OUP user menu

Editors

Darren Dahl (Editor in Chief)Eileen FischerGita JoharVicki Morwitz

3.125
5.003
16 out of 115

Consumer and Market Drivers of the Trial Probability of New Consumer Packaged Goods

Jan-Benedict E. M. Steenkamp , Katrijn Gielens
DOI: http://dx.doi.org/10.1086/378615 368-384 First published online: 1 December 2003

Abstract

We examine the effect of consumer and market factors on the trial probability of new consumer packaged goods. We distinguish between three sources of variation in consumer trial probability: (1) within new products, across consumers; (2) within new products, over time; and (3) across new products. Hypotheses are developed for the different variables concerning their likely effect on trial probability. The hypotheses are tested on weekly household-panel scanner data on the occurrence and timing of first purchases for 239 new consumer packaged goods over a 52-week period after introduction for a sample of over 3,500 consumers. We combine these household panel purchase data with consumer questionnaire data, retail scanner data, data on advertising expenditure, and expert ratings. We find support for most hypotheses. One of our main findings is that the effects of the consumers' personal makeup on the probability that they will try the new product are systematically moderated by elements of the marketing strategy associated with the new product and by category characteristics. The extensive data set provides a strong context for the generalizability of the findings.

  • Advertising
  • Diffusion, Innovation, Technology
  • Variety Seeking/Product Trial
  • Panel Data Analysis

The consumer purchase environment changes continuously. Each year, thousands of new products are introduced in the United States and elsewhere. Thus, decisions concerning whether to purchase a new product constitute an important component of consumer behavior. However, relatively little research has examined the adoption decision and the consumer and market factors influencing these individual consumer decisions (Moreau, Lehmann, and Markman 2001).

The objective of this article is to increase our understanding of the factors that affect the consumer adoption decision and, in this way, to fill an important void in our knowledge of consumer decision making. In our study, we focus on consumer packaged goods (CPGs). For these types of products, the term “adoption” sometimes refers to the first purchase, and sometimes it is interpreted more loosely to refer to the first few purchases. We focus on the decision concerning the first, or “trial,” purchase. This decision typically is associated with higher risk and lower product knowledge, while reactions by the social system are also less clear (Rogers 1995). Moreover, repeat purchases are contingent on trial purchase to have taken place, while trial is not contingent on repeat.

The contribution of this study is twofold. First, although the consumer behavior literature recognizes both consumer and market factors as important drivers of the trial probability of new products (e.g., Gatignon and Robertson 1991), adoption research has typically focused either on consumer factors or on marketing variables (e.g., models like ASSESSOR). However, ignoring either type of factor can seriously bias one's conclusions about adoption processes (e.g., Van den Bulte and Lilien 2001). We study simultaneously consumer and market factors, including their main and interaction effects, on the trial probability of new CPGs. We examine these effects in a systematic fashion, developing and testing theory-based hypotheses. The focus in this study will be on understanding and explanation rather than on forecasting or normative prescription.

Second, our data set on the occurrence and timing of first purchases for 239 new CPGs over a 52-week period after introduction for a sample of over 3,500 consumers allows for empirical generalizations regarding the factors underlying the consumer trial probability of new products, at least within the domain of CPGs. As Bass and Wind (1995, p. G1) pointed out: “empirical generalizations are the building blocks of science.”

Drivers of Consumer Trial Probability

The focal variable in our study is the probability that the consumer tries out the new CPG. We distinguish between three sources of variation in trial probability. A first source of variation in trial probability is within new products, across consumers. The trial probability of any new product is predicted to differ systematically between consumers. We consider personal and interpersonal dispositions as well as sociobehavioral covariates (Rogers 1995). There is evidence that all three types of characteristics play a role in consumer behavior with respect to CPGs (e.g., King and Summers 1970; Steenkamp and Burgess 2002). A second source of variation in trial probability is within new products, over time, because of time-varying marketing communication tactics. The intensity of various communication tools varies over time, and its value in a specific time period is expected to affect the trial probability in that period (Hanssens, Parsons, and Schultz 2001).

A third source of variation in trial probability is between new products. Some products achieve on average a higher trial probability among consumers than other products. Why is this the case? In our model, we concentrate on two groups of market factors: the marketing strategy associated with the new product and category characteristics (Gatignon and Robertson 1991). Thus, we examine two types of marketing effects: the effect of time-varying communication variables as a source of variation in the trial probability within new products, and the effect of the time-invariant (at least over the period considered; Van den Bulte 2000) marketing strategy as a source of variation in the average trial probability between new products. Both types of influences should be studied in an integrated fashion to achieve a fuller understanding of the trial process (for a similar argument, see Van den Bulte 2000). Figure 1 shows the conceptual framework guiding our research, including the specific constructs studied, and relates the constructs to the three sources of trial probability. The hypotheses underlying this framework are discussed subsequently.

Figure 1

Conceptual Framework

Note.—H1–H7 = hypotheses 1–7. Consumer characteristics vary within new products across consumers, marketing communication variables vary within new products over time, while marketing strategy variables and category characteristics vary between new products.

Variation in Trial Probability within Products

Dispositional Innovativeness

Dispositional innovativeness is the predisposition to purchase new products and brands rather than to remain with previous choices and consumption patterns (Steenkamp, ter Hofstede, and Wedel 1999). We conceptualize dispositional innovativeness as a generalized consumer trait that exerts a positive effect on the trial probability of new offerings across the broad spectrum of consumer goods and services. Its hypothesized generalized effect results from the systematic personality structure underlying this construct. Consumers who are higher on dispositional innovativeness have been found to be higher on optimum stimulation level, openness to change, independence, risk taking, venturesomeness, and tolerance of ambiguity, and lower on dogmatism and conservation (Foxall 1988; Raju 1980; Steenkamp et al. 1999). New products across the wide range of goods and services have the potential to cater to this psychological profile by providing some change of pace, stimulation, and risk through a change from established patterns (Berlyne 1960; Steenkamp and Baumgartner 1992). Previous research has related some of these general personality constructs directly to adoption behavior (e.g., Rogers 1995), but their effect on new product adoption may be primarily through dispositional innovativeness (Joachimsthaler and Lastovicka 1984; Raju 1980; Steenkamp et al. 1999).

Market Mavenism

Market mavens can be regarded as generalized opinion leaders. They are individuals “who have information about many kinds of products, places to shop, and other facets of markets, and initiate discussions with consumers and respond to requests for market information” (Feick and Price 1987, p. 85). Market mavenism can influence trial probability in two complementary ways (Feick and Price 1987). Market mavens' general marketplace expertise should lead them to earlier awareness of new products. Second, market mavens actively acquire new information about the marketplace for transmission to other consumers. New products are an important component of the changing marketplace. By purchasing and consuming the new product, the market maven acquires product knowledge that comes from actual experience (Alba and Hutchinson 1987). Product consumption is an especially important source of information if the product contains many experience attributes, as is typically the case for CPGs (Moorthy and Zhao 2000).

