Effects of prior brand usage and promotion on consumer promotional response

41
Effects of Prior Brand Usage and Promotion on Consumer Promotional Response Eileen Bridges, Richard A. Briesch, and Chi Kin (Bennett) Yim January 2006 Eileen Bridges ([email protected]) is Professor of Marketing, Kent State University, Kent, OH 44242. Richard A. Briesch ([email protected]) is Assistant Professor of Marketing, College of Business Administration, Southern Methodist University. Chi Kin (Bennett) Yim ([email protected]) is Associate Professor of Marketing, School of Business, The University of Hong Kong. We wish to express our appreciation to Editors Dhruv Grewal and Michael Levy, an anonymous Associate Editor, three anonymous reviewers, Ed Fox, Roger Kerin, and Amna Kirmani for helpful comments and suggestions. We also thank the organizations providing data, including Information Resources, Inc., A.C. Nielsen Co., and the Marketing Science Institute Library of Single-Source Data.

Transcript of Effects of prior brand usage and promotion on consumer promotional response

Effects of Prior Brand Usage and Promotion on Consumer Promotional Response

Eileen Bridges, Richard A. Briesch, and Chi Kin (Bennett) Yim

January 2006

Eileen Bridges ([email protected]) is Professor of Marketing, Kent State University, Kent, OH 44242. Richard A. Briesch ([email protected]) is Assistant Professor of Marketing, College of Business Administration, Southern Methodist University. Chi Kin (Bennett) Yim ([email protected]) is Associate Professor of Marketing, School of Business, The University of Hong Kong. We wish to express our appreciation to Editors Dhruv Grewal and Michael Levy, an anonymous Associate Editor, three anonymous reviewers, Ed Fox, Roger Kerin, and Amna Kirmani for helpful comments and suggestions. We also thank the organizations providing data, including Information Resources, Inc., A.C. Nielsen Co., and the Marketing Science Institute Library of Single-Source Data.

Effects of Prior Brand Usage and Promotion on Consumer Promotional Response

ABSTRACT

We examine how prior purchases influence consumer response to promotional activity in

brand choice decisions. To improve understanding of the nature of this influence, we separate

previous purchases into those on promotion and those not on promotion, and consider their

differential impact on subsequent brand choices. Impact may be observed at the brand and/or

category levels and we suggest circumstances in which each might occur. Across four product

categories, consumer sensitivity to price, price promotions, and feature advertisements increases

for all brands in the product category following a promotional purchase but also decreases for the

most recently purchased brand. The magnitudes of the results indicate that prior promotional

purchases influence choice more than prior brand usage does. We offer managerial

recommendations regarding promotional activities, for both retailers and manufacturers.

1

INTRODUCTION

Neslin (2002) has called for research on the related influences of prior purchases and changes in

promotional sensitivities on consumer behavior; we respond to this call by extending the

literature beyond how promotions affect consumer tendencies to exhibit loyalty by repurchasing

the same brand, to examining how various promotions affect consumer response to subsequent

marketing mix activities. Thus, we address the question of whether obtaining brand usage is

worthwhile when marketers achieve it through promotional activities.

In addition to the theoretical ramifications of running promotions, we consider the

managerial impact for both retailers and brand managers. If promotional activities result in

increased consumer price sensitivity for all products in the category, the market share of lower-

priced store or regional brands could increase following a promotion by a higher-priced national

brand.1 From a retailer’s point of view, this outcome may be beneficial, particularly if margins are

higher on the lower-priced brand. However, reductions in the promoting national brand’s share

would be an undesirable result from the brand manager’s point of view.

An existing research stream considers the costs and benefits of promotional activities

directed at consumers, but empirical results are mixed (Blattberg, Briesch, and Fox 1995).

Several studies reveal that price promotions may have adverse long-term effects on consumers’

brand choice behavior by making them more sensitive to price and/or promotions (Boulding,

Lee, and Staelin 1994; Dodson, Tybout, and Sternthal 1978; Jedidi, Mela, and Gupta 1999; Mela,

Gupta, and Lehmann 1997; Papatla and Krishnamurthi 1996; Strang 1975). Fader and McAlister

(1990) observe that some consumers actively seek out promotions for preferred brands. These

effects are not always undesirable: Shankar and Krishnamurthi (1996) find that feature

advertisements and in-store displays can differentiate a brand, making it more salient and leading

2

to reduced price elasticity. Kopalle, Mela, and Marsh (1999) find that price promotions enhance

consumer price sensitivity, but note that, under some conditions, such promotions can be

profitable to both retailers and manufacturers. Neslin (2002, p.17) concludes that non-price

promotions also influence price sensitivity, stating “there is fairly strong evidence that

promotions affect price or promotion sensitivity.” However, other studies fail to find any long-

term impact of promotions (Johnson 1984; Neslin and Shoemaker 1989; Totten and Block 1987).

Most early studies assumed any impact of promotional activities to be homogeneous

(e.g., Guadagni and Little 1983) and therefore failed to identify any differential effects. As the

research stream has developed, cross-sectional variations in consumer response to different types

of promotions have been observed across product categories, markets, and households (Bucklin

and Gupta 1992; Fader and Lodish 1990; Grover and Srinivasan 1992; Inman and McAlister

1993; Kamakura and Russell 1989; Narasimhan, Neslin, and Sen 1996). In addition, research on

“purchase event feedback” has suggested that consumers’ prior purchases might influence their

current purchase behavior and response to promotion. For example, Smith and Swinyard (1982;

1983) find that a household’s response to promotional activities can vary across purchase

occasions and Heilman, Bowman, and Wright (2000) further observe that some aspects of

purchase history can influence response to promotional offers.

Although the brand most recently purchased by a consumer, or state dependence, has

been considered in previous research (Chintagunta 1999; Erdem and Sun 2001), few authors

consider whether the most recent purchase was on promotion. A key exception comes from

Gedenk and Neslin (1999), who find that the promotional status of the previous purchase can

differentially influence brand choice, through purchase event feedback. In addition, they

indicate that the specific type of promotion (price or non-price) can influence brand preference.

3

Their results suggest that purchase event feedback following a promotional purchase generally is

more negative than that following a non-promotional purchase. Further, feedback after price

promotions is more negative than that after non-price promotions. Thus, they address the

fundamental question of how effective different types of promotions are at retaining consumers

for subsequent brand purchases.

We build on Gedenk and Neslin’s (1999) results by suggesting that the promotional

status of the most recent purchase not only leads to differences in brand loyalty but also

influences subsequent response to promotional activities and in turn, switching behavior in the

product category. Therefore, we extend their model to address additional issues, such as why a

lower-priced store brand’s market share might increase after a promotion by a national brand.

We apply a model that separates household-level effects (cross-sectional heterogeneity) from

purchase-level effects (state dependence or longitudinal heterogeneity) to test empirically our

hypotheses regarding the impact of households’ brand purchase histories on their response to

promotional activities.

Two streams of theoretical and empirical research support moderating effects of prior

brand purchases on consumer response to promotions. Usage dominance posits that prior

purchases of a brand diminish response to promotions that follow, because personal usage

dominates external information during purchase decisions. The second stream, which might be

termed “promotion enhancement,” suggests that consumers are more responsive to marketing

mix activities in a product category after they have made promotional purchases in that category,

because promotional activities are more salient when consumers are more familiar with the

product category. Our research tests for specific effects that clarify how – whether through

usage dominance or promotion enhancement – consumers’ prior purchases influence their

4

response to promotional activities both at the brand and the category levels.

