Positioning of Store Brands of store brands.pdf · We test the implications of our analysis in...
Transcript of Positioning of Store Brands of store brands.pdf · We test the implications of our analysis in...
Positioning of Store Brands
Serdar Sayman Stephen J. Hoch
Jagmohan S. Raju
April 2001
_______________________________________________________________________________
Serdar Sayman is Assistant Professor of Marketing College of Adm. Sciences, Koç University Rumeli Feneri Yolu, Sariyer Istanbul, 80910, Turkey Tel: 90– 212– 338 1616, Fax: 90– 212– 338 1653 E-mail: [email protected] Stephen J. Hoch is John J. Pomerantz Professor of Marketing The Wharton School, University of Pennsylvania, Suite 1400, Steinberg Hall - Dietrich Hall, Philadelphia, PA, 19104 - 6371. Tel: 215− 898 0233, Fax: 215− 898 2534 E-mail: [email protected]
Jagmohan S. Raju is Professor of Marketing The Wharton School, University of Pennsylvania, Suite 1400, Steinberg Hall - Dietrich Hall, Philadelphia, PA, 19104 - 6371. Tel: 215− 898 1114, Fax: 215− 898 2534 E-mail: [email protected]
Abstract
We consider the retailer’s store brand (SB) positioning problem in a market with two national brand (NB)
competitors. In the context of our model, two brands are assumed to be positioned close to each other if the
“perceptual” distance between the two brands is small, and positioning affects the degree of price competition.
Our game-theoretic model indicates that the optimal strategy for the retailer is to position the store brand as
close as possible to the leading national brand (brand with the highest base level of demand) unless the cost of
doing so is beyond a critical level. Even when facing two equally strong brands, the retailer is better off
targeting one of the national brands rather than adopting a mid-point positioning where the store brand
competes “equally” with the two national brands. Positioning the store brand closer to the leading national
brand increases the cross price sensitivity between the two, and leads to (a) lower wholesale prices from both
the leading (NB1) and secondary (NB2) national brands; (b) higher margins for the retailer; and (c) increased
category demand --- all of which adds up to increased category profit relative to other positioning strategies.
We test the implications of our analysis in three empirical studies. In a field study in two U.S. supermarket
chains, observational data (labeling, package design, color, shelf placement, etc.) showed that the probability
of a national brand being targeted by the store brand is an increasing function of its relative market share. In a
second study we estimated cross price elasticity in 19 categories as a means of assessing inter-brand
competition. In categories with high quality SB’s, the cross-price effects do suggest that the SB and NB1
compete more intensely with each other than with NB2. In categories with low quality store brands, however,
the cross-effects are more inline with the asymmetries in cross-price elasticity reported in previous
price/quality tier research. In a third product perception study, we found that although explicit targeting by
store brands influenced consumer perceptions of physical similarity, it had no influence on consumers’
perceptions of overall or product quality similarity. In fact, the SB was rated as more similar to the lower share
national brands. And so while it appears that retailers do follow a positioning strategy consistent with our
model, it meets with more limited success in changing consumer perceptions and demand side behavior.
1
1. Positioning of Store Brands
Store brands (SB’s) or private labels are created and controlled by retailers. In aggregate
they constitute about 20% of unit sales (IRI 1998) and are among the top three brands in 70%
of supermarket product categories (Quelch and Harding 1996). As is true for any brand,
positioning of the store brand can have an important influence on its performance. Unlike the
manufacturers of the national brands (NB’s), however, the downstream retailer has a different
objective function. Whereas national brand manufacturers position their products to maximize
the profits from their own products, the retailer focuses on maximizing profits from the entire
product category, including profit from store and national brands (Hoch and Lodish 1998). We
model how the retailer should position the store brand to maximize category profits within the
context of a category with two national brands, one of which is stronger. Positioning is
operationalized as the perceptual distance between two brands. Brands positioned closer exhibit
a higher cross-price elasticity. We focus on: (a) whether the store brand should target a specific
national brand or follow an “in the middle” positioning and compete to a lesser degree with
both NB’s; and (b) if targeting is better, which NB should be targeted. Although we take the
retailer’s perspective, a better understanding of store brand positioning strategy is also
important to NB manufacturers who must coexist with store brands.
Schmalensee (1978) noted that store brands often imitate the category leader,
presumably to signal comparable quality at a lower price. Although the demand for the store
brand may increase, the downside is that the demand for the targeted leading national brand
may also decrease. Since the retailer also makes money by selling the national brands, it may
not be optimal to have the store brand specifically compete against the national brand with the
largest customer base and higher margins (Corstjens and Lal 2000). Instead of targeting a
2
national brand that generates substantial profit, adopting a mid-point position where the store
brand competes to a lesser extent with both NB’s may be better. Yet, we often observe retailers
targeting leading national brands. We believe that the answer to this puzzle lies in a better
understanding of how the retailer’s objectives drives the store brand positioning decision.
The product positioning literature usually ignores the retailer. To the best of our
knowledge, there is only one study incorporating the difference in the objectives of the retailer
and the national brand manufacturers into the positioning problem. Tyagi and Raju (1998)
examine the pre-emptive positioning strategies of national brands when there is a national
brand versus a store brand entrant. We focus on the store brand’s positioning problem and
attempt to find the exact optimal location in both the symmetric case, where both NB’s are
equally strong, and the asymmetric case, where one NB is stronger than the other. We adopt a
game theoretic approach and examine a market with two incumbent national brands one of
which is stronger, and a store brand entrant. SB positioning essentially involves choosing the
appropriate perceptual distance between the SB and the NB’s. This distance in turn determines
the degree of price competition between the store brand and each of the national brands. As
such, positioning the store brand closer to one national brand results in a higher cross price
sensitivity between the two.
Our analysis suggests that the retailer should position the store brand close to the
stronger national brand unless the cost of doing so is beyond a critical level. Further, compared
to other prospective strategies, this strategy is more profitable in categories where the leading
national brand is stronger. Our results also reveal that SB targeting of the leading national
brand leads to: (a) lower wholesale prices from both the leading national brand (NB1) and to a
lesser extent the secondary brand (NB2); (b) higher margins for the retailer on national brands;
3
(c) higher profits from the store brand; and (d) increased category demand --- all of which adds
up to increased category profit relative to other positioning strategies.
We test the implications of our analysis in three empirical studies. In a field study in
two U.S. supermarket chains, we gathered observational data (labeling, package design and
color, shelf placement, etc.) regarding the targeting strategies of store brands in various
categories. We found that if the store brand follows a targeting strategy, the category leader
invariably is the target. Further, the probability of a national brand being targeted by the store
brand is an increasing function of its market share relative to its competitors. In a second study
we used store-level data from A.C. Nielsen to examine demand−price relationships in 19
categories and estimate cross-price effects as a means of assessing inter-brand competition. In
categories with high quality SB’s, the cross-price effects do suggest that the SB and NB1
compete more intensely with each other than with NB2 (Bronnenberg and Wathieu 1996). In
categories with low quality store brands, however, the cross-effects are more in line with the
asymmetries observed in price/quality tier research (Blattberg and Wisniewski 1989). In a third
study we collected product perception data. We found that consumers could detect when store
brands targeted a national brand; consumers rated the physical similarity of the store brand and
the national brand to be much higher when the targeting was explicit rather than ambiguous.
However, explicit targeting had no influence on consumers’ perceptions of the overall
similarity or product quality similarity of the SB and NB1. In fact, the store brand was rated as
more similar to the lower share national brands (NB2 and NB3). And so while it appears that
retailers do follow a positioning strategy consistent with our model, it meets with more limited
success in changing consumer perceptions and demand side behavior.
