Você está na página 1de 6

63

The Responsiveness of Food Sales to Shelf Space


Changes in Supermarkets
KEITH COX*
y Thh article tastt tha hypothetei that food product lalet are ratponsiva to changes in (half ipaca
and that "impulie" items ara relatively more rponsive than staples.

Many supermarket executives assume that sales of


most food products sold in supermarkets wiU be responsive to changes in shelf space. However, there
has been very little experimental work to test this implied hypothesis. Pessemier en^hasizes the importance
of measuring this relationship. "Ideally, if management
knew just how sensitive were the sales of each item to
the particular space allocated to it and just how large a
contribution each item produced, it would be possible
to make allocations which would retum the larg^t
profit" [12]
The objective of this paper is to measure this relationship between shelf space and product sales. One
problem with measuring the responsiveness of shelf
space to sales is the fact that this effect upon sales may
be minor, and therefore difficult to detect Also, variable
effects up(Hi sales besides test effects must be controlled
for the test results to have much validity.
An expoimental design will be used to test this relationship in the marketplace. The study will analyse
the responsiveness of product sales to shelf space from
the viewpoint of the supermarket, rather than the
manufacturer. Supermarket executives seek to maximize total sales within supermarkets, while manufacturers try to maximize sales of their brands. These two
goals are not always compatible, resulting at times in a
"battle of shelf space" between manufacturers and
retailos.
AVAILABLE TECHNIQUES
One analytical technique to use in this problem
would be to ask a sample of cusUnners what they would
btqr if the shelf space for that product was changed.
However, this method would only tdl us what customers say they would do, which may be different from
what t h ^ would actually do.
Anothn technique is the before-and-after study.
Some Steves are used as control stores, where the test
variaUe is not changed during two difiterent time
poiods. In oths states, the test variable (shelf space)
is changed from the first to die s^xmd period ci time.
* Keith Cox is assistant pnrfeaaor of taaAeOat, Kent State
UnivmUy.

Any differences between control stores in the two


time periods is subtracted from the differences between
test stores. After this adjustment, farther difference
between test stores are attributed to the test variable.
This method does not control differences between time
periods, so that the results are weakened when time
period variations are large. Also, the assumption that
variation between supermarkets is adequately controlled
by selection of the same type of supermarkets in both
the control and test groups is frequendy questionable.
From an operational viewpoint, this assumption may
be difficult to justify.
A third technique uses analysis of variance. According to Hoel [8]: "One of the most meful techniques for
increasing the sensitivity of an experiment is the desigOr
ing of the experiment in such a way that the total variation of the variable being studied can be separated into
components that are of experimental interest or importance. Splitting up the total variation in this manner
enables the experimenter to utilize statistical methods to
eliminate the effects of certain interfering variables and
thus to increase the sensitivity of his experiment. The
analysis of variance is a technique for carryii^ out the
analysis c^ an experiment designed from this point of
view."
After considerii^ these differoit techniques, the
analysis of variance was selected as the most efficient
for these e:q)eriments. A number of different experimental des^ns could be used utilizing analysis of variance. The simplest type of design is where all of the
test treatments are allocated entirely by chance within
the experiment This is called a completely randomized
desigtL Another design is randomized blocks, which is
a wig\e grouping design. One major source of variation
in addition to the treatment variation can be
M ^
from die residual error. (The residual error in analysis
of variance is diat part of the total variance that cannot
be statistically accounted for in the expenmsat. It is
considered as that part of the vari^ion due to random
flnctuadons, and is used as an unlMased estimate of the
universe variance.)
A latin square design is a doable groupiiig design,
wbere two major scmrces of variation can be controUed.
The latin square design proviiks more opportonity for

