RESEARCH PAPER NO.

1007

THE DYNAMICS OF CONSUMER RESPONSE

TO PRICE AND PROMOTION

James M. Lattin1

Randolph E. Bucklin2

January 1987
Revised August 1988

graduate School of Business

$~tanfordUniversity
Stanford, CA 94305

1. Assistant Professor of Marketing and Management Science and the James and Doris
McNamaraFaculty Fellow for 1987-1988

2. Ph.D Candidate

* The data used in this study are from a database made available for academic use by
Information Resources, Inc. This paper benefitted from comments offered by the
participants of the Stanford Marketing Seminar. The authors also wish to thank
Peter Fader, Leigh McAlister, Bill Perreault, Diane Schmalensee, V. Srinivasan,
Russ Winer, the members ofthe Marketing Science Institute Packaged Goods
Steering Group, and three anonymous JMR reviewers for helpful suggestions.
This work was supported by MSI and by the Stanford University Graduate School
of Business, through a faculty fellowship made possible by the generosity of James
and Doris McNamara.
 “The Dynamics ofConsumer Response to Price and Promotion”
)

ABSTRACT

This research examines the nature of dynamic consumer response to point-of-purchase

pricing and promotional activity. The authors begin with the premise that consumers form
expectations about future pricing and promotion based on their exposure to such activity, and
that consumer response is influenced by the disparity between the observed marketing activity
and the point of reference established by the consumer. The approach differs from existing
research in two ways: first, the model focuseson the dynamic effects ofpromotional activity
(i.e. feature and/or special display) separate from the effects ofchanges in price; second, a
threshold model is introduced to capture the formation of the consumer’s promotional reference
points. The authors calibrate a model ofbrand choice using IRI scanner panel data on ground
coffee. The findings suggest that there are significant reference effects of promotional activity
on consumer response, and that controlling for these promotional dynamics, the effects of
reference price appear to be non-significant.

KEY WORDS: Price, Promotion, Reference Price, Brand Choice, Logit.

 1. INTRODUCTION

Each year, companies spend billions ofdollars on trade promotion in an attempt to induce

retailers to offer stronger merchandising support (e.g., price reduction, feature, special display)

for theirbrands at the point of purchase. Although recent research has documented the success of

pricing and promotion in stimulating immediate sales response (e.g., Guadagni and Little 1983,

Neslin, Henderson and Quelch 1985), concerns about the long-run implications ofpromotional

activity continue to mount. Some industry experts contend that frequent price discounting over

tinie blurs the distinction between the deal price and the baseline priceof a product (Marketing
News, 1985). Ifconsumers come to expect special deals as the rule-rather than the exception,

then price promotions lose theirability to boost sales and become unprofitable. From the

perspectiveof the brand manager, the issue is how to get most from price promotions without

undermining their long-run effectiveness.

To address this issue, models of consumer response to point ofpurchase marketing

activity must be. extended to incorporate the dynamic nature of consumer response. For the most

part, recent models of brand choice and purchase incidence have not explicitly considered these

dynamic aspects. They cannot account forconsumers’ repeated exposure to price changes and

promotional activity, the likely effect of such exposure on perceptions, and any resultant changes

in consumer purchase behavior.

The purpose of this paper is to model the nature ofdynamic consumer response to price
and promotion. We define promotion to include any non-price merchandising activity at the

point-of-purchase (i.e., feature or special display). We begin with the premise that consumers

form expectations about the future marketing activity of each brand based on theirexposure to

such activity at the point of purchase. We label these expectations the reference price and the

promotional reference point for the brand. We further assume that consumers use these points of

reference as bases ofevaluation and that at each purchase opportunity consumer response is

-3-
influenced by the disparity between their reference points and the actual price and promorional

status of the brand.

Our approach extends a related streamof research that has sought to address the dynamic

nature of consumer response to price. This research is based upon the notion that the consumer
)
establishes a reference price forthe product (Monroe 1979). The reference pnce construct is

consistent with severalpsychological theories of consumer behavior and price perception,

including adaptation-level theory (Helson 1964) and assimilation-contrast theory (Sherif 1963).

Empirical work by Winer (1986) and Raman and Bass (1986) supports the existence of general

reference price effects in consumer brand choice behavior.

Our approach differs from existing research in two ways. First, our conceptualization and

empirical testing is an attempt to identify the dynamic effects of promotional activity while

controlling for the effects of changes in price. Previous research (e.g., Winer 1986) has studied
reference price effects without distinguishing between long-term changes in regularprice and

short-termpromotional price changes. Such an approach may not account for the natural

differences between regularpriceelasticity and short termpromotional price elasticity (Rao and

Sabavala 1980) which are then potentially confounded with reference price-effects. Our goal is to

show that patterns of consumer response will change with increased exposure to recent

promotional activity, and to assess the effect ofreference price while controlling for these

dynamic promotional effects.

Second, our approach differs in that we use a threshold model to capture the formation of

the consumer’s promotional reference point for a brand. We assume that when the consumer has

been exposed to little or no promotional activity at the point ofpurchase, he or she has no

expectations about the availability of special offers in the future; i.e., the consumer thinks ofthe

brand as a non-promoted brand. As exposure increases beyond a certain point, the consumer

forms a point ofreference that reflects the brand’s recent promotional activity; i.e., he or she

-4-
begins to think of the brand as a promoted brand. The consumer’s point of reference (either

“promoted” or “non-promoted”) is determined by whether or not his orher recent exposure to

promotional activity by the brand is above or below a certain threshold. The model enables us to

separate the effects of promotion frequency (which operates on exposure, determining whether

the consumer thinks of the brand as promoted or non-promoted) from depth of discount (which

determines the point-of-reference price level). In this way, the threshold model differs from the

continuous expectations models offered by Winer (1986) and Raman and Bass (1986).

Our model of consumer response suggests that promotional pricing will have a series of

implications forthe long-run. For brand choice, a consumer exposed to frequent price promotion
may become accustomed to fmding the brand available on promotion at a special price. This shift

in the consumer’s point of reference changes the way in which the consumer frames the choice

problem. The result is a diminished level of consumer response to the brand -- sort of a “wearout”

effect forpoint-of-purchase promotional pricing. Other researchers have noticed similar effects on

brand choice. Based on theirreview of several studies modeling price perceptions, Sawyer and
Dickson (1984) point out that “temporary decreases from the list price... serve to keep the

reference price low and postpone consumers’ adaptation to the higher list price.” For

manufacturers, this suggests that overly intensive promotion may increase expectations and

ultimately undermine consumer response.