4
2. The Model
We consider a market consisting of two NB’s, each offering one brand sold through a
common retailer. The retailer can introduce a store brand if it results in higher total category
profits for the retailer. Our model extends previous work in this area by allowing the retailer to
also decide how the store brand is positioned relative to the two national brands. For example,
the retailer may choose to position the store brand "in-between" the two national brands, or
may decide to target a particular national brand.
Before describing the model in detail, we discuss why we have chosen this particular
modelling approach. Previous studies of brand competition where there is heterogeneity in
consumer preferences has, by and large, used two distinct modelling approaches.
1. Models such as DEFENDER (Hauser and Shugan 1983) allow for horizontal
differentiation. No brand is uniformly better than the other brands; different consumers
buy different brands because of differences in tastes. Difference in tastes are often
modelled by allowing for different consumer ideal points in a Hotelling type framework
(Hotelling,1929). Competition amongst NB’s can be construed as competition among
horizontally differentiated brands.
2. Moorthy (1985) studies differences among vertically differentiated brands. In this
context, if prices are the same, all consumers prefer the brand with higher quality.
Competition between a SB and a NB is more like competition between vertically
differentiated brands.
Our positioning problem requires us to simultaneously model the competition among NB’s, as
well as the competition between a SB and NB’s. The problem is further complicated by the fact
that the competitive parties reside at two different levels of the distribution channel, with each
5
party maximizing its respective profits. To tackle this complex problem, we utilize a reduced
form modelling approach where heterogeneity in tastes and differences among brands are
captured through a parsimonious demand model well grounded in utility theory.
2.1. Demand Structure without the Store Brand.
The demand for national brand i, denoted by qi, i = 1, 2, is assumed to be as follows:
[ ])(11211
211 pppa
aaq −+−
+= θ (1)
[ ])(12122
212 pppa
aaq −+−
+= θ (2)
where pi is the price of national brand i, ai ∈ (0, 1) is the base level of demand of national brand
i, and θ ∈ (0, 1) is the cross-price sensitivity representing the degree of price competition
between the two national brands. The proposed linear demand function is consistent with utility
maximizing consumers with quadratic utility functions (see Shubik and Levitan 1980). A
different utility function could lead to another demand function; however, market data seems to
be consistent with linear demand (e.g., Brodie and de Kluvyer 1984, Bolton 1989).
The demand structure outlined in (1) and (2) generalizes the demand model used in
Raju, Sethuraman, and Dhar (1995) as it allows the base level of demand of the two national
brands to be different. Overall category demand equals 1 when p1 = p2 = 0 implying that there
is a bound on how much consumers will buy. We also assume that the marginal cost of NB’s to
the manufacturers is 0; so, prices are additional to the marginal cost.
2.2. Demand Structure with the Store Brand.
In addition to the two national brands, we now include the store brand denoted by the
subscript s in (3)-(5).
6
{ }
−+−+−++
= )()(211
11121121
1 pppppaaaa
q ss
δθ (3)
{ }
−+−+−++
= )()(211
22212221
2 pppppaaaa
q ss
δθ (4)
{ }
−+−+−++
= )()(211
221121
sssss
s pppppaaaa
q δδ (5)
where pS is the price of the store brand and aS ∈ (0, 1) is the base level of demand of the store
brand. As in (1) and (2), θ is the cross-price sensitivity between the two national brands. In
addition, δi ∈ (0, 1), the price sensitivity between the store brand and national brand i, captures
the extent to which the store brand competes with national brand i. As we shall discuss later in
more detail, the δi 's are affected by the positioning of the store brand. As in the case with only
two NB’s, overall category demand equals 1 when pl = p2 = pS = 0. However, we will show that
in equilibrium the introduction of a store brand leads to an increase in category volume due to a
lowering of the average price in the category.
Note that (3)-(5) have two price difference terms where as (1)-(2) contain only one price
difference term. The 1/2 outside the weighted sum of the price difference terms in (3)-(5) is a
normalization constant to ensure that the mere addition of another brand does not result in a
higher demand. This normalization also results in a structure where demand is affected by own
price and the difference between own price and the (weighted) average price of the competing
brands in the product category.
To keep the model tractable, we use the same parameter δl in (3) as well as (5), and δ2
in (4) as well as (5) implying that cross-price sensitivities are symmetric. In other words, a unit
price difference between NB1 and the SB has the same effect on NB1 demand as it has on SB
demand. This is consistent with the findings in Sethuraman, Srinivasan and Kim (1999), where
7
cross-price effects, when measured in absolute terms, are by and large equal. We are not,
however, assuming that the elasticities are symmetric because elasticities also depend on the
base level of demand. The linear demand functions impose a particular structure on the effect
of changes in wholesale prices on retail prices as noted in Tyagi (1999). We acknowledge that
other functional forms may lead to different outcomes, a limitation that we hope will be taken
care of in future research.
2.3. Modeling the Store Brand Positioning Decision.
Recall that the parameters δ1 and δ2 in (3)-(5) capture the extent to which the store
brand competes with the two national brands. In our framework, positioning corresponds to
choosing δ1 and δ2 so as to maximize the retailer’s category profits. δ1 and δ2 are determined by
the perceptual distance between the SB and the two NB’s, respectively.
In order to formally model the positioning decision, we assume that brands are located
in an n-dimensional perceptual space. Let f(d) map the distance between the two brands into
cross-price sensitivity. We assume that f(d) has the following characteristics.
1. f(d) should be a non-increasing function of d: This property implies that as the distance between the two brands increases, the cross-price sensitivity decreases. That is, d is inversely related to the δi’s.
2. As d tends to ∞∞∞∞, f(d) should approach 0: This property implies that if the two brands are
positioned very far apart, they do not compete with one another. 3. As d tends to 0, f(d) should approach 1: This puts an upper bound on δi. Recall that the
upper bound on θ is also 1. 4. The same change in d should lead to a greater change in f(d) when d is small: This
property assumes that a unit change in the store brand's position will have a greater impact on cross-price sensitivity when the store brand is closer to a national brand than when it is farther away. For example, imagine a product category characterized by a single perceptual dimension, and a brand is located at the origin. It seems reasonable to expect that a move by a second brand from 5 units of distance away to 4 units of distance would have a smaller impact on price competition than a move from 2 units away to 1 unit.
8
In addition to these four properties, if we restrict f(d) to be monotonic and continuous, then it
follows that f(d) must be a strictly convex function of d. The assumption that f(d) is convex is
easy to justify in situations where NB’s are positioned around clusters of consumer ideal points
and Lee and Staelin (2000) also show that f(d) is convex even when consumer ideal points are
distributed uniformly.1
2.4. Effect of SB Positioning on the Marginal Cost of the Store Brand.
The marginal cost of the store brand can be affected by its intrinsic quality (the
ingredients) and also by packaging, labelling, and the like. While these differences are
important, we combine the two into one marginal cost parameter in our model. Recall that our
model assumes that the marginal cost of the national brands equals zero. For parsimony, we
assume that when the store brand targets a national brand, the marginal cost of the store brand
to the retailer also equals zero. However, when it is positioned in the “middle”, we assume that
the store brand has a cost advantage over the national brands. More specifically, we assume
that the marginal cost of the store brand equals -km when the SB is positioned in the middle.
We also study a case where the cost of targeting depends on which national brand is being
targeted.
