Accepted Manuscript

Title: Exposure Order Effects and Advertising Competition

Author: Oksana Loginova

PII: S0167-2681(09)00114-0
DOI: doi:10.1016/j.jebo.2009.04.012
Reference: JEBO 2373

To appear in: Journal of Economic Behavior & Organization

Received date: 11-1-2007

Revised date: 17-4-2009
Accepted date: 20-4-2009

Please cite this article as: Loginova, O., Exposure Order Effects and Advertising
Competition, Journal of Economic Behavior and Organization (2008),
doi:10.1016/j.jebo.2009.04.012

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Blinded Manuscript (NO Author Details)

Exposure Order Effects and Advertising Competition

t
ip
April 17, 2009

cr
Abstract

This paper applies the theories of exposure order effects, developed in the psychology litera-

us
ture, to an industrial organization model to explore their role in advertising competition. There
are two firms and infinitely many identical consumers. The firms produce a homogeneous prod-
uct and distribute their brands through a common retailer. Consumers randomly arrive at the

an
retailer and buy their most preferred brands. The order in which a consumer sees the advertising
messages affects his brand preferences. Under the primacy effect the consumer prefers the brand
he first saw advertised, under the recency – the last encountered brand. The equilibrium of
the advertising game is characterized separately under the primacy and the recency effects. In
M
the first setting all consumers are initially unaware of the product existence. The equilibrium
advertising intensities, remarkably, do not depend on the type of exposure order effect. In the
other two settings some consumers have already formed their brand preferences. The primacy
d

and the recency effects give rise to different equilibrium outcomes.

te

JEL classifications: C73, D11, D43, L13, M37.

Keywords: advertising order effects, primacy, recency.
p
ce
Ac

1
Page 1 of 22
1 Introduction
Traditionally, advertising has been studied by two very different subfields of social science, economics
and psychology.1 The economics literature on advertising has been concerned with such important
questions as the informational content of ads, the socially optimal level of advertising, and firm
rivalry. The psychology literature, among other things, has explored the connection between in-

t
formation processing and advertising effectiveness. The goal of this paper is to develop a model

ip
that incorporates economics and psychology to examine how information processing biases affect
advertising competition between firms.

cr
Past research in psychology has shown that the order in which consumers are exposed to brands
influences consumer preferences. Two advertising order effects have received attention. The pri-
macy effect is characterized by greater persuasion consequence of the first advertising message. On

us
the other hand, the recency effect occurs when the last encountered brand is preferred. Whether
exposure order results in the primacy effect or in the recency depends on different factors, such as

an
overall involvement, motivation, attitude strength, and the time between information exposure and
preference construction.
Lana (1963) showed that subjects greatly interested in the topic exhibited the primacy effect,
whereas subjects with minimal interest exhibited the recency effect. The primacy effect occurs
M
when participants are motivated to elaborate the initial message and show critical thinking toward
later information, whereas the recency effect occurs when motivation is low (Haugtvedt and Wegner
1994). Brunel and Nelson (2003) examined how advertising order effects connect to gender dif-
d

ferences. Their results suggested that under conditions of low involvement, females exhibited the
te

primacy effect, males – the recency effect. Under conditions of higher involvement, females contin-
ued to exhibit the primacy effect; the recency effect with males disappeared when the advertisement
matched their values. Niedrich and Swain (2008) found the primacy effect when the delay between
p

attribute encoding and preference construction is long (a few days). On the other hand, the recency
ce

effect is more likely to occur when the delay is short (a few minutes).
The model presented here applies the primacy and the recency effects to advertising competition.
The agents of the model are two firms and infinitely many identical consumers of total mass one.
Ac

The firms produce a homogeneous product, firm A produces brand A and firm B produces brand B.
The firms distribute their brands through a common retailer. Each consumer randomly (according
to a Poisson process) arrives at the retailer and buys his most preferred brand. Consumers learn
about the product existence and form their brand preferences through the firms’ advertising. The
order in which a consumer sees the advertising messages affects his brand preferences. Under the
primacy effect the consumer purchases the brand he first saw advertised, whereas under the recency
effect – whichever brand he most recently saw advertised.
The firms compete in advertising intensities – the rate parameters of Poisson processes that
advertising messages follow. The equilibrium of the advertising game is characterized separately
1
Of course, advertising is one of the primary subjects studied in the field of marketing. Most papers in marketing
concerning advertising, however, can be classified as having either economic or psychologic foundations.

2
Page 2 of 22
under the primacy and the recency effects. Three setting are considered. In Setting I (New Market)
consumers initially are unaware of the product existence. In this setup the equilibrium advertising
intensities are symmetric and, remarkably, do not depend on the type of exposure order effect.
In Setting II (Growing Market) it is assumed that at the beginning of the game a number of
consumers believe brand A is better, the equal number prefer brand B, and the rest are unaware of
the product existence. In this setup the two exposure order effects give rise to different equilibrium

t
outcomes. Results show that the firms choose higher advertising intensities under the recency effect

ip
than under the primacy. In Setting III firm A has advantage over firm B. Specifically, at the
beginning of the game a number of consumers prefer brand A, and the rest are unaware of the

cr
product existence. Under the primacy effect the equilibrium advertising intensities are symmetric,
whereas under the recency effect firm B chooses higher advertising intensity than firm A. Various

us
welfare implications are investigated.
This paper contributes to the small but growing body of economic literature on advertising
grounded in psychological research. Krähmer (2004) developed a monopolistic model in which the

an
buyer may not recall correctly his past experience with the product. Advertising can activate
memory and help consumers to recollect their past experiences, or distort actual consumption ex-
periences. Shapiro (2006) explored two alternative advertising mechanisms that yield very different
M
predictions. In the first, advertising converts memories of bad experiences into memories of good
ones. In the second, advertising makes favorable experiences more likely to be remembered. Brekke
and Rege (2007) captured availability heuristic phenomenon, according to which consumers cannot
d

distinguish between the recommendations or real people and fictitious characters in advertisements.
Their analysis showed that even if a person knows that his observations of others may be distorted
te

by advertising, it is still rational for him to choose whichever product he has observed most often.
In the next section, the formal model is presented. The three settings are analyzed in sections 3
p

through 6. Concluding remarks appear in Section 7. All proofs are relegated to the Appendix.
ce