Consumer Susceptibility to Normative Influence

Consumers differ in the degree to which they are influenced by the norms of the social system, that is, in their susceptibility to normative influence. This refers to a consumer's tendency to conform to the expectations of others (Bearden, Netemeyer, and Teel 1989). Burt's (1973) sociological research suggests that individuals who are more susceptible to normative influence are less willing to make an adoption decision until it is clear that a majority of relevant others supports the new concept. Based on the above argumentation, we hypothesize:

  • H1: The trial probability of a new product is higher among consumers who are

    • (a) higher on dispositional innovativeness;

    • (b) higher on market mavenism; and

    • (c) lower on susceptibility to normative influence.

What are the relations between these three consumer dispositions? Market mavens should generally have a higher inclination to purchase new products in order to acquire and transmit new product information and usage experience to others (Feick and Price 1987), while consumers high on dispositional innovativeness tend to be less concerned about social approval (Midgley and Dowling 1978). Market mavens should take care not to violate the norms of the social system in order to be able to perform their social communication role effectively (Rogers 1995). This suggests that market mavenism is positively correlated with dispositional innovativeness and susceptibility to normative influence, while the latter two constructs are negatively correlated. However, because of different and partially contradictory underlying motivations, we neither expect these correlations to be so high that the three constructs are expressions of the same construct nor that they are necessarily causally related. Consumers high on dispositional innovativeness tend to be inner-directed, independent, venturesome, and risk and stimulation seeking; consumers high on market mavenism tend to be other-directed, well integrated into peer groups, and high on social activity; and consumers high on susceptibility to normative influence are low on self-esteem and high on motivation to comply and on attention to social comparison.

The Effects of Time-Varying Marketing Communication

We examine the role of two key marketing communication instruments that typically vary considerably over time (Hanssens et al. 2001): mass advertising, and feature and in-store display. Mass advertising is effective in building awareness, conveying product information, and countering the adverse effect of forgetting (memory decay; Naik, Mantrala, and Sawyer 1998). Consequently, heavier recent advertising should positively affect trial probability. Feature advertising (advertising in the store flyer and local door-to-door newspapers) and in-store displays build new product awareness and influence the trial decision because they are present at the point of purchase (Papatla and Krishnamurthi 1996).

  • H2: The trial probability of a new product is positively affected by

    • (a) more intensive recent advertising; and

    • (b) more intensive recent feature and display promotion.

Variation in Trial Probability between Products: The Role of the Marketing Strategy

Goldenberg, Lehmann, and Mazursky (2001) found that market success—defined as substantial positive financial results—is lowest when novelty is low and highest if the new product is of moderate novelty. Although positive financial results are not the same as trial probability, a higher trial probability may be expected to translate into better financial results. Hence, their work suggests an inverted U-shaped relation. Goldenberg et al. (2001) explained this finding by hypothesizing that really novel products are too difficult to understand (too complex), while incrementally new products are too trivial (offer too little advantage).

A strong brand name commanding high levels of awareness and positive associations, high sustained advertising expenditure, and feature and display efforts can all three act as signal for the quality of a newly introduced product (Milgrom and Roberts 1986), and quality is a key factor in consumer purchase decisions (Shankar, Carpenter, and Krishnamurthi 1998). The rationale behind these default-independent signals (Kirmani and Rao 2000) is that the firm spends money up front, expecting to recover it through future sales. Marketing investments in them will be lost if the firm cheats on its assurance of quality, and therefore rational consumers can reliably use these signals to infer the new product's quality (Klein and Leffler 1981; Milgrom and Roberts 1986).

All things equal, economic theory suggests that the higher the average price of the new product relative to alternatives in the product category, the lower the inclination of consumers to buy the product. Finally, the higher the average distribution coverage (i.e., percentage of the retail outlets selling the new product during the introduction period), the higher the new product's trial probability is expected to be. In sum, we hypothesize:

  • H3: The average trial probability of a new product is

    • (a) related to its novelty according to an inverted U;

    • (b) higher when it is introduced using a strong brand;

    • (c) higher when it is supported by heavy average advertising expenditure;

    • (d) higher when it is supported by heavy average feature and display promotion;

    • (e) lower when its average relative price is high; and

    • (f) higher when it has high average distribution coverage.

Variation in Trial Probability between Products: The Effects of Category Characteristics

Companies find it increasingly difficult to create awareness for their new products because of the clutter of brands in the marketplace. Weber's law suggests that the larger the number of brands in the product category, the less the consumer notices changes in that category, including changes in category composition resulting from new product introductions. Indirect evidence is provided by Broniarczyk, Hoyer, and McAlister (1998). They found that consumers did not notice moderate changes (i.e., a reduction) in the number of items offered in a category.

Intense competitive reactions to advertising attacks by competing brands may produce interference effects in memory as consumers find it more difficult to remember which ad is associated with which brand in the product category. Keller (1987) showed experimentally that exposure to ads for competing items reduces consumers' ability to remember the advertised item. Memory interference especially hurts new products since awareness and knowledge still need to be created rather than “merely” maintained. Thus, in categories characterized by substantial reactive conduct to advertising attacks by other brands, trial probability should be lower.

In categories characterized by high advertising intensity, perceived product differentiation is higher, which should make consumers less responsive to other offerings, including new products (Mela, Gupta, and Jedidi 1998). In addition, high category advertising spending constitutes a severe barrier to entry, as it increases the capital required to create awareness (Shankar 1999). Hence, it will be more difficult to gain a foothold, and trial probability should be lower for products introduced in such categories.

Categories that are typically bought on impulse might exhibit, on average, a higher trial probability. These categories typically require less cognitive elaboration and planning prior to purchase. The decision is often made in the store on a whim, where there is an immediate opportunity to act on the urge (Narasimhan, Neslin, and Sen 1996). Finally, interpurchase cycles tend to be longer for categories that can be stockpiled more easily because of higher inventory levels (Narasimhan et al. 1996). Thus, consumers are less often “in the market,” which should reduce trial probability. To summarize:

  • H4: The trial probability of a new product in a category is higher

    • (a) the lower the number of brands in that category;

    • (b) the less intense the competitive reactions toward advertising actions in that category;

    • (c) the lower the advertising intensity in that category;

    • (d) if the category is characterized by a higher degree of impulse buying; and

    • (e) if the category is less easy to stockpile.

Moderating Role of Marketing Strategy and Category Characteristics on Consumer Dispositions

Dispositional Innovativeness

Stimuli high on novelty have a higher potential to provide stimulation, risk, and change from established patterns than do incrementally novel stimuli (Berlyne 1960). This appeals especially to consumers higher on dispositional innovativeness who are less risk averse, more open to change, and seek more stimulation (Steenkamp et al. 1999). Thus, one would expect that the more novel the product, the greater the difference in trial probability between more and less innovative consumers.

Wiggins and Lane (1983) demonstrated analytically that risk-averse consumers should purchase heavily advertised products because high advertising outlays not only signal higher expected quality but also lower quality risk. Consumers high on innovativeness are less risk averse (Foxall 1988), and hence we expect that the difference in effect between more and less innovative consumers will be stronger, the lower the average advertising expenditure for the new product.

