The impact of location-specific factors on attractiveness and performance of product categories K.Campo 1 E.Gijsbrechts 2 T.Goossens 3 A.Verhetsel 4 1 Postdoctoral Fellow of the Fund for Scientific Research - Flanders; and UFSIA 2 Professor of Marketing at UFSIA and FUCAM 3 Research assistant at UFSIA 4 Professor of Economic Geography at UFSIA Acknowledgments The authors thank Mike Hanssens, Lee Cooper, and Randy Bucklin for their useful suggestions and comments.
40
Embed
The impact of location-specific factors on attractiveness ... market potential and buying power can be related to population characteristics like fam ily size, ... variables affecting
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The impact of location-specific factors on
attractiveness and performance of
product categories
K.Campo1
E.Gijsbrechts2
T.Goossens3 A.Verhetsel4
1 Postdoctoral Fellow of the Fund for Scientific Research - Flanders; and UFSIA
2 Professor of Marketing at UFSIA and FUCAM 3 Research assistant at UFSIA
4 Professor of Economic Geography at UFSIA Acknowledgments The authors thank Mike Hanssens, Lee Cooper, and Randy Bucklin for their useful suggestions and comments.
The impact of location-specific factors on
attractiveness and performance of
product categories
I. Introduction
The topic of geo-marketing has received increasing attention from academics as well as
practitioners. The use of geographic information in marketing has many potential applications,
and research in the area is rapidly evolving (see e.g. Longley and Clarke 1995). Interesting papers
by Hoch et al. (1995), Montgomery (1997), Kalyanam and Putler (1997) and Mulhern et al.
(1998) demonstrate that geo-demographic profiles of a store’s trading area may strongly affect
SKU-item market shares or sales, and open up opportunities for highly profitable store-specific
marketing actions. Geo-marketing has also found its way to practice. Examples of companies that
incorporate geographical technology in their marketing strategy are Levi Strauss and IKEA. Levi
Strauss starts from spatial data to determine the combination of products to be offered in
individual stores, while IKEA uses a geographical information system to optimise the distribution
of its catalogue (Kotler et al. 1996, Longley and Clarke 1996).
In this paper, we are interested in the impact of location characteristics on the attractiveness and
performance of different product categories. Today, many retail chains offer a broad assortment
of products at various locations. It is clear that the overall performance of each outlet depends on
location characteristics like the socio-economic profile of local inhabitants, and competitive
conditions in the area. Less obvious, however, is the potentially differential impact of these
1
location characteristics on the various product categories in the retailer’s assortment. Depending
on the type of location, some product departments or categories within the store may be more or
less attractive, and hence contribute more or less strongly to overall outlet performance.
Insights into the relationship between relative category attractiveness and location characteristics
may be crucial to chain managers for many reasons. First, location differences in category
attractiveness may provide a basis to develop more efficient assortment strategies. During the last
two decades, competition in the retail sector has grown substantially (see e.g. Corstjens and
Corstjens 1995). In order to maintain their competitive position, retailers need to manage costs
carefully, and in many cases are forced to reduce their assortments. Yet, the increasing demand
for variety by a heterogeneous and variety seeking customer base implies that assortment
reductions entail high risks of dissatisfying and losing customers (see e.g. Johnson 1997, Kahn
1998). Adjusting assortment composition to location characteristics could provide a means to cut
costs down while limiting the risk of negative customer reactions. Second, retail companies face
the problem of allocating scarce resources available at each outlet to the different categories.
Examples of such resources are floor space, promotional budgets, or local personnel. Efficient
allocation of these resources requires insight into category attractiveness. To the extent that
category appeal varies with the trading area’s geo-demographic profile, adjusting for these
differences may allow for more efficient allocation decisions. Moreover, the problem of location-
specific assortment and allocation strategies is not only relevant to retail chains. Service
companies (like financial institutions) and not-for-profit institutions (like hospitals or cultural
organisations) could also benefit from the insights of this type of analysis. Yet, to our knowledge,
little systematic research has been conducted on the impact of location characteristics on category
attractiveness. This study attempts to shed light on the issue.
2
This research has three major objectives. First, we want to investigate the impact of location-
specific factors on the relative attractiveness of various categories within the store. A different
way of putting this question is: do location factors significantly affect the ‘mix’ of a store’s
business. In a second stage, we study implications for overall store performance. More
specifically, we analyze whether differential effects of location features on performance of
various product categories cancel out, or produce a net overall effect on store returns or store
margins. Third, we examine whether location-specific differences in category attractiveness
provide a rationale for micro-marketing strategies, that is, marketing assortment strategies at the
store level as opposed to chain wide assortment strategies.