64

reducing the residual error than the randomized blocks


or completely randomized design, when there are at
least two major sources of variation in the experiment
not subject to direct control. In previous supermarket
tests [5, 10], variations between stores and variatirais
between time periods were indeed difficult to control.
Hence, the latin square design was selected for these
series of tests.
RESEARCH METHODOLOGY
In setting up the experimental tests to measure the
relationship between self space and product sales, all
other variations besides variation between treatments
(shelf spaces), variation between stores, and variation
between time periods needed to be controlled (kept constant). Prices of test products remained the same
throughout the testing period. The same shelf level in
each store at the start of the experiments was used during the tests. No sales promotional materials or advertising of the different test products were used. The
testing period comprised the summer of 1962, so that
any effect upon sales due to seasonal variation was
minimized.
Selection of Size of Latin Squares
Four factors were considered in choosing the size of
the latin square design. These factors were (1) precision desired from the test results, (2) total cost of
collecting the data, (3) time required to collect the
data, and (4) number of shelf treatments being tested.
It was decided that a 6 by 6 latin square design, using
six supermarkets over six weeks of time, was the most
desirable size for this study.
Selection of Test Products
Ideally, different food products should be tested in
measuring the relationship between shelf space and
sales. A number of limiting factors prevented many
products from being chosen for the tests. Products where
price specials were used had to be discarded, since any
variation in sales due to price changes could not be
controlled. Products such as coffee, sugar, com, ffour,
pickles, and peas feU into this category. Another group
of products, including frozen foods, milk, and meat,
were rejected because of the limited shelf space available for testing. Candy, cookies, and nuts were eliminated because of the difficulty of performing a physical
audit of the unit sales of these products. Salt was discarded because one of the proposed test supermarkets
had three times more shelf space for salt than any of
the other proposed supermarkets.
After such considerations, baking soda. Tang,
hoiAiny, and powdered coffee cream were selected as
test products. Baking soda was classified as a staple
product, while Tang, hominy, and powdered coffee
cream were classified as impulse products. The hypothesis was advanced that staple goods would be

JOURNAL OF MARKETING RESEARCH, MAY 1964

relatively unresponsive to changes in shelf spacs while


impulse goods would berelativelyresponsive.
Selection of Test Stores
The desirability of selecting the six supermarkets
randomly from all supermarkets in the area was c<isidered. The advantage of picking such a random
sample was the possible greater generality of the results.
The disadvantages were: (1) it would be very difficult
to obtain permission for the shelf tests with six different
supermarkets picked at random; (2) variations due to
price changes, sales promotions, and other extraneous
variations would be easier to control using six stores
from one supermarket chain in the experiment rather
than six different supermarkets selected at random; and
(3) selection of aU test stores from a supermarket
chain with a large volume of sales would probably
increase the power of the experimental results.
Because of the exploratory nature of the study, and
since full cooperation from the participating supermarkets was essential, it was decided to work intensively with one supermarket chain.
The district manager of the largest supermarket chain
in Austin, Texas, was approached about using six
supermarkets. Since two of the stores had crowded shelf
conditions which would make testing of shelf treatments
difficult, he suggested that these two stores be dropped
from the experiment. Accordingly, two additional supermarkets from a local chain were added. The study was
then conducted with four supermarkets from the regional chain and two supermarkets from the local chain.
Selection of Shelf Treatments
In selecting the shelf treatments (number of shelf
spaces), four different factors were considered. It was
desirable to space the shelf treatments at equal intervals,
thereby simplifying the statistical analysis. Second, to
make the tests more powerful, it was desirable to make
the differences between the shelf treatments as large as
possible. Third, differences between shelf treatments
were limited by the preferences of the supermarket managers. Fourth, the shelf treatment should not deviate
too greatly from the shelf spaces that the test products
occupied prior to the testing. Considering these factors,
the shelf treatments enumerated in Table 1 were
selected.
Table I
SHELF TREATMENTS FOR FOUR 6 BY 6
LATIN SQUARE DESIGNS
Product line
Baking soda
Hominy
Tang
Powdered coffee cream

SheHf raonfi enfs ( Number of sheH tpacm)