In the next section, we review some theoretical and empirical literature that supports

usage dominance and promotion enhancement and describe recent advances that address how

prior purchases can influence consumer sensitivity to promotions. We then detail our modeling

and empirical testing, and report our results. Finally, we conclude with managerial implications,

limitations, and directions for further research.

BRAND USAGE AND RESPONSE TO PROMOTIONAL ACTIVITIES

Prior theoretical and empirical studies suggest that brand usage history can affect consumer

response to temporary price reductions and other promotional activities. However, as we noted

previously, two streams of research posit opposing influences, which may occur at the brand

and/or the category level. Our goal is to enhance understanding of the effects of promotional and

non-promotional purchasing on consumer response to subsequent price and non-price

promotional activities.

Usage Dominance

Literature in support of the usage dominance concept suggests that, after purchase and

use of a brand, consumers become less responsive to promotional activities for that brand

because their direct experience dominates external information (e.g. marketing mix activities) in

their purchase decisions. For example, some researchers find that, if a consumer has access to

multiple sources of information, a specific input is more likely to influence his or her buying

decision if it is more accessible in memory and provides more diagnostic information (Alba,

Hutchinson, and Lynch 1991; Anderson 1971). This idea, that a decision maker having multiple

signals available will weight the signal with less noise more heavily, is also supported by the

information economics literature (e.g. Banker and Datar 1989).2 Because a consumer’s own

5

usage experience is more diagnostic than external information (Fazio and Zanna 1978; Fazio,

Chen, McDonel, and Sherman 1982; Fazio, Powell, and Williams 1989), consumers who have

purchased a brand before are likely to rely on their internal information when making a purchase

decision. This proposal is consistent with observations by Smith and Swinyard (1982; 1983) that

direct experiences tend to be the primary information sources for forming attitudes that increase

consumers’ commitment to buy.

As consumers become more aware of their own likes and dislikes, as well as the

performance of various brands (through purchase and use), their choices are more likely to be

driven by non-price factors (Heilman et al. 2000). Kopalle and Lehmann (1995) agree that, after

consumers have experienced a brand, they tend to rely on their internal information more heavily

than external cues such as advertising to update their impressions of the brand, which influences

their brand choice (Kopalle and Lehmann 2006). That is, usage dominance suggests that

consumers who focus on their personal experience are less responsive to marketing mix activities

for the most recently purchased brand and, consequently, more likely to repurchase the brand

after a promotion has ended.

Promotion Enhancement

Literature favoring promotion enhancement states that the impact of marketing mix

activities increases when the consumer’s most recent purchase of any brand in the category was

on promotion. Empirical support for this theory has been provided by many above-mentioned

studies, which indicate that promotions reduce subsequent brand loyalty. Petty and Cacioppo’s

(1986) elaboration likelihood model offers theoretical support in suggesting that consumers are

more motivated to process information that has greater personal relevance. Thus, cognitive

elaboration is likely to be richer for product categories with which consumers have more

6

extensive usage histories in different contexts (Cacioppo and Petty 1985). In turn, promotional

activities for such products are more likely to be mentally processed and influence choice. Also,

from a search perspective, as a consumer’s experience with a product category increases, so does

his or her ability to distinguish when a brand in the category offers a better deal, which leads to a

greater incentive to buy the brand on promotion (Bettman and Park 1980; Johnson and Russo

1984; Moorthy, Ratchford, and Talukdar 1997). Therefore, due to their increased sensitivity to

price and promotional activities, consumers are more likely to purchase lower-priced or

promoted brands when they make a subsequent purchase.

In summary, promotion enhancement suggests that consumers are more responsive to

price and promotional activity for all brands in a category after they have made a promotional

purchase of any brand. Thus, while usage dominance suggests that consumer responsiveness to

promotional activities for a recently purchased brand decreases, promotion enhancement

simultaneously indicates increases in response to marketing mix activities for all brands in the

category. We investigate this conflict further by measuring observed effects.

MODEL DEVELOPMENT

The objective of our study is to test empirically for the effects of usage dominance and/or

promotion enhancement on consumer response to price and promotional activities. We begin by

describing a theoretical model, and then discuss our model estimation approach.

Theoretical Model Development

Following the discrete choice literature, we define a consumer’s indirect utility for a

brand to be a function of four marketing mix variables (regular price, deal depth or amount of

temporary price cut, presence of in-store display, and presence of feature advertising), state

dependence, time since last purchase, and brand-specific intercept terms. We are fortunate to

7

have access to the depth, rather than simply the presence or absence, of any deals, because deal

depth is directly comparable to the price of the item purchased (both are measured on the same

scale). Briesch, Chintagunta, and Matzkin (2002) find that deal depth provides more accurate

choice predictions than does deal presence. Further, several studies have observed that

consumers respond differently to deals than to regular prices (Blattberg et al. 1995; Briesch et al.

2002; Van Heerde, Leeflang, and Wittink 2000; 2001). Thus, our model is given by

(1) thititithbtithbtithbtithbtithb

ththbththbthhshbthi

DFFDARETSDSDU

,,,,,,,8,,,,7,,,,6,,,,5,,,,4

,,,,,3,,,,2,,1,,0,,

εβββββββββ

++++++

+++=

where Ui,h,t is the utility of item i for household h in period t. Note that i (i = 1...I) represents a

brand-size item, b is an indicator of the brand (b = 1...B), and s represents package size (s =

1...S). SDb,h,t is a binary vector set to one if brand b was purchased on the last occasion and ETh,t

is the elapsed time since household h purchased in the category. The marketing mix for item i in

period t is described by Ri,t, the regular price, Ai,t, the depth of any promotional price cut, Di,t, a

binary vector set to one if there is an in-store display, and Fi,t, a binary vector set to one if there is

a feature advertisement. For brands that are feature advertised and displayed in the same week,

we include an interaction term to control for any omitted variable bias. The intercepts β0,b,h and

β1,s,h (for brand and size, respectively) allow for correlation in consumer preferences (i.e., all

items of the same brand share the same brand intercept and all items of the same size share the

same size intercept). We make the standard assumptions about utility and error to obtain a

multinomial logit model as our empirical model.

The state dependence term captures the main effect of usage dominance on brand choice:

if positive, consumers are more likely to repurchase the same brand on successive occasions. An

interaction term with time since the last purchase allows the impact of usage dominance to decay

as the time between purchases increases (Chintagunta 1999).

8

To take into account whether the previous purchase was promotional, let Ph,t be a binary

variable set to one if the alternative that household h selected on the prior occasion was

displayed, feature advertised, and/or offered at a promotional price. We hypothesize that, if the

household purchased on promotion, its response to the marketing mix and tendency to

repurchase will be affected. To test this, we use hierarchical equations. We modify the variables

for state dependence (coefficients β2,h,t and β3,h,t), replacing them with the following:

(2) thkthkkthk P ,,,,1,0,, ζγγβ ++=

where k = 2,3. Note that ζk,h,t represents a household-specific effect and is the deviation from the

mean response in the market after adjusting for a prior promotional purchase by household h..

We assume that all of the ζk,h,t are normally distributed (over households and time) with mean

zero and standard deviation σk. Considering the findings of Gedenk and Neslin (1999) and the

hypothesized impact of promotion enhancement, we anticipate that γ1,2 will be negative.

Various theories suggest that consumer response to the marketing mix is affected by the

brand selected on the prior occasion and the promotional status of that brand at the time. Thus,

we rewrite the coefficients β4,b,h,t through β8,b,h,t using the following equation:

(3) thbkthbkthkkthbk SDP ,,,,,,2,,1,0,,, ζγγγβ +++=

where k = 4,5,6,7,8. γ1,k represents a category-level effect indicating sensitivity to promotions,

γ2,k represents a brand-specific effect that depends on whether the household is repurchasing the

same brand, and thbk ,,,ζ represents a household-, time-, and brand- specific response. As an

example, if γ2,4 is positive, consumers are less price sensitive for the brand they purchased on the

last occasion; if it is negative, they are more price sensitive for the previously purchased brand.