3
Previous research has paid some attention to micro-marketing implications of demographic and
competitive environmental store variables (Hoch et al. 1995, Montgomery 1997, Kalyanam and
Putler 1997, Mulhern et al. 1998). While these studies emphasize the importance of micro-
marketing, and provide interesting basic insights for our study, they differ from the present
analysis in several ways. The previous studies concentrate on price and promotion issues, and
systematically link price responsiveness observed in different stores to geo-demographic data.
Price sensitivity is assessed at the level of product items or SKU’s, for a selected set of items
offered by the store. Even though these studies measure price response in several categories (2
categories in Montgomery 1997, and Kalyanam and Putler 1997, 18 categories in Hoch et
al.1995, 4 categories in Mulhern et al. 1998), their focus is on price response differences between
stores. Our analysis takes a different perspective. We primarily seek to explain differences in
intrinsic attractiveness of product categories or departments, and to analyze all categories offered
by the store. Our ultimate interest is in implications of these differences in intrinsic category
appeal for decisions at the store level, like allocation of marketing efforts and resources to each of
the departments in which the store is active. Naturally, given this focus, our models should
account for the role that these categories play in producing store level results, by allowing for
flexible category interactions, and distinguishing between within store shifts and store expansion
effects. Besides this difference in scope and resulting model structure, our analysis is somewhat
different from previous research in the category and location factors that it includes. While price
and promotion do not come into play, space allocations across categories are included. Also, the
models incorporate additional competitive indicators, and account for the potential business from
people not living in the trading zone. This leads to additional insights on location effects and
micro-marketing opportunities.
The remainder of the discussion is organized as follows. Section II sheds light on the
methodology adopted to study location effects on the composition and overall level of store
results. Section III describes the data sets, variables and estimation results of category and store
performance models. Implications for micro-marketing strategies are discussed in section IV.
Section V, finally, provides conclusions and indicates areas for future research.
II. Methodology
As outlined in the introduction, we are interested in the impact of location factors on store
performance. Previous research has demonstrated that overall store sales increase with local
market potential and buying power, and decrease with level of local competition. In turn, local
market potential and buying power can be related to population characteristics like family size,
age distribution, income level, and ethnicity (see e.g. Johnson 1997). The effects mentioned so far
are typically studied and obtained at the store level, based on associations between overall store
performance measures on the one hand, and characteristics of the trading area on the other hand.
4
As such, they reflect an ‘average’ influence of location characteristics for the different store
activities, or a location impact ‘common’ to the store’s different sources of business or categories.
We refer to these effects as ‘direct’ effects of location factors on store performance: they are
represented by the bold arrows in figure 1.
In addition to these direct effects, location factors may affect store results indirectly through their
impact on the relative attractiveness of various categories in the store. Differently stated, location
characteristics may determine the composition of store sales. This effect is represented by the
dotted arrow in figure 1, and is referred to as the ‘indirect’ impact of location characteristics on
store sales. Even if overall store sales remain unchanged, the share of return captured by various
categories may have a significant influence on overall store margins, given that different margins
apply to different categories within the store. In this paper, we intend to shed more light on (i) the
differential effects of location factors on category performance, and (ii) on how these effects
indirectly influence store sales and margins. In turn, we discuss the category performance model,
the store performance model and the variables affecting category and store performance.
Store Sales
return share 1
return share n
Location Factors
Store characteristics
surfaceshare 1
other var cat 1
surfaceshare n
other var cat n
Direct
(Return) ... ...IndirectFigure 1: Direct and indirect effects of location factors on store performance
5
A. The category performance model
To assess the differential effect of location characteristics on various product categories offered
by the store, we estimate a model linking the relative attractiveness of a category in a store, to
factors characterising that store’s location. Relative category attractiveness is measured as the
category’s share in total outlet return. By concentrating on the share that each category represents
in the outlet’s total sales value, we are able to detect any differential effects of location features
on intrinsic category attractiveness1. If location variables have no significant impact on
categories’ ‘share of return’, this indicates that all categories undergo a similar influence of these
location variables, which can be completely captured at the level of the outlet as a whole. If
location factors do affect category share of return, this points to differential effects of location
features on the attractiveness of various product categories. As stated before, these location
effects may prove crucial for efficient resource allocation across categories at the store level, as a
function of the store’s location type.