Consumers higher on dispositional innovativeness are more impulsive (Foxall 1988) and, hence, can be expected to engage more often in impulse buying. Impulse buying can be highly stimulating (Rook 1987), which is also appreciated by these consumers (Steenkamp and Baumgartner 1992). Thus, we expect that the effect of innovativeness on actual trial behavior will be stronger in categories in which purchases are often made without advance intention or plan. Further, consumers have the most difficult time resisting the impulse buying urge in the moments following the encounter with the product (Rook 1987). Feature and display activities are effective instruments to stimulate impulse buying. They are among the most important means for bringing the product to the consumer's attention in the store (Papatla and Krishnamurthi 1996), where the consumer has the opportunity to act immediately on the urge. Therefore, we expect that feature and display activity will be more effective in increasing the trial probability of people high on dispositional innovativeness than among less innovative people.

Finally, it is expected that for more innovative consumers a higher purchase price is less of a barrier toward purchasing the new product. The utility of the new product will be higher for more innovative consumers because, apart from its intrinsic qualities, it offers potential for stimulation and relief from boredom. A higher relative price also entails higher perceived risk (Derbaix 1983), which should reduce the trial probability of risk-averse, less innovative consumers.

  • H5: The positive effect of a consumer's dispositional innovativeness on trial probability is

    • (a) stronger for more novel products;

    • (b) weaker when average advertising expenditure for the new product is higher;

    • (c) stronger in categories characterized by a higher degree of impulse buying;

    • (d) stronger when average feature and display activity for the new product is higher; and

    • (e) stronger when the relative price of the new product is higher.

Market Mavenism

A key role played by market mavens is that of transmitter of information on the marketplace. New product purchases are an important source of such information. However, given constraints associated with time, cognition, and finances, each new product is not equally likely to be tried out by market mavens. From a cost-benefit perspective (Ratchford 1982), the likelihood of trial purchase is greater if the information acquired through purchase is more valuable to others in the peer group. We submit that the new information is more valuable when it concerns a product introduced under a strong brand and less valuable when the product is introduced in an impulse category. Strong brands command higher awareness, and more consumers will have a relationship with—and interest in—such a brand (Keller 1993). New information about such brands will be valued relatively more by the peer group. Hence, market mavens have a stronger incentive to purchase new products introduced under a strong brand. Purchases in “impulse” categories are typically made without much careful thought or use of available information (Narasimhan et al. 1996). Therefore, in such categories, the information transmitter function of market mavens may be valued less by other consumers. Consequently, from a cost-benefit perspective, market mavens will have less incentive to purchase new products in these categories.

Market mavens' involvement with the marketplace and their role as information transmitter will lead them to give more attention to marketing communications for the new product than will consumers low on market mavenism (Feick and Price 1987). This implies that more advertising, or feature and display, are needed for consumers low on market mavenism to achieve a given likelihood of purchase. Or, to put it differently, a given level of advertising or feature and display effort will have a greater effect on trial probability of market mavens.

  • H6: The positive effect of a consumer's market mavenism on trial probability is

    • (a) stronger when the product is introduced under a strong brand name;

    • (b) weaker in categories characterized by a higher degree of impulse buying;

    • (c) stronger when the average advertising expenditure for the new product is higher; and

    • (d) stronger when the average feature and display activity for the new product is higher.

Consumer Susceptibility to Normative Influence

Consumers high on susceptibility to normative influence are hypothesized to be less likely to try out a new product. This effect is expected to be stronger if the purchase is perceived to be riskier (Bearden and Etzel 1982). Products higher on novelty are typically more risky as their social acceptance is less obvious, and they have more potential of violating norms (Steenkamp et al. 1999). Hence, we hypothesize that the negative effect of susceptibility to normative influence on trial probability will be stronger (i.e., more negative) when novelty of the new product is higher.

We further theorize that heavy sustained advertising is interpreted by consumers as a signal for approval of the new product by the social system. After all, how can the marketer have confidence in the sales potential of the new product if the product is not accepted socially? In line with this argument, Kirmani and Wright (1989, p. 344) posited that heavy advertising expenditure may be perceived by consumers as “a sign of … probable social acceptance.” Given the important role of advertising in contemporary society, heavy advertising may also “legitimize” the new product vis-à-vis one's social system. Hence, we expect that the negative effect of susceptibility to normative influence on trial probability will be weakened for more heavily advertised new products.

  • H7: The negative effect of a consumer's susceptibility to normative influence intensity on trial probability is

    • (a) stronger for more novel products; and

    • (b) weaker when average advertising expenditure for the new product is higher.

Covariates

Five covariates are included in our framework (fig. 1) and analyses. Heavy users of the category have heavier ongoing category needs. Hence they are more often “in the market,” which should increase trial probability (Gatignon and Robertson 1991). Three key sociodemographics identified in the literature (Gatignon and Robertson 1985)—age, household income, and level of education—are also included. Previous research suggests that these sociodemographics might affect adoption behavior, although the direction of their effect is quite inconsistent across studies (Rogers 1983). Finally, we control for time-varying advertising activity by competing brands in the category. Heavy contemporaneous competitive advertising should reduce the trial probability for any new product introduced in the marketplace (for a similar argument, see Carpenter et al. 1988). Controlling for these covariates provides a stronger test of our hypotheses and produces more accurate parameter estimates for our focal constructs.

Method

The consumer characteristics and the (time-varying) marketing communication tactics vary within new products, whereas the marketing strategy and category characteristics are constant within new products but vary between new products. Hence, we proceed in two stages. First, a hazard model is estimated for each new product introduction separately to test hypotheses concerning within-product drivers of the trial probability (hypotheses 1–2). Second, estimates of parameters obtained in the first stage of the hazard probability analysis (in this case, baseline hazard, effects of consumer characteristics) are regressed on marketing strategy instruments and category characteristics to test the hypotheses concerning differences in trial probabilities between new products (hypotheses 3–4), to assess the moderating effects of marketing strategy and category characteristics on consumer dispositions (hypotheses 5–7), and to identify moderating effects on the consumer covariates in an exploratory way (see below).

Estimation of Within-Product Effects on Trial Probability

In a first stage, a proportional parametric exponential hazard model (Jain and Vilcassim 1991) was estimated for every new product.1 This results in a one-parameter, time-invariant specification of the hazard probability. The trial probability λij(t) is expressed as an exponential function (to ensure nonnegativity) of consumer characteristics and time-varying marketing communication variables: Embedded Image where λ0ij represents the baseline hazard with a prescribed distribution, and i, j, and t refer to new product introduction (i = 1, …, 239), consumer (j = 1, …, 3,687), and time period (t = 1, …, 52), respectively. The labels DI, MM, and CSNI are dispositional innovativeness, market mavenism, and consumer susceptibility to normative influence, respectively. The labels UsaInt, Age, Income, and Educ represent the sociobehavioral covariates category usage intensity, age, household income, and education, respectively; Advert denotes advertising spending; FDt refers to feature and display effort; and CompAdvt represents the covariate competitive advertising in period t. We mean-centered all predictors within products so that the baseline hazard can be interpreted as the mean trial probability of the new product (for a similar argument, see Van den Bulte 2000). The parameter estimates for the factors influencing the trial probability are obtained through maximum likelihood estimation whereby information is taken into account on both completed and censored observations.

