnson, M. D. (1984). Consumer choice strategies for comparing noncomparable alternatives. Journal of Consumer Research, 11,741-753.

Joiner, C., & Loken, B. (1994). Consumer inferences and family branding strategies: A demonstration of category-based induction. In C. Allen & D. R. John (Eds.), Advances in consumer research (Vol. 21, pp. 188-194). Provo, UT: Association for Consumer Research.

Kardes, F. (1993, February). Category-bused induction andproduct-line coherence. Paper presented at the American Marketing Association Winter Educators' Conference, Newport Beach, CA.

Keller, K. L. (1993). Conceptualizing, measuring, and managing customer-based brand equity. Journal ($Marketing, 57, 1-22.

Lakoff, G. (1987). Women, jire, and dangerous things: What categories reveal about the mind. Chicago: University of Chicago Press.

Loken, B., & John, D. R. (1993). Diluting brand beliefs: When do brand extensions have a negative impact? Journal (~Marketing, 57, 71-84.

Nakamoto, K., Maclnnis, D. J., & Jung, H. (1993). Advertising claims and evidence as bases for brand equity and consumer evaluations of brand extensions. In D. A. Aaker & A. L. Biel (Eds.), Brand equity and advertising (pp. 281-297). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Osherson, D. N., Smith, E. E., & Shafir, E. B. (1986). Some origins of belief. Cognition, 24, 197-224. Osherson, D. N., Smith, E. E., Wilkie, O., Lopez, A,, & Shafir, E. (1990). Category-based induction.

Psychological Review, 97, 185-200. Park, C. W., Milberg, S., & Lawson, R. (1991). Evaluation of brand extensions: The role of product

feature similarity and brand concept consistency. Journal of Consumer Research, 18, 185-193. Peters, J. T., Hammond, K. R., & Summers, D. A. (1974). A note on intuitive vs. analytic thinking.

Organizational Behavior and Human Performance, 12, 125-1 3 1. Rips, L. (1975). Inductive judgments about natural categories. Journal of Verbal Learning and Verbal

Behavior, 14, 665481. Romeo, J. B. (1990). The effect of negative information on the evaluation of brand extensions and the

family brand. In R. H. Holman & M. R. Solomon (Eds.), Advances in consumer research (Vol. 18, pp. 399-406). Provo, UT: Association for Consumer Research.

Rosch, E., & Memis, C. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7, 573405.

Rothbart, M., & Lewis, S. (1988). Infening category attributes from exemplar attributes: Geometric shapes and social categories. Journal of Personality and Social Psychology, 55, 861-872.

Schwarz, N., & Strack. F. (1991). Context effects in attitude surveys: Applyingcognitive theory to social research. In W. Stroebe & M. Hewstone (Eds.), European review of social psychology (Vol. 2, pp. 31-50). Chichester, England: Wiley.

Shafir, E. B., Smith, E. E., & Osherson, D. N. (1990). Typicality and reasoning fallacies. Memory and Cognition, 18, 229-329.

Sloman, S. A. (1993). Feature-based induction. Cognitive Psychology, 25, 231-280. Smith, D. C., & Park, C. W. (1992). The effects of brand extensions on market share and advertising

efficiency. Journal of Marketing Research, 29, 296-313.

THE INCLUSION EFFECT 129

Smith, E. E., Lopez, A,, & Osherson, D. N. (1992). Category membership, similarity, and naive induction. In A. F. Healy, S. M. Kosslyn, & R. M. Shiffrin (Eds.), From learning processes to cognitive processes: Essays in honor of William K. Estes (Vol. 2, pp. 181-206). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Tversky, A., & Kahneman, D. (1982). Judgments of and by representativeness. In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgmeni under uncertainty: Heuristics and biases (pp. 84-98). New York: Cambridge University Press.

Accepted by Dipankar Chakravarti.