2.5. Additional Model Features.
In our model formulation positioning affects cross-price elasticity but is not assumed to
affect the base level of demand. One could argue that as the SB is positioned closer to a NB, its
base level of demand should increase as compared to when it is positioned in the middle
because the NB’s are probably positioned closer to consumer ideal points. On the other hand,
9
Lee and Staelin (2000) point out that the base level of demand may in fact decrease in these
circumstances. We decouple positioning from changes in base level of demand partly because
the effects are not clear, but mainly because we want to partial out the effect of positioning on
base demand and focus primarily on the effect of positioning on cross-price sensitivity and the
resultant effects on wholesale prices, demands, and profits.
Our model ignores store competition. Store competition may lower margins on leading
national brands as stores price these brands competitively to attract new shoppers. Hence, the
additional mileage that a retailer may gain by positioning the SB optimally may be lower as the
retailer margins on NB will be lower due to store competition. On the other hand, store
competition has the potential to increase the power of the NB manufacturers vis-à-vis a
particular retailer. Therefore, the importance of SB positioning as a means to discipline the NB
manufacturer may increase once one takes into account store competition. Overall, store
competition is an important issue but we do not account for it in our model.
2.6. Sequence of Decisions.
The assumed sequence of decisions is as follows.
• Stage 1: The retailer positions the store brand (δ1 and δ2 are determined).
• Stage 2: National brand manufacturers choose their respective wholesale prices w1 and w2 to maximize their respective profits.
• Stage 3: The retailer chooses retail prices p1, p2 and ps to maximize category profits. We assume that the national brand positions are fixed. Although NB’s may prefer to reposition,
it does not happen often even over the long run (Halstead and Ward, 1995). We also do not
account for the effect of other marketing variables such as advertising or personal selling.
1 Assume f(d) is linear in d ∈ (0, ∞). In this case the slope of f(d) has to be negative and finite to satisfy the above
conditions. But then it has to intersect with the d axis. Similarly any strictly concave function has to intersect with the d axis, given the above conditions. Only a strictly convex function can satisfy the four conditions above.
10
3. The Analysis
The analysis consists of two parts. First we consider the symmetric case where we
assume that the base level of demand for all three brands (NB1, NB2, and SB) are all equal
(i.e., we set a1 = a2 = aS = 1). The main purpose of the first part is to study how positioning of
the store brand affects overall retailer profits as well as its components so that one can get a
deeper insight into why positioning strategy is different when adopting a retailer versus
manufacturer perspective. In the second part of our analysis, we consider the asymmetric case
where one of the national brands is stronger than the other. More specifically, we consider the
case when a1 > a2 , aS.
3. 1. Symmetric Brands (a1 = a2 = aS = 1)
To focus only on demand side issues, we first begin by analyzing a scenario where the
retailer does not incur any additional cost when it is positioned to target a national brand. In
other words, we begin by assuming that km =0.
Lemma 1 The optimal store brand position lies on a line segment connecting the two national brands. Proof: See Technical Appendix A.
While the proof is in the Appendix, it may be worthwhile here to outline how this result
is derived. Let Πr denote the retailer's profit. We show in the Appendix that 01
>∂∏∂δ
r , and
02
>∂∏∂δ
r . Any point that is not on the line-segment joining the two national brands is
dominated by a point on the line-segment that joins the two national brands because by moving
to the line segment we can increase at least one of the δi's without reducing the other.
11
We believe that Lemma 1 has important and non-obvious implications. The fact that
retailer profits increase with δ1 and δ2 implies that the retailer gains by increasing the
competition between the store brand and the national brands. While it is commonly believed
that one should position a brand to minimize competition, it does not hold when we are talking
about the retailer. By appropriate positioning of the store brand, the retailer increases the
competition at the retail level. More competition at the retail level lowers NB – which is good
for the retailer. As we shall see later, this basic intuition can also help us further understand the
precise positioning strategy of the SB.
Proposition 1 When the brands are symmetric (a1 = a2 = aS = 1), it is optimal for the store brand to target either one of the national brands than to be positioned elsewhere on the line segment joining the two national brands, as long as f(d) is reasonably convex. Proof: See Technical Appendix A.
The intuition is as follows. The retailer's profits increase as the δi’s increase. However,
from Lemma 1, the optimal position is on the line segment joining the two national brands. If
we were to move the SB on the line segment joining the two national brands, δ1 and δ2 cannot
increase simultaneously. Therefore, the optimal positioning essentially boils down to
determining what is the best combination of δ1 and δ2. Since Πr is symmetric with respect to δ1
and δ2, the solution is either targeting or mid-point positioning. As f(d) is convex, it is best to
increase one of the δi’s to the maximum and the other to its lowest level by targeting one of the
national brands. Note that Proposition 1 depends on the convexity of the distance function. If
f(d) is linear, the optimal position is in the middle – which corresponds to δ1 = δ2 = (1+θ )/ 2.
How much convexity is needed? The answer is that while linear f(d) is not good
enough, only a small degree of convexity produces the results (see proof of Proposition 1 in
Technical Appendix A). The optimal positioning of the store brand is not a continuous function
12
of the convexity of f(d). As soon as the convexity of f(d) increases beyond a critical level, and
this critical level is not that high, it is optimal for the SB to target a NB rather than position at
any other point on the line segment joining the two brands.
The retail and wholesale prices of both national brands decrease with the introduction of
the store brand but the targeted national brand experiences a greater decrease.2 This is
consistent with Halstead and Ward (1995) who report that the most common response of NB’s
to the increasing SB threat is to decrease their prices. The introduction of the SB leads to
increased retailer margins on the NB’s. Furthermore, this increase is larger on the targeted
brand. The profits of both NB’s decline after store brand entry; and the decrease is larger for
the targeted NB. Finally, the equilibrium category demand increases, and demand for both
NB’s decreases with the introduction of the SB because the overall category demand is
relatively inelastic. However, the targeted NB does not lose as much demand as the non-
targeted NB because of its lower equilibrium price. Overall, targeting a NB results in better
terms of trade for the retailer and a more desirable allocation of demand across the three brands
that leads to higher profits – higher than the profits that would be obtained by positioning the
SB anywhere else.
We now allow for SB marginal cost to depend on its positioning strategy. Recall that
the marginal cost of the NB’s is assumed to equal 0. To keep the number of parameters to a
minimum, we capture cost of positioning by assuming that store brand marginal cost is equal to
the marginal cost of the national brands (i.e., 0) when it targets a NB, but it is –km when it is
positioned in the middle. The main result is reported in Proposition 2.
2 The equilibrium expressions are reported in Table A1 in Technical Appendix A.
13
Proposition 2: When the brands are symmetric (a1 = a2 = aS = 1), it is optimal for the store brand to target either one of the national brands, as long as f(d) is sufficiently convex and km is small. Proof: See Technical Appendix A.
If f(d) is more convex, km can be higher, and yet it is optimal for the SB to target a NB. The
maximum value of km (defined as km*) that can be tolerated before it becomes better to position
in the middle, increases as f(d) becomes more convex.
3.2. Asymmetric Brands (a1 > a2 , aS)
Analysis of the symmetric case results in the following key insights:
• The optimal position of the store brand is on the line segment joining the two national brands.
• As long as the distance function is reasonably convex, and the cost of targeting a NB is below a threshold, it is best to target one of the national brands. More convex the distance function, higher is the cost threshold.
As the two national brands were assumed to be symmetric, targeting one is equivalent to
targeting the other. Once we allow one national brand to be stronger than the other as well as
the SB (a1 > a2, aS), we can resolve the issue of whether it is better to target the strong or the
weak national brand. The main result is summarized in Proposition 3. Note that comparison of
profits from targeting NB1 or NB2 is independent of the cost advantage of positioning at the
midpoint.