We recognize that even though the underlying trial process is continuous, our data do not record the exact duration until trial but only the interval, that is, the week after introduction in which trial took place or was censored. The contribution to the likelihood function of a trial that occurred after tij weeks is therefore given by the difference of the survivor functions S(tij − 1) − S(tij) (rather than the density function). The likelihood contribution of an observation censored in interval [tij − 1, tij) is taken to be S(tij − 1). Assuming the trial probability to be exponentially distributed, the expression for the log-likelihood function for a new product introduction i for a sample of N consumers becomes: Embedded Image where dij is an indicator variable taking the value of one when the observation is censored and zero otherwise, and tij represents the observed duration of trying new product i by consumer j.

To control for heterogeneity not accounted for by the predictors, we allow the baseline hazard λ0ij to vary across different consumers according to a gamma distribution g0ij; ri, ai); E0ij) is given by ri/ai, which is a dependent variable in the second stage analysis. Mixing equation 2 with the gamma distribution, the following expression for the log-likelihood function is obtained (Vanhuele et al. 1995): where Xij(tij) represents the set of fixed and time-varying predictors as specified in equation 1, and Bij(tij) = ∑tijk = 1 eXij(k)], Embedded Image

Estimation of Between-Product Effects on Trial Probability

To test hypotheses 3–7, we related the baseline hazard and the parameter estimates for the consumer dispositions obtained in 239 individual hazard probability analyses to marketing-strategy variables and category characteristics (eqq. 4–7). For the consumer covariates, we examined interactions with marketing strategy and category characteristics in an exploratory way. Exploratory interactions were retained if they were theoretically (potentially) meaningful and statistically significant. These exploratory interactions are shown in equations 8–11. Embedded Image Novel, BrandStr, A̅d̅v̅e̅r̅, F̅D̅, P̅r̅i̅c̅e̅, and D̅i̅s̅t̅r̅ refer to the (time-invariant) marketing-strategy variables: that is, degree of novelty of the product, brand strength, average budget spent on advertising, average feature and display effort, average relative price, and average distribution coverage, respectively. NrBrand, CatReacAdv, CatAdInt, Impulse, and Stock denote the product category characteristics number of brands, competitive advertising reactivity, advertising intensity, the propensity of impulse buying, and ability to stockpile, respectively. Parameters β1i, β2i, and β3i represent, respectively, the estimated effect of dispositional innovativeness, market mavenism, and susceptibility to normative influence on the trial probability of new product i. Parameters β4i, β5i, β6i, and β7i are the estimated effects of the covariates usage intensity, age, income, and education. The term ϵki (k = 0, …, 7) refers to the normally and independently distributed error term for new product i. Because the dependent variables in equations 4–11 are estimated parameters, characterized by a differing degree of estimation accuracy, we weighted both the dependent and independent variables by the inverse of the estimated standard error of the dependent variable (see Narasimhan et al. [1996] and Shankar et al. [1998] for a similar approach). To correct for possible correlations between corresponding disturbances from the different equations, we estimated equations 4–11 jointly using seemingly unrelated regressions. The correlations between residuals across equations 4–11 range from −.05 to .35.

Data

Sample Description

The hypotheses were tested using a large set of 239 new CPGs that were introduced in the Netherlands in 1997–98. These product introductions occurred in 93 categories, which cover a wide range of human foods, pet foods, beverages, and personal care and household care products. Consumer purchases for each of the 239 new products were monitored in a household panel of 3,687 consumers during a period of 12 months after the new product was introduced. Industry analysts consider the first 12 months after introduction critical for success or failure in the CPG industry (Ernst & Young and AC Nielsen 2000). For every new product, we recorded in which week after introduction each of the consumers in our panel purchased the new product for the first time. When a consumer did not purchase the product during the observation period of 52 weeks, we right-censored the observation, and the relevant information on these consumers is therefore also incorporated in the analysis. Purchase records were provided by the market research agency Europanel.

Measures

Source of the Data

Data on the three consumer dispositions were collected in a special questionnaire administered to the household panel members. Data on the sociobehavioral variables were supplied by Europanel. Expert judgments on brand strength and novelty were provided by category management experts of Europanel and IRI. Both agencies have extensive knowledge on CPGs. Advertising data were acquired from the Bureau Budget Controle advertising research agency. With two exceptions (category impulse buying and ability to stockpile), all other information on marketing mix and category characteristics was obtained from IRI's retail scanner data.

Consumer Characteristics

The measurement instrument for dispositional innovativeness was a revised and extended version of the scale used by Steenkamp et al. (1999). Susceptibility to normative influence was measured by the eight-item scale developed by Bearden et al. (1989). Market mavenism was measured by four items, adapted from Feick and Price (1987). See the appendix for the items. Items were rated on five-point scales, with end poles of “completely disagree” and “completely agree.” The covariate category usage intensity was measured by the volume (in 100 g or its liquid equivalent in milliliters) a person bought in the category over the past 12 months, based on his or her household scanner purchase record. The age of the respondent was measured in years. The total monthly household income was measured in euros. The level of education was categorized as high or low, based on the Dutch educational system.

Marketing Mix Variables

Information on the degree of novelty of the new product was collected among a group of experts from IRI and Europanel. Experts also have been used in other recent research to assess the degree of novelty of a new product (e.g., Goldenberg et al. 2001). The degree of novelty of a new product was measured by two bipolar items, referring to the extent to which the product was new and unique (Henard and Szymanski 2001). Each new product was rated independently by a varying group of two-to-six experts (out of a total group of 25 experts). The ratings for each item were averaged across experts. The novelty score of a new product was obtained by averaging the means on the two items. The correlation between the average scores on the two items was .91.

Brand strength was evaluated by six experts from IRI and Europanel. Each expert independently rated the brand on two seven-point bipolar items, pertaining to whether the brand commands high awareness and has many strong, positive associations (Keller 1993). The interjudge reliability was .92. The ratings for each item were averaged across experts. The brand strength score was obtained by averaging the means on the two items. The correlation between the average scores on the two items was .91.

The advertising expenditure on the new product was expressed in €1,000 and covered television, radio, and print media. Feature and display effort was measured by the percentage of total volume of the new product sold by retail outlets offering feature and in-store display support. The distribution coverage was operationalized as the percentage of the retail outlets selling the new product (weighted by market share in the category). The relative price was computed as the average shelf price (weighted by distribution) of the new product divided by the average price of the category. These latter three measures are standard IRI measures (e.g., Nijs et al. 2001). All these marketing variables were available on a weekly basis. We created time-invariant predictors by averaging a particular marketing mix variable across the 52 weeks considered in this study (Van den Bulte 2000). The averaged (time-invariant) predictors are used in equations 4–11.