The balance of the paper follows in several sections. In section 2, we discuss the dynamic
nature of consumer response to price and promotion and present ourproposed models of

reference point formation and brand choice behavior. In section 3, we present the results ofour

empirical tests. To obtain cross-sectional estimates of the model parameters, we use IRI scanner

panel data on ground coffee. In section 4, we discuss the results from our study, and in
section 5, we conclude the paper and offer some suggestions forfuture research.

-5-
 2. MODELING DYNAMIC RESPONSE TO PRICE PROMOTION

In the following development, we focus on conditional brand choice behavior. In other

words, once the consumer has decided to buy from the product category, how do price and
promotion affect his or her decision to purchase a particular brand? We begin in section 2.1 with

a discussion of reference price effects, and extend the framework to characterize the reference

effects ofpromotional activity. We then propose in section 2.2 a model ofconsumer utility

incorporating these reference effects, and relate utility to consumer response by means of a
multinomial logit model. Finally, in section 2.3, wepresent models ofreference price and

promotional reference point formation.

2. 1 Reference Effects of Price and Promotion.

Our model rests on the premise that a consumer assesses the utility of a brand for purchase

by comparison to some expectation or point of reference based on previous experience with the

brand. This notion of reference point comparison is embodied in severalpsychological theories

of consumer behavior, widely used with respect to price perception (for a review, see Sawyer and
Dickson 1984). Helson’s (1964) adaptation-level theory suggests that consumer response to

price may depend upon established comparison prices. These points of comparison may change

over time as consumers adapt to prevailing conditions. Assimilation/ contrast theory (Sherif

1963) holds that consumer response to price may depend on whether or not the price ofthe item

falls within a certain “latitude ofacceptance.” This reference range ofprice may be influenced by

recent price variation for the item over time or across stores. Empirical work by Gabor and

Granger (1964) supports the notion that consumers establish some range of acceptable price.

Winer (1986) also investigates the nature ofreference price effects on brand choice. He
proposes a linear probability model in which the probability of purchase for brand i is a function

of the observed price of brand i (expressed relative to the prices ofother brands) and the

-6-
difference betweenthe reference price and the observed price for the brand. This latter term is

designed to capture the positive or negative effects ofprice expectations (referred to by Winer as a
“sticker shock” effect for non-durable goods). For example, when the observed priceof the

brand is higher than the pricethe consumer expects to pay (i.e., when reference price is less than

observed price), there is a negative impact on purchase probability. Winerfinds evidence that

consumer response is significantly related to the disparity between reference price and observed

price.

While consistent with the hypothesized patterns suggested by reference price theory,

Winer’s model may be picking up something else. Frequently, short-term price changes are

accompanied by special promotional support from the retailer (e.g., an end-of-aisle display or a

featured advertisement). These promotional events often elicit much greater consumer response

than unpromoted changes in the regularprice of the product, a phenomenon noted by Rao and

Sabavala (1980) and empirically documented by Guadagni and Little (1983). In many categories
point-of-purchase promotions are irregularly scheduled and difficult to anticipate, resulting in

large discrepancies between observed price and reference price. Thus, it is possible that the

“sticker shock” effect uncovered by Winer is in fact confounded with the difference in elasticity

between promoted price discounts and changes in regular price.

In order to sort out these effects, we need to do two things. First, we must specify a

functional form for consumer response that will accommodate the differences between non-

promotional and promotional price response. And second, before we can claim to have evidence

of reference price effects, we must first control for the existence ofany promotional reference

effects. The existence of such effects is clearly plausible. Recent research suggests that

promotional activity has considerable impact on consumer response (e.g., Guadagni and Little

1983, Gupta 1986, Fader and McAlister 1988) and is at least as salient to consumers as a change

in the price of the brand. In a study of purchase behavior and priceperceptions, Dickson and

Sawyer (1986) report that nearly 50% of consumers correctly identified the deal status of the

-7-
brand purchased, closely comparable to the proportion of consumers who could correctly recall

the purchase price of the brand. Furthermore, there is a sizable body of research examining the

effects of prior promotional purchase on subsequent choice behavior (e.g. Scott 1976, Dodson,
Tybout and Sternthal 1978, Jones and Zufiyden 1981, Guadagni and Little 1983, Ortmeyer,

Lattin and Montgomery 1987).

We assume that the consumer develops a promotional reference point (PRP) based on his

or her prior exposure to the brand at the point of purchase. Ifthe consumer hasrelatively little
exposure to the brand on promotion (i.e., if there has been no promotional activity by the brand

on recent purchase occasions), then PRP=O and we assume the consumer expects to find the

brand off promotion. If the consumer’s exposure to promotional activity increases beyond a
threshold level, then PRP=1 and we assume the consumer expects to find the brand on

promotion.

Reference price (RP) is also determined by the consumer’s exposure to the brand at the

point ofpurchase. Thus, there is likely to be a relationship between PRP and RP. When there

has been little orno promotional activity by the brand during the consumer’s-recent purchase

occasions (i.e., PRP=0), RP will reflect the regular price of the brand on those previous
occasions. When the consumer has experienced considerable promotional activity by the brand

(i.e., PRP=1), RP will reflect the price of the brand on those promotional occasions, which will

likely involve special discounts from the regularprice. Thus, to the extent that promotional

activity is accompanied by special price discounts, PRP and RP will exhibit negative covariance.

-8-
2.2 Reference Effects on Consumer Response.

We propose the following utility model to capture the reference effects of price and

promotion in consumer response:

h h h h

~ = ~.tit+ + (Yi + Y2 Z1~)P~+ ö~(~ ‘~-‘~)

+ 6~
- (Zit PRP~) - (1)

where t = a subscript denoting a purchase occasion for consumer h,

= a utility function, reflecting the utility to consumer h ofpurchasing brand i on

purchase occasion t,
= consumer h’s base-level utility forbrand i at time t (influenced by non-marketing

factors such as brand preference, choice inertia, etc.),

= the purchase price of brand i at time t,

Zj~= an indicator of promotional activity at time t, and

PRP~, RP~ consumer h’s promotional reference point and reference price,

respectively, forbrand i at time t.