2
4
6
6

4
6
9
9

6
8
12
12

8
10
15
IS

10
12
18
18

12
14
21
21

THE RBPONSiyB4ESS OF FOCH> SALES TO SUPBtNiARKEt SHBf SPACE CHAN(^S

Inventory and Audit Records


In estimating weekty sales of each of the four product
lines in the six test supermarkets, a procedure was
originally considered of subtracting the ending shelf
inventory and the ending back-room inventory from the
sum of the beginning shelf inventory, the beginning
back-room inventory, and purchases during the week.
However, this mediod was not acceptable for two
reasons. A great amount of time and sldll were needed
to record accurately actual purchases during the week.
Also, the back-room inventory was almost impossible
to count in some of the supermarkets. Because inaccurate recording of either back-room inventory or
weekly purchases would distort the experimental results,
the following alternative system of counting inventory
was devised and used:
Beginning Shelf , Additions Ending Shelf _ Weekly
Inventory ~^ to Shelf
Inventory
Sales
An audit was taken four times every week for all
four products in all six of the test supermarkets. Besides
taking this physical inventory, additions of new stock
to the shelves were made whenever possible. Other
additions to the shelves by grocery clerks could generally be verified by the clerks. Additions to the shelves
were easy to count, since the stock was always added
in full-case lots of 12, 24, or 48 units at one time.
Testing Procedures
Using the staple and impulse assumptions concerning
shelf space and supermarket sales, the following hypotheses were tested:
1) There is no significant relationship between the
amount of shelf space given to baking soda and
total unit sales of baking soda.
2) There is a significant relationship between the
amount of shelf space given to hominy. Tang, and
powdered coffee cream, and total unit sales of each
of these products.

65

tween time periods, variaticm between treatments, and


tlw residual variation. In terms of the usual notation:
The total sum of squares (total SS) is equal to the row
sum of squares (row SS) plus the column sum of
squares (column ^ ) plus the treatment sum of squares
(treatment SS) plus the residual sum of squares (residual SS).

2(Yt - Y)* + 2 ( Y - Y, - Yk - Y;-f-2 Y ) '


where Y := experimental variable,
Y = overall mean,
Y, = row mean,
Yk = column mean,
Yt = treatment mean.
Each sum of squares, when divided by the appropriate degrees of freedom, can be used as an estimate
of the variance, the results being the so-called mean
squares.
In the analysis, both a regression analysis test and an
analysis of variance test can be used in measuring the
effect of the treatment shelf spaces upon product sales.
Regression analysis is appropriate in this study because
the different treatments are quantitative differences of
one factor (shelf space). The treatment means can be
tested for linearity by fitting a linear regression line to
the treatment means [11], though the linear relationship can be assumed only within the range of the actual
shelf treatments. The treatment sum of squares with five
degrees of freedom is subdivided into a linear regression
sum of squares with one degree of freedom and the
deviation from linearity sum of squares with four degrees of freedom. The null hypothesis should be rejected
when the F ratio is greater than 4.35 using a S percent
level of significance with 1 and 20 degrees of freedom,
where

In all latm square tests, the total variation is subdivided into the variation between stores, variation be-

Linear regression mean square


" Residual error mean square

Table 2
MEAN SQUARES AND F RATIOS FOR FOUR PRODUCTS TESTED

Source

freedom

5
5
5

Rows (stores)
Columns (weab)
Treatments (shelf spaces]
Linear regrsion
Deviation from linearity
Rasidual airar