The (mean-centered) weekly observations on feature and display and advertising can be used as time-varying communication variables in equation 1. To capture contemporaneous and delayed advertising effects we used an adstock specification, which incorporates short-run lagged effects of advertising (Hanssens et al. 2001). It allows us to estimate a single coefficient of short-run advertising effectiveness (eq. 1). To account for the impact of the past and current advertising on trial, we created a weighted variable, following the procedure of Carpenter et al. (1988). We first calculated the cross-correlations between the aggregated trial series for every new product and the relevant advertising expenditure series. Almost consistently, we found positive, significant cross-correlations for the current advertising spending and for the first lag. Moreover, these effects were approximately equal in size. We therefore constructed the adstock variable using an equal weighting scheme for current and one-period-lagged advertising spending. A similar specification was used by Carpenter et al. (1988). Thus, in equation 1, the predictor Adver was operationalized as the average of the advertising expenditure in weeks t and t − 1. For the time-varying competitive advertising covariate, we followed a similar adstock approach. For feature and display, no stock variable was created as their effect largely occurs in the week in which the promotion takes place (Van Heerde, Leeflang, and Wittink 2000).

Category Characteristics

The product categories were based on IRI's classification of product types. The assignment of new products was typically straightforward but was discussed in detail with managers of IRI and Europanel. For the degree of category impulse buying and ability to stockpile, we used the scores developed by Narasimhan et al. (1996). The other category characteristics were taken from Nijs et al. (2001). These authors provide an extensive rationale for their operationalization and give technical details on their computation. Briefly, the number of brands in a category refers to the number of brands with a market share exceeding 1% over at least a three-month period. The magnitude of competitive advertising reactivity was based on the extent to which the top five brands in a category react to each other's advertising attacks with advertising of their own, expressed as elasticities. The reactivity variable was coded so that a higher score implies a stronger competitive response. The category advertising activity was captured by the advertising-to-sales ratio, both measured at the category level. Table 1 summarizes the descriptive statistics for the measures.2

View this table:
Table 1

Descriptive Statistics

VariableUnit of measurementMeanSD
Consumer characteristics:
  Dispositional innovativenessFive-point scale3.0.7
  Market mavenismFive-point scale2.9.6
  Susceptibility to normative influenceFive-point scale4.0.6
  Usage intensity100 ml/gram3,247.414,073.4
  AgeYears46.214.5
  Education: 0 = low; 1 = high0–119.1%-
  Monthly household incomeEuro1,591.7611.8
Marketing communication (time-varying):a
  Advertising variationEuros (in 1,000s)83.265.3
  Feature and display variation%12.47.3
  Competitive advertising variationEuros (in 1,000s)150.7114.1
Marketing strategy:
  Novelty of the productSeven-point scale3.31.1
  Brand strengthSeven-point scale4.71.4
  Average advertisingEuros (in 1,000s)49.053.8
  Average feature and display%6.15.9
  Average relative priceIndex1.1.2
  Average distribution coverage%42.821.7
Category characteristics:
  No. of brands8.24.0
  Competitive reaction to advertising attacksElasticity.3.5
  Category advertising intensity%.07.1
  Category specific impulse buyingFactor score.03.3
  Category specific ability to stockpileFactor score.2.5
Trial rate:
  Consumers who tried product after 52 weeks after introduction%6.24.1
  • a To illustrate the variation over the 52-week period of the time-varying advertising and feature and display measures, we report the standard deviation over this observation period. For our empirical analyses the measures are considered at disaggregate week level.

Results

Relations between the Three Consumer Dispositions

We examined whether the items used to measure the three consumer dispositions are essentially measuring the same underlying construct or whether they are related but distinct dispositions. The fit of the one-factor model was very bad: χ2(168) = 12,776.71 (p < .001), Comparative Fit Index (CFI) = .78, Tucker-Lewis Index (TLI) = .75, and Root Mean Square Error of Approximation (RMSEA) = .208. Thus, the one-factor model was clearly rejected. Next, we estimated the hypothesized three-factor model. The fit of this model was good: χ2(165) = 2,214.01 (p < .001), CFI = .96, TLI = .96, and RMSEA = .064.3 Moreover, the fit of the three-factor model was dramatically and significantly (Δχ2(3) = 10,562.70, p < .001) better than the fit of the one-factor model. All factor loadings were significant (all t-values were larger than 20), and all standardized loadings were greater than .40, the average standardized loading being .67. These findings support the convergent validity of the measures. The correlations between the constructs were consistent with theoretical expectations: r(dispositional innovativeness, mavenism) = .573 (p < .001), r(mavenism, susceptibility to normative influence) = .123 (p < .001), and r(dispositional innovativeness, susceptibility to normative influence) = −.231 (p < .001). All correlations were significantly (p < .0001) and substantially below unity, which supports the discriminant validity of the constructs and is the reason for the poor fit of the one-factor model.

Another conceptualization is that the three constructs are first-order expressions of a single underlying second-order construct. This second-order factor model will yield the same fit as the above model specifying three correlated factors. However, the nonuniform pattern of interconstruct correlations (varying between .123 and .573) and the small magnitude for two of three correlations is inconsistent with the second-order factor model (Byrne 1998). Hence, our data support our conceptualization of the three consumer dispositions as correlated but distinct constructs. Cronbach's alphas were .87, .89, and .68, respectively. A composite score for each construct was obtained by averaging the appropriate scale items.

Hypotheses Testing

The average pseudo R2 (Jain and Vilcassim 1991) for the hazard model (eq. 1) was 44.3%, while the system-weighted R2 for the second stage analysis (eqq. 4–11) was 48.1%. The system-weighted R2 was highly significant (F(32, 1,976) = 42.3, p < .001), indicating that the set interactions as specified in equations 4–11 collectively were significantly different from zero. The R2 for the hypothesized moderating effects on baseline hazard and consumer dispositions (eqq. 4–7) varied between 19.3% (moderating effects on susceptibility to normative influence) and 69.5% (baseline hazard), while R2 for the exploratory moderating effects on the sociobehavioral covariates (eqq. 8–11) varied between 5.2% (education) and 74.4% (usage intensity). All R2s were significant (p < .001). Hence, the variables in our conceptual framework have substantial explanatory power.

Table 2 reports the parameter estimates. For each construct, the unstandardized coefficient, t-value, and effect size r is given. For the effect of the within-product constructs, we discuss the average coefficients, t-values, and effect sizes across the 239 product introductions.