2 The Model
2.1 Firms and Consumers
Ac

Two firms, A and B, produce a homogeneous product using zero marginal cost technology. The
firms distribute their brands through a common retailer. The demand side consists of infinitely
many identical consumers of total mass 1. Consumers are risk-neutral, possess continuous-time
discount rate r > 0, and have a sequence of unit demands for the product. In particular, a consumer
receives a gross payoff of 1 whenever he arrives at the retailer and buys his most preferred brand.
The number of times the consumer visits the retailer follows a homogenous Poisson process with
rate parameter σ > 0. That is, in a small time interval ∆t the consumer visits the retailer with
infinitesimal probability σ∆t + o(∆t).2
2
Formally, let Nt denote the number of times a consumer visits the retailer by time t. Nt is a Poisson process with
rate parameter σ > 0 if and only if (i) the number of visits during one time interval is independent of the number of
visits during a different non-overlapping time interval; (ii) in a small time interval ∆t the consumer visits the retailer

3
Page 3 of 22
 Consumers learn about the product existence and form their brand preferences through adver-
tising. The number of ads describing brand A a consumer receives is a homogeneous Poisson process
with rate parameter αA ; the number of ads describing brand B he receives follows a Poisson process
with rate parameter αB . It is assumed that consumers receive advertising messages independently
from each other, as they watch/ listen to different TV and radio programs at different times.
Advertising intensities, αA and αB , are chosen by the firms. The discounted cost needed to

t
achieve intensity α ≥ 0 is c(α)/r. Function c(·) is convex and increasing, c0 > 0 and c00 > 0. The

ip
standard boundary assumptions, c(0) = c0 (0) = 0 and limα→∞ c0 (α) = ∞, guarantee the existence
of an interior solution.

cr
2.2 Advertising Order Effects

us
The order in which a consumer sees the advertising messages affects his brand preferences. Two
main effects are considered: primacy and recency. The primacy effect is characterized by greater
persuasion consequence of the first advertisement. In the context of the present model, the primacy

an
effect implies that the consumer purchases whichever brand he first saw advertised. The recency
effect, on the other hand, occurs when the last encountered brand is preferred. That is, the consumer
buys whichever brand he most recently saw advertised.
M
Let xA (t) denote the number of consumers who at time t believe brand A is better. Similarly,
xB (t) is the number of consumers who prefer brand B. The rest of the consumers, x∅ (t) = 1−xA (t)−
xB (t), are unaware of the product existence. These consumers will be referred to as “ignorant”
d

consumers, for the lack of a better term. How do xA (t), xB (t), and x∅ (t) evolve under the primacy
te

and the recency effects given advertising intensities αA and αB ?

Consider the primacy effect. The probability that a consumer sees firm A’s advertisement in a
small time interval ∆t is αA ∆t + o(∆t). Hence, the number of ignorant consumers who see firm A’s
p

ad is
ce

αA x∅ (t)∆t + o(∆t).

Similarly, the number of ignorant consumers who see firm B’s ad is

Ac

αB x∅ (t)∆t + o(∆t).

Therefore, xA (t) increases at instantaneous rate αA x∅ (t); xB (t) increases at rate αB x∅ (t); the
number of ignorant consumers decreases at rate (αA + αB )x∅ (t). Algebraically,

d

 dt x∅ (t) = −(αA + αB )x∅ (t),
d (1)
 dt xA (t) = αA x∅ (t),
d
dt xB (t) = αB x∅ (t).


The recency effect differs from the primacy effect in that consumers switch from one brand to
with probability Pr{Nt+∆t = Nt + 1} = σ∆t + o(∆t), and he visits the retailer more than one time with probability
o(∆t). On Poisson processes see, for example, Doob (1953).

4
Page 4 of 22
the other when they see an advertisement for the latter. Thus, the number of consumers who switch
from brand A to brand B is
αB xA (t)∆t + o(∆t).

Similarly, the number of consumers who switch from brand B to brand A is

αA xB (t)∆t + o(∆t).

t
ip
Therefore, xA (t) increases at rate αA (x∅ (t)+xB (t)) and at the same time decreases at rate αB xA (t);
xB (t) increases at rate αB (x∅ (t) + xA (t)) and at the same time decreases at rate αA xB (t). Alge-

cr
braically, 
d

 dt x∅ (t) = −(αA + αB )x∅ (t),

us
d
dt xA (t) = αA (x∅ (t) + xB (t)) − αB xA (t), (2)

d
dt xB (t) = αB (x∅ (t) + xA (t)) − αA xB (t).


The transition matrices (infinitesimal generators) for the primacy and the recency effects are,

an
respectively,  
−(αA + αB ) 0 0
P = αA 0 0
 
M
αB 0 0
and  
−(αA + αB ) 0 0
d

R= αA −αB αA  .
 

αB αB −αA
te

Note that the column sums of P and R are equal to zero, the nondiagonal entries are nonnegative,
and the diagonal entries are nonpositive. Let
p

 
x∅ (t)
ce

x̄(t) = xA (t) .