20

Total

3E

1
4

Hominy

Baking toda

Tang

Powdered
coffee cream

Mean square F ratio

Mean square F ratio

Mean square F ratio

Mean square F ratio

8.891.2
690.1
1,426.2
4,230.7
725.1
335.7

25.00
1.94
4.01
11.89
2.04

I.657J
1963
79.0
113.2
705
74.4

22.28
2.64

\M
1.52

.95

1^38.4
63.2
32v4
145.3
4.1
49.6

24.97
1.27

.65
2.93

.08

1,295.2
105.9
101.7
159.7
87.2
63.9

20.27
1.66
1.59
2.50
1.36

JOURNAL OF AAARKETING REKARCH, MAY 1964

66
TEST RESULTS
The results oi the hominy shelf space experiment are
given in Table 2 and the data appendix. In analyzing
die results in the data appendix, the shelf space treatments are given in parenthesis for each cell. The other
number in each cell is the actual unit sales of hominy.
For example, there were 140 cans of hominy sold in
store 1 during the first test week with twelve shelf spaces
ased in the shelf display. In Table 2, the F ratio is seen
to be 11.89, which exceeds the 5 percent level of
significance of 4.35. The null hypothesis is therefore
rejected, and the alternative hypothesis accepted of
significant differences between the average treatment
sales of hominy.
The baking soda results are also given in Table 2 and
the data appendix. From the results in Table 2, the F
ratio is 1.52, which is less than the 5 percent level of
significance of 4.35, and the null hypothesis is accepted
of no significant difference between the average treatment sales of baking soda. Similarly, no significant
difference due to shelf space was obtained for Tang and
for powdered coffee cream.

DATA APPENDIX
PRIMARY RESULTS FROM 6 BY 6 LATIN SQUARE TESTS
HOMINY

Weeks
Stores

(12)
140
(14)
131
(8)
67
(4)
109
(i)
58
(10)
37

(10)
150
(&)
126
(14)
94
(12)
134
(8)
71
(4)
36

(4)
71
(8)
130
(A)
49
(10)
96
(14)
59
(12)
52

2
3
4
5
6

(8)
III
(12)
188
(4)
93
(6)
123
(10)
62
(14)
58

(6)
121
(4)
133
(10)
112
(14)
127
(12)
49
(8)
38

(14)
127
(10)
154
(12)
161
(8)
84
(4)
27
(6)
51

(8)
40
(10)
83
(12)
63
(4)
65
(2)
27
(6)
31

(4)
42
(8)
67
(2)
40
(10)
43
(6)
35
(12)
37

(2)
36
(12)
84
(10)
69
(6)
65
(8)
17
(4)
36

30
(18)
65
(15)
54
(12)
41
(21)
26
(9)
9

(9)
35
(15)
62
(21)
54
(6)
29
(12)
25
(18)
25

(15)
25
(12)
43
(6)
47
(18)
33
(9)
II
(21)
22

(21)
28
(9)
37
(6)
38
(15)
37
(18)
18
(12)
19

(9)
22
(6)
23
(12)
48
(18)
50
(21)
17
(15)
22

BAKING SODA
1
2
3

IMPLICATIONS
Hominy was the only test product for which the alternative hypothesis was accepted. The test results
therefore reject the original hypothesis that impulse
items respond more to variations in.shelf space than do
staples. Also, the assumption that sales of food products
will be responsive to changes in shelf space is open to
serious question. For many food products, increasing
the amount of shelf space may be an inefficient way of
increasing food product sales in supermarkets. From
the viewpoint of the retailer, shelf allocation decisions
may be infiuenced more by minimum restraints such as
(1) oat of stock policies, (2) full-case stocking to
minimiTft labor costs, and (3) assortment policies of the
retailer. One practical solution to stocking fast-moving
products may be to stock the products more frequentiy
within a snuller amount of shelf space, rather than
increasing the amount of shelf space for the product.
For manufacturers of food products, the results
suggest that the cost of gaining additional shelf space
f(H their product may not increase sales sufficiendy to
justify tte expense. Farther research in this area is
reccnnmended for both retailers and manufacturers.
Finally, this study shows how distorting variations
between test stor and between different time periods
can be statistically eliminated by asing latin square
deigns. This type of mearch design should be useful in
otha types of supomarket testing, such as for selecticm
of better point-of-puichase d i s ^ y s , making packaging
annparisons, testing the most profitable price deals, and
toting the effects d difiooit assortment policies.