View this table:
Table 2

Determinants of Trial Probability

PredictorExpectedUnstandardized coefficientt-valueEffect size r
Consumer characteristics:
  Dispositions (hypothesis 1):
    Dispositional innovativeness (β1i)+.196**13.81.23
    Market mavenism (β2i)+−.034−.91.01
    Susceptibility to normative influence (β3i)−.041**−5.04.06
  Sociobehavioral covariates:
    Usage intensity (β4i)NA.026**3.26.12
    Age (β5i)NA−.002−1.45.02
    Income (β6i)NA.020**4.51.11
    Education (β7i)NA.027.98.01
Marketing communication (time-varying):
  Product support (hypothesis 2):
    Advertising (β8i)+.096**5.99.15
    Feature and display (β9i)+.113**7.73.17
  Covariate: competitive advertising (β10i)NA−.002+−1.76.03
Marketing strategy (hypothesis 3):
  Novelty (γ01)−.192**−4.04.26
  Novelty202).076**3.31.21
  Brand strength (γ03)+.068+1.85.12
  Average advertising (γ04)+.007**4.72.30
  Average feature and display (γ05)+.050**5.78.36
  Average relative price (γ06)−.076**−4.18.27
  Average distribution coverage (γ07)+.004*2.31.15
Category characteristics (hypothesis 4):
  Number of brands (γ08)−.021*−2.03.13
  Competitive reaction to advertising (γ09)−.078+−1.65.11
  Category advertising intensity (γ010)−.036−.64.04
  Category impulse buying (γ011)+.343*2.24.15
  Category ability to stockpile (γ012)−.168**−3.02.20
Interactions with consumer characteristics (hypotheses 5–7):
  Dispositional innovativeness (hypothesis 5):
    × Novelty (γ11)+.064**6.43.39
    × Average advertising (γ12)−.002**−4.00.25
    × Category impulse buying (γ13)+.054**2.67.17
    × Average feature and display (γ14)+.008**6.03.37
    × Average relative price (γ15)+.057**6.44.39
  Market mavenism (hypothesis 6):
    × Brand strength (γ21)+.027**3.86.24
    × Category impulse buying (γ22)−.042*−2.02.13
    × Average advertising (γ23)+.001+1.81.12
    × Average feature and display (γ24)+.003*2.13.14
  Susceptibility to normative influence (hypothesis 7):
    × Novelty (γ31)−.043**−5.36.33
    × Average advertising (γ32)+.001*2.19.14
Exploring interactions:
  Usage intensity:
    × Average feature and display (γ41)NA.025*2.06.13
    × Competitive reaction to advertising (γ42)NA−.213**−4.75.30
    × Novelty (γ43)NA−1.192**−6.72.40
    × Brand strength (γ44)NA.275**4.19.27
    × No. of brands (γ45)NA.304**8.48.49
  Age:
    × Category impulse buying (γ51)NA−.009**−4.09.26
  Income:
    × Novelty (γ61)NA.018**3.98.25
    × Average relative price (γ62)NA.012+1.88.12
  Education:
    × No. of brands (γ71)NA.003*2.01.13

Note.—For the consumer and marketing communication effects (which vary within products), the average coefficient and average effect size across 239 new product introductions are reported. The significance of the generalized within-product effects is evaluated by applying a weighted t-test on the set of 239 parameter estimates, with their standard errors serving as weights. The effect size r is computed as [t2/(t2 + df)].5. NA = not applicable (no hypothesis offered).

  • + p < .10 (two-sided).

  • * p < .05 (two-sided).

  • ** p < .01 (two-sided).

Consumer Dispositions

We found strong support for hypothesis 1a, which posits that consumers who are higher on dispositional innovativeness exhibit a higher trial probability for a new product (β̅1̅ = .196, p < .01). The effect of market mavenism was not significant (β̅2̅ = −.034, p > .10). Thus, hypothesis 1b was not supported. However, in line with hypothesis 1c, the impact of susceptibility to normative influence was negative (β̅3̅ = −.041, p < .01).

Time-Varying Marketing Communication

More intensive recent advertising and feature and display are effective instruments to increase trial probability immediately. Across all new product introductions, both had a significant effect on the trial probability: advertising, β̅8̅ = .096 (p < .01); feature and display, β̅9̅ = .113 (p < .01). Thus, hypotheses 2a–2b were supported.

Marketing Strategy

A nonlinear relationship was found between product novelty and trial probability. However, contrary to expectations (hypothesis 3a), we obtained a U-shaped effect rather than an inverted U-shaped effect of product novelty on the baseline hazard (γ01 = −.192, p < .01; γ02 = .076, p < .01). This indicates that new products of intermediate novelty generated lower trial probability than either new products of incremental or high novelty. Introducing a new product with a strong brand name increased trial probability (γ03 = .068, p < .10). As such, hypothesis 3b was corroborated. The trial probability of a new product was also higher when average advertising expenditure (γ04 = .007, p < .01) and average feature and display effort (γ05 = .050, p < .01) were higher. This supported hypotheses 3c–3d. The higher the average price of the new product relative to the category, the lower the trial probability (γ06 = −.076, p < .01), which is in line with hypothesis 3e. As proposed in hypothesis 3f, distribution coverage for the new product increased the trial probability (γ07 = .004, p < .05).

Category Characteristics

Consistent with hypothesis 4a, we found that the trial probability was lower in categories with many existing brands (γ08 = −.021, p < .05). In line with hypothesis 4b, trial probability was lower in categories characterized by more aggressive competitive reactions with advertising to advertising attacks by competing brands (γ09 = −.078, p < .10). The effect of category advertising intensity on the trial rate (hypothesis 4c) of a new product introduced in that category did not reach statistical significance (γ010 = −.036, p > .10). Finally, the trial probability was higher in categories high on impulsive buying (γ011 = .343, p < .05), and lower in categories that are easier to stockpile (γ012 = −.168, p < .01). This supported hypotheses 4d–4e.

Moderating Effects on Consumer Dispositions

With respect to dispositional innovativeness, we found that novelty shifts its impact upward (γ11 = .064, p < .01), thereby corroborating hypothesis 5a. Consistent with hypothesis 5b, higher levels of advertising spending for the new product reduced the impact of innovativeness (γ12 = −.002, p < .01), whereas feature and display (hypothesis 5d) increased its effect (γ14 = .008, p < .01). In categories high on impulse buying, the positive effect of dispositional innovativeness was stronger (γ13 = .054, p < .01), which is in line with hypothesis 5c. Finally, as predicted by hypothesis 5e, the effect of dispositional innovativeness was stronger the higher the relative price of the product (γ15 = .057, p < .01).

The effect of market mavenism was stronger when the product was introduced under a strong brand name (γ21 = .027, p < .01) but weaker if the product was introduced in a category characterized by high impulse buying (γ22 = −.042, p < .05). This is consistent with hypotheses 6a–6b. Sustained intensive advertising (γ23 = .001, p < .10) and feature and display support (γ24 = .003, p < .05) for the new product shifted the effect of mavenism upward, supporting hypotheses 6c–6d.

In line with hypotheses 7a–7b, the negative impact of susceptibility to normative influence on trial probability was more pronounced (negative) the more novel the product (γ31 = −.043, p < .01), but was attenuated when average advertising was higher (γ32 = .001, p < .05).

Covariates

Main Effects

Table 2 also reports the results for the covariates. The trial probability was higher among heavy users of the category (β̅4̅ = .026, p < .01). Gatignon and Robertson (1985) proposed that trial probability should increase with income and education and decrease with age. We found support for a generalized effect of income across the broad range of CPGs (β̅6̅ = .020, p < .01), but neither age nor level of education exhibited such a generalized effect on trial probability. However, previous evidence is inconclusive, with many studies showing contradictory effects for age and education (Rogers 1983). It appears that the effects of age and education may only work in particular contexts. Below, we will examine some moderators. Finally, heavy contemporaneous advertising for competing products in the category reduced the trial rate (β̅1̅0̅ = −.002, p < .10).

