Posted: December 17th, 2014

Quantitative Marketing and Economics, 1, 93–110, 2003.# 2003 Kluwer Academic Publishers. Manufactured in The Netherlands.Effects of $9 Price Endings on Retail Sales:

Quantitative Marketing and Economics, 1, 93–110, 2003.# 2003 Kluwer Academic Publishers. Manufactured in The Netherlands.Effects of $9 Price Endings on Retail Sales:

Evidence from Field Experiments
ERIC T. ANDERSON*
University of Chicago, Graduate School of Business, 1101 E. 58th Street, Chicago, IL 60637, USA
E-mail: [email protected]
DUNCAN I. SIMESTER
Sloan School of Management, MIT, 38 Memorial Drive, E56-305, Cambridge, MA 02142
E-mail: [email protected]
Abstract. Although the use of $9 price endings is widespread amongst US retailers there is little evidence
of their effectiveness. In this paper, we present a series of three ?eld-studies in which price endings were
experimentally manipulated. The data yield two conclusions. First, use of a $9 price ending increased
demand in all three experiments. Second, the increase in demand was stronger for new items than for items
that the retailer had sold in previous years. There is also some evidence that $9 price endings are less
effective when retailers use ‘‘Sale’’ cues. Together, these results suggest that $9-endings may be more
effective when customers have limited information, which may in turn help to explain why retailers do not
use $9 price endings on every item.
Key words. price ending, odd-pricing, pricing, catalogs
JEL Classi?cation:
C81, C93, D12, D8, L11, L15, M3
1. Introduction
Published studies report that 30% to 65% of all prices end in the digit 9 (Stiving and
Winer, 1997; Schindler and Kirby, 1997; Daily Mail, 2000). Despite interest from
economists reaching back over 65 years (Ginzberg, 1936) and the current widespread
use of this practice amongst retailers, empirical evidence that price endings affect
demand is limited. In this paper, we present a series of three ?eld experiments in
which we vary the price endings of many products. The three studies were conducted
with two different national mail-order companies that sell moderately priced
women’s clothing.
The study was motivated in part by a pilot study in which one of the catalogs
agreed to mail three versions of a catalog to separate randomly selected customer
samples. The prices of four dresses were manipulated across catalog versions. The
current policy of the catalog was to use a $9 ending and we refer to this as the
Control version. In the two treatments, the price was either raised or lowered by $5,
*Corresponding author.
94
ANDERSON AND SIMESTER
which removed the $9 ending from each dress. The prices and demand for the four
dresses in each of the versions are summarized in the table below. A total of 66
dresses were sold in the $9 ending conditions, compared to 46 units in the $5 lower
conditions and 45 units in the $5 higher conditions. The $9 ending yielded a demand
increase of approximately 40% for these four dresses, while the $10 price difference
between the two Treatment conditions resulted in effectively no difference in
demand. This outcome persuaded catalog managers to conduct additional studies to
replicate the ?ndings on a larger sample of items. The ?ndings also raise important
questions about the contexts in which a $9 ending effect occurs. The three studies
reported in this paper investigate whether the $9 ending effect varies depending upon
how often an item has appeared in the catalog, and whether the $9 ending effect is
moderated by ‘‘Sale’’ cues claiming that an item is discounted.
Pilot study: design and results.
Prices
Number of units purchased
Item
Control
Test A
Test B
Control
Test A
Test B
1
2
3
4
$39
$49
$59
$79
$44
$44
$54
$84
$34
$54
$64
$74
21
14
7
24
17
8
7
12
16
10
6
15
The marketing and economics literatures have long recognized the phenomenon of
price endings but have produced few conclusive empirical studies. Prior research has
focused on consumer-packaged goods or has involved very small price changes.
Stiving and Winer (1997) offer a recent example of research conducted with
consumer-packaged goods. They analyzed demand for canned tuna and yogurt and
found con?icting evidence about the effect of 9-cent price endings (see also Blattberg
and Neslin, 1990, p. 349). Examples of studies investigating very small price changes
include Ginzberg (1936) and Schindler and Kibarian (1996). Ginzberg (1936) offers a
brief description of a price-ending study conducted by a national mail-order catalog.
In this study, round prices (i.e., 0-cent ending) were compared against ‘‘just under’’
prices of one to two cents less. Overall, the results were varied, with demand
increasing for some items and decreasing for other items. Schindler and Kibarian
(1996) conducted a study comparing 88-cent, 99-cent and 00-cent price endings in a
clearance version of a mail-order catalog that sold women’s clothing. Their ?ndings
regarding the impact on demand were also inconclusive.