 

xB (t)
Ac

Then the systems of differential equations (1) and (2) can be rewritten in a matrix form,

d
x̄(t) = P x̄(t)
dt

and
d
x̄(t) = Rx̄(t).
dt
The solutions are given by
x̄(t) = etP x̄(0) (3)

and
x̄(t) = etR x̄(0), (4)

5
Page 5 of 22
where x̄(0) is the vector of initial conditions.3

2.3 Profit Functions of the Firms

The diagonalization4 is used to compute the exponentials etP and etR ,
 
e−(αA +αB )t 0 0

t
 α −α e−(αA +αB )t
etP = 

(5)

ip

A A
αA +αB 1 0
αB −αB e−(αA +αB )t
αA +αB 0 1

cr
and  
e−(αA +αB )t 0 0
 α −α e−(αA +αB )t αA +αB e−(αA +αB )t αA −αA e−(αA +αB )t 
etR = 


us
. (6)
A A
 αA +αB αA +αB αA +αB
αB −αB e−(αA +αB )t αB −αB e−(αA +αB )t αB +αA e−(αA +αB )t
αA +αB αA +αB αA +αB

Straightforward but tedious calculations are relegated to the Appendix. Substituting the above

an
expressions into (3) and (4) allows to compute x̄(t) under the primacy and the recency effects, hence
the number of consumers who prefer brand A, xA (t), and the number of consumers who believe
brand B is better, xB (t).
M
There is no price competition between the firms, as, by assumption, each consumer buys
whichever brand he believes is better. The firms set their prices equal to 1, consumer valuation
for the product. Hence, the profit functions of the firms are
d

Z ∞
c(αA )
ΠA (αA , αB ) = e−rt xA (t)σ dt −
te

0 r

and Z ∞
c(αB )
p

ΠB (αA , αB ) = e−rt xB (t)σ dt − .

0 r
ce

2.4 Three Different Settings

The firms choose their advertising intensities αA and αB simultaneously at the beginning of the game.
Ac

The equilibrium concept employed is Nash equilibrium. Three different settings are considered. In
3
The exponential etA is defined as
∞
X (tA)n
etA = .
n=0
n!
4
The first step is to diagonalize the matrix, A = QDQ−1 , where D = diag(d1 , d2 , . . .). Once A is diagonalized it is
easy to compute exponential etA ,
∞
X Q(tD)n Q−1  
etA = = Q diag ed1 t , ed2 t , . . . Q−1 .
n=0
n!

6
Page 6 of 22
Setting I it is assumed that at time t = 0 consumers are unaware of the product existence,
 
1
x̄(0) = 0 .
 

In Setting II, γ consumers are ignorant, (1 − γ)/2 believe brand A is better, and (1 − γ)/2 prefer

t
ip
brand B. That is,  
γ
x̄(0) =  12 (1 − γ) .
 

cr
1
2 (1 − γ)
In Setting III firm A has advantage over firm B. In particular, it is assumed that at the beginning of

us
the game γ consumers are unaware of the product existence, and the rest believe brand A is better.
Algebraically,  
γ

an
x̄(0) = 1 − γ  .
 

0
M
In the next three sections each setting is analyzed separately under the primacy and the recency
effects.
d

3 Setting I: New Market

te

Consider the setting in which initially consumers are unaware of the product existence. Under the
primacy effect
p

    
e−(αA +αB )t 0 0 1 e−(αA +αB )t
 α −α e−(αA +αB )t     α −α e−(αA +αB )t 
ce

x̄(t) =  A A
α +α 1 0 0 = 
 A A
αA +αB
.
 A B 
αB −αB e−(αA +αB )t αB −αB e−(αA +αB )t
αA +αB 0 1 0 αA +αB
Ac

Hence,
αA − αA e−(αA +αB )t
xA (t) = (7)
αA + αB
and
αB − αB e−(αA +αB )t
xB (t) = . (8)
αA + αB
Observe that the first column of matrix (5) coincides with the first column of matrix (6). Hence,
functions xA (t) and xB (t) are the same under the recency effect as under the primacy. This appar-
ently surprising result can be easily explained. Indeed, the number of consumers who switch from
brand B to brand A in a small time interval ∆t is

αA xB (t)∆t + o(∆t).

7
Page 7 of 22
Similarly,
αB xA (t)∆t + o(∆t)

consumers switch in the opposite direction – from brand A to brand B. It follows from (7) and (8)
that for any given αA and αB
αA xB (t) = αB xA (t)

t
holds for all values of t. Therefore, consumer switching does not affect xA (t) and xB (t) obtained

ip
under the primacy effect.
Substituting (7) and (8) into the firms’ profit functions and computing the integrals yield

cr
σαA c(αA )
ΠA (αA , αB ) = −
r(αA + αB + r) r

us
and
σαB c(αB )
ΠB (αA , αB ) = − .
r(αA + αB + r) r

an
∗ and α∗ denote the advertising intensities chosen by the firms in equilibrium.
Let αA B

∗ =
Proposition 1 (Equilibrium in Setting I). Under both the primacy and the recency effects αA
M
∗ = α∗ , where α∗ is implicitly defined by
αB

σ (α∗ + r)
= c0 (α∗ ) .
(2α∗ + r)2
d

How do the equilibrium levels of advertising intensities compare with the social optimum? The
te

social welfare is given by

Z ∞
1
p

W (αA , αB ) ≡ e−rt (xA (t) + xB (t))σ dt − (c(αA ) + c(αB )).