In addition to location factors, a category’s share of the store’s return may depend on category-
specific variables and store characteristics (e.g. marketing variables). These factors constitute the
explanatory variables in the return share models.
6
To model category share of return, we use an asymmetric MCI attraction structure suggested by
Carpenter, Cooper, Hanssens and Midgley (1988). Besides being robust, this model allows for
flexible interrelationships between categories. Category sales within a store are not mutually
independent, but may exhibit important complementary and substitution relationships, which
cannot be captured by symmetric attraction models. Ignoring category interrelationships may lead
to biased estimates. Using a flexible model like the one developed by Carpenter et al. (1988),
allows to avoid these biases. In our application, Carpenter et al.’s model takes the following form:
P = Cross
L * S * P * = Att
Cross * Att
Cross * Att = SR
rj,ir
mj,i,
tj,t
sj,s
rj,i,r
i0,ji,
mj,c,Cm
jc,c
mj,i,Cm
ji,
ji,
rm,i,
3,i,ts2,i,r1,i,
c
i
,
)1(
γ
βββ
ε
ε
β
Π
ΠΠΠ
ΠΣΠ
where :
SRij = share of return of product category i in store j Attij = intrinsic attractiveness of category i in store j Crossi,j,m = cross-category effect of category m on category i in store j Pi,j,r = value of product category characteristic r for category i in store j Sj,s = value of store characteristic s for store j Lj,t = value of location characteristic t for store j
Ci = set of product categories with potential asymmetric cross-category effects on the return share of category i
∃0,i , β1,ι,r ,β2,i s , β3,i t ,γi,m,r = parameters
B. The (overall) store performance model
The impact of location characteristics on overall store sales is assessed by means of a
multiplicative model with overall store return as the dependent variable, and store characteristics
as well as location characteristics as explanatory variables. The coefficients of these latter
variables capture what we referred to as the direct location effects. In addition, we include ‘total
store attraction’ as an explanatory variable. Using a procedure similar to Mason (1990) and
Bultez et al. (1995), we use the denominator of the asymmetric MCI model as a measure of total
store attraction. This measure accounts for the effect of location factors on the attractiveness of
product categories within the store, and therefore captures any indirect location effects on store
sales that might exist. The store return model then takes the following form:
7)2()()()( Totatt* L * S * = R jtj,
tsj,
s0j
2,ts1, µδδδ ∏∏
where Rj = total returns in store j Totattj = total attraction of store j (denominator of the attraction model (1)) δ0, δ1,s, δ2,t, µ = parameters
From the foregoing discussion, it is clear that our perspective and modeling approach is different
from that in previous micro-marketing studies. Instead of considering a subset of store categories,
and modeling sales or choice processes for specific items in these categories separately, we
essentially start from the perspective of the store as a whole. Category performance is assumed to
result from (i) the store’s overall capability of attracting business, and (ii) the category’s ability of
securing a portion of this business, taking its interactions with other categories into account. Note
that taken together, models [1] and [2] are a generalization of many category return models used
elsewhere. For instance, when µ equals 1, the product of [1] and [2] reduces to a double
logarithmic category return model similar to the one used by Hoch et al. (1995) and Montgomery
(1997), with possible cross-category effects for all category pairs.
C. Variables affecting category and store performance
As indicated in figure 1, three types of variables determine store performance and category return
share : overall store characteristics, location-specific variables, and variables related to separate
categories within each store. We discuss each of these variable types in turn.
Store Characteristics
A variety of store characteristics might affect performance at the store and category level. Store
chain image can influence overall store return as well as performance of specific categories
(Desmet and Renaudin 1998). In a similar vein, the store’s retail format and strategy (assortment
depth, quality level, service, price and promotion policy) is sure to affect store and category
8
performance (Bell and Lattin 1998, Desmet and Renaudin 1998, Sirohi et al. 1998, Dhar and
Hoch 1997). Specific outlet characteristics like store size influence performance in several
respects. Store size may be positively related to assortment width and depth, service level,
convenience, and lower likelihood of stockouts, and hence positively contribute to overall store
performance (Sinigaglia 1997). In addition, store size may differentially affect the performance in
various departments. Following Desmet and Renaudin (1998), large stores attract a higher
percentage of customers with lower store loyalty, higher assortment sensitivity and more impulse
buying – these customers are likely to have different basket compositions and across-category
spendings.