We assume that thbk ,,,ζ has a normal distribution (over time, brands and households) with mean

zero and standard deviation σk.

9

This functional form is similar to the one used by Jedidi et al. (1999) to test for the long-

term effects of promotions, advertising, and loyalty on consumers’ marketing mix response.

However, there are several key differences in our approaches. First, we use an indicator (Ph,t) to

identify whether the previous brand purchased was on promotion at the time, whereas Jedidi et

al. use the brand’s long-run marketing mix activity, formed as a geometric series of average

market values. Second, we allow promotions to lead to increases in marketing mix sensitivities

for all brands in the category; they allow promotions to affect only marketing mix sensitivities

for the promoted brand. Third, we identify state dependence, whereas they use a loyalty term

constructed from the previous four non-promoted purchases. Although both studies test for the

effects of previous promotional activity on current purchases, our study focuses on the short-term

impact of promotions on response to price and promotional activities and ties promotional

sensitivities to actual purchasing behavior rather than market averages.

Separating the effects of a previous purchase on promotion from those of a previous

purchase not on promotion enables us to test for the presence of usage dominance and promotion

enhancement effects. As we have described these theories, they suggest hypothesized directions

of effects, given by the signs of the hierarchical coefficients in Table 1.

[INSERT TABLE 1 ABOUT HERE]

The directions of the effects described in Table 1 are consistent with our theoretical

discussion of usage dominance and promotion enhancement. Specifically, promotion

enhancement implies a reduced likelihood to buy the previously purchased brand, simultaneous

with an increase in the impact of promotional activities for all brands in the category. Usage

dominance suggests a positive main effect of state dependence but does not imply any particular

interactions with the marketing mix; thus, we do not include coefficient γ2,k in this model.

10

(However, for the previously purchased brand, usage dominance implies that consumers are less

sensitive to both price and promotional activities.) Both theoretical effects may act

simultaneously and their relative magnitudes then would determine whether consumers are more

or less responsive to marketing mix activities for available brands. Because promotion

enhancement operates at the category level, we should find that consumers are more sensitive to

price and promotional activities for all brands in the category following a purchase on

promotion.

Model Estimation

In our estimation, we account for consumer heterogeneity by assuming that response

coefficients have multivariate normal distributions; thus, if Ζ is the vector of heterogeneity

parameters, then Z ~ MVN(0,Σ). We then estimate the parameters of this multivariate normal

distribution of coefficients. We would prefer to use maximum likelihood estimation, but because

of the many dimensions in the parameter space, computation using numerical integration

methods is not practical. Therefore, we employ a quasi-Monte Carlo method to select a sample

of points in the parameter space and compute the integrand at those points. We then apply this

simulated maximum likelihood estimation (SMLE) to construct each household’s likelihood

function, where each household’s response parameters are drawn from a multivariate normal

distribution with an estimated mean and covariance matrix (Hajivassilios and Rudd 1994). We

provide additional details of the estimation method in Appendix A.

Description of Data

We used the ERIM data from four different product categories (peanut butter, tuna, stick

margarine and tissue) to test our hypotheses, as given by the expected directions of effects in

Table 1. For each product category, we selected the best selling brand-sizes to ensure that

11

enough of the category activity would be captured by the model. After selecting the brand-sizes,

we screened households for inclusion on the basis of two criteria: they must have (1) made at

least three purchases and (2) purchased only the selected brands. The first condition is necessary

to identify lagged variables (state dependence and whether the previous purchase was on

promotion for each household) and heterogeneity, and the second ensures that we model the

correct effects. For each product category, we randomly assigned each observation to either the

estimation or the holdout sample, with a probability of 50% either way.

We provide descriptive statistics for the observations assigned to the estimation sample in

Table 2, which shows that the total shares of selected brand sizes account for at least 79.9% of

the choices within each product category. Our loyalty measure indicates that most households

purchase the same brand most of the time. In addition, the proportion of purchases made on

promotion varies among the different categories, from 32% in peanut butter to almost 69% in

tuna. This variation may be related to the frequency of promotion that occurs in each category.

[INSERT TABLE 2 ABOUT HERE]

As an example of how purchase event feedback may depend on the marketing mix at the

time of the prior purchase, consider observed changes in price sensitivity. Promotion

enhancement would suggest that, following a promotion by one brand, consumer sensitivities to

price and promotional activities increase for all brands in the category. We can determine

whether consumer price sensitivity meets this expectation by examining consumer switching

behavior in the tuna category. Following are two tabulations that show Period t and Period t+1

purchases of national and local brands. Specifically, in these switching matrices, “national”

brands include Starkist and Chicken of the Sea and “local” brands include 3 Diamond and the

control.

12

We begin by considering situations in which no promotion is offered during Period t or

Period t+1. We assume that consumers who buy a national brand at the regular price prefer that

brand, and select only those consumers who purchased a national brand during Period t, which

enables us to eliminate those consumers who tend to switch to national brands only during price

promotions. Thus, we include all consumers who buy a national brand in Period t, and examine

their purchases in Period t+1, to determine the baseline transition rate for consumers switching

to a local brand. The following tabulation shows the subsequent purchases by consumers who

purchased a national brand in Period t, considering all observations for which there are no

promotions in either period.

Period t+1 Purchases National Local Total Period t Purchases National 2975 266 3241 Percentage 92% 8%

Thus, 8% of the population switches away from national brands and to local brands when there

are no promotions; this can be considered the baseline transition probability.

To examine price sensitivity, we compare this baseline transition rate to the switching

behavior that occurs following a national brand promotion in Period t+1. Thus, we retain those

consumers who purchased a national brand for two consecutive periods, when there was no

promotion in Period t and a national brand promotion in Period t+1, then examine their

purchases in Period t+2. These consumers must be at least as loyal as the consumers in the

baseline situation because they have not only purchased a national brand when it was not on

promotion but also followed up with another national brand purchase in Period t+1. We assume

that when they buy the national brand on promotion in Period t+1, their loyalty to the brand does

not diminish.3 Therefore, if the rate of switching to a local brand increases among these

13

consumers when there is no promotion in Period t+2, the change is due to increased price

sensitivity caused by the promotion. The switching matrix shows the buying behavior for the

retained consumers in Period t+2, when there are no promotions in effect.

Period t+2 Purchases National Local Total Period t+1 Purchases National 1157 140 1297 Percentage 89% 11%

The local brands’ market share among this group of consumers increased significantly (t = 3.7;

p<0.01), from 8% to 11%, after the promotion by a national brand in Period t+1, as compared to

the no-promotion scenario. Therefore, this example provides evidence of increased price

sensitivity following a promotion by one of the national brands.

RESULTS

The key questions addressed by this research are as follows. (1) Are consumers’ responses to the

marketing mix influenced by the brand they purchased most recently? (2) Does it matter

whether the most recently purchased brand was on promotion at the time of the purchase? (3) If

both usage dominance and promotion enhancement effects occur, what are their relative

magnitudes? In this section, we examine the results of the model estimation and draw

conclusions regarding the relative strengths of each of these hypothesized effects.

Model Selection

To provide a comparison of model fit, we estimated a restricted model in which we set

the hierarchical coefficients (γ1,k and γ2,k) to zero but still incorporated continuous consumer

heterogeneity. This restricted model offers an appropriate benchmark because it represents what

our model would reduce to if the hypothesized effects were not observed. Among recent

research utilizing similar benchmark models, we note that close variations are employed by

14

Chintagunta and Dubé (2005) and by Bell, Bonfrer, and Chintagunta (2005). In Table 3, we

provide both the estimation and the holdout sample performance for both models and all four

product categories.