The specification in (1) differs from existing models embodying price effects in two ways. First,

the form explicitly accounts for different consumer response to promoted and non-promoted price

changes. The coefficient ‘Yo captures the direct effect ofpromotion, ~i captures the direct effect of

price, and Y2 captures any interaction between promotion and price. Second, the model includes
reference effects forpromotion as well as price. The coefficient 6~captures the reference effects

of price while &2 captures the reference effects due to promotion. For parsimony, we have

assumed that the parameters ~O’~ ~2’ ~1’ and 6~are the same across consumers, and that term
accommodates the remaining variance in response across consumers and over time.

By linking utility to consumer purchase behavior, we establish a relationship between the

reference effects ofprice and promotion and consumer response to the brand. If we assume that

the consumer buys the brand offering maximum utility and that any random effects on utility are

-9-
independently and identically distributed across brands according to the double-exponential

distribution, then the likelihood ofpurchasing brand i may be written:

h h h
= exp(u~)/~
exp(uk~) (2)

where p~= the probability that consumer h chooses brand i on purchase occasion t. The

multinomial logit model in (2) enables us to express purchase behaviortoward brand i as a

function ofconsumer response to all available brands. Thus, we avoid the need to employ a

constructed relative price -- like that used by Winer (1986) -- to capture competitive pricing

effects. While the consumer’s utility forpurchasing brand i is simply a function ofprice,
promotion and reference effects for brand i alone, the consumer’s response toward brand i is a

function of price, promotion and reference effects for all available brands.

2.3 Reference Point Formation

Because. consumerreference points are difficult if not impossible to measure directly, we

present the following models to capture these constructs indirectly. The models are based on the

premise that consumer’s exposure to the brand’s point-of-purchase price and promotional activity

determines his or her point ofreference forevaluating future marketing activity by the brand. We
begin by focusing on the formation of the consumer’s promotional reference point (PRP). We

assume that consumers view point-of-purchase promotional activity as either present or absent.
Thus, it makes sense to think of this construct in discrete terms: either the consumer expects to

find the brand available on promotion (PRP=1) or not (PRP=0). The continuous

adaptation/expectation models proposed by Winer (1986) and Raman and Bass (1986) are not

directly appropriate in this context, because they do not permit an either/or representation.

We therefore propose a threshold model of PRP formation, determined by the consumer’s

level ofexposure to the brand at the point of purchase. Each time the consumer encounters a

-10-
given brand in the store (i.e., has the opportunity to notice it and actively process information

about it), we will say that he or she is exposed to the brand. If consumer h is exposed to brand i
only when it is in the absence offeature or special display, then he or she has no reason to

develop strong expectations about future promotion activity. Thus, at low levels of exposure to

promotion, the consumer thinks of the brand as a non-promoted brand and establishes a PRP=0.

Ifconsumer h encounters brand i predominantly on promotion, he or she is more likely to use the
recent promotional status ofthe brand as a basis for evaluation on future choice occasions. Thus,

at high levels of exposure to promotion, the consumer thinks of the brand as a promoted brand

and establishes a PRP =1. At some point the exposure to the brand on promotion “outweighs”

the exposure to the brand offpromotion, and the consumer’s point of reference shifts.

We model consumer h’s exposure to promotion by brand i as an exponentially weighted

average of its promotional status on previous purchase occasions. Thus, we let

= ?~.x~ + (1-A.) Z it-i (3)

where x~= consumer h’s exposure to promotional activity by brand i at purchase occasion t,

= 1 if brand i was on promotion on purchase occasion t- 1 and zero otherwise, and

A. = a smoothing parameter (between 0 and 1 and constant across consumers) reflecting

the carryover of exposure from time t- 1 to time t.

Note that if consumer h is never exposed to brand i on discount, x~approaches zero over

time. Ifconsumer h encounters brand i only when it is on discount, x~approaches one.

We have assumed that consumer h is exposed to point-of-purchase information about the

product category only on those occasions when a purchase is made. Since consumers may

evaluate price information and decide to postpone purchase:(perhaps-in hopes-of holding out for a

better deal), this is a simplification. With most sources of scanner panel data it is impossible to

—11—
determine whether or not a shopper has had the opportunity to activelyconsider a brand on any
given visit to the store. We have also assumed that consumer h is exposed to any price discount

activity by brand i on every purchase occasion, whether or not the consumer buys brand i or

anotherbrand in the category. Research by Dickson and Sawyer (1986) suggests that this may

overstate exposure, since consumers are not always aware of discount availability even forthe

brands they buy. Nevertheless, equation (3) has precedent (see Srinivasan and Kesavan 1976,

Guadagni and Little 1983, and Lattin 1987) and parsimoniously captures the first-order effects of

exposure.

We now define consumer h’s promotional reference point forbrand i at time t (PRP~)

as follows:

( . h

h Ii if
~ I if x1~
h <O

where 9= a constant parameter between 0 and 1 had constant across consumers. Note that (4)
h h
specifies PRP~= 1 when exposure exceeds the threshold and PRP~=0 when exposure falls

below threshold.

Unlike the promotional reference point, it is reasonable to model reference price as a

continuous construct determined by the consumer’s exposure to the price of the brand on previous

purchase occasions. While one might argue that consumers are capable of distinguishing special

deal prices from the regular base price of the brand (in effect, maintaining two points ofreference

forprice), the evidence presented by Dickson and Sawyer (1986) suggests that the consumer’s

awareness ofprice is not that finely tuned. Also, it will be desirable to t~eatprice as a separate
construct from promotion (since the two constitute different types ofmarketing activity). We

therefore model consumer h’s reference price for brand i at time t (RP~)using an exponentially

weighted average like the one used in equation (2):

-12-
 RP~= A. RP~1+ (1 A.) 11t-1
- (5)

A special case of the models in equations (3) - (5) arises when the carryover parameterA.

equals zero. In this case, the terms x~and RP~are completely determined by the point-of-
purchase marketing activity on the previous purchase occasion. Thus, exposure at time t is either

0 or 1 (depending on whether or not brand i was promoted at time t-1), and the reference price is

the last observed price for brand i. Under this condition the threshold parameter 9 becomes
unnecessary, and equations (4) and (5) simplify to:

PRP~= Z~ (6)

RP~= Pit-i (7)

In this form, ourmodel ofreference price is very similar to the specifications usedby Raman and

Bass (1986) and Winer (1986). In fact, equation (7) is a special case of the extrapolative

expectations regression model offered by Winer, in which the coefficient of Pa-i equals 1 and the

trend component is absent. Our model has the added benefit that when A.> 0, there is some
“stickiness” in the reference terms over time.