Location Characteristics
Three types of location features can be distinguished: variables related to the competition, socio-
economic characteristics of the trading zone, and urbanization variables.
Supermarket competition in the trading zone has been identified as a potential determinant of
store performance in several studies, with conflicting expectations and results. The presence of
more or larger retailers in the trading area may increase competitive pressure and exert a negative
impact on store performance (Dhar and Hoch 1997, Hoch et al. 1995). At the same time,
supermarket presence may point to high economic potential and buying power in the trading zone
(Ingene 1984), and hence be positively related to store performance. The impact of supermarket
competition may also vary by category. For example, consumers are more likely to ‘shop around’
in different supermarkets for beauty care items than for basic groceries like rice or salt. Also, the
chain’s competitive strength (quality/price ratio) compared to other supermarkets may strongly
vary by category (Dhar and Hoch 1997).
9
The impact of socio-economic variables on retail performance is widely discussed and
documented. Previous studies point to significant effects of variables like age (Montgomery
1997, Dhar and Hoch 1997, Johnson 1997, Webster 1965), education (Bawa and Shoemaker
1987, Narasimhan 1984), employment status (Bawa and Shoemaker 1987, Kim and Park 1997),
ethnicity (Dhar and Hoch 1997, Johnson 1997, Hoch et al. 1995, Kim and Park 1997), income
(Bawa and Shoemaker 1987, Chiang 1995, Kim and Park 1997, Zeithaml 1985, Johnson 1997),
family composition and household size (Kim and Park 1997, Bawa and Shoemaker 1987, Gupta
1988, Chiang 1995). These socio-economic variables influence store and category performance in
three main ways. First, they constitute indicators of buying power, and hence affect overall
spending levels as well as the allocation of resources over more versus less income-elastic
product categories. Second, they influence the pattern of needs of the population in the trading
zone. Obvious examples are a higher demand for children clothes in large family areas, or a
lower meat consumption among certain ethnic groups. Third, they clearly affect consumers
shopping behavior and store selection. Through their influence on price sensitivity, mobility and
time cost; socio-economic variables exert a major influence on the type of supermarket most
likely visited (e.g. EDLP vs. Hi-Lo: see Bell and Lattin 1998, Kim and Park 1997). Whether
consumers engage in one stop shopping, or tend to buy certain items/categories in specialty stores
rather than supermarkets, also depends on their socio-economic profiles (Kim and Park 1997,
Dellaert et al. 1998).
10
Degree of urbanization of the trading area may influence performance at the store and category
level in two main ways. First, consumer lifestyles - and hence category consumption patterns –
are expected to be different in urban versus rural areas. For example, the demand for outdoor
leisure products or garden equipment will be higher in rural environments. Degree of
urbanization is also linked to consumer mobility and distribution density, variables found to
clearly affect shopping behavior (Ingene 1984, Hoch et al. 1995). So far, few studies have
systematically accounted for the potential impact of ‘degree of urbanization’ on store or category
performance (Walters and Bommer 1996, Bearden et al. 1978).
Local versus non-local sources of business : An important issue is that the retail store’s clientele,
though mainly recruited among local inhabitants, does not exclusively consist of people living in
the trading zone. Fragmented surveys suggest that, at least in some locations, people working but
not living in the trading area may constitute a non-negligible source of business. There are two
reasons to distinguish between ‘local’ versus ‘non-local’ sources of business. First, as the group
of workers-commuters in a trading area becomes more important, the socio-demographic profile
of local inhabitants should receive less weight in the model. In addition, apart from having a
possibly different socio-demographic profile, workers-commuters may have different buying
patterns in the supermarket than local shoppers. As they typically shop during lunch hours or
after work, they are more likely to engage in minor or fill-in shopping trips; and/or experience
more severe time constraints. To our knowledge, the distinction between local and non-local
business- though potentially important - has not been taken up in previous research.
Category variables
Category performance is expected to vary from location to location as a result of differences in
category-specific competition. The presence of more specialty stores in the category’s line of
business might act as an ‘attraction’ mechanism, and stimulate category shoppers to purchase in
the trading area (Ingene 1984). At the same time, more category stores will increase category
competition and put pressure on sales. From the supermarket’s perspective, the latter negative
effect is expected to prevail: category return in the outlet is likely to decrease with the number of
specialty stores in the trading zone.