[INSERT TABLE 3 ABOUT HERE]

In Table 3, we offer four tests that compare the performance of the restricted and full

models: the likelihood ratio (LR), Akaike’s Information Criterion (AIC), Schwartz’s Information

Criterion (SIC), and the holdout sample likelihood. For all four product categories, the AIC and

holdout sample results recommend the full model; the LR test also indicates that the full model is

best (p<0.0001). The LR test is appropriate in this situation because the models are nested; that

is, the restricted model tests whether or not the key coefficients are zero. The SIC recommends

the full model for three of the four product categories, but for tissue, it indicates that the

restricted model is best. We investigated this puzzling result and found it is an artifact that

occurs because we included the hierarchical equation for the display × feature ad interaction

variable. If we remove the hierarchical equation for this interaction (i.e., set γ1,8 and γ2,8 to zero),

the SIC indicates that the full model is best for all product categories. Although the full model

was not initially selected for tissue because of the number of insignificant parameters associated

with the hierarchical equation for the interaction term, on the basis of these findings, we

conclude that it is appropriate to select the full model for all four product categories.

Discussion of Estimation Results

In Table 4, we provide the coefficient values obtained using the hierarchical equations for

each product category.4 Our results exhibit face validity because all four categories have the

appropriate signs and significances for the main effects of the marketing mix variables (price,

deal depth, feature advertising, and display). The coefficients for state dependence are positive

15

and significant, which indicates that consumers are inertial in their purchasing behavior, and tend

to repurchase the same brands. Further, the significant negative coefficient for the interaction

between state dependence and time since last purchase suggests that this inertial behavior wears

off as the time between purchases increases. Thus, our results support a main effect of usage

dominance.

[INSERT TABLE 4 ABOUT HERE]

In all four product categories, our results support the promotion enhancement effect, in

that consumers’ tendencies to repurchase the same brand diminish if the last purchase occurred

on promotion (although the effect is significant in only three of four categories). This finding is

consistent with empirical analyses in the literature (e.g., Van Heerde, Gupta, and Wittink 2003)

that indicate that a significant portion of the increased volume for a brand on promotion is due to

brand switching. Thus, it is not surprising that many consumers tend to switch to another brand

following a promotional purchase.

Regarding consumer response to regular price following a promotional purchase, all four

categories exhibit negative coefficients, supporting promotion enhancement, although the

coefficient for stick margarine is only marginally significant (p<0.06). Only two of the four

categories (tissue and stick margarine) provide support for effects of usage dominance. In terms

of response to price promotions, for deal depth, both usage dominance and promotion

enhancement are supported. The negative state dependence variables for all four categories

(significant for three of four, and marginally significant for peanut butter) support usage

dominance effects. Further, consumers become more sensitive to price promotions in a product

category after purchasing a brand on promotion, as indicated by positive and significant

coefficients in all four categories.

16

With regard to non-price promotions, we find support for the promotion enhancement

effect of feature advertisements; coefficients in all four product categories have the correct sign

and three are significant. We also find some support for the usage dominance effect through

features, in that three of four categories are significant and have the correct sign. However,

neither usage dominance nor promotion enhancement effects clearly emerge for displays, which

indicates that consumer response to displays is not affected by a purchase in the previous period

(whether promoted or not). This is consistent with the findings of Seetharaman, Ainslie, and

Chintagunta (1999), who do not find any significant impact of in-store displays.

Managerial Implications of Results

On the basis of our results, we can make several managerial recommendations. However,

we first must consider the portion of changes in total choice elasticities due to usage dominance

and promotion enhancement.5 Because we use hierarchical equations, the elasticity associated

with usage dominance is the derivative of the choice probability w.r.t. γ1,k and the elasticity

associated with promotion enhancement is the derivative of the choice probability w.r.t. γ2,k

(k=4,5,6,7).6 The total choice elasticity is thus the sum of the elasticities w.r.t. γ0,k, γ1,k, and γ2,k.

This enables us to calculate the portion of the effects due to each cause as its relative share of the

total elasticity. We provide these estimates in Table 5.

Table 5 reflects the impact of usage dominance (UD%), based on how much the total

choice elasticity is modified by the interaction between state dependence and each of the

variables in the model. Similarly, the impact of promotion enhancement (PE%) is given by the

degree to which the total elasticity is altered by the interaction between a prior promotional

purchase and each variable. The results in Table 5 suggest that margarine is the least elastic of

the four product categories for all marketing mix elements. Tuna, peanut butter, and tissue have

17

similar elasticities for regular price, deal depth, and feature ads, as we anticipated, but displays

do not engender any consistent effects.

[INSERT TABLE 5 ABOUT HERE]

Considering regular price elasticity, we find patterns of effects due to usage dominance

and promotion enhancement. Specifically, the impact of usage dominance (UD%) on price

elasticity for margarine is much greater than that for the other categories, whereas the magnitude

of the effects of promotion enhancement is greater for the other categories. Deal depth

elasticities suggest that deals have the greatest impact in the tuna category, and the least impact

for margarine. Further, changes in this type of elasticity due to state dependence are greatest for

margarine, but those due to the promotion enhancement effect are greater for the other three

categories. Thus, regular price and price-based promotions have similar state dependence and

promotion enhancement effects. Moreover, promotion enhancement influences all brands in the

product category.

Feature advertisements lead to sales increases in all four product categories and reflect

the expected direction of effects in three of four categories for usage dominance, and in all four

categories for promotion enhancement. The results for margarine again indicate very strong

effects of state dependence, whereas the strongest impact of promotion enhancement occurs in

the tissue category. Thus, by decomposing the contributions to elasticity of usage dominance

and promotion enhancement effects, we find that using promotions to obtain purchases may be

very effective in the margarine category, because state dependence effects consistently outweigh

promotion enhancement effects. However, for the other three product categories, marketers must

consider what promotions may be expected to do with regard to other brands in the category,

before they make a decision about what promotions to implement.

18

To summarize, for regular price, price promotions, and feature advertisements, the effects

of promotion enhancement are greater than those of usage dominance, in three of four product

categories. (No such pattern emerges for in-store displays.) We do not have clear results to

explain why margarine behaves differently than the other three product categories, and no

relationships emerge readily from the category data provided in Table 2. We suspect that the

differences may be related to the shelf life of the product category; margarine stock must be

turned over more rapidly than stocks of the other three categories. Our results also show the

importance of including both usage dominance and promotion enhancement effects in the model.

If we were to ignore usage dominance, switching behavior would be overstated because of the

increases in price and promotional sensitivities, and the brand’s ability to retain new consumers

following a promotional trial would be understated. By contrast, were we to neglect promotion

enhancement, switching behavior would be underestimated and the ability of a brand to retain

consumers after a promotion would be overestimated.

Our findings support several promotional strategy implications, for both retailers and

manufacturers. Retailers often use national brand promotions to draw consumers to their stores,

even though such promotions work against the retailer’s house brands (Dhar and Hoch 1997).

However, because promotion enhancement outweighs usage dominance, national brand

promotions could improve retailers’ profits by sensitizing consumers to price and promotions.

Increased price sensitivity, in particular, might benefit small-share and/or store brands, for which

retailers usually enjoy higher margins, leading to higher profits. As a cautionary note, retailers

should monitor the competition among and between national and store brands carefully when

developing a product category promotion strategy, because increased price competition also may

lead to reductions in overall category profits (Raju, Sethuraman, and Dhar 1995). Our results

19

offer a second explanation as to why retailers successfully can offer price promotions of national

brands to increase their store profits: such promotions erode consumer loyalty to national brands,

which could make it easier for private labels to gain market share and thus obtain increases in

store profits (see also Gedenk and Neslin 1999).