-13-
 3. ESTIMATION AND TESTING

In this section, we test our hypotheses regarding the nature ofdynamic consumer response

to price and promotion. We begin in section 3,1 by describing the data. In section 3.2, we

propose a cross-sectional operationalization ofthe brand choice model and in section 3.3 we

present and discuss the results ofour empirical tests.

3.1 Data

We calibrate our proposed model ofbrand choice using WI scanner data for ground coffee.

We began by identifying the items (i.e. stockkeeping units denoted by a unique UPC) purchased

most frequently by over 1000 fRI scanner panelists from six stores in Pittsfield, Massachusetts.

During a 75-week period (weeks 61 to 135 ofthe study), nearly 80% of ground caffeinated coffee

volume was accounted for by 10 different items (all 16 oz. size). Because items of the same size

and brand are usually priced and promoted together, we combined these 10 items into four choice

alternatives (as shown in Table 1): Hills Brothers, Folgers, Maxwell House, and Chock Full

O’Nuts. These data will be used to initialize and calibrate our model.

[Table 1 here]

Table 2 summarizes the pricing and promotional activity foreach of the four brands across

all six stores over the last 50 weeks of data, (the calibration period forthe proposed model). These

data show considerable variation in pricing activity (across all brands and stores, price ranged from

a low of $1.58 to a high of $2.91), as well as relatively frequent promotion (defined as the

occurrence of either feature or special display for any one ofthe UPC items which make up the
brand). The data also show a strong negative correlation between price and promotion: i.e., a

special discount in price is almost always accompanied by feature ordisplay.

-14-
 [Table 2 here]

We further refined our sample by selecting only those panelists who made at least three

purchases of any ofthe four brands in any of the six stores. We also eliminated as

nonrepresentative a small number of panelists who purchased more than 100 lbs. ofcoffee. The

result was 577 panelists who purchased ground coffee on 4720 purchase occasions during the

model calibration period. We divided this remainder randomly into two groups: 300 panelists

formed the calibration sample (2420 purchase occasions) and 277 panelists formed the hold-out

sample (2300purchase occasions) for purposes of model validation. -

3.2 Cross-Sectional Operationalization

Because the purchase histories ofmost panelists are not sufficiently long to permit reliable

individual-level calibration, we propose to calibrate our consumer response model cross-sectionally

(see also Lattin 1987). To accomplish this, we must first provide a specification of base-level

brand response . to capture the heterogeneity across consumers and over time. One such

approach, proposed by Guadagni and Little (1983), is to create a measure of brand loyalty using an

exponential smoothing model of the past-purchase behavior ofeach panelist. While this loyalty

measure is able to track the differences in purchase behavior across consumers and over time, it is

not capable of separating these two components of variance (i.e., cross-sectional and longitudinal).
As pointed out by Lattin (1987), using a single loyalty term implicitly assumes that differences

across consumers and differences over time contribute equally to the heterogeneity in base-level

utility. If such an assumption is inappropriate, it could have a distorting effect on our reference

price and reference promotion terms, since these constructs are also designed to capture

longitudinal variance in consumer response.

-15-
 We therefore propose the following model of base-level utility using two measures of past

purchase behavior (cf. the variance decomposition approachproposed by Ortmeyer, Lattin and

Montgomery 1987):

= a1 + [3~BL i~2LP (8)

where aj = the brand-specific constant (i.e. intercept term) for brand i.

BL = consumer h’s preference for (or loyalty to) brand i, measured by the proportion

ofpurchase occasions in which consumer h selected brand i during an initialization period. The
term BL ~, which is constant over time, is designed to capture the heterogeneity in base-level

response across consumers.

LP~ = the relativeimpact ofrecent choice behavior, measured by whether or not

consumer h last purchased brand i on choice occasion t-1. This term, which in expectation is

centered around BL~L,is designed to capture the heterogeneity in base-level response over time
within consumer.

For parsimony, we assume that the parameters aj, ~ and ~2 are constant across

consumers. Note that when ~i =0, our approach for modeling base-level utility simplifies to the

approach used by Winer (1986) in his linear probability model.

Substituting equation (8) into equation (1) and then (1) into (2) yields our complete model

of consumer response. In order to provide the measures required by the model, we divide the data

into two periods. We use weeks 60 through 85 to calculate the measure BL and to initialize x~.
We use the next 50 weeks of data (weeks 86-135) to calibrate the logit model in equation (2) by

choosing parameters ~ ~1’~2, ~ ‘Y~?2’ ~ and ~2 to maximize the following log likelihood

function:
h fh\
LL = ~h ~t ~k ~kt in ~ (9)

-16-
where ~t~t=1 if consumer h purchased brand k on purchase occasion t and zero otherwise.

3.3 Empirical Results

We begin ourtests with the assumption that the parameter A., which determines the long-

term carryover effect ofprice and promotion, is equal to zero. Thus, reference levels forPRP and
RP are determined by equations (6) and (7) respectively. In order to illustrate the impact ofvarious

terms, we calibrate a number of nested models in which subsets ofcoefficients are constrained to
zero. The goodness-of-fit of each model is given by the U2 coefficient measured relative to a

model in which only the brand-specific coefficients eq are estimated (see Guadagni and Little 1983;

see also R~as reported by Winer 1986). The estimated coefficients foreach of the nested models

are presented in Table 3.

[Table 3 here]

Price and Reference Price Only (Model I). The first model reflects only the direct

and reference effects of price on consumer response; the coefficients ofall terms describingthe

effects of promotion on response are constrained to zero (i.e., ~ =o~72=0, and 62=0). Though

our developmentis different, Model I captures essentially the same pricing phenomena as the

reference price model of brand choice proposed by Winer (1986). Just as Winer found, we find

that the effect ofprior purchase (LP ~ ) is positive and and significant (I~2 0.84, t
= = 14.4). We

also find that consumer preference (B141) plays at least as strong a role in determining base-level
utility (~i = 1.58, t = 16.8). By including both terms in the model, we can say that controlling for

the heterogeneity in tastes across consumers, there is a tendency for a household to repurchase the

brand bought last.