0 r
ce

It coincides with the sum of the firms’ profits, as consumer surplus is zero in the present model.
Hence, in Setting I the social welfare equals
Ac

σ(αA + αB ) 1
W (αA , αB ) = − (c(αA ) + c(αB )).
r(αA + αB + r) r

† †
Let αA and αB denote the socially optimal advertising intensities. The next proposition shows
that the firms over-advertise in equilibrium. Intuitively, each firm cares only about the number
of consumers that prefer its own brand. The social planner, on the other hand, aims to increase
the number of consumers that learn about the product existence through the firms’ advertising
messages, xA (t) + xB (t).

Proposition 2 (Social optimum in Setting I). The socially optimal levels of advertising intensities

8
Page 8 of 22
 † †
are αA = αB = α† , where α† is implicitly defined by

σr  
2 = c0 α† .
(2α† + r)

Advertising is socially excessive, α∗ > α† .

t
In the simple setting with x∅ (0) = 1 the equilibrium outcome is the same under the primacy

ip
and the recency effects. However, if at t = 0 a number of consumers have already formed their
preferences about the two brands, as in Setting II, or if one of the firms has competitive advantage,

cr
as in Setting III, the two advertising order effects have different strategic implications.

us
4 Setting II: Growing Market
Consider the setting in which at the beginning of the game (1 − γ)/2 consumers believe brand A
is better, (1 − γ)/2 consumers prefer brand B, and the rest are ignorant consumers. Under the

an
primacy effect  
e−(αA +αB )t

0 0 γ
 α −α e−(αA +αB )t 
1
x̄(t) =  1 0  2 (1 − γ) .
A A

M
 α +α A B
αB −αB e−(αA +αB )t 1
αA +αB 0 1 2 (1 − γ)

Hence,
αA − αA e−(αA +αB )t 1
d

xPA (t) = γ + (1 − γ) (9)

αA + αB 2
te

and
αB − αB e−(αA +αB )t 1
xPB (t) = γ + (1 − γ). (10)
αA + αB 2
p

(Superscript “P” stands for primacy.) Substituting (9) and (10) into the firms’ profit functions and
ce

computing the integrals yield

σαA 1 σ c(αA )
ΠPA (αA , αB ) = γ + (1 − γ) − (11)
r(αA + αB + r) 2 r r
Ac

and
σαB 1 σ c(αB )
ΠPB (αA , αB ) = γ + (1 − γ) − . (12)
r(αA + αB + r) 2 r r
These formulas deserve some discussion. At the beginning of the game, (1 − γ)/2 consumers
prefer brand A and under the primacy effect will never switch to brand B (they are “locked up” by
firm A). Firm A earns
1 σ
(1 − γ)
2 r
from these consumers. Similarly, (1 − γ)/2 consumers are locked up by firm B. Firm B earns

1 σ
(1 − γ)
2 r

9
Page 9 of 22
from these consumers. Finally, γ consumers are unaware of the product existence. If an ignorant
consumer first sees an ad describing brand A, he becomes firm A’s customer. If the consumer first
sees firm B’s ad, he becomes firm B’s customer. Firms A and B earn, respectively,

σαA
γ
r(αA + αB + r)

t
and

ip
σαB
γ
r(αA + αB + r)

cr
from γ initially ignorant consumers.
In contrast to Setting I, the recency effect in Setting II gives rise to a different vector x̄(t).5 The
profits of the firms are given by

us
σαA 1 σ 1 σ(αA − αB ) c(αA )
ΠR
A (αA , αB ) = γ + (1 − γ) + (1 − γ) − (13)
r(αA + αB + r) 2 r 2 r(αA + αB + r) r

an
and

σαB 1 σ 1 σ(αB − αA ) c(αB )

ΠR
B (αA , αB ) = γ + (1 − γ) + (1 − γ) − . (14)
r(αA + αB + r) 2 r 2 r(αA + αB + r) r
M
(Superscript “R” stands for recency.) It follows that firm A’s profit function under the recency effect
(13) differs from that under the primacy effect (11) by
d

1 σ(αA − αB )
(1 − γ) . (15)
te

2 r(αA + αB + r)

This term reflects consumer switching between the brands. It enters with negative sign firm B’s
p

profit function (14).

P ∗ and αP ∗ denote the advertising intensities chosen by the firms in equilibrium under
Let αA
ce

B
R∗ and αR∗ denote the equilibrium intensities chosen under the
the primacy effect. Similarly, let αA B
recency effect. Proposition 3 shows that the equilibrium advertising intensities are higher under the
recency effect. This result is driven by (15), which creates extra incentives for the firms to advertise.
Ac

Proposition 3 (Equilibrium in Setting II).

P ∗ = αP ∗ = αP ∗ , where αP ∗ is implicitly defined by
(i) Under the primacy effect αA B

σ αP ∗ + r


= c0 αP ∗ .


γ 2
(2αP ∗ + r)
5
Recall that in Setting I, αA xB (t) = αB xA (t) holds for any values of αA and αB . Here, αA xP P
B (t) = αB xA (t) holds
only for αA = αB .

10
Page 10 of 22
 R∗ = αR∗ = αR∗ , where αR∗ is implicitly defined by
(ii) Under the recency effect αA B

σ αR∗ + r


1 σ
= c0 αR∗ .