Unlike retailers, whose primary concern must be the profitability of entire product

categories, manufacturers are interested in maintaining or improving only their brands’ market

shares without sacrificing profitability (Basuroy, Mantrala, and Walters 2001). Our results

suggest that there may be differences by product category in the relative strengths of usage

dominance and promotion enhancement effects. Manufacturers can use our research to develop

a clear understanding of each brand’s promotional effectiveness, by decomposing the

contributions of usage dominance and promotion enhancement. This effort would help

manufacturers to better use their resources (e.g., trade incentives) to move retailers toward

desirable pricing and promotional strategies for each brand.

To assess how market share estimates and the resulting managerial implications might be

different if we had not included the effects of promotion enhancement and usage dominance in

our model, we performed a simulation. Two product categories were selected, margarine

because it exhibits strong effects of usage dominance, and tuna because it shows greater impact

from promotion enhancement. We simulated a promotion of the leading brand in Period 1, and

tracked the resulting market shares in subsequent periods. In addition, we calculated the

simulated shares by market segment, which we assigned based on whether a consumer had

purchased the brand in the previous period. We then obtained brand shares by summing across

the segments. We also gathered simulated shares for the restricted model, for comparison. We

present the results in Table 6.

20

[INSERT TABLE 6 ABOUT HERE]

The results show that the restricted model severely under-predicts market shares for smaller

and house brands, compared to the full model. Thus, if we do not consider state dependence and

promotional effects, the market shares of national brands are overstated, which will affect

promotional strategy recommendations. Specifically, the share of the promoting brand might be

overstated because the effects of state dependence on consumer response have not been included.

(This possibility is reflected in the negative values for the leading brand, and positive values for

other brands, in Period 1.) Kopalle et al. (1999) report a similar phenomenon, in which brand

managers failed to account for the dynamic effects of discounting on sales. We also note that the

shares of the smallest two brands are dramatically understated in both Periods 1 and 2. Thus,

during a promotion, if store brands are more profitable to a retailer than national brands, this

profitability may be understated, particularly in categories that exhibit stronger promotion

enhancement effects.

CONCLUSION

The primary purpose of this research has been to observe any differences in consumer response

to marketing mix activities for brands in a product category after regular or promotional

purchases in the category. Further, we hope to improve understanding of whether recent

consumer purchases tend to lead to usage dominance, promotion enhancement, or a combination

of these effects. The unique features of our estimated model enable us to tease out each of these

theoretical drivers of promotional response, as well as their combined effects.

Prior usage of a brand and prior promotional activities can both play roles in driving

consumer promotional sensitivities. In general, households that previously purchased a non-

promoted brand are more likely to buy it again, while those that bought on promotion are less

21

likely to repeat buy. More specifically, we observe both usage dominance and promotion

enhancement effects in all four product categories that we investigate. The influence of

promotion enhancement outweighs that of usage dominance in three categories (c.f. margarine),

so the typical aggregate-level observed behavior appears more consistent with promotion

enhancement. Between-category differences may explain this result; for example tuna, peanut

butter, and tissue manufacturers may offer more innovations than margarine makers, or

differences in product shelf lives may explain differences in buying behavior.

We note that the effects of promotion enhancement and usage dominance typically

coexist and influence brand choice jointly. Further, when the effects of promotion enhancement

are greater than the effects of usage dominance, national brand managers should avoid promoting

too often, because the resulting increased price sensitivity can drive consumers to lower-priced

local and regional brands. However, retailers may prefer frequent promotions of national brands,

because the increases in consumers’ price sensitivities lead to increased sales of local and

regional brands. When the effects of usage dominance are greater than those of promotion

enhancement, promotions are a very effective tool for building market share, though we qualify

this conclusion in two respects: (1) the brand manager should perform a profitability analysis

prior to offering a promotion and (2) extensive promotion could lead to a situation in which the

profitability for all brands is diminished. As an example, consider the tuna and tissue categories,

both of which have greater promotion enhancement than usage dominance effects. For both

categories, more than 50% of all purchases occur on promotion, which implies that a category

can be over-promoted, harming the national brands. Therefore, further research should

determine specifically which factors drive promotion enhancement and usage dominance, and

identify the “tipping point” at which one effect begins to overtake the other.

22

Our results help explain the reasons for mixed findings in previous literature. If a

researcher focuses only on the effect of promotions on the promoted brand, the relative

magnitudes of promotion enhancement and usage dominance determine the sign and duration of

the observed effects. However, researchers and managers might overlook the larger impact of

increasing consumer sensitivity to prices and promotions of all brands in the category, when the

focus is at the brand level. Our results take a broader viewpoint and suggest a potential negative

impact of promotions that enhance consumer sensitivity for all brands after a promotion for any

brand in the category.

Several challenging opportunities remain for future research. Although we are confident

that our findings are generalizable because they are theory-driven and show generally consistent

results across four product categories, further testing in other, more varied product categories

would be appropriate. The present methodology also could be tested in other choice contexts in

which purchase behavior may change over time; for example, the situational need for a shopping

trip (e.g., quick versus planned trip) may affect consumer buying. The consumer’s reason for

buying also could influence his or her response to a promotion; if a consumer is purchasing for

someone else, whether as a gift-giver or a service provider, this position could impact his or her

buying behavior as well as the relationship between preferences and response to marketing mix

variables. As we have noted, additional research that clarifies the conditions under which we

might expect promotion enhancement to overtake usage dominance, and vice versa, would be

useful.

One limitation of the current study is that we do not take into account the possibility of

asymmetric effects across brands. This may prove to be a fruitful area for future research.

Despite this limitation and the ongoing need for more research, we believe our research increases

23

understanding of the factors that drive differences in consumer price and promotion sensitivities

across different circumstances, and thus provides valuable theoretical knowledge. Further, we

believe this topic is managerially relevant and that the increased understanding is an important

step toward refining strategies for temporary price reductions and other promotional activities.

24

REFERENCES

Alba, Joseph W., J. Wesley Hutchinson, and John G. Lynch (1991), “Memory and Decision

Making,” in Handbook of Consumer Behavior, H. Kassarjian and T. Robertson eds.

Englewood Cliffs, NJ: Prentice-Hall, 1-49.

Anderson, N.H. (1971), “Integration Theory and Attitude Change,” Psychological Review. 78(3).

171-206.

Banker, Rajiv D. and Srikant M. Datar (1989), “Sensitivity, Precision, and Linear Aggregation of

Signals for Performance Evaluation,” Journal of Accounting Research. 27(1). 21-39.

Basuroy, Suman, Murali K. Mantrala, and Rockney G. Walters (2001), “The Impact of Category

Management on Retailer Prices and Performance: Theory and Evidence,” Journal of

Marketing. 65(4). 16-32.

Bell, David R., André Bonfrer, and Pradeep K. Chintagunta (2005), “Recovering Stockkeeping-

Unit-Level Preferences and Response Sensitivities from Market Share Models Estimated

on Item Aggregates,” Journal of Marketing Research. 42(3). 368-379.

Bettman, James R. and C. Whan Park (1980), “Effects of Prior Knowledge on Experience and

Phase of the Choice Process on Consumers’ Decision Processes: A Protocol Analysis,”

Journal of Consumer Research. 7(3). 234-248.

Blattberg, Robert C., Richard A. Briesch and Edward C. Fox (1995), “How Consumer Sales

Promotions Work,” Marketing Science. 14(3/2). G122-G132.