The effects ofprice and reference pricein Model I also mirror Winer’s findings. As

anticipated, price has a strong negative impact on utility ~ = -3.58, t = -21.5). Furthermore, the

-17-
disparity between price and reference pricehas an additional negative effect (6i= -0.32, t = -2.6).
Thus, there appears to be evidence of the “sticker shock” effect in ground coffee; when observed

price is above the expected price, consumer response to the brand decreases.

Jnsiuding Effects of Promotion (Model II). We now relax the constraints on the
terms involving the effects ofpromotion. This enables us to explicitly separate promotional from

non-promotional priceresponse and to account for the effect of a promotional reference point.

According to a likelihood ratio test of nested models, the improvement in fit of Model H over

Model I is highly significant (~= 275, p<O.OOl). This result is hardly surprising, given the

overwhelming empirical evidence regarding the importance ofmerchandising support such as

feature and display.

At first glance, the sign of the coefficient describing the effect ofpromotion appears to be

incorrect (yo = -1.80, t = -1.8); however, this term is interpretable as the effect ofpromotion at

zero price, well beyond the range of the data. Figure 1 shows the nature of the direct effect of

promotional and non-promotional price changes on consumer utility over the relevant range of

price (using consumer utility for a non-promoted price of $2.91 as base case). For example, a

promotional price cut from $2.49 (unpromoted) to $1.99 leads to an increase in consumer utility

from 1.5 to 3.7, of which 0.3 (about 14%) is attributable to the direct effect of feature and/or
display. The figure confirms our expectations: consumer response to promoted price discount is

greater than consumer response to non-promotedprice discount over nearly the entire range of

observed price. While it appears that response to non-promoted price is greater at very low levels

of price, there are in fact no instances in which a price below $1.79 was not accompanied by either

feature or display. The interaction term ~ = ~ t =2.6) suggests that promotion has a greater

overall effect at higher levels of price.

[Figure 1 here]

-18-
With respect to reference effects, the disparity between the promotional status of the brand (i.e.
whether a brand is on oroff promotion) and the consumer’s promotional reference point has a
8
significant positive impact on consumer utility ( i = 0.54, t = 6.5). In the absence ofreference
effects, as shown in Figure 1, consumer utility for a brand previously available at $1.99 will
increase by 0.30 when the brand is put on promotion (with no change in price). Considering
reference effects, the overall impact on consumer utility may be as high as 0.84 ifthe consumer has

not had prior exposure to the brand on deal (i.e. PRP = 0).

When reference effects due to promotion are controlled for in the model, we find that the

effect ofreference price is not significant ~ = 0.21, t= 1.3). In fact, the coefficient estimate has
the wrong sign (we expect a positive disparity in PRP to have a negative impact on consumer

utility), which is likely due to collinearity with promotional reference effects.

Controlling for Prior Promotional Purchase (Model IIfl. Other researchers have
used different approaches in their attempts to explain the carryover effects of promotion by looking
at the effects of prior promotional purchase on consumer response. In particular, Dodson, Tybout
and Sternthal (1978), Guadagni and Little (1983), and Ortmeyer, Lattin and Montgomery (1988)
have empirically tested the proposition that following aprior promotional purchase, ahousehold is
less likely to repurchase the brand on the next occasion than following a non-promotional
purchase. In all cases, the findings suggest that a prior promotional purchase has weaker impact

on reinforcing future purchase of the brand than a prior purchase off promotion.

In order to provide more convincing empirical support for our framework, we must show
that the reference effect of promotion due to exposure persists even in the presence of an indicator
capturing the effect ofprior promotional purchase. Otherwise, one might conclude that the
coefficient ~2is significant only because the term (z1~ PRP~)accounts for the effect of prior
-

-19-
promotional purchase already identified by other researchers. We therefore modify the

specification of in equation (8) as follows:

= eq + ~ BL~’+ 132LP~+ 133LPP~ (10)

where LPP~= 1 if consumer h purchased brand i on promotion at purchase occasion t- 1, and zero

otherwise. Substituting equation (10) into (1) and (1) into (2) yields the full specification of

Model Ill.

As shown in Table 3, the estimate for f33 is negative and highly significant

(~= -0.99, t = -7.2). The result suggests that a prior purchase of a brand offpromotion has a
much stronger impact on subsequent purchase (givenby I~2= 1.67) than does a prior purchase on

promotion (where the net effect is ~2 + ~33= 0.69). In either case, the effect of prior purchase (on

or offpromotion) is still positive and significant, consistent with the finding ofGuadagni and Little

(1983).

More importantly, we find that controlling for prior promotional purchase, the reference

effect of promotion remains significant (62 = 0.30, t = 3.5). Thus, we are able to distinguish

between the prior promotional purchase and prior exposure to promotion at the point ofpurchase- in

their impact on consumer response. Note that the reference effects ofpromotion better explain the

variance in consumer response than reference price; as in Model II, 6~is non-significant (and

incorrectly signed).

Setting A.> 0 (Model IV). Models I III have all assumed that the smoothing
-

parameter in the exponential smoothing model of exposure is equal to zero. While we have

obtained good fits and strong empirical support forthis special case of our proposed framework, it
may be possible to improve these results by relaxing the restriction that A. =0. Unfortunately, this

additional flexibility comes at a computational cost. Because ofdiscrete nature of the promotional

-20-
reference point, it is necessary to determine the threshold parameter 9 using a grid search for a

given value ofA..

In order to illustrate the estimating procedure of the threshold model with non-zero

parameter A., we begin by setting A. = 0.70. This value suggests that an exposure to a brand on

promotion has a half-life of about two purchase occasions (i.e., on the second occasion following

an encounter with a promoted brand, the effect ofthat encounter on exposure is diminished by

half). While it might be possible to improve model fit by further refining the value of A., our goal
is to provide an illustration of the proposed framework with A. >0 and to-compare the findings to

the previous model in which A. =0.

To calibrate this model, we begin by setting a value of 9 and calculating PRP~according to

equation (4). We then estimate the remaining model parameters by maximum likelihood according

to equation (9). We repeat this process for all values of 9 between 0.20 and 0.80 at intervals of
0.01. Figure 2 shows partial results of the grid search for the threshold parameter 9. The

functional form seems reasonably well behaved. The maximum likelihood value of -2055.8 occurs

at 8 = 0.51. This value for the threshold parameterhas intuitive appeal: it suggests that when the

weighted average of past exposure to the brand on promotion is more than one-half, the consumer

thinks of the brand as a promoted brand (i.e. PRP = 1).