γ + (1 − γ) R∗
2
(2αR∗ + r) 2 2α + r

(iii) There is more advertising under the recency effect, αR∗ > αP ∗ .

t
The social welfare is the same under the primacy and the recency effects,

ip
σ(αA + αB ) σ 1
W (αA , αB ) = γ + (1 − γ) − (c(αA ) + c(αB )).
r(αA + αB + r) r r

cr
Indeed, it does not matter for the social planner whether a consumer switches from one brand to

us
the other after he observes an advertising message for the latter (the recency effect), or not (the
primacy effect). The next proposition shows that the firms over-advertise under the primacy effect,
and, hence, under the recency effect.

an
Proposition 4 (Social optimum in Setting II). The socially optimal levels of advertising intensities
† †
are αA = αB = α† , where α† is implicitly defined by

σr
M
γ = c0 (α† ).
(2α† + r)2

Advertising is socially excessive, αR∗ > αP ∗ > α† .

d

5 Setting III: One Firm Has Advantage

te

Consider the setting in which at the beginning of the game γ consumers are ignorant and the rest
p

believe brand A is better. As in Setting II, the primacy and the recency effects give rise to different
equilibrium outcomes. Under the primacy effect the profit functions of the firms are given by
ce

σαA σ c(αA )
ΠPA (αA , αB ) = γ + (1 − γ) − (16)
r(αA + αB + r) r r
Ac

and
σαB c(αB )
ΠPB (αA , αB ) = γ − . (17)
r(αA + αB + r) r
Under the recency effect

σαA σ(αA + r) c(αA )

ΠR
A (αA , αB ) = γ + (1 − γ) − (18)
r(αA + αB + r) r(αA + αB + r) r

and
σαB c(αB )
ΠR
B (αA , αB ) = − . (19)
r(αA + αB + r) r
Formulas (16) through (19) deserve some discussion. Consider first γ initially ignorant con-
sumers. The profits that firm A and firm B earn from these consumers are the same under the

11
Page 11 of 22
recency effect as under the primacy (the result obtained in Setting I). In particular, firm A earns

σαA
γ
r(αA + αB + r)

and firm B earns

σαB
γ .
r(αA + αB + r)

t
ip
Next, consider the rest of the consumers that at time t = 0 prefer brand A. These consumers are
locked up by firm A under the primacy effect, hence firm A earns

cr
σ
(1 − γ) (20)
r

us
and firm B earns 0. Under the recency effect firm A loses part of (20) to firm B. In particular, firm
A earns
σ(αA + r)
(1 − γ)
r(αA + αB + r)

an
and firm B earns
σαB
(1 − γ) .
r(αA + αB + r)
M
Proposition 5 (Equilibrium in Setting III).
P ∗ = αP ∗ = αP ∗ , where αP ∗ is implicitly defined by
(i) Under the primacy effect αA B
d

σ αP ∗ + r


= c0 αP ∗ .


γ 2
te

(2αP ∗ + r)

R∗ and αR∗ are implicitly defined by

(ii) Under the recency effect αA
p

σ (αR∗
B +γr )

= c0 αA
ce

R∗ ,


 2
αR∗ R∗

(A +αB +r )
σ αR∗(
A +r )
= c0 αB
R∗ .


 2
αA +αR∗
(R∗
B +r )
Ac

R∗ < αR∗ < αR∗ + (1 − γ)r.

Also, αA B A

(iii) Firm B chooses higher advertising intensity under the recency effect than under the primacy,
R∗ > αP ∗ .
αB

The first result of Proposition 5 is easily understood. Under the primacy effect (1−γ) consumers
are locked up by firm A. Firm A and firm B are symmetric with respect to γ initially ignorant
customers, for which they compete. Hence, the firms choose the same advertising intensities in
P ∗ = αP ∗ .
equilibrium, αA B
The result that firm B chooses higher advertising intensity than firm A under the recency effect
is also intuitive. At the beginning of the game firm A’s advertising messages affect only γ ignorant
consumers, as the rest of the consumers already believe brand A is better. Firm B’s messages, on

12
Page 12 of 22
the other hand, affect all consumers – ignorant as well as those who initially preferred brand A.
Thus, firm B’s incentives to advertise are stronger than firm A’s, especially at the beginning. This
R∗ > αR∗ .
results in αB A
Finally, Proposition 5 shows that firm B chooses higher advertising intensity under the recency
R∗ > αP ∗ = αP ∗ . Firm A’s equilibrium advertising intensities αR∗
effect than under the primacy, αB B A
P ∗ = αP ∗ , however, cannot be ranked because Firm A’s best reply under the recency effect is
and αA

t
non-monotonic in αB .

ip
The social welfare function in the current setting coincides with the one obtained in the previous
setting,

cr
σ(αA + αB ) σ 1
W (αA , αB ) = γ + (1 − γ) − (c(αA ) + c(αB )).
r(αA + αB + r) r r

us
The next proposition shows that under the primacy effect both firms over-advertise, and, hence,
under the recency effect firm B over-advertises (α† and αA
R∗ cannot be ranked).

Proposition 6 (Social Optimum in Setting III). The socially optimal levels of advertising intensities

an
† †
are αA = αB = α† , where α† is implicitly defined by

σr  
γ 2 = c0 α† .
(2α† + r)
M
R∗ > αP ∗ > α† .
Moreover, αB
d

6 Setting II versus Setting III

te

The social welfare function is the same in Setting II and Setting III. This makes comparison between
the equilibria in the two settings legitimate and, therefore, valuable.
p

Proposition 7 (Equilibrium in Setting II vs. equilibrium in Setting III).

ce

(i) Under the primacy effect the firms’ equilibrium advertising intensities in Setting II are the same
as in Setting III,
P∗ P∗ P∗ P∗
Ac

αA [S2] = αB [S2] = αA [S3] = αB [S3] .

(ii) Under the recency effect firm B’s equilibrium advertising intensity is higher in Setting III than
in Setting II,
R∗ R∗
αB [S3] > αB [S2] .