Exploratory Interactions Involving Consumer Covariates

There is no firm theoretical basis to develop a priori predictions concerning interactions involving the sociobehavioral covariates. However, the rich data set allows us to derive exploratory empirical generalizations involving these covariates that can subsequently be used to develop generalized theoretical explanations (see Bass [1995] for an elaboration of the Empirical-Theoretical-Empirical-Theoretical philosophy). We offer brief post hoc explanations for these interactions as possible starting points for theoretical investigations.

The effect of usage intensity on trial probability was higher when the new product was supported by heavy sustained feature and display (γ41 = .025, p < .05), while its effect was reduced if there were strong competitive reactions to advertising attacks (γ42 = −.213, p < .01). Heavy users will be more focused on making purchases in the category, because of heavier ongoing category needs. Hence, they are likely to give more attention to marketing communications for that category, which should increase the effect of feature and display information for the new product.4 By contrast, more attention to advertising increases the chances of memory interference if brands react strongly to each other's advertising attacks (Keller 1987).

The other three interactions involving usage intensity (table 2) might be explained based on category knowledge. Knowledge might be thought of in terms of familiarity and expertise. Product usage constitutes an important component of familiarity, while “in general, increased product familiarity results in increased consumer expertise” (Alba and Hutchinson 1987, p. 411). We found a negative moderating effect of novelty on the effect of usage intensity (γ43 = −1.192, p < .01). Experts will find it easier to process and comprehend new products that fit into existing cognitive knowledge structures, but these deep, interconnected structures may decrease comprehension of new products that are high on novelty (Moreau et al. 2001). We further found that the difference in effect between light and heavy category users on trial probability was more pronounced for strong brands (γ44 = .275, p < .01). Strong brands are characterized by extensive knowledge structures (Keller 1993), which are more accessible and easier to process by more knowledgeable consumers (Alba and Hutchinson 1987). Finally, a large number of brands in a category are a barrier to trial because consumers may not even notice changes in category composition. However, this may be less of a problem for knowledgeable consumers who, because of frequent purchases in the category, typically should be more aware of the different options offered (γ45 = .304, p < .01).

In impulse categories, new products achieved a higher trial probability among young consumers (γ51 = −.009, p < .01). This is consistent with the notion that impulse buying can be highly stimulating (Rook 1987) and that stimulation seeking and age are negatively correlated (Raju 1980). The positive effect of income on trial probability was stronger for more novel (γ61 = .018, p < .01) and relatively more expensive products (γ62 = .012, p < .10). Both types of products are more risky, which should be easier to bear for well-to-do consumers. Finally, in categories with many brands, trial probability was higher among higher-educated consumers (γ71 = .003, p < .05). Information processing in such markets is cognitively more demanding (Malhotra 1982), which should pose more difficulties to less-educated consumers.

Validation

To assess the validity of our empirical findings, several cross-validation analyses (Cooil, Winer, and Rados 1987) were conducted. First, we randomly split the sample of consumers into a calibration sample (two-thirds of the sample) and a validation sample (one-third of the sample). We estimated the hazard model (eq. 1) for a randomly chosen product on both samples separately and used the coefficients of the calibration sample to predict (1) which consumers in the validation sample bought the new product in the observation period of 52 weeks and (2) the time to trial for those consumers in the validation sample who bought the product in the observation period. We repeated these analyses for 60 randomly chosen products. On average, 66.5% of the consumers who bought the new product were correctly predicted (p < .001), while the average correlation between predicted and actual time to trial was .619 (p < .001).

Second, we cross-validated the level-2 results by randomly splitting the set of 239 new products into a calibration set (two-thirds of the sample) and a validation set (one-third of the sample). Equations 4–11 were estimated on the calibration set and the validation set separately. Next, the calibration estimates were used to predict the dependent variable of these equations for the validation sample. Two cross-validation shrinkage statistics (Cooil et al. 1987; Steenkamp and Wedel 1993) were computed for each equation: (1) the adjusted percentage of variance in the validation sample explained by the parameter estimates of the calibration sample, R2(c)a, divided by the adjusted percentage of variance in the validation sample explained by the parameter estimates of the validation sample, R2(v)a; and (2) the root mean square error between the “true” and predicted value of the dependent variable in the validation sample, based on the calibration sample estimates, RMSE(c), divided by the same statistic based on the validation sample estimates, RMSE(v). The analyses were repeated 10 times, on 10 different sample splits. For both shrinkage statistics, values close to one (zero) indicate high (low) predictive validity (Steenkamp and Wedel 1993). The average value of R2(c)a/R2(v)a across the eight equations was a high .787 (range: .621–.992), while the average value of RMSE(c)/RMSE(v) was .923 (range: .866–.960). In sum, our cross-validation analyses support the validity of the results reported in table 2.

Discussion

Recently, Mick (2003, p. 459) called for more research “combining mental phenomena with actual behavior.” The present study heeds this call by studying the joint effect of consumer dispositions and market factors on actual new product purchase behavior. In general, we found support for our hypotheses. Two consumer dispositions, innovativeness and susceptibility to normative influence, had a generalized effect on the trial probability of a new product over a wide range of CPGs. Trial probability was also predictably related to (time-varying) marketing communication, (time-invariant) marketing strategy, and category characteristics. The effects of the three consumer dispositions were systematically and predictably moderated by market factors. Systematic moderating effects were also found for the sociobehavioral covariates.

The substantial magnitude of many interactions (r̅ = .25) indicates that the effect of consumer characteristics is heavily influenced by marketing and category variables. The new product adoption literature has typically focused on either consumer or market variables, but our findings indicate that detailed analysis of their interplay is necessary to understand the complexity of adoption processes fully. For several constructs (market mavenism, age, and education), the moderating role of market factors is so strong that no evidence was found for a generalized main effect. Rather, the direction of the effect was completely dependent on market factors. This may explain the contradictory findings often reported for sociodemographics (Rogers 1983).

Contrary to expectations, no support was found for a generalized effect of market mavenism. This indicates that the role of market mavenism is completely context dependent. Only in specific situations, such as when the benefits of information acquisition through new product purchase exceed the costs, is this construct positively associated with trial of new products. At least two explanations can be forwarded for the absence of a generalized effect of market mavenism. First, the construct reliability was borderline acceptable (.68), which may have adversely affected the results. Second, market mavenism may be too broad a construct to study informational influences across a wide range of new product contexts. One might argue that these patterns are more product or category specific and that, therefore, product or category specific measures are more powerful (King and Summers 1970). Future research could examine the validity of these explanations.

Contrary to previous work by Goldenberg et al. (2001), new products of intermediate novelty generate a lower rather than a higher trial probability than incrementally new or really new products. Goldenberg et al. explained their findings by hypothesizing that “radical changes are likely to be rejected and minor ones ignored” (2001, p. 78). However, their sample of 127 new consumer products (many being non-CPGs) came from books containing case studies on new product failures and successes. It is plausible that cases published in books, on average, will be higher on novelty than our set of CPGs. The illustrative set of new products provided in their article supports this point of view. Combining Goldenberg et al.'s and our range of novelty gives a fuller representation of the total range of novelty of new products introduced in the consumer marketplace. This would suggest that the relation between novelty and trial probability might be better described by a cosine-shaped relation. In figure 2, we depict this function and provide illustrative new product examples, taken from our set of products and from Goldenberg et al. (2001, app. B). We also provide possible underlying causes of the effect of novelty in terms of the two factors outlined by Goldenberg et al., namely, complexity and relative advantage. We speculate that both complexity and relative advantage tend to increase with higher levels of novelty, albeit not in a linear fashion. In case of radical innovations, the relative advantage is likely to be high but uncertain, as it may require extensive behavioral changes that might not be acceptable or whose consequences might not be clear.