The estimates ofthe remaining coefficients for Model IV (forA. = 0.7 and 9 = 0.51) are

shown in Table 3. As in Model Ill, we find that the reference effect of promotion is significant

(62 = 0.38, t = 5.1). We also find that the effect of reference price is incorrectly signed, although

unlike Model Ill the estimated coefficient is in this case significant ~ = 1.00, t = 3.1). While the

incorrect sign of is likely to be attributable to negative collinearity between RP and PRP, it is

difficult to explain the statistical significance ofthis counter-intuitive finding. The factremains that

the anticipated reference effectson consumer response appearto be due to promotion rather than

price.

-21-
 Predictive Validation (Model V’). Both Models Ill and IV indicate support for our

notions regarding the reference effects of promotion. However, it is also important for us to

establish the generalizabiity ofthese findings. We do this by testing the ability of calibrated

Models Ill and N to predict the purchase behavior of the hold-out sample of277 households,

compared to a model in which the reference effects ofprice and promotion have been constrained
to zero (this is Model V, shown in column 5 ofTable 3). According to the likelihood ratio test,

Model Ill (X~= 11.0, p < .01) and Model IV (X~= 24.8, p <.01) provide a significant

improvement in fit over Model V. Our goal is to show that across different groups ofpanelists, the

predictive performance ofthe models embodying reference effects is superior to the model without
reference effects; i.e., that the improvementin fit forthe proposed models is not simply attributable

to capitalizing on idiosyncratic sampling variance.

The 277 households in the hold-out sample engaged in a total of2300 purchase occasions

between weeks 86 and 135. Using the estimated coefficients of Model Ill to predict their purchase

behaviorresults in a log likelihood value of -2012.3; for Model IV, the fit is -2018.2. In both

cases, the predictiveperformance in the hold-out group is superior to Model V, which has a fit of

-2023.3. This is evidence in support of the predictive validity of the reference effects models.

Although both reference effects models are superior to Model V, we note that the
improvement in fit of Model IV over Model Ill in the calibration sample does not hold up to

validation. This may be due to some capitalization or chance associated with the incorrectly signed

coefficient of reference price. It therefore seems appropriate to use the more parsimonious model

(Model III) as a basis for further discussion.

-22-
 4. DISCUSSION

We illustrate the implications of dynamic consumer response by looking at a “slice” of data

from the estimation period. During the three-week period from week 121 to week 123, five of the
six stores in the Pittsfield areaoffered promotions and price discounts on Folgers ground

caffeinated coffee (see Table 4). Consumers purchasing coffee during this period were exposed to
the point-of-purchase marketing activity in these stores, which potentially influenced theirpoints of
reference with respect to price and promotion. We can assess the overall impact of the pricing and

promotional activity during this period by comparing prior purchase behavior, points ofreference,

(i.e., PRP and RP), and purchase probability measured before and afterthe period.

[Table 4 here]

Promotional activity by Folgers may affect subsequent consumer response in two ways.

On the one hand, features and displays (accompanied in some cases by special price discounts)
increase utility and the likelihood of consumer purchases; these choices positively reinforce

subsequent choice of Folgers. On the other hand, exposure to the brand on promotion will tend to

offset this reinforcing effect. First, consumers who last purchased Folgers on promotion are

significantly less likely to repurchase Folgers than those who last purchased offpromotion. To the

extent that regular buyers of Folgers (offpromotion) are induced to buy on promotion, their
probability of subsequent repurchase will decline. Second, exposure to Folgers on promotion

increases the number ofconsumers with PRP = 1, which also has a negative impact on the

probability of subsequent repurchase.

Out of 577 households in the full sample, 256 made 380 purchases ofcoffee between

weeks 121 and 123. Folgers share of incidence during this three-week period was just over 45%,

compared with an average market share of less than 27% during the entire 50-week calibration

period. Clearly, the promotional activity by Folgers during weeks 12 1-123 was quite successful in

-23-
inducing consumers to purchase the brand. Of the 256 households who bought coffee during this

period, only 14% (36 households) had last purchased Folgers before the promotion. Afterwards, *

)
this proportion increased to just over 50% (129 households), a significant increase (t = 10.1).

Most of these purchases, however, were on promotion. At the beginning of the promotional

period, only 42% of the households who had last purchased Folgers (15 out of 36) had purchased
on promotion. At the end of the period, over 90% of prior Folgers purchases were made on

promotion (118 out of 129 households). Exposure to the brand on promotion (i.e. households

with PRP = 1) also increased significantly, going from 36% to just over 60% (t = 5.5), and

reference price declined slightly, from $2.36 to $2.23 (t = 5.8).

-

We can assess the net effect of these countervailing forces (i.e. the positive reinforcement

of past purchase behavior versus the negative effects ofpast promotional purchase and exposure)

by using Model Ill to calculate the average probability of purchase before and after the three-week

promotional period. Our calculations assume that each brand is available at the same price ($2.49)

in each of the six stores, and that there is no promotional activity. Under these conditions, our
model suggests that average purchase probability increases slightly but significantly from 0.23

before the promotional period to 0.27 afterwards (t = 4.0). Thus, for the case ofFolgers
promoting in weeks 121 to 123, the positive effects ofpromotion more than offset the negative

ones. However, this does not have to be true in general. For a brand with a stronger franchise
(i.e. with a large number of consumers willing to buy the unpromoted brand at regular price), the

incremental effects of promotion may not be offset by the negative dynamics in subsequent

periods.

Note that an increase in average brand choice probability does not necessarily imply that we

should expect to see increased sales ofFolgers in the post-promotional period. Our model captures

only brand choice behavior, conditional on category purchase. The increase in average choice
probability from 0.23 to 0.27 is calculated over all 256 households making a purchase in the

promotional period (the choice probabilities of the remaining households do not change).

-24-
However, not all of these households are equally likely to be in the market forcoffee before and

after the period. In fact, those who favor Folgers are likely to be disproportionately represented
among households taking advantage of a Folgers promotion. These households are

correspondingly less likely to be in the market for coffee following the promotional period, due to
inventory effects. Thus, any change in the level ofFolgers sales before and after a promotional

period depends not only on the changes in brand choice probabilities, but also on any change in the

mix of households in the market for coffee.