Proposition 3 If a1 > a2, aS, the retailer's profit is higher if the store brand targets NB1 than when it targets NB2. Proof: See Appendix.
Proposition 3 is consistent with Schmalensee (1978). To understand the intuition, we consider a
specific case where a1 > a2 = aS. We show in the Appendix that targeting the stronger brand
results in lower total national brand demand than targeting the weaker national brand.
14
Furthermore, targeting the stronger national brand results in a lower average wholesale price.
These two effects combined result in lower total manufacturer profits on national brands when
the SB targets NB1 as opposed to NB2. Hence, targeting NB1 leads to a greater profit pressure
on the manufacturers, allowing the retailer to capture some of what is given up by the
manufacturers. Therefore, it is not surprising that we find that, as long as the SB is not too
weak, the retailer’s combined profits from the two NB’s are higher when the strong brand is
targeted as opposed to the weak brand. Furthermore, retailer’s profit from the SB is higher
when it is targeted at the strong brand as opposed to the weak brand.
Table 1 Equilibria in a Category with Asymmetric Brands
Without the Store Brand With the Store Brand
NB1 wholesale price (w1*) 2 2
4 8 32
2
+ +
+ +
θ θθ θ
a
2 4 4
24 32 72
2
( )+ +
+ +
θ θθ θ
a
NB2 wholesale price (w2*) 2 2
4 8 32 2
2
a a+ +
+ +
θ θθ θ
2 6 2
24 32 72 2
2
( )a a+ +
+ +
θ θθ θ
NB1 retail price (p1*) w a1 2
2
1
2 4
*+
+ +
+
θ θθ
)3148(2
428262
*2
22
2221
θθθθθθθθ
+++++++++
+aaaaaw SSS
NB2 retail price (p2*) w a a2 2 2
2 2 4
*+
+ +
+
θ θθ
)3148(2
64482
*2
22
22222
θθθθθθθθ
++++++++
+aaaaaw SS
SB retail price (pS*) )3148(2
482622
22
222
θθθθθθθθ
+++++++++ aaaaa SSS
NB1 demand (q1*) ( )( )
( )( )
1 2 2
1 8 16 62
22
+ + +
+ + +
θ θ θθ θ
aa
)73224)(1(2
)44)(3(2
2
2
θθθθθ++++
+++
Saaa
NB2 demand (q2*) ( )( )
( )( )
1 2 2
1 8 16 62 2
22
+ + +
+ + +
θ θ θθ θ
a aa
)73224)(1(2
)26)(22(2
2
22
θθθθθ
+++++++
Saaaa
SB demand (qS*) )73224)(1(2
2744)1(2 2
2
22
22
2 θθθθθθ
++++++++
+++ SS
S
aaaa
aaa
Retailer’s profit (Πro*, Πr*) q1*(p1*−w1*) + q2*(p2*−w2*) q1* (p1* − w1*) + q2* (p2* − w2*) + qS* pS*
15
Table 1 summarizes the equilibrium expressions for the asymmetric case with the SB
targeting the stronger national brand (note that a1 is set to 1, without any loss of generality). So
far, we have assumed that SB cost is the same whether it targets NB1 or NB2. This is consistent
with our basic assumption that the marginal production cost of NB’s is equal. Yet what happens
if the marginal cost of SB is lower when it targets NB2? Assume that the SB cost decreases to –
k2 < 0 when it targets NB2. What we find is that targeting NB1 is more profitable for the
retailer than targeting NB2, as long as k2 is small. The intuition is straightforward.
It is also of interest to understand how the advantage of targeting NB1 is affected by the
relative strength of the two NB’s. This result is useful from an empirical perspective. Define
∏ sr to be the retailer’s profit when the SB targets NB1, ∏ w
r to be the profit when it targets
NB2, and Π rm is the profit if the SB is positioned at the mid-point between the NB’s.
Proposition 4 The profit advantage of targeting the strong national brand is greater compared to other prospective strategies, if the second national brand is weaker. More specifically:
(i) [ ] .0
2
<∂
∏ ∏−∂a
wr
sr (ii)
∂∂
[ ]Π Πrs
rm
a−
<2
0
Proof: See Appendix.
Proposition 4 states that in categories where NB1 is stronger relative to NB2, targeting the NB1
is more advantageous compared to either targeting NB2 or the mid-point positioning. We note
that we resorted to numerical analysis to prove (ii), since Π rm depends on f(d) (or δm = δ1 = δ2 ).
We varied a2, aS, θ, and δm in relevant ranges and checked the sign of the derivative. Our
numerical analyses also showed that Proposition 4 holds when the cost of SB is lower at the
mid-point or when it targets NB2.
16
In the symmetric case, we were able to show that targeting one of the NB’s is better
than being anywhere else on the line (Proposition 1). In the asymmetric case, while we have
demonstrated analytically that targeting NB1 leads to higher profits than targeting NB2
(Proposition 3), we were not able to show analytically whether positioning the store brand on
the line segment results in lower profits than targeting NB1. Retailer profits from targeting
NB1 or NB2 do not depend on the precise distance function. However, profits from other
positioning choices on the line joining the two national brands do depend on the assumed
distance function, and comparison of mid-point positioning versus targeting is not sufficient.
Unlike the symmetric case, we were not able to derive a general result analytically for the
asymmetric case. Therefore, for the asymmetric case, assuming a number of distance functions
(e.g., f(d) = –exp(d), and f(d) = 1 / (1 + d) satisfy the conditions above), we conducted
numerical analysis, and in each of these cases, it turned out that targeting NB1 results in higher
profits than positioning anywhere else on the line segment joining the two NB’s. Even for the
linear distance function, which corresponds to the limiting case for convexity, targeting NB1 is
optimal unless NB2 is sufficiently strong. This implies that the convexity condition on the
distance function weakens when NB’s are asymmetric.
4. Empirical Studies
In this section we present three empirical studies that test implications of our analysis.
4.1. Study 1: Observational Data on Store Brand Positioning
A key finding in the theoretical analysis is that the store brand should locate next to the
leading national brand (Proposition 2) unless the cost of targeting is beyond a critical level.
Because of differences in cost of targeting, we may not observe this strategy in all product
categories. But we do not have access to marginal cost (and cost of targeting) data. However,
17
Proposition 4 suggests that other things being equal, the relative profitability of targeting NB1
is an increasing function of asymmetries in national brand strength, hence the probability that a
store brand targets the leading national brand should be higher in categories where the leader is
stronger.
4.1.1. Data and Methodology. We collected observational data. In two leading US
grocery chains, two observers collected data regarding the positioning / targeting strategies of
the store brands. To maintain consistency, the product categorization scheme provided in the
IRI Marketing Fact Book (1998) was utilized. 75 categories were randomly selected ranging
from dry grocery to frozen/refrigerated foods and health and beauty aids. Data were eventually
collected for 64 of these 75 categories in one chain (call this Store A) and for 56 categories in
the other (Store B). For the remaining categories, there was no store brand alternative
available. 53 of these categories were common to both stores.
Store brand products are easily identified by their brand names. Observers evaluated the
available extrinsic cues, and judged the positioning strategy of the store brand products. The
specific extrinsic dimensions used in the evaluation were: (i) package design; (ii) labeling /
color; (iii) shelf placement; and (iv) shelf talkers (“Compare and Save” signage) − if any. Prior
research has found that consumers evaluate store brands based on such extrinsic cues
(Richardson, Dick, and Jain, 1994). Each observer independently made a judgment of whether
SB was targeting one particular NB based upon a close match on all four extrinsic dimensions
listed above.3 Observers agreed 85% of the time. When both observers agreed that the store
brand was trying to compete with a specific national brand, the brand name of the targeted
3 The two observers were not aware to the market shares of the national brands and also unaware as to which of
the national brands was the leading national brand in a product category when they collected these data.