11
A considerable number of studies document a positive impact of space allocated to a category on
category performance (Corstjens and Doyle 1981, Bultez and Naert 1988, Bultez et al. 1989,
Desmet and Renaudin 1998). This positive effect can result from increased visibility (Desmet and
Renaudin 1998), or follow from the fact that more stocking space on the shelves reduces
stockouts (Borin et al. 1994). Visibility depends on absolute space allocated to the category. To
the extent that the category competes with other categories for customer attention, visibility also
depends on the category’s share in the store’s total sales. Similarly, reduced likelihood of
stockouts in a given store is linked to absolute stocking space. Yet, to the extent that larger stores
have higher overall demand, and hence, higher rotation rates, stockouts for a given absolute space
vary with store size. It follows that, if space share is included as an explanatory factor of
performance – as is done in most studies on shelf impact- overall store size should also be
included to capture the absolute space effect.
III. Empirical Application
In this section, we provide an empirical analysis of location differences in relative category
attractiveness and their impact on overall store performance. We start with a description of the
data set and variables, and then present estimation results of the category and store performance
models developed in the previous section. Micro-marketing implications are studied in section
IV.
A. Data and Measures
Our data set covers information on 55 supermarkets of a European retail chain. For each of these
outlets, the retail chain has delineated the trading area based on extensive surveys. Data provided
by the retail chain on store characteristics and performance is supplemented with information
from two other sources. Information on competition in the store’s trading zone is constructed
12
from the yellow pages’ electronic database. Socio-demographic characteristics of inhabitants
living in each store’s trading area are derived from census data. As this information is collected
once every 10 years, we used data of one year only and performed a cross-sectional analysis of
the relationship between location characteristics and category and store performance.
Store Characteristics
Overall store performance is measured by the outlet’s yearly return (R). The outlet’s total sales
surface is used as a measure of store size (STSIZE). Store chain image and retail format are not
included in the analysis: as all the stores in our sample belong to one chain and format, these
variables show no variation.
Location Characteristics
Intensity of supermarket competition is captured by STCOMP, defined as the store's own sales
surface relative to that of competing supermarkets in the trading zone. Thirty indicators are
available on socio-economic characteristics of the trading area. These indicators relate to age,
sex, income, family size and structure, employment, nationality, type of habitat, mobility,
education levels, and spending patterns of the households living in the trading area. As these
socio-demographic variables are highly correlated and cannot possibly be incorporated
simultaneously into the models, we performed a principal component analysis on these data.
Results of the analysis are reported in appendix B. Four principal components are retained from
the 30 original variables, accounting for about 84 % of the variation. The first factor (FACTOR1)
represents the presence of young households with two or more children. Factor two (FACTOR2)
loads heavily on single households, and low income families. Factor three (FACTOR3) has a
high score in locations dominated by >middle class’ families, where middle class refers to
13
intermediate income levels and social status. Factor 4 (FACTOR4), finally, is high in regions
where single child households are predominantly present.
While our census data do not allow to assess the socio-demographic profile of people working
but not living in the trading area, we do have indications on the number of persons who work in
each trading area without being a local inhabitant (EMPLOY). We will use this number as an
indicator of the potential source of business coming from outside the trading zone.
Our measure for the degree of urbanization is derived from the typology of Van Hecke and
Cardyn (1995), which classifies regions into several groups with different degree of urbanization,
based on their service functions and density. For reasons of parsimony, the original classification
was reduced to a 2-level classification2. Degree of urbanization is thus incorporated in the models
as a dummy variable, which equals 1 when the store is located in a highly urbanized city, and 0
otherwise.
Taken together, the ‘location’ variables used to characterize the trading zones (i.e. the socio-
economic variables, and ‘degree of urbanization’) are comparable to those used in the
development of neighbourhood classifications like ACORN or MOSAIC (see, e.g., Johnson
(1997) for a basic description and additional references).
Category Variables
For each store, information is available on 17 product categories. The second column of table 1
describes the composition of these product categories, as defined by the chain’s managers.