(The corresponding settings are specified in square brackets.) The first result of Proposition
7 follows directly from Proposition 3(i) and Proposition 5(i). Intuitively, in both settings under
the primacy effect the firms compete for γ initially ignorant consumers. Whether the rest of the
consumers are locked up by both firms, as in Setting II, or by firm A, as in Setting III, does not
change the competition between the firms. Hence, the equilibrium advertising intensities in Setting
II are the same as in Setting III.

13
Page 13 of 22
 Next, consider the recency effect. In Setting III, at the beginning of the game firm B’s advertising
messages affect all consumers – γ ignorant consumers and (1 − γ) consumers who prefer brand A.
However, in Setting II firm B’s messages have actual effect only on γ +(1−γ)/2 consumers – ignorant
and those who initially preferred brand A. Thus, firm B’s incentives to advertise are stronger in
Setting III. As a result, firm B chooses higher advertising intensity in Setting III than in Setting II.

t
ip
7 Conclusion
This paper has applied the theories of exposure order effects, developed in psychology literature, to

cr
an industrial organization model to explore their role in advertising competition. The equilibrium
advertising intensities were characterized separately under the primacy and the recency effects in

us
three different settings. It was shown that in a new market (Setting I) the advertising equilibrium
does not depend on the type of exposure order effect.
However, in a growing market where a number of consumers have already formed their preferences

an
about the two brands (Setting II), or if one of the firms has more customers than the other (Setting
III), the primacy and the recency effects give rise to different equilibrium outcomes. Under the
primacy effect the consumers who prefer a particular brand from the outset will never switch to
M
the other brand. The firms compete for initially ignorant consumers, and, therefore, choose the
same advertising intensities across both settings. Under the recency effect, on the other hand, each
firm’s advertising messages affect ignorant consumers as well as those who prefer the rival brand.
d

In Setting II the firms choose higher advertising intensities under the recency effect than under the
primacy. In Setting III firm B chooses higher advertising intensity than firm A. It was also shown
te

that in all three settings the firms over-advertise in equilibrium.

One of the limitations of the present model is that the firms’ advertising intensities are fixed
p

throughout the game. A dynamic version of the advertising game, in which the firms choose their
advertising intensities continuously, does not have an analytical solution. This is because the differ-
ce

ential equations that describe the primacy and the recency effects are not linear in the advertising
intensities.6 Analyzing the static game under different initial conditions, however, uncovers some
features of the dynamic setup. The number of ignorant consumers, γ, decreases with the passage of
Ac

time. Therefore, under the primacy effect the equilibrium advertising intensities gradually converge
to zero. Under the recency effect the firms advertise more at the beginning, but their equilibrium
advertising intensities will always be bounded away from zero.

Appendix
Derivations of (5) and (6). Consider matrix P . This matrix has eigenvalues 0, −(αA + αB ), 0.
Hence,
P = QPeQ−1 ,
6
On differential games see, for example, Zaccour (2002) and Erickson (2003).

14
Page 14 of 22
where  
0 0 0
Pe = 0 −(αA + αB ) 0 ,
 

0 0 0
   
0 1 0 1 1 1
αA αA 1 −1
Q = α − αA +αB , Q = 1 0 0 .
   

t
A +αB αA +αB 
αB
− αAα+α

ip
1
αA +αB
B
B
− αA +α B
0 αB −αA

The columns of Q are right eigenvectors of P and the rows of Q−1 are left eigenvectors. The

cr
eigenvectors are unique up to a multiplicative constant. The constant in the right eigenvector for
eigenvalue 1 is chosen so that it is the invariant probability distribution for P ,

us
 
0
 αA 
 αA +αB  .
αB

an
αA +αB

Next, consider matrix R. This matrix has eigenvalues 0, −(αA + αB ), −(αA + αB ). Hence,
M
e −1 ,
R = QRQ

where  
0 0 0
d

R = 0 −(αA + αB ) 0 ,
e  
te

0 0 −(αA + αB )
Q is the same as above.
Once P and R are diagonalized, it is easy to compute exponentials etP and etR ,
p

 
ce

e−(αA +αB )t
n  
0 0

∞ Q tP
e Q−1 1 0 0
X  α −α e−(αA +αB )t
tP
0 Q−1 = 

e = = Q 0 e
 −(α A +α B )t  A A
1 0
n!  αA +αB 
n=0 αB −αB e−(αA +αB )t
0 0 1 α +α 0 1
A B
Ac

and
 n  
∞ e Q−1
Q tR 1 0 0
X  −1
etR = = Q 0 e−(αA +αB )t 0 Q

n!
n=0
0 0 e−(αA +αB )t
 
e−(αA +αB )t 0 0
 α −α e−(αA +αB )t α +α e−(αA +αB )t α −α e−(αA +αB )t 
=
A A
αA +αB
A B
αA +αB
A A
αA +αB
.

αB −αB e−(αA +αB )t αB −αB e−(αA +αB )t αB +αA e−(αA +αB )t
αA +αB αA +αB αA +αB

15
Page 15 of 22
Proof of Proposition 1. The profit function of firm i = A, B is
Z ∞ Z ∞
−rt c(αi ) σαi −rt −(αA +αB +r)t
 c(αi )
Πi (αA , αB ) = e xi (t)σ dt − = e −e dt −
r α + α r
0
 A B 0
σαi 1 1 c(αi ) σαA c(αi )
= − − = − .
αA + αB r αA + αB + r r r(αA + αB + r) r

t
Firm i’s best reply is implicitly defined by the first order condition for the profit maximization

ip
problem,
∂ σ(αj + r) c0 (αi )
Πi (αA , αB ) = − = 0,
∂αi r(αA + αB + r)2 r

cr
or
σ(αj + r)
= c0 (αi ),

us
(αA + αB + r)2
i 6= j. Because
∂2 −2σ(αj + r) c00 (αi )
Π i (αA , αB ) = − < 0,
∂αi2 r(αA + αB + r)3 r
the first order condition is sufficient.
Therefore, equilibrium advertising intensities satisfy
an
M
σ (α∗B +r)