Our studies offer new insights on the role and effectiveness of price endings. First,
we present ?ndings investigating $9 ending effects in three large-scale studies
conducted in two different catalogs. The studies involved large price manipulations
(generally $1 to $5) and revealed a consistent signi?cant effect across all three
studies, providing conclusive evidence that $9 price endings can increase demand.
Second, we demonstrate that $9-endings are more effective on newer items. There is
EFFECTS OF $9 PRICE ENDINGS ON RETAIL SALES
95
also some evidence that their effectiveness varies in the presence of ‘‘Sale’’ cues. The
evidence that the effectiveness of price endings is moderated by the context in which
they are used, may offer an explanation for the inconclusive ?ndings in past studies.
The remainder of the paper is organized as follows. In Section 2, we present an
overview of the studies and a brief description of the catalog companies. In Section 3,
we present detailed descriptions and analyses of each study. In Section 4, we discuss
the ?ndings and conclude the paper.
2. Overview of studies
In this section, we provide an overview of the three studies. Because the details vary
across studies, we provide a more precise description of each study in Section 3. The
studies were conducted in two different mail-order catalog titles (companies). The
titles, which we will call Grace’s and Sandi’s, are both owned by the same
corporation. Although both catalogs sell moderately priced women’s clothing, there
is little overlap in their target markets. In both catalogs all of the items carry the
catalog’s own brand (private label) and have an average price of approximately $50.
Over 52% of Grace’s prices and 68% of Sandi’s have $9 endings. The next most
common price ending is $4, with 20% of Grace’s prices and 28% of Sandi’s prices
ending in $4. Active customers typically purchase approximately twice a year, and
may receive up to 20 catalogs per year.
In each study we simultaneously varied the use of $9 price endings on identical
items in different catalog versions mailed to randomly selected customer samples.
Each study had a Control version, in which the catalog’s standard pricing policy was
used. The Control version acted as a control for other unrelated tests, and so we were
unable to vary the prices in this version. One or more Treatment versions were also
created and prices were manipulated in these versions. As we explain in detail in the
next section, the current pricing policy (i.e., Control version) for these catalogs was
to use $9 endings on most items. Thus, most of the price changes in the Treatment
versions resulted in the removal of a $9 ending. Customers were randomly
designated to receive either the Control or one of the Treatment versions.
Randomization was based on zip þ 4 postal codes. Analysis of the frequency of
orders by zip code revealed that over 99.7% of orders were from unique zip codes
and that no more than two orders were received from the same zip þ 4 postal code.
Measuring purchasing behavior of real customers offers greater external validity
than laboratory experiments. The opportunity to charge different prices for the
same products also overcomes potential endogeneity issues that arise when
analyzing historical data. However, varying the content of catalogs is expensive
and preserving the cooperation of a catalog retailer restricts discretion over the
experimental design. In each study, catalog managers restricted the range of prices
that we could vary. For example, in Study 1 we were only able to adjust prices on
the ?rst and last ten pages of the catalog. The catalog printer treated these twenty
96
ANDERSON AND SIMESTER
pages as a ‘‘block’’ and the cost of changing the prices on any of these pages was
?xed. Changing prices on other pages would have resulted in additional costs. The
managers also placed some restrictions on the price-levels. In Study 1, price changes
were limited to $1 increments, while in Study 2 most of the price changes were
limited to $5 increments.
The market for women’s apparel is competitive and both Grace’s and Sandi’s
compete with other retailers and mail-order ?rms. Although we have no
information about competitor’s activities, we would expect these activities to affect
customers in the Control and Treatment versions similarly. More generally, the
experimental design allows us to exclude alternative explanations arising from
intervening events.
When customers call to place an order they are asked for a code printed on the
back of the catalog that they are ordering from. This code allows the ?rm to identify
which catalog the customer is purchasing from and, where appropriate, which
catalog version. In each study, we received aggregate sales data for each version at
the item-level. Items returned or cancelled by customers are netted out of the
analysis.
The direct marketing industry has a long history of conducting ‘‘split-sample’’
experiments, in which randomly selected customer samples are mailed modi?ed
versions of otherwise identical catalogs. Wisdom in the industry suggests that
pro?tability depends upon a ?rm’s ability to design, implement and analyze splitsample studies and then appropriately disseminate the ?ndings within the
organization. Prices are a common focus of these studies; over 31% of catalog ?rms
reported that they conducted split sample experiments of pricing strategies in 1999
(Direct Marketing Association, 2000). Other common experiments include testing
the demand for new products, and the creative design of catalog covers, page
layouts, and copy.