18
product was recorded.4 Examples of clear targeting and ambiguous targeting (or not targeting)
are shown in Figure 1.
Figure 1: Examples of Categories with Ambiguous and Clear Targeting
Ambiguous Targeting Clear NB Targeting
4.1.2.Results. In 25 of the 64 (39%) categories in Store A and 18 of 56 (32%)
categories in Store B, the store brand followed a targeting strategy. In the remaining categories,
the store brand either did not follow a targeting strategy or targeted multiple NB’s. Although at
first glance the overall level of targeting seems fairly low, it should be pointed out that the
observers concluded that a SB was targeting a particular NB only if all four criteria were
satisfied. Also, one needs to recognize that it is not that easy for the SB to differentially target
only one NB in cases where the two NB look quite similar.
We identified that in 21 out of 25 product categories of Store A, and in 15 out of 18
categories of Store B, the target was the category leader (identified from the Fact Book). This
finding is consistent with our first prediction that store brands generally target the category
leader if and when they follow a targeting strategy.
To test our second prediction (Proposition 4), we estimated separate logit models for
4 We also repeated the analysis only on the categories where both observers agreed a-priori. The results were
qualitatively similar.
19
Stores A and B based on the following variables:
Targeting Strategy of the Store Brand (TS). We use a dichotomous variable to represent whether the store brand in the category targets the leading national brand or not. Hence, for Store A, TS = 1 for 21 categories, and TS = 0 for 43 categories. The respective numbers for Store B are 15 and 41.
Base Level of Demand for the Major National Brands: We use national market shares as surrogates for base level of demands.5 In the context our model, NB with the higher base level of demand also, by and large, has the higher equilibrium market share. We identified the top two NB’s based on unit volume shares from the Marketing Fact Book for each of the product categories employed. The underlying assumption is that the effect of other/weaker NB on the store brand positioning is limited. MS1 and MS2 represent the unit shares of the highest and second highest share national brands respectively. Number of National Brand in the Category (#NB) and Category Size (M). We include these covariates to take into account other plausible explanations of the retailer’s store brand strategy. One can argue that targeting a specific national brand may not be the optimal if there is a large number of national brands in the category. The number of national brands in the category (#NB) is obtained from the Fact Book by identifying the number of distinct brand names. We do not have a strong prior regarding the effect of the category size. However, incremental profits obtained by the targeting strategy are higher in a large category (although the relative profitability does not depend on the category size). Therefore, in large categories the retailer may formulate the positioning strategy more carefully, and the store brand is more likely to target the leading national brand. Category size (M) is obtained by multiplying the category volume and the average price in the category.
We estimated binary logit models for both Stores A and B. In both specifications, the
log-odds ratio pertains to TS which represents whether the store brand targets the leading
national brand (TS = 1) or not (TS = 0). In the first model we use MS1 as a predictor variable.
However, when there are more than two national brands, relative base strength of the leading
national brand with respect to the secondary brand may be a more realistic measure. MS1/MS2 is
5 Note that the market shares for a particular retailer may be different than the national level market shares. To
make sure this was not a serious issue, we computed correlation in ranks (based on market share) computed at the local level and at the national level for 19 different product categories (see more on these data in Study 2). The correlation across all was 0.5. We then repeated the analysis for only the top five brands and the correlation was XXX. We also had access to data specific to area where one of the two stores used in Study 1 is located, though the data are not from the same time period during which we conducted Study 1. The leading national brands in this market is the same as the leading national brand nationally in XX% of the categories. While shares differ
20
used as a predictor variable in the second model. There is a positive correlation (+0.42 and
+0.73 for Stores A and B respectively) between MS1 and MS1/MS2. M and #NB serve as
covariates. Table 2 shows the results.
Table 2 Results of Study 1
Chain A Chain B
Model 1 Model 2 Model 1 Model 2
Variable Coefficient (p value)
Coefficient (p value) Coefficient
(p value) Coefficient
(p value)
Intercept −2.8797 (0.047)
−2.4232 (0.036)
−4.1662 (0.006)
−2.8421 (0.010)
M: Category size 0.0000 (0.867)
0.0001 (0.299)
−0.0001 (0.389)
−0.0000 (0.733)
#NB: Number of NB’s −0.0692 (0.440)
−0.1526 (0.095)
0.1160 (0.159)
0.0634 (0.401)
MS1: Share of the leading NB 7.9858 (0.012)
7.2619 (0.026)
MS1 / MS2: Relative market share 1.3209 (0.006)
0.6652 (0.052)
In the first model MS1 is the only significant variable for both data sets, and its effect is
in the hypothesized direction. In categories where the market share of the leading NB is higher,
the probability of observing a store brand that targets the leader increases. The effect of
MS1/MS2 is significant in the second model, although the significance is smaller for Store B.
This implies that if the leading national brand is stronger relative to the underdog, it is more
likely to be a target for the store brand. The effect of number of national brands and category
across stores, using national level data on shares in place of store level data in this particular case does not seem to introduce all that much error.
21
size have little impact on store brand positioning.
Estimating separate models for each chain may not be ideal as a number of omitted
variables relevant to both stores may result in correlated errors. We estimated this correlation
and it was 0.32 for the first model and 0.35 for the second model. We therefore estimated a
probit model that allows correlated errors across the two stores. This was done only for the 53
product categories that were common to both stores. We allowed store intercepts and the β
coefficients to be different for the two stores. But it turned out that we could not reject the
hypothesis that the β coefficients are the same across stores and so we estimated a probit with
pooled data model that allows for different store intercepts and correlated errors. The results are
very similar to the ones reported in Table 2.
4.1.3. Discussion. Study 1 provides evidence that if the store brand follows a targeting
strategy, the target is, indeed, the leading national brand. We also found that SB targeting
strategy depends on the (relative) market share of the leading national brand. A limitation of
this first study is that it considered only two grocery chains and may not generalize to other
retailers. More importantly, however, the data do not provide any direct evidence on whether
the targeting strategy has the intended influence on either consumers’ perceptions of store
brands or their buying behavior. A positioning strategy is a means for the retailer, and whether
or not a store brand is successful in competing with the leading brand is still an empirical
question. The next two studies address these issues.
4.2. Study 2: Inferring SB-NB Competitive Relationships from Secondary Data
Our model predicts that it is optimal for the store brand to choose a position closer to
NB1 thereby resulting in greater competition between NB1 and the SB than between NB2 and
the SB or NB1 and NB2.
22
4.2.1.Predicted Pattern of Cross-Price Effects: The three demand equations outlined
in (3)-(5) can be rewritten as follows:
++++−++
= ss
pppaaaa
q22
)22
1(1 121
11
211
δθδθ (6)
+++−+++
= ss
pppaaaa
q2
)22
1(2
1 22
212
212
δδθθ (7)
++−++++
= sss
s pppaaaa
q )22
1(22
1 212
21
1
21
δδδδ (8)
Let us define βij to be the effect of price of Brand i on the demand of Brand j. For example, β1
s
represents the effect of NB1’s price on SB demand. Recall that when the store brand targets
the leading national brand, it follows that δ1=1, δ2=θ. Furthermore, θ is less than 1 and so
δ1 > δ2 = θ. Keeping these in mind, (6)-(8) suggest the inequality relationships summarized in
Column 2 of Table 4.