The choseti illustration reflects only the incremental effects of a four-week pricepromotion

by a single brand. A strength ofthe model is that it can be used to assess the potential effects of a

substantial change in the promotional pricing environment. For example, the model could be used

in simulation to explore the implications ofdecreasing the frequency ofpromotion for a given

brand or for the category as a whole. Because ourmodel is adaptive (i.e., consumers’ points of
reference adjust to changes in the depth and frequency ofprice promotion in the category), it may

prove useful in forecasting outside the range of the existing data.

-25-
 S. CONCLUSION

)
We have proposed and tested a model ofconsumer response incorporating the reference

effects of price and promotion. Our results support the notion that consumers form expectations
based on theirexposure to pricing and promotional activity, and that those expectations influence

the patterns of brand choice. By including both price and promotional variables in our model of

consumer response, we are able to explicitly characterize the differences between-promotional: and

non-promotional price elasticity, and to separate these effects from the reference effects of price

and promotion. Our results suggest that when the reference effects ofpromotions are held

constant, the reference effects of price are not significant.

We also provide a new rationale to explain the carry-overeffects of promotions on

consumer response. Other researchers have focused on the differences between prior purchase and

prior promotional purchase on subsequent brand choice. Guadagni and Little (1983) and

Ortmeyer, Lattin and Montgomery (1988) suggest that a purchase on promotion (under less

competitive conditions) is not as strong an indicator of brand loyalty than a non-promotional

purchase. Others (e.g. Dodson, Tybout and Sternthal 1978) use attribution theory to explain why

a consumer, afterhaving purchased a brand on promotion, might be less likely to repurchase the

brand in the absence of promotion. Note that both ofthese rationales require that the promotional

brand actually be purchased by the consumer. Our results suggest that controlling for prior

promotional purchase, there is still a significant impact on consumer response from exposure to the

brand on promotion.

We have also presented a threshold model of reference point formation, a special case of

which is broadly consistent with the expectations models in the existing literature. For two

different parameterizations of the reference point formation model (A. =0 and A.> 0), our models

incorporating reference effects outperformed a model without reference effects, both in terms offit

and predictive validity.

-26-
 Several interesting and important questions still remain for future research. First, we need

to develop a better understanding of exposure and its relationship to the construct of brand loyalty.

Certainly the two are closely linked, since exposure to promotional activity in the product category

is often accompanied by purchase, which reinforces brand loyalty. However, we still do not

understand exactly what constitutes exposure, nor is it clear whether or not we can measure it.
Perhaps controlled experimental research, using techniques such as those used by Dickson and

Sawyer (1986), might help to improve our understanding. We may also want to entertain other
models of brand loyalty, in which promotional purchases are treated differently than non-

promotional purchases.

Second, we need to extend ourmodel to consider the reference effects ofprice and
promotion on category purchase incidence. Consumers who are exposed to the product category

during periods of high category value (i.e. when favored brands are offered on promotion at

discountedprices) may establish expectations of high value in the future. In response, these

consuniers may be more likely to time theirpurchase to coincide with deal periods, effectively
redistributing demand over time.

-27-
 TABLE 1 )

Description of the 10 Stockkeeping Units Comprising the Brands

Brand UPC Number Item Description Size

1. Hills Bros 1840000137 HILLS BROS REG COFFEE 16 oz

1840000138 HILLS BROS AUTO DRIP 16 oz

2. Folgers 2550000004 FOLGERS REGULAR COFFEE 16 oz

2550000005 FOLGERS ELECTRIC PERK 16 oz
2550000006 FOLGERS DRIP COFFEE 16 oz

3. Maxwell House 4300070134 MAX HSE REGULAR COFFEE 16 oz

4300070135 MAX HSE DRIP COFFEE 16 oz
4300070733 MAX HSE ELECTRA PERK 16 oz
4300070780 MAX HSE AUTO DRIP 16 oz

4. Chock Full 7103800001 CHK FLL 0 NUTS COFFEE 16 oz

0’ Nuts

..28..
 TABLE 2

Summary of Pricing and Promotional Activity

by Four Brands in Six Stores

Brand 1 Brand 2 Brand 3 Brand 4

Store 11 0.18 (a) 0.23 0.19 0.33

~11(b) .16 .20 .45
_,63(c) — .73 —.62 —.84

Store 12 0.23 0.33 0.22 0.29

.16 .45 .32 .35
—.73 —.84 —.60 —.83

Store 13 .19 .22 .09 .18

.20 .32 .12 .39
—.62 —.60 .03 —.73

Store 14 .33 .29 .18 .23

.45 .35 .39 .39
—.84 —.83 —.73 —.62

Store 15 .31 .28 .06 0.24

.51 .27 .39 .28
—.56 —.73 —.15 —.58

Store 16 .22 .18 .06 .21

.32 .39 .13 .25
—.60 —.73 —.12 —.58

(a) Standard deviation of price (in dollars)

(b) Proportion of weeks on promotion
(c) Correlation between price and promotion

—29—
 TABLE 3
Parameter Estimates for Nested Models of Brand Choice

MODEL 1 MODEL II MODEL III MODEL IV MODEL V

a1 0.678 0.768 .808 .870 .807
(8.27) (9.4) (9.45) (9.37) (9.70)
a~ 0.746 0.739 .724 .789 .699
(9.69) (9.35) (9.30) (9.48) (9.08)
a3 1.411 1.369 1.348 1.495 1.297
(16.41) (14.88) (14.71) (13.40) (15.50)
f3~ 1.581 1.687 1.665 1.675 1.674
(16.82) (17,21) (16.81) (16.85) (16.82)

1~2 0.836 1.010 1.674 1.702 1.750

(14.41) (16.29) (14.98) (15.35) (15.85)

70 -- -1.802 -1.581 -1.653 -1.336

-- (-1.84) (-1.62) (-1.69) (-1.36)

~‘i -3.584 -3.762 -3.839 -4.579 -3.608

(-21.46) (-8.98) (-9.20) (-9.25) (-9.08)

12 -- 1.053 1.078 1.078 1.096

(2.61) (2.68) (2.68) (2.70)
6~ -0.324 0.209 .271 1.004
(-2.57) (1.33) (1.72) (3.13)