14
The first 7 product categories are situated in the food, the other 10 in the non-food sector. For
purposes of interpretation, the categories defined by the chain are linked to the classification of
store assortments suggested by Van der Ster and Van Wissen (1993). The latter distinguish 4
types of product categories: the core assortment, image-enhancing assortment, varying
assortment, and profit-increasing assortment. Results of a correspondence analysis mapping
product categories and stores together largely confirms this classification (see appendix C). The
core assortment (cat. 1, 2 and 4) is clustered in the centre, and more to the right we find a group
of loyalty improving product categories (which are part of the image-enhancing assortment
according to Van der Ster and Van Wissen, 1993; cat. 3, 5, 6 and 7). For the varying assortment, a
distinction can be made between clothing categories grouped together in the upper left corner
(cat. 9, 10, 11 and 12), and luxury products situated somewhat below this group (cat. 14, 15, 16
and 17). The category ‘Health and Beauty Care’ (cat. 8) is positioned in between the core
assortment and the luxury assortment, which can be explained by the fact that it contains a
mixture of products. Health and Beauty Care includes core products such as toothpaste, as well as
luxury products such as higher-priced cosmetics and perfume.
15
Category performance is measured by its share in the store's total return (SR; based on yearly
figures to avoid seasonality influences). Relative shelf space allocated to the category is measured
by its share in overall store sales surface (SS). Discussions with several chain managers reveal
that the allocation of store space to categories is decided upon when the store is set up, and is not
systematically adapted over time. In addition, space allocation is said to be rather ad hoc, and not
formally linked to characteristics of the location. This anecdotal information is confirmed by
preliminary regressions which reveal no significant links between category space share (as the
dependent variable) and location features (as explanatory variables). At the same time, variation
in space shares for given categories across stores is found to be large. These insights are
important in the estimation stage: they suggest that while space shares are potential explanations
for return share, we need not worry about reversed causal effects.
Table 1: Product category classification
Cat. N° Category
Type
1 Groceries
Core assortment
2 M eat
C ore assortment
3 F ruit & Vegetables L oyalty improving assortment 4 D airy & Fine Meat - self service C ore assortment 5 D airy & Fine Meat - counter L oyalty improving assortment 6 F ish L oyalty improving assortment 7 B ake-Off L oyalty improving assortment 8 H ealth & Beauty Care B etween Core – Luxury assortment 9 M en's Wear & Underwear C lothing 10 L ady's Wear C lothing 11 C hildren's Wear C lothing 12 S hoes C lothing 13 A udio, video, micro-electronics L uxury products 14 H ousehold L uxury products 15 F abrics (Interior) L uxury products 16 L eisure (outdoor) L uxury products 17 Leisure (indoor) Luxury products
Category-specific competition in the store’s trading zone (CCOMP) is derived from the Yellow
Pages data base. Per product category offered in the store, a category-specific competition index
is constructed measuring the number of stores in the trading area in direct competition with the
considered product category (e.g. the competition variable for the fish category is equal to the
number of fish stores located in the trading area, relative to the number of inhabitants).
B. Differential effects of location characteristics on category performance
16
Explanatory variables in the asymmetric attraction model are indicated in the left column of
tables 2 and 3: they cover store size, competition at the store level, the four principal components
reflecting socio-demographic profiles of inhabitants of the trading area, degree of urbanisation,
number of people employed but not living in the area, level of category-specific competition in
the area and the share of surface of the product category.
Estimation of the asymmetric attraction model is carried out in three steps (see Carpenter et al.
1988). First, a symmetric model version is estimated, containing - besides store and location
characteristics - the category's own surface share as an explanatory variable for its attraction.
Since store and location features are not category-specific, we can only assess their differential
impact on category attraction in comparison with a reference category (see e.g. Maddala 1990).
We select category 1 (Groceries) as the reference category, a choice inspired by the fact that in
the correspondence analysis, category 1 was positioned near the origin representing the average
return share profile. Our choice of groceries as the reference category leads to the following
interpretation for the store and location feature coefficients: a positive (negative) coefficient for a
given variable and category means that the variable affects this category's share of return more
(less) strongly than the share of return of groceries. Share of surface and category-level
competition are category-specific measures, the impact of which can be estimated in an absolute
way rather than compared with groceries.
Having estimated the symmetric model, a next step is to look for significant cross-category
effects. Using an approach similar to Carpenter et al. (1988), we investigate to what extent the
share of return residuals of the symmetric model can be explained by other categories' share of
surface variables. After identification of significant cross-effects, the third step involves
estimation of an asymmetric model, including those selected cross-category effects. The results of
this final estimation step are included in tables 2 and 3.