0 ∗
2 = c (αA ) ,

(α∗A +α∗B +r)

σ α∗ +r
 ∗( A∗ ) 2 = c0 (αB
 ∗ ).
(αA +αB +r)
d

The equilibrium is unique and symmetric. Indeed, it follows from above that
te

∗
(αA + r) c0 (αA
∗ ∗
) = (αB + r) c0 (αB
∗
).
p

Because (α + r)c0 (α) is an increasing function of α, it must be the case that αA

∗ = α∗ = α∗ , where
B
ce

α∗ satisfies
σ (α∗ + r) 0 ∗
∗ 2 = c (α ) . (21)
(2α + r)
Ac

Next, observe that the left-hand-side of (21) is decreasing and the right-hand-side is increasing in
α∗ . Hence, the solution is unique.

Proof of Proposition 2. The first order conditions for the social planner’s optimization problem are
  
†

  σr 2 = c0 αA ,
α†A +α†B +r

 
†

  σr 2 = c0 αB .
α†A +α†B +r


16
Page 16 of 22
The matrix of second derivatives
c00 (αA )
!
2σ 2σ
− (αA +α B +r)
3 − r − (αA +α B +r)
3
2σ 2σ c00 (αB )
− (αA +α B +r)
3 − (αA +αB +r)
3 − r

is negative semidefinite, which implies the sufficiency of the first-order conditions. By symmetry,
† †
αA = αB = α† , where α† satisfies

t
σr  

ip
2 = c0 α† . (22)
(2α† + r)
It follows from (21), (22), and the convexity of c(α) that α∗ > α† .

cr
Proof of Proposition 3. Each part is proven in turn.

us
(i) Consider the primacy effect. Firm A’s best reply is given by the first order condition

σ(αB + r)
γ = c0 (αA ),

an
(αA + αB + r)2

firm B’s – by
σ(αA + r)
γ = c0 (αB ).
M
(αA + αB + r)2
As in the proof of Proposition 1, it can be shown that the first-order conditions are sufficient
P ∗ = αP ∗ = αP ∗ , where αP ∗ satisfies
and that the equilibrium is unique and symmetric, αA B
d

σ αP ∗ + r


= c0 αP ∗ .


γ (23)
te

2
(2αP ∗ + r)

(ii) Consider the recency effect. Firm A’s best reply is given by the first order condition
p

σ(αB + r) 1 σ(2αB + r)
ce

γ 2
+ (1 − γ) = c0 (αA ),
(αA + αB + r) 2 (αA + αB + r)2

firm B’s – by
σ(αA + r) 1 σ(2αA + r)
Ac

γ 2
+ (1 − γ) = c0 (αB ).
(αA + αB + r) 2 (αA + αB + r)2
It is straightforward to show that the first-order conditions are sufficient and that the equilib-
R∗ = αR∗ = αR∗ , where αR∗ satisfies
rium is unique and symmetric, αA B

σ αR∗ + r


1 σ
= c0 αR∗ .


γ + (1 − γ) R∗
2 (24)
(2αR∗ + r) 2 2α + r

(iii) It follows from (23), (24), and the convexity of c(α) that αR∗ > αP ∗ .

17
Page 17 of 22
Proof of Proposition 4. The first order conditions for the social planner’s optimization problem are
  
†
 γ

 σr 2 = c0 αA ,
α†A +α†B +r
 
†

 σr
 γ α† +α† +r2 = c0 αB .
A B

t
† †
By symmetry, αA = αB = α† , where α† satisfies

ip
σr  
γ 2 = c0 α† . (25)
(2α† + r)

cr
It follows from (23), (24), (25), and the convexity of c(α) that αR∗ > αP ∗ > α† .

us
Proof of Proposition 5. Each part is proven in turn.

(i) Consider the primacy effect. Firm A’s best reply is given by the first order condition

an
σ(αB + r)
γ = c0 (αA ),
(αA + αB + r)2
M
firm B’s – by
σ(αA + r)
γ = c0 (αB ).
(αA + αB + r)2
P ∗ = αP ∗ = αP ∗ , where αP ∗ satisfies
The equilibrium is unique and symmetric, αA
d
B

σ αP ∗ + r


te

= c0 αP ∗ .


γ 2 (26)
(2αP ∗ + r)
p

(ii) Consider the recency effect. Firm A’s best reply is given by the first order condition
ce

σ(αB + γr)
γ = c0 (αA ),
(αA + αB + r)2

firm B’s – by
Ac

σ(αA + r)
= c0 (αB ).
(αA + αB + r)2
R∗ and αR∗ satisfy
Therefore, αA B

σ (αR∗
B +γr )

0 αR∗ ,


 = c A 2
(αR∗ R∗

A +αB +r )
R∗ +r (27)
(
 R∗ A R∗ ) 2 = c0 αB
 σ α R∗ .