Table 3 Predicted Pattern of Cross-Price Effects
Our Model
Price/Quality Tier Research
βs1 vs βs
2 > < βs
1 vs β2s > <
βs1 vs β2
1 > < βs
1 vs β12 > <
β1s vs β2
s > > β1
s vs βs2 > >
β1s vs β1
2 > > β1
s vs β21 > >
It is worthwhile to contrast the predictions of our model with the predictions from price/quality
tier research (e.g., Blattberg and Wisniewski 1989; Allenby and Rossi 1991). Assuming that
NB1 is in the top tier, NB2 is in the next quality tier, and the store brand is in the lowest quality
23
tier, price/quality tier research would make the predictions shown in Column 3. In 4 out of 8
cases, our model makes the same prediction as the price/quality tier research and in the other 4
cases the predictions are different. If we assume that NB1 and NB2 are in the same
price/quality tier, then two of the inequalities change to equalities, specifically βs1=βs
2 and
β1s=β2
s.
4.2.2. Data and Methodology: We utilized syndicated sales data from A.C. Nielsen for
19 product categories and 122 retailers operating in the top 50 US markets. In each of the
categories, a store brand was sold by more than 50% of the retailers. For each retailer, the data
base includes brand level information (unit sales, prices, and promotions) on a four week basis
for each category over 30 periods (February 1993 through May 1995). As in Study 1, we
assume that smaller share brands have limited influence on the store brand strategy. To take
into account the effects of other national brands, we combined them into an omnibus “third”
national brand.
We assume that the store brand positioning strategy is determined at the retailer level
and not modified over the 30 four week periods. Therefore we identified the category leader,
the secondary brand, and other national brands separately for each retailer based on unit sales
over the 30 periods. Hence, what we are measuring does not pertain to specific national
brands, but to the leader and the secondary national brand for each retailer. For each of the 19
categories, we estimated demand functions for the store brand, and the leading and secondary
national brands using a linear specification.6 The demand of a product is a function of the retail
prices and sales promotions.
We use the following notation. Let i = 1, 2, and 3 refer to national brands, and S refers
24
to the store brand. The subscript r = 1, ..., 122 refers to the 122 different retailers in the data.
Finally, the subscript t = 1, ..., 30 represents each of the four-week periods over which we have
the data. Our key measures are defined next.
Demand for National Brand i (Qirt). We use equivalent units as the basis for demands. Hence, the demand for a national brand is the pound (or ounce) sales. i = 1 designates the leading national brand based on the sales in retailer r, and i = 2 is the secondary national brand. The demand for national brand i = 3 is the sum of the demands for the other national brand products in the category.
Demand for the Store Brand (QSrt). Likewise, we use the equivalent unit sales of the store brand in the analysis. In the 19 categories employed here, store brands are present in 75% of the retailer-category combinations. Conditional on the presence, average store brand share across categories and retailers is 28%.
Retail Prices (Pirt, PSrt). Retail prices are also based on equivalent units. Equivalent unit prices are obtained by dividing the dollar sales by the number of equivalent units. As is standard when computing prices for brands made up of numerous individual sku’s, the prices are defined as the geometric share average or the Divisia price index of all the UPCs that make up that brand. P3rt is the weighted average price of the other national brands. The prices are effective prices net of promotions rather than regular prices.
Indicator for Sales Promotions (Dirt, DSrt). These dummy variables represent whether the corresponding brand was promoted in the particular retailer and period. Because sales promotions are often accompanied by a price reduction, there is some degree of negative correlation between the sales promotion indicator variables and the corresponding price. However, these simple correlations are less than 0.2 in magnitude. We note that D3rt is the percentage of the promoted brands in the “other brands” basket, since i = 3 consists of multiple products.
Indicator for Retailers (Rr). We included these dummy variables to account for the variation in demand across retailers. Basically Rr = 1 if the data point comes from retailer r, and 0 otherwise. More detailed discussion will be presented below.
We estimate the following models for each of the 19 product categories:
rtrtrtSrtSrtrtrtSrtSr
rrrt DDDDPPPPRQ 3132
121
11
13
132
121
11
111 ααααββββλ ++++++++= ∑ (9)
rtrtrtSrtSrtrtrtSrtSr
rrrt DDDDPPPPRQ 3232
221
21
23
232
221
21
222 ααααββββλ ++++++++=∑ (10)
6 Results using a double-log specification produced similar results.
25
rtS
rtS
rtS
SrtSSrt
Srt
Srt
SSrt
SS
rr
SrSrt DDDDPPPPRQ 332211332211 ααααββββλ ++++++++=∑ (11)
where Q, P and D represent demand, price and availability of sales promotions respectively.
Equations (9), (10), and (11) are estimated separately, and they designate the demands for the
leading national brand, the secondary brand, and the store brand. In the empirical model, the
demand and price terms are normalized with respect to their averages within respective
retailers. This allows us to interpret the coefficients as elasticity estimates. Retailer indicator
variables capture the differences in base levels of demand due to possible differences in
consumer demographics, retail competition etc.
Prior research (Bronnenberg and Wathieu 1996; Dhar and Hoch 1997) has shown that
the quality of the store brand is a key determinant of performance. Therefore we a-priori
divided the categories into high (n = 10) and low quality (n = 9) groups using data from Hoch
and Banerji (1993).7 Our main interest here is the jiβ terms (j = 1, 2, S, and i = 1, 2, 3, S). These
represent the effect of brand is price on the demand for brand j. In order to summarize the
results for the high and low store brand quality groups, we combined the estimates for each
group by weighting each coefficient according to its precision (the inverse of its standard error
squared). In all, we obtained a total of twelve jiβ ’s for each group. To avoid potential
econometric difficulties that could arise due to aggregating across heterogeneous retailers, we
also estimated models at the individual retailer. Due to sparse data (30 observations/retail
account), we identified the five largest multi-market retailers that operated in a minimum of
five different markets (30 periods x 5+ markets=150+ observations/retailer). We estimated the
models in Eq. 9-11 separately for each retailer and included market level dummies.
26
4.2.3. Results: The average adjusted R2 from the regression models is 0.72. Note that
the dependent variables in Equations (9)-(11) are ratios to the average demands. We also
calculated the estimated demands by multiplying the estimations from the above models with
the corresponding averages. The average simple correlation between the actual demands and
the estimated demands is 0.94. The models explain the data quite well thereby providing
support for our assumed demand structure. The price and promotion variables together explain
around 0.25 of the variation.
Table 4: Average Estimated Cross-Price Elasticities in Study 2
Aggregate Results Top 5 Multi-Market Retailers Our
Model Price/Quality Tier Research
High Quality Store Brands
Low Quality Store Brands
High Quality Store Brands
Low Quality Store Brands
βs1 vs βs
2 > < .165=.157 -.024<.090 .237=.226 .068<.110
βs1 vs β2
s > < .165>.117 -.024<.126 .237>.053 .068<.107
βs1 vs β2
1 > < .165>.035 -.024<.035 .237>.023 .068<.073
βs1 vs β1
2 > < .165>.065 -.024<.017 .237>.123 .068<.259
β1s vs β2
s > > .146>.117 .225>.126 .206>.053 .186>.107
β1s vs βs
2 > > .146=.157 .225>.090 .206<.226 .186>.110
β1s vs β1
2 > > .146>.065 .225>.017 .206>.123 .186<.259
β1s vs β2
1 > > .146>.035 .225>.035 .206>.023 .186<.216
The estimated cross-price effects are compared with the main predictions of our model
in Table 4 and as can be seen the pattern depends crucially on store brand quality. All pairs of
coefficients are significantly different from each other (p<.05) except when there is an = sign.