As can be seen from these tables, the model exhibits significant explanatory power (R5 values
17
between 0.24 and 0.80), predominantly positive own surface share coefficients, and positive as
well as negative asymmetric surface share interdependencies. Also, the tables point to differential
influences of store and location features on the return share of various categories. To get a better
grasp on the importance and nature of these location influences on categories’ share of return, we
determined the improvement in fit compared to a model without location variables, and computed
return share elasticities3 for various categories and location variables. Table 4 compares R²adj
values for various share of return models. While allowing for category-specific (differential)
space effects and asymmetric interactions greatly improves fit over a base model not accounting
for these effects, including store and location variables leads to a substantial additional
improvement. On average, the adjusted R² increases from .27 to .51.
% of single households % of 2 person households % of 3 person households % of households with 4 or more persons % of population under 14 years of age % of population between 15 and 64 years of age % of population over 64 years of age % employed % unemployed % in military service % of population under 18 years of age % of students % of population who has never had a profession % retired % of population with no profession % unmarried % married % widowed % divorced % of households without children % of households with 1 child % of households with 2 children % of households with more than 2 children % of European Mediterranean nationality % of Islamitic origin % with income between 2500-6250 Euroa % with income between 6250-12500 Euroa % with income between 12500-17500 Euroa
% with income between 17500-25000 Euroa % with income over 25000 Euroa
-.698 -.151 .533 .853 .788 .702
-.936 .146 .114 .326 .816 .235 .359
-.893 .098 .351 .414
-.925 -.853 -.843 -.378 .922 .501 .016 .103
-.113 -.455 .102 .237 .128
.608 -.858 -.518 -.271 .167
-.261 .011
-.759 .525
-.657 .146 .368 .028
-.388 .467 .872
-.878 .090 .431
-.438 .186
-.157 .591 .584 .675 .881 .711 .026
-.346 -.438
-.077 -.005 .112 .038 .266
-.018 -.186 -.508 .761
-.319 .244
-.801 .553
-.004 .732
-.133 .095 .046
-.024 -.112 .303
-.197 .127 .449 .379
-.006 .415 .951 .754
-.845
.101 -.105 .194
-.056 -.456 .264 .200 .006 .201
-.088 -.412 .144 .095 .052 .073
-.170 -.036 .280 .155 .029 .718 .096
-.597 .329
-.496 -.109 -.049 -.121 .132 .077
Cumulative % of variance explained 34.12 63.83 78.26 83.61 a Household income indicated in tax declarations
variable indicates the percentage of the population belonging to the corresponding income
class. A description of the other variables can be found in the second column of table 7.
In order to reduce these variables to a manageable set, we performed a principal components
analysis. Based on a scree-test, a solution with 4 factors was selected, explaining 84% of total
variation. Factor loadings – obtained after a varimax rotation – are displayed in column 3 to 6
of table 7.
33
Appendix C
Correspondence analysis of product categories and outlets To facilitate interpretation of the results, we sought for a re-grouping of the 17 product
categories into larger classes with similar characteristics. In view of the research objective,
product categories belonging to the same class should exhibit a similar variation in relative
attractiveness over various outlets of the supermarket chain. To derive clear management
implications from the results, categories in the same product class should also serve similar
functions within the retail store. The classification of Van der Ster and Van Wissen (1993) is
based on the second criterium, and distinguishes between 4 broad product classes: the core
assortment consisting of the product categories that are essential to the store’s business, the
image-enhancing assortment which is made up of product categories that may improve the
store’s image (such as service goods, loyalty-improving products, high-quality products, and
innovative products), the varying assortment comprising temporary product offerings (such as
seasonal products), and the profit-increasing assortment consisting of products with a higher
than average profit margin to compensate for the low margins on core and other products.
To check whether a re-grouping of the 17 product categories along these principles yields
product classes with similar attractiveness profile, we performed a correspondence analysis
mapping product categories and stores together, based on the stores’ distribution of sales over
the 17 categories. The two-dimensional configuration is displayed in figure 6. As this figure
demonstrates and was discussed above, the results largely confirm Van der Ster and Van
Wissen’s classification. The major exceptions are that (1) there are two different types of
varying product classes (clothing and ‘luxury’ products), and (2) one product category is
situated in between two category clusters (Health and Beauty Care).
34
-1.5 -1 -0.5 0 0.5 1 1.5-1
-0.5
0
0.5
1
1
23
4
5
6
7
8
9
10
11
12
13
14
1516
17
X-Axis
Y-A
xis
Figure 6: Correspondence analysis of product categories by stores
35
References
Bawa, K. and R.W. Shoemaker (1987), “The Coupon-Prone Consumer: Some Findings Based on Purchase
Behavior Across Product Classes”, Journal of Marketing, 51(4), 99-110.