(αA +αB +r)

R∗ < αR∗ < αR∗ + (1 − γ)r. By contrary, suppose αR∗ ≥ αR∗ . Then
It is left to show that αA B A A B

18
Page 18 of 22
 c0 αA
R∗ ≥ c0 αR∗ . It follows from (27) that



B

R∗ + γr R∗ + r



σ αB σ αA
R∗ + αR∗ + r

2 ≥ R∗ + αR∗ + r

2 ,
αA B αA B

or
R∗ R∗
αB ≥ αA + (1 − γ)r.

t
ip
R∗ < αR∗ .
This contradicts the supposition. Hence, αA B

Next, suppose αB ≥ αA + (1 − γ)r. Then c αB > c0 αA

R∗ R∗ 0 R∗ R∗ . It follows from (27) that

cr
R∗ + r R∗ + γr



σ αA σ αB
R∗ + αR∗ + r

2 > R∗ + αR∗ + r

2 ,

us
αA B αA B

or
R∗ R∗
αB < αA + (1 − γ)r.

an
R∗ < αR∗ + (1 − γ)r.
This contradicts the supposition. Hence, αB A

(iii) By contrary, suppose αP ∗ ≥ αB

R∗ . Then c0 αP ∗ ≥ c0 αR∗ . It follows from (26) and the



B
M
second equation in (27) that

σ αP ∗ + r R∗ + r



σ αA
γ ≥
2 . (28)
(2αP ∗ + r)2 R∗ + αR∗ + r
αA
d

The right-hand-side of the above inequality is greater than

te

R∗ + r


σ αB

.
p

R∗ + r 2
2αB
ce

Indeed,
R∗ + r R∗ + r



σ αA σ αB

−
R∗ + αR∗ + r 2 R∗ + r 2


αA B 2αB
Ac

can be rewritten as
R∗ − αR∗ R∗ αR∗ + r − αR∗ + r 2 + αR∗ r




σ αB A αB A B A
> 0.
R∗ + αR∗ + r )2 ( 2αR∗ + r 2


αA B B

Therefore, (28) implies

σ αP ∗ + r R∗ + r



σ αB
γ ≥
.
(2αP ∗ + r)2 R∗ + r 2
2αB
R∗ > αP ∗ . This
Because (α + r)/(2α + r)2 is a decreasing function of α, it follows that αB
R∗ > αP ∗ .
contradicts the supposition. Hence, αB

19
Page 19 of 22
Proof of Proposition 6. The result follows directly from Proposition 3(i), Proposition 4, and Propo-
sition 5(i,iii).

Proof of Proposition 7. Each part is proven in turn.

(i) The result follows directly from Proposition 3(i) and Proposition 5(i).

t
R∗ [S2] ≥ αR∗ [S3]. Then c0 αR∗ [S2] ≥ c0 αR∗ [S3] . It follows from



(ii) By contrary, suppose αB B B B

ip
(24) and the second equation in (27) that

R∗ [S2] + r R∗ [S3] + r



σ αB 1 σ σ αA

cr
γ
+ (1 − γ) R∗
R∗ [S2] + r 2
≥
.
R∗ [S3] + αR∗ [S3] + r 2
(29)
2αB 2 2αB [S2] + r αA B

us
The right-hand-side of the above inequality is greater than

R∗ [S3] + r


σ αB

,
R∗ [S3] + r 2
2αB

an
as was shown in the proof of Proposition 5. Therefore, (29) implies

R∗ [S2] + r R∗ [S3] + r


M
σ αB 1 σ σ αB
γ
+ (1 − γ) R∗
R∗ [S2] + r 2
≥
,
R∗ [S3] + r 2
2αB 2 2αB [S2] + r 2αB

or
R∗ [S2] + r R∗ [S3] + r



d
σ αB 1 σr σ αB

− (1 − γ)
R∗ [S2] + r 2

≥
R∗ [S2] + r 2

.
R∗ [S3] + r 2
2αB 2 2αB 2αB
te

R∗ [S3] > αR∗ [S2].

Because (α + r)/(2α + r)2 is a decreasing function of α, it follows that αB B
R∗ [S3] > αR∗ [S2].
This contradicts the supposition. Hence, αB B
p
ce

References
Ac

Brekke, K. A., Rege, M., 2007. Advertising as a distortion of social learning. The B. E. Journal of
Theoretical Economics, 7, art. 38.

Brunel, F. F., Nelson, M. R., 2003. Message order effects and gender differences in advertising
persuasion. Journal of Advertising Research, 43, 330-341.

Doob, J. L., 1990. Stochastic Processes. New York: Wiley Classics Library.

Erickson, G. M., 2003. Dynamic Models of Advertising Competition, 2nd edition. Norwell, MA:
Kluwer Academic Publishers.

20
Page 20 of 22
Haugtvedt, C. P., Wegener, D. T., 1994. Message order effects in persuasion: An attitude strength
perspective. Journal of Consumer Research, 21, 205-218.

Krähmer, D., 2004. Advertising and consumer memory. University College London mimeograph.

Lana, R. E., 1963. Interest, media, and order effects in persuasive communications. Journal of

t
Psychology, 56, 9-13.

ip
Niedrich, R. W., Swain, S. D., 2008. The effects of exposure-order and market entry-information
on brand preference: A dual process model. Journal of the Academy of Marketing Science, 36,

cr
309-321.

us
Shapiro, J. M., 2006. A “memory-jamming” theory of advertising. SSRN working paper available
at http://ssrn.com/abstract=903474.

an
Zaccour, G. (ed.), 2002. Optimal Control and Differential Games. Advances in Computational
Management Science, Vol. 5. Norwell, MA: Kluwer Academic Publishers.
M
d
p te
ce
Ac

21
Page 21 of 22
Title Page (with Full Author Details)

Exposure Order Effects and Advertising Competition

Oksana Loginova∗

t
ip
Department of Economics
University of Missouri-Columbia

cr
118 Professional Building
Columbia, MO 65211, USA

us
April 17, 2009

an
M
d
p te
ce
Ac

∗
Tel.: +1-573-882-0063; fax: +1-573-882-2697. E-mail: loginovao@missouri.edu.

1
Page 22 of 22