Boulding, William, Eunkyu Lee, and Richard Staelin (1994), “Mastering the Mix: Do

Advertising, Promotion, and Sales Force Activities Lead to Differentiation?” Journal of

Marketing Research. 31(2). 159-172.

25

Briesch, Richard A., Pradeep K. Chintagunta and Rosa L. Matzkin (2002), “Semiparametric

Estimation of Brand Choice Behavior,” Journal of the American Statistical Society.

97(December). 973-982.

Bucklin, Randolph E. and Sunil Gupta (1992), “Brand Choice, Purchase Incidence and

Segmentation: An Integrated Modeling Approach,” Journal of Marketing Research.

29(2). 201-15.

Cacioppo, John T. and Richard E. Petty (1985), “Central and Peripheral Routes to Persuasion:

The Role of Message Repetition,” in Psychological Processes and Advertising Effects:

Theory, Research, and Applications, L.F. Alwitt and A.A. Mitchell eds. Hillsdale, NJ:

Lawrence Erlbaum Associates, 91-111.

Chintagunta, Pradeep K. (1999), “Variety Seeking, Purchase Timing, and the ‘Lightning Bolt’

Brand Choice Model,” Management Science. 45(4). 486-498.

Chintagunta, Pradeep K., and Jean-Pierre Dubé (2005), “Estimating a Stockkeeping Unit-Level

Brand Choice Model That Combines Household Panel Data and Store Data,” Journal of

Marketing Research. 42(3). 368-379.

Dhar, Sanjay K., and Stephen J. Hoch (1997), “Why Store Brand Penetration Varies by

Retailer,” Marketing Science. 16(3). 208-227.

Dodson, Joe A., Alice M. Tybout and Brian Sternthal (1978), “Impact of Deals and Deal

Retraction on Brand Switching,” Journal of Marketing Research. 15(1). 72-81.

Erdem, Tülin and Baohong Sun (2001), “Testing for Choice Dynamics in Panel Data,” Journal

of Business and Economic Statistics. 19(2). 142-152.

26

Fader, Peter S. and Leonard M. Lodish (1990), “A Cross-Category Analysis of Category

Structure and Promotional Activity for Grocery Products,” Journal of Marketing. 54(4).

52-65.

Fader, Peter S. and Leigh McAlister (1990), “An Elimination by Aspects Model of Consumer

Response to Promotion Calibrated on UPC Scanner Data,” Journal of Marketing

Research. 27(3). 322-332.

Fazio, Russell H., Jeaw-Mei Chen, Elizabeth C. McDonel and Steven J. Sherman (1982),

“Attitude Accessibility, Attitude Behavior Consistency, and the Strength of the Object

Evaluation Association,” Journal of Experimental Social Psychology. 18(July). 339-357.

Fazio, Russell H., Martha C. Powell and Carol J. Williams (1989), “The Role of Attitude

Accessibility in the Attitude-to-Behavior Process,” Journal of Consumer Research. 16(3).

280-288.

Fazio, Russell H. and Mark P. Zanna (1978), “Attitudinal Qualities Relating to the Strength of

the Attitude-Behavior Relationship,” Journal of Experimental Social Psychology.

14(July). 398-408.

Gedenk, Karen and Scott A. Neslin (1999), “The Role of Retail Promotion in Determining

Future Brand Loyalty: Its Effect on Purchase Event Feedback,” Journal of Retailing.

75(4). 433-459.

Grover, Rajiv and V. Srinivasan (1992), “Evaluating the Multiple Effects of Retail Promotions

on Brand Loyal and Brand Switching Segments,” Journal of Marketing Research. 29(1).

76-89.

27

Guadagni, Peter M. and John D.C. Little (1983), “A Logit Model of Brand Choice Calibrated on

Scanner Data,” Marketing Science. 2(3). 203-238.

Hajivassilios, V.A. and P.A. Rudd (1994), “Classical Estimation Methods for LDV Models

Using Simulation,” Pp 2384-2441 in Handbook of Econometrics. R.F. Engle and D.L.

McFadden eds. Holland: Elsevier.

Heilman, Carrie M., Douglas Bowman and Gordon P. Wright (2000), “The Evolution of Brand

Preferences and Choice Behaviors of Consumers New to a Market,” Journal of

Marketing Research. 37(2). 139-155.

Inman, J. Jeffrey and Leigh McAlister (1993), “A Retailer Promotion Policy Model Considering

Promotion Sensitivity,” Marketing Science. 12(4). 339-356.

Jedidi, Kamel, Carl F. Mela and Sunil Gupta (1999), “Managing Advertising and Promotion for

Long-Run Profitability,” Marketing Science. 18(1). 1-22.

Johnson, Eric J. and J. Edward Russo (1984), “Product Familiarity and Learning New

Information,” Journal of Consumer Research. 11(1). 542-550.

Johnson, Tod (1984), “The Myth of Declining Brand Loyalty,” Journal of Advertising Research.

24(1). 9-17.

Kamakura, Wagner A. and Gary J. Russell (1989), “A Probabilistic Choice Model for Market

Segmentation and Elasticity Structure,” Journal of Marketing Research. 26(4). 379-390.

Kopalle, Praveen K. and Donald R. Lehmann (1995), “The Effects of Advertised and Observed

Quality on Expectations about New Product Quality,” Journal of Marketing Research.

32(3). 280-290.

28

Kopalle, Praveen K. and Donald R. Lehmann (2006), “Setting Quality Expectations When

Entering a Market: What Should the Promise Be?” Marketing Science. Forthcoming.

Kopalle, Praveen K., Carl F. Mela, and Lawrence Marsh (1999), “The Dynamic Effect of

Discounting on Sales: Empirical Analysis and Normative Pricing Implications,”

Marketing Science. 18(3). 317-332.

Mela, Carl F., Sunil Gupta and Donald R. Lehmann (1997), “The Long-Term Impact of

Promotion and Advertising on Consumer Brand Choice,” Journal of Marketing Research.

34(2). 248-261.

Moorthy, Sridhar, Brian T. Ratchford and Debabrata Talukdar (1997), “Consumer Information

Search Revisited: Theory and Empirical Analysis,” Journal of Consumer Research.

23(4). 263-277.

Narasimhan, Chakravarthi, Scott A. Neslin and Subrata K. Sen (1996), “Promotional Elasticities

and Category Characteristics,” Journal of Marketing. 60(2). 17-30.

Neslin, Scott A. (2002), Sales Promotion. Cambridge, MA: Marketing Science Institute.

Neslin, Scott A. and Robert W. Shoemaker (1989), “An Alternative Explanation for Lower

Repeat Rates after Promotion Purchases,” Journal of Marketing Research. 26(2). 205-

213.

Papatla, Purushottam and Lakshman Krishnamurthi (1996), “Measuring the Dynamic Effects of

Promotions on Brand Choice,” Journal of Marketing Research. 33(1). 20-35.

Petty, Richard E. and John T. Cacioppo (1986), Communication and Persuasion: Central and

Peripheral Routes to Attitude Change. NY: Springer-Verlag.

29

Raju, Jagmohan S., Raj Sethuraman, and Sanjay K. Dhar (1995), “The Introduction and

Performance of Store Brands,” Management Science. 41(6). 957-978.

Seetharaman, P.B., Andrew Ainslie and Pradeep K. Chintagunta (1999), “Investigating

Household State Dependence Effects Across Categories,” Journal of Marketing

Research. 36(4). 488-500.

Shankar, Venkatesh and Lakshman Krisnamurthi (1996), “Relating Price Sensitivity to Retailer

Promotional Variables and Pricing Policy: An Empirical Analysis,” Journal of Retailing.