Bell, D.R. and J.M. Lattin (1998), “Shopping Behavior and Consumer Preference for Store Price Format: Why
Kim, B. and K. Park (1997), “Studying Patterns of Consumer’s Grocery Shopping Trip”, Journal of Retailing,
73(4), 501-517.
Kotler P., G. Armstrong, J. Saunders and V. Wong (1996), Principles of Marketing : The European Edition,
Prentice Hall, London.
Longley, P. and G. Clarke (eds.) (1995), “GIS for Business and Service Planning”, Bell and Bain, Glasgow.
Longley, P. and G. Clarke (eds.) (1996), “Spatial Analysis: Modelling in a GIS Environment”, Bell and Bain,
Glasgow.
Maddala, G. (1990), “Limited Dependent and Qualitative Variables in Econometrics”, Cambridge: Cambridge
University Press.
Mason, C. (1990), “New Product Entries and Product Class Demand”, Marketing Science, 9(4), 158-173.
McDonald, M. and I. Dunbar (1995), Market segmentation: a step-by-step approach to creating profitable market
segments, MacMillan : Houndmills.
Messinger, P.R. and Ch.Narasimhan (1998), “A Model of Retail Formats Based on Consumers’ Economizing on
Shopping Time”, Marketing Science, 16(1), 1-23.
37
Montgomery, A. (1997), ΑCreating Micro-Marketing Pricing Strategies Using Supermarket Scanner Data≅,
Marketing Science, 16(4), 315-337.
Mulhern, F.J., J.D.Williams and R.P.Leone (1998), “Variability of Brand Price Elasticities across Retail Stores:
Ethnic, Income and Brand Determinants”, Journal of Retailing, 74(3), 427-446.
Narasimhan, C. (1984), “A Price Discrimination Theory of Coupons”, Marketing Science, 3(2), 128-147.
Narasimhan, C., Neslin S.A. and S.K. Sen (1996), “Promotional Elasticities and Category Characteristics”,
Journal of Marketing, 60(2), 17-30.
Sinigaglia, N. (1997), Measuring Retail Units efficiency : a Technical approach, Doctoral Dissertation CREER,
FUCAM, MONS.
Sirohi, N., E. McLaughlin and D. Wittink (1998), “A Model of consumer Perceptions and Store Loyalty
Intentions for a Supermarket Retailer”, Journal of Retailing, 74(2), 223-245.
Van Hecke E. and C. Cardyn (1995), ΑGemeentelijke hiërarchie op basis van de verzorgende functies en de
intensiteit van de relaties≅ in Stativaria 10, Vlaamse Steden en Gemeenten, Wegwijzer in bestaande
typologieën, adminstrastie Planning en Statistiek, 41-44.
Van der Ster W. and P. Van Wissen (1993), “Marketing en Detailhandel”, Groningen: Wolters-Noordhoff,.
Walters, R.G. and W. Bommer (1996), “Measuring the Impact of Product and Promotion-Related Factors on
Product Category Price Elasticities”, Journal of Business Research, 36, 203-216.
Webster, F.E. jr. (1965), “The “Deal-Prone” Consumer”, Journal of Marketing Research, 2(2), 186-189.
Zeithaml, V.A. (1985), “The New Demographics and Market Fragmentation”, Journal of Marketing, 49(3), 64-
75.
38
39
Footnotes
1 Price differences between product categories should have no impact on the results, as we concentrate on variation in return shares over stores (and not over product categories), and uniform pricing strategies are used across stores. 2 Preliminary analysis indicated that a higher-level classification does not substantially improve model results. 3 For the dummy variable >degree of urbanization=, the strength of effect was measured as the percentage difference in share of return observed for the product category in urban (dummy 1) vs rural (dummy 0) areas. In addition, elasticities of socio-demographic variables cannot be evaluated at the mean value, since each factor has zero mean. For this reason, we computed the % change in return share, resulting from a marginal change in the factor value.
4 As an additional check on the stability of the results, we re-estimated the model using a new set of data on store characteristics and category variables (that is, for a subsequent year), together with our base information on location characteristics (which is not available year by year, and assumed stable over time). The re-estimated model exhibited similar location effects, except for the clothing category where deviations were observed. A discussion session with the chain’s managers revealed that these changes could be attributed to a (temporary) turnaround in the management team for clothing. Overall, the check confirmed our findings for all other categories.