Testing of prices and other strategies is far more common among direct marketing
?rms than among traditional retailers. In part, this re?ects the bene?ts of conducting
split-sample tests in a catalog setting. The experimental versions of the catalogs can
be distributed at the same time to an identical sample of customers. Tests conducted
in retail stores generally require differences in strategies over time or differences
across stores. This introduces the potential for alternative explanations due to
intervening events or systematic differences between stores. Second, the number and
identity of catalog customers who are exposed to the different experimental versions
is known. In a retail store it is much more dif?cult to track the total number and
identity of customers who visit a store. Third, stock-outs can distort measurement of
demand in a traditional retail setting. For example, there is generally no record of
customers who searched for an item and then departed when they could not ?nd it,
or customers who were never aware of an item but would have purchased it if it had
been on display. In a catalog setting customers initiate orders in writing or via
telephone before learning whether the item is available. The decision to cancel an
order, substitute an alternative item or backorder an item is also explicit and
therefore observable.
97
EFFECTS OF $9 PRICE ENDINGS ON RETAIL SALES
3. Results
In this section, we present a detailed description of each study. Study 1 involved the
fewest items and the narrowest variation in prices. In Study 2, we were able to
investigate whether the results from the ?rst study generalize to a different catalog
title. We were also offered more ?exibility over the number of prices that we could
change and the magnitude of the price changes. This increased ?exibility enabled us
to investigate whether the $9-ending effect depends on whether an item is new to the
catalog or has appeared in previous seasons. Study 3 extends these results by
considering the effect of a $9-ending when a ‘‘Sale’’ sign is present. We begin our
discussion with Study 1.
Study 1
Study 1 was conducted with a Grace’s catalog and involved a Control version and
two Treatment versions that we label Version A and Version B. The three catalog
versions were distributed to separate, randomly chosen customer samples, with
20,000 customers receiving each version. All of the customers had purchased from
Grace’s in the previous 18 months. The test involved varying the prices of items on
the ?rst and last ten pages of the catalog. There were a total of 73 items in these two
ten page sections and 146 other items in the catalog.
We varied the prices of 27 items, all of which had $9 price-endings in the Control.
On six items we increased the price in Version A and reduced the price in Version B.
On ten items this manipulation was reversed; prices were decreased in Version A and
increased in Version B. Finally, for a further eleven items we only varied the prices in
Version B, increasing the price of four items and decreasing the price of seven items.
All price changes were in $1 increments, so that price increases yielded price-endings
of $0 and price decreases yielded price-endings of $8. A frequency distribution of
price endings and summary statistics for all three versions are provided in Table 1.
Management of Grace’s provided us with aggregate statistics for the three catalogs
along with item-level demand data for the 73 items on the twenty test pages. Sales of
each item (in a given version) ranged from zero to nineteen with an average demand
of 2.6 units per item. This count of the number of units ordered could be expected to
follow a Poisson distribution. In particular, we will assume that the number of units
of item i ordered from version j is drawn from a Poisson distribution with parameter
lij :1


Prob Q ¼ qi; j

q
eli; j li;i;jj
;
¼
qi; j !
qi; j ¼ 0; 1; 2; . . .
1 For all three studies, we also modeled demand using a semi-log speci?cation, qi; j ¼
and obtained similar results to those reported in the paper.
ð1Þ
P
k
bk lnðXk;i;j Þ,
98
ANDERSON AND SIMESTER
Table 1. Summary of price endings and prices offered in Study 1.
Test items
Control
Version A
Version B
Other items
Price Endings
$0
$4
$8
$9
Other
Total
22
50
1
73
Average price
Std deviation
$54.44
$21.11
6
22
10
34
1
73
$54.38
$21.10
14
22
13
23
1
73
$54.45
$21.05
37
103
6
146
$56.01
$19.93
where:
J 1
I1
X
X


ln li; j ¼ a0 þ
aj Versionj þ
wi Itemi þ b1 ln Pricei; j þ b2 Ninei; j :
j¼1
ð2Þ
i¼1
The aj ; wi ; b1 and b2 terms are estimated parameters and the variables are de?ned as
follows:
qi,j
Versionj
Itemi
Pricei,j
Ninei,j
Number of units of item i purchased from version j.
Dummy variables identifying the version.
Dummy variables identifying the item.
The price of item i in version j.
1 if the price of item i in version j ends in $9; 0 otherwise.