4
5
6

(12)
22
(6)
65
(4)
58
(2)
43
(10)
40
(8)
38

(10)
30
(4)
61
(6)
42
(8)
41
(12)
26
(2)
23

(6)
36
(2)
74
(8)
73
(12)
47
(4)
33
(10)
42

TAN
1
2
3
4
5
6

(12)
25
(9)
59
(18)
36
(21)
39
(15)
23
(6)
22

(18)
38
(21)
48
(12)
48
()
19

(M

17
(15)
18

(21)
31
(6)
47
(9)
55
(15)
27
(18)
24
(12)
19

POWDERED COFFEE CREAM


1
2
3
4
5
6

(IS)
27
(21)
34
(9)
39
(12)
40
(15)
15
(6)
16

(15)
14
(18)
31
(21)
67
(i)
57
(12)
15
(9)
15

(12)
18
(15)
34
(18)
31
(9)
39
(6)
II
(21)
14

[t>)
35
(12)
46
(15)
49
(21)
70
(9)
9
(18)
12

THE REa>ONSIVENESS OF FOOD SALES TO SUPBtMARKET SHBF SPACE CHANGES


REFERENCES
1. Sejmuxir Banks, "Marketing Experiments," Joumal of Advertising Researeh, 3 (Maidi 1963), 34-41.
2. WiUiam Brown and W. T. Tucker, "Hie Marketing Center:
Vanishing Shelf Space," Atlanta Economic Review, 11
(October 1961), 9-13.
3. Max Brunk and Walter Federer, "Experimental Doigns
and Probability Sampling in Marketing Research," Journal of the AmeHcetn Statistical Association, 48 (SeptembCT
1953), 440-452.
4. William Cockran and Gertrude Cox, Experimental Designs,
New YOTk: John WUey & Sons, Inc, 1954.
5. Bennett Dominick, Jr., "An Illustration of the Use of the
Latin Square in Measuring the Effectiveness of Retail Merchandising Practices," Methods of Research in Marketing,
Paper #2, Cornell University, June 1952.
6. Churdull Eisenhart, "The Assumptions Underlying the
Analysis of Variance," Biometrika, 3 (March 1947), 1-21.
7. Rcmald Fisher, The Design of Experiments, 6th ed. revised,
: OUver & Boyd, 1951.

67

8. Paul Hoel, Introduction to Mathematical Statistics, New


York: John Wiley & Sons, Inc., 1954.
9. Jerome Li, Introduction to Statistical lnferetice, Ann Arbor: J. W. Edwaids, Publisher, Inc., 1957.
10. Murray MacOregor, "Uniformity Trial Experinmits in
Marketing Research," Methods af Research in Marketing,
Paper 6, Cornell University, September 1958.
11. Bernard Ostle, Statistics in Research, Ames: Iowa SUte
University Press, 1954.
12. Edgar Pessemier, "Applying Supermarket Techniques to
Non-Food Retailing," Joumal of Retailing, 36 (Summer
1960), 108-113.
13. James Shaffer, "The Influence of Impulse Buying' of InThe-Store Deciuons on Consumers' Food Purchases,"
Journal of Farm Economics, 42 (May 1960), in-3U.
14. U. S. Department of Agriculture, Better Utilization of Selling Space in Food Stores, Marketing Research Report #30,
November 1952.

R^rints of aD artides in this irane are arailaUe (as long as snpidy lasts) at the

Single reprint

$ 1.00

Two reprints

1.50

First 100

40.00

1.80

Additional lOO's

20.00

Three

reprints

Four to 99, each

.50

Covers are available on orders of 100 or more at a surcharge of 15 percent.


Special prices are available for large quantities.
Send your order to:

AMERICAN AAARKETING ASSOCIATION


230 North Michigan Avenue
Chicago, Illinois 60601
Duplication, reprinting, or republication of any portion of the JOURNAL OF
MARKETING RESEARCH is strictly prohibited unless the written consent of
the American Marketing Association is first obtained.

Você também pode gostar