72(3). 249-272.

Smith, Robert E. and William R. Swinyard (1982), “Information Response Models: An

Integrated Approach,” Journal of Marketing. 46(1). 81-93.

Smith, Robert E. and William R. Swinyard (1983), “Attitude-Behavior Consistency: The Impact

of Product Trial versus Advertising,” Journal of Marketing Research. 20(3). 257-267.

Strang, Roger A. (1975), The Relationship between Advertising and Promotion in Brand

Strategy. Cambridge, MA: Marketing Science Institute.

Totten, John and Martin Block (1987), Analyzing Sales Promotion: Test and Cases. Chicago, IL:

Commerce Communications.

Van Heerde, Harald J., Sachin Gupta and Dick R. Wittink (2003), “Is ¾ of the Sales Promotion

Bump Due to Brand Switching? No, it is 1/3,” Journal of Marketing Research. 40(4).

481-491.

Van Heerde, Harald J., Peter S.H. Leeflang and Dick R. Wittink (2000), “The Estimation of Pre-

and Postpromotion Dips with Store-Level Scanner Data,” Journal of Marketing

Research. 37(3). 383-395.

30

Van Heerde, Harald J., Peter S.H. Leeflang and Dick R. Wittink (2001), “Semiparametric

Analysis to Estimate the Deal Effect Curve,” Journal of Marketing Research. 38(2). 197-

216.

31

Table 1: Hypothesized Directions of Effects for Promotion Enhancement and Usage Dominance

Promotion Enhancement (Sign of γ1,k) Usage Dominance (Sign of γ2,k)

State Dependence Negative. No coefficient γ2,k. Regular Price

Negative. Consumers are more responsive to price for all brands in the category.

Positive. Consumers are less price sensitive for brands they purchase off promotion.

Deal Depth

Positive. Purchases on promotion increase attention to price cut promotions.

Negative. Consumers are less sensitive to promotions for previously purchased brands.

Feature Advertisement

Positive/zero. Purchases on promotion may increase response to non-price promotions.

Negative. Consumers are less sensitive to promotions for previously purchased brands.

In-Store Display

Positive/zero. Purchases on promotion may increase response to non-price promotions.

Negative. Consumers are less sensitive to promotions for previously purchased brands.

32

Table 2: Descriptive Statistics for ERIM Category Data

Tuna Peanut Butter

Stick Margarine Tissue

Number of Households 3,081 2,115 1,780 1,370 Number of Brand-Sizes 4 6 4 5 Avg. Purchases/Household 21.3 16.4 24.8 27.2 % Purchases on Promotion 68.8% 32.1% 41.4% 51.7% Avg. Interpurchase Time 40.9 56.1 37.0 29.0 Loyalty 64.7% 69.1% 72.6% 66.3% Average % Time Featured 15.3% 3.2% 12.4% 5.1% Average % Time Displayed 6.4% 2.4% 5.6% 4.7% Size Metric Oz Oz Oz Rolls Brand-Sizes and Shares Starkist-6.5 Peter Pan-18 Parkay-16 Scott-4 52.7% 29.8% 47.1% 34.6% Chicken of Sea-6.5 Control-18 Control-16 Northern-4 30.5% 25.5% 17.3% 28.1% Control-6.5 Jif-18 B-B-16 Charmin-4 9.5% 12.9% 12.7% 11.4% 3 Diamond-6.5 Skippy-18 Fleischman-16 Northern-6 3.2% 7.7% 10.6% 3.2% Peter Pan-28 Scott-6 7.0% 2.6% Jif-28 6.4% Total Share 95.9% 89.3% 87.7% 79.9%

Note: We calculate average % of time featured and displayed by weighting the items’ frequency percentages by their relative market shares.

33

Table 3: Estimation Results

Tuna Peanut Butter Margarine Tissue

Rstr Full Rstr Full Rstr Full Rstr Full

-LL 8,173 8,101 9,186 9,131 5,751 5,639 7,220 7,174

Families

3,061

2,115

1,780

1,352

Choices

18,252

9,254

10,919

9,634 Parameters

20

32

22

34

20

32

20

32

Likelihood Ratio

144

110

226

93

Prob.(LR)

0.0000

0.0000

0.0000

0.0000

AIC

16,386

16,266

18,416

18,330

11,543

11,341

14,481

14,412

Estimation Sample

SIC

16,542

16,516

18,573

18,572

11,689

11,575

14,624

14,641

-LL

8,138

8,033

9,105

9,058

5,644

5,542

6,833

6,806

Families

3,081

2,104

1,778

1,370 Holdout Sample

Choices

18,018

8,959

10,703

9,358 Notes: Rstr = restricted model, LR = likelihood ratio, AIC = Akaike’s information criterion, SIC = Schwartz’s information criterion, and –LL = negative log likelihood. Boldface entries indicate best fit model.

34

Table 4: Parameter Estimates for Hierarchical Equations

Peanut Stick Butter Margarine Tissue Tuna

Main (γ0,2) 2.30 (0.20) 0.56 (0.29) 1.24 (0.23) 1.37 (0.30)Promotion (γ1,2) -1.17 (0.24) -0.90 (0.24) -0.15 (0.22) -0.61 (0.18)Heterogeneity 0.11 (0.10) 10.26 (1.09) 2.94 (0.39) 0.80 (0.29)

State Dependence

Distribution (ζ2)

Main (γ0,3) -0.15 (0.04) -0.11 (0.05) -0.16 (0.05) -0.19 (0.04)Promotion (γ1,3) 0.19 (0.07) 0.05 (0.07) -0.09 (0.07) 0.02 (0.06)Heterogeneity 0.31 (0.04) 0.01 (0.01) 0.01 (0.01) 0.03 (0.02)

State Dependence

* Time Distribution (ζ3)

Main (γ0,4) -2.82 (0.17) -5.39 (0.36) -4.06 (0.26) -9.32 (0.49)State Dep. (γ2,4) -0.12 (0.08) 2.96 (0.33) 0.35 (0.13) 0.49 (0.35)Promotion (γ1,4) -0.47 (0.08) -0.42 (0.25) -0.60 (0.14) -1.89 (0.37)Heterogeneity 1.44 (0.46) 20.94 (3.93) 22.77 (2.41) 14.33 (4.09)

Regular Price

Distribution (ζ4)

Main (γ0,5) 5.84 (0.33) 12.56 (0.71) 7.96 (0.38) 12.33 (0.49)State Dep. (γ2,5) -0.71 (0.49) -2.26 (0.88) -0.85 (0.42) -1.61 (0.37)Promotion (γ1,5) 1.60 (0.49) 1.55 (0.82) 0.84 (0.40) 3.25 (0.51)Heterogeneity 31.20 (8.12) 135.74 (27.32) 58.17 (10.08) 60.80 (11.08)

Deal Depth

Distribution (ζ5)

Main (γ0,6) 2.32 (0.30) 1.01 (0.18) 0.78 (0.22) 1.73 (0.17)State Dep. (γ2,6) -0.71 (0.39) -0.45 (0.24) 0.42 (0.32) -0.36 (0.21)Promotion (γ1,6) 0.43 (0.38) 0.85 (0.24) 0.56 (0.28) 0.54 (0.19)Heterogeneity 12.10 (7.57) 0.42 (0.54) 2.87 (2.19) 3.71 (1.26)

Feature Ad

Distribution (ζ6)

Main (γ0,7) 2.13 (0.23) 1.19 (0.25) 1.31 (0.20) 3.64 (0.27)State Dep. (γ2,7) 0.38 (0.38) -0.15 (0.35) 0.02 (0.23) -0.79 (0.28)Promotion (γ1,7) -0.09 (0.36) 0.05 (0.34) 0.33 (0.23) 0.43 (0.31)Heterogeneity 3.00 (1.48) 0.01 (0.07) 0.57 (0.96) 3.58 (1.25)

Display

Distribution (ζ7)

Main (γ0,8) -1.15 (0.99) -0.49 (0.41) -1.65 (0.46) -2.13 (0.56)State Dep. (γ2,8) 2.72 (2.19) -0.03 (0.63) -0.02 (0.60) 0.21 (0.65)Promotion (γ1,8) -0.47 (1.33) -0.61 (0.62) -0.17 (0.56) -0.75 (0.68)

Feature *Display

Heterogeneity 1.32 (3.19) 0.10 (0.38) 1.73 (1.63) 0.80 (1.71) Distribution (ζ8)

Notes: Standard errors are in parentheses. Bolded items are significant (p<0.05, one-tailed test) and bolded, italicized items are significant to a lesser degree (p<0.10, one-tailed test).