First, let us focus on the cross-price effects for categories with higher quality store brands
(columns 4 and 6 of Table 4). For both the aggregate model and the models for the top 5
individual retailers, the results are reasonably consistent with our model. The only sign
7 Hoch and Banerji (1993)’s measure of store brand quality is based on a survey of retail experts (quality assurance
managers) from 50 leading chains and wholesalers.
27
reversal occurs for the individual models, where β1s < βs
2. The most striking result is that the
effect of SB price on NB1 demand (βs1) is more than both the effect of NB2 price on NB1
demand (β21) and NB1 price on NB2 demand (β1
2), i.e., {row 3, column 4 and 6} and {row 4,
column 4 and 6}. This is consistent with our conjecture that the SB should be positioned closer
to NB1 and is counter to the prediction of the price/quality tier theory. Overall, the effects for
high quality store brands are by and large consistent with the predictions of our model.
It is equally striking that the results for the categories with low quality store brands do
not support our model, and in fact are in line with previous price/quality tier research.
Specifically, prices of both NB’s have a greater impact on SB demand than vice versa.
Moreover, the SB appears to compete more with NB2 than with NB1.
4.2.4. Discussion. Overall, Study 2 offers some limited support for our model. In
categories with higher quality store brands, it does appear that the SB and NB1 compete with
each other to a greater extent than they do with NB2. Such is not the case for the low quality
store brand categories. What might explain the difference? One possibility is that retailers
pursue different positioning strategies depending on the quality of the store brand that they can
procure. When they can buy a store brand that is comparable to national brand quality, they
follow the predictions of our model and position against the leading national brand. When
store brand quality cannot match that offered by the NB’s, the retailer treats the SB as an
inferior good and positions it against the weaker NB’s. Alternatively, let us assume that the
retailer follows the dictates of our model irrespective of store brand quality, always positioning
against the leading NB. The observed results for low quality categories could also arise if the
consumer simply does not accept the position that the retailer stakes out for their SB. In this
case consumers may readily perceive the retailer’s intent to position the SB against NB1 based
28
on extrinsic characteristics but still not accept that the SB offers a similar level of intrinsic
product quality. We address these issues in Study 3.
4.3. Study 3: Store Brand Positioning and Consumer Perceptions
4.3.1. Method. The task required consumers to judge the similarity between the top 3
national brands and the store brand in 8 different product categories. Respondents were 102
primary shoppers in households recruited through a local PTA. The categories were yogurt,
tomato sauce, toilet paper, canned peaches, canned tuna, chocolate syrup, peanut butter, and
skin lotion. The store brands came from two local supermarket chains; 90% of respondents
indicated that they had shopped at both stores within the past year. For each category
respondents saw color pictures of the brands lined up four across the page (see Figure 1) and
then made one of three similarity ratings for each of the {4 choose 2}=6 pairs: (1) “How similar
overall are the following pairs of brands?”; (2) “How similar are the following pairs of brands
in terms of product quality?”; and (3) “How similar are the following pairs of brands in terms
of physical appearance?”. Similarity rating were assessed on a 1=very similar to 7=very
dissimilar scale; type of rating task was manipulated between subjects.
There also were three within subjects manipulations. The first variable was whether or
not the SB targeted one specific NB. Using data from Study 1 we selected four categories
where the store brand clearly targeted one of the NB’s, in all cases here NB1; in the other four
categories, there was no clear target. The second variable was the location of the SB relative to
NB1 in the stimuli . Either the SB was put adjacent to NB1 (NB2, NB1, SB, NB3) or it was
separated (NB1, NB2, NB3, SB). Although this manipulation had absolutely no impact, we
thought a priori that adjacency might increase similarity. Finally, we manipulated the price
differential between NB1 and the SB; either the SB sold at a 15% or 30% discount to NB1.
29
Again, although this variable also had no impact on similarity ratings, we thought that
consumers would be more likely to believe that SB quality was comparable to that of NB1
when the price differential was smaller. To summarize, the overall design was an 3 rating task
x 6 brand pairs x 2 level of targeting x 2 level of location x 2 level of price mixed design where
the four categories for each level of targeting were rotated across the 2 levels of location and
price according to a Latin square. Rating task was a between subjects variable and level of
targeting, price differential, and location were within subjects. We also collected supplemental
information about store brand familiarity, usage and attitudes.
4.3.2. Results. Although the design is complicated, the results are robust, simple, and
therefore easy to interpret. The results are displayed graphically in Figure 2. The six pairwise
similarity judgments are broken down by type of rating task: overall similarity, product quality
similarity, and physical appearance similarity. For purposes of comparing individual means,
the critical range is 0.35. As mentioned previously, there are no significant main effects or
interactions effects due to the price or location manipulations. The key effect is a significant
interaction between rating task, brand pair, and the targeting variable, F10,4794=5.64, p<.0001
The pattern of similarity ratings differ systematically depending on the rating task and whether
SB targeting is explicit or ambiguous. Both the overall similarity and product quality similarity
ratings produce comparable results and do not differ systematically depending on the targeting
variable. Specifically, all of the NB’s are seen as fairly similar (mean of 3.0 for overall and 2.7
for quality similarity). In contrast, the SB is viewed as less similar to each of the NB’s
(mean=4.2), especially NB1 (mean=4.5). The results for overall and product quality similarity
do not differ much whether the SB explicitly targeted NB1 or targeting was ambiguous.
30
Figure 2: Similarity Ratings Depending on Explicit Targeting of NB1 from Study 3
It is only in the case of physical appearance that explicit targeting has any impact.
Specifically, when the SB purposefully targets NB1, consumers readily detect the similarity in
physical appearance (3.0 vs 5.1). In addition, targeting reduces perceived similarity between
the SB and both NB2 (4.9 vs 5.2) and NB3 (4.1 vs 4.7). Finally, because perception data are
influenced by relative context, targeting also reduces perceptions of similarity between NB1
NB NB NB SB3.7
4.0
4.2
1 2 3 3.3
3.5
2.8NB NB NB SB
4.
4.5
4.7
1 2 3 2.9
2.9
2.6
NB NB NB SB3.8
4.2
4.5
1 2 3 3.3
3.1
2.5NB NB NB SB
4.1
4.1
4.7
1 2 3 2.6
2.7
2.1
Explicit Targeting Ambiguous Targeting
NB NB NB SB4.1
4.9
3.0
1 2 3 3.7
4.2
4.3NB NB NB SB
4.7
5.2
5.1
1 2 3 3.5
3.9
3.1
Overall Similarity
Product Quality
Physical Appearance
31
and both NB2 (3.1 vs 4.3) and NB3 (3.7 vs 4.2). Table 5 shows a simplified view of the data
focusing on similarity ratings of SB and NB1 and the average of the similarity ratings of the SB
with NB2 and NB3. Again the 3-way interaction between similarity type, brand pair, and
targeting is significant, F2,1530=6.92, p<.001. Explicit targeting of NB1 by the SB has two
effects: respondents see the SB as more similar to NB1 and less similar to NB2 and NB3, but
only in the case of physical appearance. In contrast, explicit targeting does not produce the
intended effect on perception of overall similarity or product quality, as the SB is viewed as
slightly more similar to the secondary brands than to NB1. This is also the case when the
targeting is ambiguous.