0.535 .299 .377

(6.52) (3.46) (5.12)

133 -.987 -1.072 -1.163

(-7.18) (-8.36) (-9.44)
A. 0.7
8 0.51
Fit Statistics:
LL -2225.2 -2087.3 -2062.7 -2055.8 -2068.2
u’2 .328 0.369 0.377 0.379 0.375

Validation:

LL -2012.3 -2018.2 -2023.3

-30-
 Table 4

Pricing and Promotional Activity

by Folgers during weeks 121 to 123

Stnre
11 12 13 14 15 16

Week
121 (a) 2.49 2.49 2.39 1.86 1.86 2.49
(b) 0.00 0.00 1.00 1.00 1.00 0.00

122 (a) 2.39 2.29 1.99 1.86 1.86 2.49

(b) 1.00 1.00 1.00 1.00 1.00 0.00

123 (a) 2.49 2.49 2.49 1.86 1.86 2.49

(b) 0.00 0.00 0.00 1.00 1.00 0.00

(a) Price (in dollars)

(b) Promotion (= 1 if feature or display, 0 otherwise)

—31—
 FIGURE 1

Consumer Utility as a function of Price and Promotion,

Showing the Effect of a Promoted Price Change

from $2.49 (unpromoted) to $1.99

(In LI ~7”(

PRiCE

—32—
 FIGURE 2

Results of the Grid Search for Threshold Parameter 8:

Maximum Likelihood of —2055.8 occurs at 8 = 0.51

LL •~2055
,,.,‘•‘~ni:•p
~n vJ
I_v ~ /

•~2058
~ ~C

... 2060
‘•206 1.

0. 45 C~ r
V. ~)
(,~
v. rr 0. 6
e

—33—
 REFERENCES

Ben-Aidva, M. and S. Lerman (1979) “Disaggregate travel and mobility choice models and
measures of accessibility,” in Hensher and Stopher (eds) Behavioral Travel
Modeling, Croom Helm, London.

(1986) Discrete Choice Analysis, MIT Press, Cambridge, MA.

Bucklin, R. and J. Lattin (1986) “A two-state model of brand choice and purchase
incidence,” unpublished working paper, Stanford University.
Dickson, P. and A. Sawyer (1986) “Point-of-purchase behavior and price perceptions of
supermarket shippers,” Report No. 86-102, Marketing Science Institute,
Cambridge, MA.

Dodson, J., A. Tybout, and B. Stemthal (1978) “Impact of deals and dealretraction on
brand switching,” Journal ofMarketing Research, 15 (Februaiy),72-8 1.

Fader, P. and L. McAlister (1988) “Promotionally-Induced Choice Restriction,”

unpublished working paper, University of Texas at Austin.

Gabor, A. and C. Granger (1964) “Price sensitivity ofthe consumer,” Journal of

Advertising Research, 4 (December), 40-44.

Guadagni, P. and J.D.C. Little (1983) “Alogit model ofbrand choice calibrated on scanner
data,” Marketing Science, 2 (Summer), 203-238. )
Gupta, S. (1986) “An integrated model of interpurchase time, brand choice, and purchase
quantity,” unpublished Ph.D. dissertation, Columbia University.

Helson, H. (1964) Adaptation-Level Theory, Harper and Row, New York, NY.

Jones, M. and F. Zufryden (1981) “Relating deal purchases and consumer characteristics
to repeat purchase probability,” Journal ofthe Market Research Society. 23, 8499.
Lattin, J. (1987) “A Model ofBalanced Choice Behavior,” Marketing Science, 6 (Winter),
48-65.
Marketing News (1985) “Panelists offer pricing strategy advice for consumer and industrial
products,” 19 (#3), p. 1.
McAlister, L. (1983) “Marketresponse to promotions: a model,” unpublished working
paper, Massachusetts Institute of Technology.
McFadden, D. (1974) “Conditional logit analysis ofqualitative choice behavior,” in
P. Zarembka (ed) Frontiers ofEconometrics. Academic Press, New York, NY.
(1981) “Econometric models of probabilistic choice,” in C. Manski and D.
McFadden (eds) Structural Analysis ofDiscrete Data with Econometric
Applications. MIT Press, Cambridge, MA.

Monroe, K. (1979) Pricing: Making Profitable Decisions, McGraw-Hill, New York, NY.

-34-
Muth, J. (1961) “Rational expectations and the theory ofprice movements,” Economeirica,
72 (July), 315-335.

Neslin, S., C. Henderson and J. Quelch (1985) “Consumer Promotions and the
Acceleration of Product Purchases,” Marketing Science 4 (Spring), 147-165.

Ortmeyer, 0., J. Lattin, and D. Montgomery (1987) “Routinized choice behavior, brand
commitment, and consumer response to promotions,” unpublished working paper,
Stanford University.

Raman, K. and F. Bass (1986) “A general test of reference price theory in the presence of
threshold effects,” unpublished working paper, University of Texas, Dallas, TX.
Rao. V. and D. Sabavala (1980) “Allocation of marketing resources; the role of price
promotions,” in R. Leone (ed) Proceedings ofthe Second Market Measurement
Conference, TIMS College of Marketing Providence, RI.

Sawyer, A. and P. Dickson (1984) “Psychologicalperspectives on consumer response to

sales promotion,” in K. Jocz (ed) Research on Sales Promotion: CollectedPapers,
Marketing Science Institute, Cambridge, MA.

Scott, C. (1976) “The Effects of Trial and Incentives on Repeat Purchase Behavior,”
Journal ofMarketing Research, 13 (August), 263-269.
Sherif, C. (1963) “Social categorization as a function oflatitude of acceptance and series
range,” Journal of Abnormal Psychology. 67 (August), 148-156.

Shoemaker, R. and F. Shoaf (1977) “Repeat rates of deal purchases,” Journal of

Advertising Research, 17 (April), 47-53.

Srinivásan, V. and R. Kesavan (1976) “An Alternative Interpretation of the Linear

Learning Model of Brand Choice,” Journal of Consumer Research, 3 (September),
76-83.
Winer, R. (1986) “A reference price model ofbrand choice for frequently purchased
products,” Journal of Consumer Research, 13 (Spring), 250-256.

Wright, P. (1975) “Consumer choice strategies: simplifying vs. optimizing,” Journal of

Marketing Research, 12 (February), 60-67.

-35-