35

Table 5: Elasticity Decomposition for Price and Promotional Activities

Tuna Peanut Butter Margarine Tissue

Regular Price Total Elasticity -3.2 -3.2 -1.1 -3.0 (0.18) (0.22) (0.10) (0.24) UD % -3% 2% -56% -5% PE % 13% 5% 4% 7% Deal Depth Total Elasticity 1.5 1.1 0.5 1.3 (0.02) (0.02) (0.02) (0.01) UD % -6% -6% -10% -5% PE % 18% 12% 7% 7% Feature Ad Total Elasticity 0.4 0.8 0.2 0.6 (0.03) (0.12) (0.06) (0.11) UD % -6% -13% -66% 4% PE % 16% 3% 27% 40% Display Total Elasticity 1.5 2.7 0.0 0.9 (0.08) (0.43) (0.01) (0.12) UD % -13% 52% -20% 0% PE % 8% 0% -29% 17% Notes: To access the actual elasticities obtained in the decomposition, multiply the percentage by the total elasticity. For example, the UD elasticity for regular price in the margarine category is given by (-56%) × (-1.1) = 0.6. The average overall price elasticity is -1.1, and the average decrease in price elasticity due to usage dominance is 0.6. Therefore, if usage dominance effects did not occur, the average price elasticity would be -1.7.

36

Table 6: Results of Marketplace Simulation

Tuna Product Category: Full Model Restricted Model Market Share Market Share

Period Starkist Chicken of Sea Control

3 Diamond

Starkist

Chicken of Sea Control

3 Diamond

0 57.2% 24.2% 11.9% 6.7% 57.2% 24.2% 11.9% 6.7%1 82.3% 10.5% 4.7% 2.5% 86.4% 8.5% 3.4% 1.8%2 60.5% 19.2% 13.3% 6.9% 62.3% 22.1% 10.0% 5.7%

Difference 1 -4.1% 2.0% 1.3% 0.8% 2 -1.8% -2.8% 3.4% 1.2%

Percentage Difference

1 -5.0% 19.1% 28.3% 30.5% 2 -2.9% -14.8% 25.2% 18.1%

Margarine Product Category:

Full Model Restricted Model Market Share Market Share

Period Parkay Control B-B

Fleisch-man Parkay Control B-B

Fleisch-man

0 53.7% 19.8% 14.5% 12.0% 53.7% 19.8% 14.5% 12.0%1 66.8% 14.2% 10.9% 8.1% 76.1% 13.9% 6.5% 3.5%2 59.2% 19.9% 12.9% 8.0% 72.5% 17.1% 7.0% 3.3%

Difference 1 -9.3% 0.3% 4.4% 4.6% 2 -13.4% 2.8% 5.9% 4.7%

Percentage Difference

1 -13.9% 1.9% 40.1% 57.3% 2 -22.6% 14.1% 45.6% 58.6%

37

APPENDIX A: DETAILS OF ESTIMATION METHOD

In this appendix, we provide more details of the estimation methodology used in this research.

First, we define the likelihood function for all of household h’s purchases (also called its

purchase string), conditional on household h’s parameter vector θh, as:

(A.1) ( ) htih yh

hti

T

t

N

i

hh

h yYL θθ ;1Pr1 1

== ∏∑= =

where Ys is the vector of item choices, Th is the number of choice occasions, and N is the number

of items available.7 Rather than estimating a separate parameter vector for each household, we

assume that the distribution of parameter vectors across households is multivariate normal; that

is, has a multivariate normal distribution with mean θ0 and variance Σ. We wish to estimate

the parameters of this multivariate normal distribution. Under the distributional assumptions

described, we can designate the likelihood function for all households as:

(A.2) θθθθ dfYLYLH

hh

h );();(),,(1

Σ=Σ ∏∫=

where );( Σθf is the distribution of the parameter vector θ conditional on the covariance matrix

Σ. In our case, we assume the distribution to be multivariate normal. The integral in equation

(A.2) can then be numerically integrated using simulated maximum likelihood estimation

(SMLE), as described by Hajivassilios and Rudd (1994).

Rather than drawing numbers from a normal distribution, we use quasi-random Halton

sequences to perform the numerical integration, which enables us to use considerably fewer

draws. For example, Bhat (2001) finds that 100 draws from a Halton sequence is equivalent to

approximately 1,000 – 1,500 draws from a normal sequence. In our application, we use 200

draws from a Halton sequence.8

38

We further assume that Σ is a diagonal matrix or that the unobserved heterogeneity is

uncorrelated among the predictors (which implies an additional K parameters, where K = number

of elements in θ). The additional k parameters are the variance terms of the unobserved

heterogeneity. Note that we allow correlation in the predictors through the relationships with

prior period promotions and state dependence. Although SMLE can be used to estimate the

vector of parameters and the covariance matrix, it is difficult to estimate such a large multivariate

normal distribution empirically; the number of parameters is o(n2) (Chintagunta 1999).

To perform the numerical integration, we calculate each household’s likelihood

separately (using equation A.1), then combine the households’ likelihoods (as in equation A.2) to

obtain the total likelihood. We define R as a matrix (ND = number of draws, by K) of random

draws and C as a lower-triangular K by K matrix of covariance terms (in our case, C is a diagonal

matrix), such that CC’ = Σ.9 Thus, θ0 + RC’ provides ND instances of the parameter vector. For

the numerical integration, we take the average household likelihood across the ND instances of

the parameter vector.

39

FOOTNOTES 1 Note that this argument does not necessarily imply that increased price sensitivity will be

observed at the aggregate level, as only those consumers who purchase on promotion are

expected to exhibit increased sensitivity when they next purchase.

2 We thank an anonymous reviewer for suggesting this connection.

3 This assumption is consistent with the findings of Gedenk and Neslin (1999), who observe that

consumer loyalty increases when a brand is purchased, regardless of whether it is on promotion,

although the increase is smaller when the brand is on promotion.

4 Detailed brand, size, and associated heterogeneity parameters are available from the authors

upon request.

5 Because feature advertising and displays are indicated by discrete variables, we calculate arc

elasticities; these values provide the change in probability when the promotion is present versus

absent.

6 The elasticities for feature advertising and display include the interaction effect.

7 For expository ease, we assume that N is not time-varying, although the method extends to this

situation. We further assume that all of the parameters of q contain unobserved consumer

heterogeneity, although the algorithm allows non-heterogeneous parameters similar to those in

the hierarchical equations. The non-heterogeneous parameters are incorporated by setting their

rows in Σ to zero.

8 SAS IML code is available from the authors upon request.

9 The matrix C is also known as the Cholesky matrix and the unconstrained elements of C are

optimization parameters.

40