Table 5: Similarity Ratings of the SB with NB1 and NB2/NB3
Explicit NB1 Targeting Ambiguous Targeting Type of Similarity NB1 vs SB NB2 & NB3 vs SB NB1 vs SB NB2 & NB3 vs SB
Overall 4.7 4.3 4.2 3.9 Product Quality 4.8 4.2 4.5 4.0
Physical Appearance 3.0 4.5 5.1 5.0 Note: low numbers indicate greater perceived similarity
4.3.4. Discussion. The overall picture that emerges from Study 3 is as follows. When
the retailer specifically targets the leading national brand, such attempts do succeed to the
extent that consumers easily perceive the proximal position of the SB relative to NB1 in
physical similarity space. Moreover, this positioning tactic also tends to distance both the SB
and NB1 from the other NB’s. However, it appears that consumers interpret this positioning in
a very literal and narrow manner. And so what does not happen is that there is any discernible
carry over from physical appearance space to perceptions of either product quality similarity or
overall similarity. Explicit targeting had no influence on consumers’ perceptions of the SB and
NB1 in terms of overall similarity or product quality similarity. In fact, the SB was rated as
32
more similar to the lower share NB2 and NB3.
5. Conclusions
Retailers are, or at least should be, interested in category profits rather than the profit
from any specific brand. In this paper, we examined how the retailer’s objective function
reveals itself in the optimal positioning strategy of the store brands. Our contribution is
twofold. On the theoretical side, we address the store brand positioning problem. On the
empirical side, we provide evidence that store brands in fact aspire to compete with the
category leader; although there is mixed evidence that they are successful in doing so.
We frame the retailer’s positioning decision as choosing the degree of competition
between the store brand(s) and each of the national brands in the product category. Assuming a
category with two national brands, we find that the store brand should be positioned closer to
the stronger national brand. Targeting is relatively more profitable in categories where the
leading national brand is stronger. We tested the implications of the analysis with three
empirical studies. We provide evidence that store brands indeed target the leading national
brand in the category (Study 1). We also analyze demand-price relationships in 19 categories
and find support for our conjecture that differential positioning leads to greater competition
between the store brand and the leading national brand but only for categories with high quality
SB alternatives (Study 2). We find that even though consumers can readily detect retailers’
efforts to use extrinsic cues to position against the leading national brand, this does not
necessarily translate into consumer perceptions that the SB offers comparable intrinsic quality
(Study 3).
Our analysis indicates that the retailer prefers to have a store brand which competes
heavily with the national brands. The basic premise here is that it can not increase both cross
33
price sensitivities δ1 and δ2 at the same time. The conceptual space in which the positioning
game takes place allows us to represent the tradeoff between δ1 and δ2. In the presence of this
tradeoff, it is better to have a high δ1 than a high δ2. Hence, there is a rationale for the tendency
of store brands to imitate the category leader.
There is empirical evidence that store brands do particularly well in categories with
high concentration (Dhar and Hoch, 1997). Rubel (1995) suggests that store brands do better
because it is easier for consumers to compare the store brand when there is a distinct category
leader. Dhar and Hoch (1997) argue that store brand can pursue a focused positioning strategy
in a concentrated market characterized by less heterogeneity in tastes, and offer an attractive
alternative with a lower price. Our analysis is in line with Dhar and Hoch (1997), and we claim
that the focus of that positioning strategy should be the leading national brand.
One can devise scenarios in which targeting the category leader may not be the optimal
strategy. For example, if the secondary national brand provides a much lower margin than the
leader, the retailer may be better of by diverting the sales of the secondary national brand to the
store brand. Alternatively, in some categories targeting strategy may lead to negative
inferences; consumers may prefer to buy the “real thing” rather than the “lower quality
copycat”. In this case the retailer may elect to make its brand as distinct as possible. It is also
possible that being closer to the customer may help the retailer identify the unfulfilled needs or
a niche market, thus leading to a differentiated product strategy.
34
References
Allenby, G.M., and Rossi, P.E. (1991), “Quality Perceptions and Asymmetric Switching
Between Brands,” Marketing Science, 10 (3), 185-204.
Blattberg, R.C., and Wisniewski, K.J., (1989), “Price-Induced Patterns of Competition,”
Marketing Science, 8 (4), 291-309.
Bolton, R.N., (1984), “The Robustness of Retail-Level Price Elasticity Estimates”, Journal of
Retailing, 65 (2), 193-219.
Brodie, R., and de Kluvyer, C.A., (1984), “Attraction versus Linear and Multiplicative Market
Share Models: An Empirical Evaluation”, Journal of Marketing Research, 21 (2), 194-
201.
Bronnenberg, B.J., and Wathieu, L., (1996), “Asymmetric Promotion Effects and Brand
Positioning”, Marketing Science, 15 (4), 379-394.
Corstjens, M., and Lal, R., (2000), “Building Store Loyalty Through Store Brands”, Journal of
Marketing Research, (37) 3, 281-291.
Dhar, S.K and S.J. Hoch (1997), “Why Store Brand Penetration Varies by Retailer,” Marketing
Science, 16 (3), 208-227.
Halstead, D., and Ward, C.B., (1995), “Assessing the Vulnerability of Private Label Brands,”
Journal of Product & Brand Management, (4) 3, 38-48.
Hauser, J.R., and Shugan, S.M., (1983), “Defensive Marketing Strategies”, Marketing Science,
(2) 4, 319-360.
Hoch, S.J., and Banerji, S., (1993), “When Do Private Labels Succeed?,” Sloan Management
Review, 34 (4), 57-67.
Hoch, S.J., and Lodish, L.M. (1998), “Store Brands and Category Management,” Working
Paper, The Wharton School, University of Pennsylvania, Philadelphia, PA.
Hotelling, H., (1929), “Stability in Competition”, The Economic Journal, 39, 41-57.
Lee, E. and Staelin R., (2000), “ A General Theory of Demand in a Multi-Product Multi-Outlet
Market, Marketing Science Conference, Syracuse.
35
Marketing Fact Book, (1998), Information Resources, Inc, Chicago, IL.
Moorthy, K.S., (1985), “Using Game Theory to Model Competition”, Journal Of Marketing
Research, (22) 3, 262-282.
Quelch, J.A., and Harding, D., (1996), “Brands versus Private Labels : Fighting to Win,”
Harvard Business Review, 74 (1), 99-109.
Raju, J.S., Sethuraman, R., and Dhar, S.K., (1995), “The Introduction and Performance of Store
Brands,” Management Science, 41 (6), 957-978.
Richardson, P.S., Dick, A.S., and Jain, A.K., (1994), “Extrinsic and Intrinsic Cue Effects on
Perceptions of Store Brand Quality,” Journal of Marketing, 58 (4), 28-36.
Rubel, C., (1995), “Price, Quality Important for Private Label Goods,” Marketing News, 29 (1),
24.
Schmalensee, R., (1978), “Entry Deterrence in the Ready-to-Eat Breakfast Cereal Industry”,
Bell Journal of Economics, 9, 305-327.
Shubik, M., and Levitan, R., (1980), Market Structure and Behavior, Harvard University Press,
Cambridge, Mass.
Sethuraman, R., Srinivasan, V., and Kim, D., (1999), “Asymmetric and Neighborhood Cross-
Price Effects: Some Empirical Generalizations”, Marketing Science, 18 (1), 23-41.
Tyagi, R.K., (1999), “A Characterization of Retailer Response to Manufacturer Trade Deals”,
Journal of Marketing Research, 36 (4), 510-516.
Tyagi, R.K., and Raju, J.S., (1998), “The Effect of Entrant Brand's Ownership on National
Brand Positioning Strategies”, Working Paper, Graduate School of Management,
University of California, Irvine.