Figure 2

A Revised Relationship between Novelty of New Products and Trial Probability

Note.—Single asterisk refers to this study; double asterisk refers to Goldenberg, Lehmann, and Mazursky (2001).

Our item-based measure of dispositional innovativeness requires separate data collection. An alternative measure of dispositional innovativeness could be based on actual behavior. If a consumer purchases many new products across a wide range of categories, she or he is likely to be high on the trait of innovativeness. Such a measure can be constructed from household panels. We compared the relative performance of our attitudinal measure with the behavioral measure, using jackknifing. For each new product we calculated what percentage of the other 238 new products was bought by the consumer and subsequently estimated two bivariate hazard models, using the consumer's trial probability for the holdout product, λij(t), as the criterion variable. The behavioral and attitudinal measures served as alternative predictors. The average pseudo R2 was about the same: 5.4% and 5.6% for the attitudinal and behavioral measure, respectively. Conceptually, the behavioral measure is less appealing as it measures the trait of innate innovativeness indirectly. Many factors can intervene between disposition and behavior. Further, because behavioral measures are more affected by contextual factors, one may expect them to be less stable. Indeed, the one-year stability of the behavioral measure of innovativeness (1997 vs. 1998) was .71, while the three-year stability of the dispositional innovativeness measure on the Europanel household panel was .85. On the practical side, the applicability of the behavioral measure is limited to panelists, while the attitudinal measure can be applied to the entire population.

We found strong effects for two communication instruments: advertising and feature and display. Previous experimental research (Kirmani 1990) indicated that excessive marketing communication (advertising) effort has a negative effect on brand perceptions (and by implication on purchase likelihood) as consumers interpret this as a sign of desperation. We examined whether there is support for an inverted-U relation by adding quadratic terms to equation 4. Both quadratic terms were negative but not significant (p's > .10). Thus, our data suggest that marketers indeed can overadvertise or overpromote, but in practice, their behavior ensures that they do not move into the excessive region of the curve.

Limitations and Directions for Future Research

Our study has several limitations, which offer avenues for future research. Our hypotheses were tested on CPGs. These tend to be marketing intensive and relatively lower in risk. The relative importance of various drivers of trial may be different for consumer durables or services. We speculate that the effect of interpersonal variables might be relatively more important for these types of goods, which are often higher in social visibility. It is also useful to study the role of novelty in other classes of products. Although major novel products are introduced in CPGs, radically new high-tech products are typically of a different kind.

We used an aggregate (market-level) measure of feature and display, based on observed purchases. Future research could refine our measure by using store-level information and by accounting for the quality and actual level of in-store feature and display support. Further, our measures of novelty and brand strength are based on expert judgments. It would be preferable to collect this information among those consumers whose adoption decisions are being analyzed.

For some of our hypotheses, we drew on the signaling literature. With few exceptions (e.g., Kirmani 1990; Kirmani and Wright 1989), the empirical evidence whether consumers perceive the signals and (intuitively) use the rationale underlying signaling theory in making quality judgments is limited. More research is needed on developing and testing a framework concerning the categories, consumers, and information contexts in which these quality inferences are more or less likely to occur.

The predictors of trial probability were treated as exogenous, while it is possible that some marketing variables also are influenced by the (lack of) market success of the product. Our data did not allow us to estimate simultaneous effects. Further, we estimated our effects using a two-step procedure. It is in principle possible to specify one overall model with sets of covariates capturing systematic variation in trial rates across categories, consumers, and time, and with two sets of “crossed” random effects to capture heterogeneity across consumers and across categories. Such a model was not feasible to estimate with an iterative procedure as it would contain about 40 million records in the data set. However, we corrected for differing degrees of estimation accuracy in the second stage, using the inverse of the estimated standard errors of the dependent variable as weights (Narasimhan et al. 1996; Shankar et al. 1998).

Other areas for future research are open as well. In our work, we focused on advertising expenditure. As a next step, one should consider advertising content. Ads used for new products could be evaluated on their match to advertising creativity templates (Goldenberg, Mazursky, and Solomon 1999) or on their originality (Pieters, Warlop, and Wedel 2002). This information could be included in the explanation of the baseline hazard (eq. 4). In a similar vein, one could study whether particular new product ads are more effective among particular consumer segments (eqq. 5–11). Future research could test the validity of the proposed cosine-shaped relation between novelty and trial probability, and of its underlying factors as an integrated explanation for the effect of novelty on trial probability in the consumer domain. Research is also needed as to how novelty perceptions change over time.

Appendix Measurement of Consumer Constructs

Dispositional innovativeness (α = .87):

  • When I see a new product on the shelf, I'm reluctant to give it a try. (*)

  • In general, I am among the first to buy new products when they appear on the market.

  • If I like a brand, I rarely switch from it just to try something new. (*)

  • I am very cautious in trying new and different products. (*)

  • I am usually among the first to try new brands.

  • I rarely buy brands about which I am uncertain how they will perform. (*)

  • I enjoy taking chances in buying new products.

  • I do not like to buy a new product before other people do. (*)

Market mavenism (α = .68):

  • I like introducing new brands and products to my friends.

  • I don't talk to friends about the products that I buy. (*)

  • My friends and neighbors often come to me for advice.

  • People seldom ask me for my opinion about new products. (*)

Consumer susceptibility to normative influence (α = .89):

  • If I want to be like someone, I often try to buy the same brands that they buy.

  • It is important that others like the products and brands I buy.

  • I rarely purchase the latest fashion styles until I am sure my friends approve of them.

  • I often identify with other people by purchasing the same products and brands they purchase.

  • When buying products, I generally purchase those brands that I think others will approve of.

  • I like to know what brands and products make good impressions on others.

  • If other people can see me using a product, I often purchase the brand they expect me to buy.

  • I achieve a sense of belonging by purchasing the same products and brands that others purchase.

An asterisk (*) indicates reverse scored item. Items were scored on a five-point scale ranging from completely disagree = 1 to completely agree = 5.

Footnotes

  • 1 We compared the exponential, Weibull, and nonmonotonic specifications using the log-likelihood test (Jain and Vilcassim 1991). In the large majority of cases (about 85%), support was found for the exponential specification.

  • 2 The correlation matrix among all predictors can be obtained by contacting the authors.

  • 3 In both models, two correlated error terms were introduced because these had very large modification indices. Specifying these correlated errors made sense because of similarity in item wording (Byrne 1998). The factor loadings were not affected, though. In both instances, the correlation between the loadings was above .90.

  • 4 For the same reason, one might expect that advertising by the new product would be more effective among heavy users. We indeed found a positive moderating effect of advertising effort on the effect of usage intensity, but it did not reach statistical significance.

References

View Abstract