The Versionj terms control for version-speci?c effects (including any differences in
the distribution of customers that received each version), the Itemi terms control for
item-speci?c effects, b1 and b2 measure the price and $9 ending effects. We estimated
this Poisson regression model through maximum likelihood using data describing the
demand for the 73 test items in the three versions (219 observations). The ?ndings
are reported in Table 2 (for ease of exposition we omit the coef?cients describing the
version and item effects). As a benchmark we also report models in which we omit
either the $9 ending or the price variable.
The ?ndings in Models 2 and 3 offer support for a $9 ending effect. The
coef?cients for the $9 ending variable are positive and statistically signi?cant
ðp < 0.05Þ. Expected demand is given by qE½qij jxij  ¼ li; j and the marginal effect is
qE½qij jxij =qxij ¼ blij . Thus, the percentage change in demand from using a $9 ending
is qE½qij jxij =qxij =E½qij jxij  ¼ b. We conclude that using a $9 ending in Study 1 led to
99
EFFECTS OF $9 PRICE ENDINGS ON RETAIL SALES
Table 2. Nine ending model results for Study 1.
Model 1
$9 ending
Price (b1)
Sample size
Log likelihood
 1.954
(6.406)
219
 210.9
Model 2
Model 3
0.341*
(0.170)
0.351*
(0.172)
 3.182
(6.383)
219
 208.9
219
 209.1
Notes. Item and version-speci?c effects are omitted. Asymptotic standard errors are in parentheses.
*Signi?cantly different from zero (p < 0.05).
an increase in demand of approximately 35%, which translates into 0.9 additional
units per item.
While we ?nd evidence of a positive $9 ending effect, the estimated price
coef?cient is not signi?cant in either model. This is somewhat consistent with the
?ndings from the pilot study where aggregate demand did not vary between the high
and low price conditions. We might be tempted to conclude that customers are more
sensitive to the $9 price cue than they are to actual price changes. However, the
variance available to identify a price coef?cient is small. The item-level effects
account for variation in prices between items and so the price coef?cient is identi?ed
by the price changes on the 27 test items. These same price variations are used to
identify the $9-ending effect. In Study 2 we were able to mail catalogs to a larger
sample of customers and introduce greater price variation. This enabled us to better
separate the price and $9-ending coef?cients and increased the statistical power of
the study.
Study 2
Study 2 was conducted in a Sandi’s catalog. The test was designed to investigate
whether the results from Study 1 could be replicated in a different catalog title. We
were also interested in whether the ?nding would survive greater variation in the
prices and whether the effect varies depending on whether an item is new to the
catalog.
Two versions of a catalog, which we label Treatment and Control, were each
mailed to randomly selected samples of 31,250 customers. Between the two versions
we manipulated the prices of 120 items, out of a total of 211 in each version. The
prices of the 120 test items were all lower in the Treatment version, with the price
decreases ranging from $3 to $5. The price changes resulted in the following
distribution of price endings across the two versions:
100
ANDERSON AND SIMESTER
Table 3. Summary of price endings and prices offered in Study 2.
Test items
Control
Treatment
Other items
Price ending
$0
$2
$4
$5
$6
$7
$8
$9
Total
6
6
22
11
3
8
71
2
4
8
75
120
24
120
Average price
Std deviation
$57.75
$18.95
$53.05
$18.59
16
5
1
1
68
91
$48.25
$24.11
. Seventy-?ve test items had $9 price endings in the Control version. Seventy-one of
these items had $4 price-endings in the Treatment, two had $5 price endings, and
two had $6 price endings.
. Twenty-four test items had $9 price endings in the Treatment. Twenty-two of these
items had $4 price-endings in the Control version, and 2 had $2 price-endings.
. Twenty-one test items did not have a $9 ending in either the Treatment or the
Control.
The average prices across the two versions and the distribution of price endings are
summarized in Table 3.
We used the same approach to analyze the ?ndings as we employed in Study 1. In
particular, we again assumed that the number of units of item i ordered from
Versionj follows the Poisson distribution described in equations (1) and (2). We
estimated the model using data for all 211 items in the two versions (422
observations). Because the study only had two experimental versions, each itemspeci?c effect is identi?ed by just two observations. However, this is analogous to
any study in which an outcome is compared across Treatment and Control samples.
The Control sample provides the extra degree of freedom required to control for
variance that is common to the two samples. The ?ndings are reported in Table 4,
where we also report the benchmark models in which the $9 ending or price variable
was omitted.
Additional data provided by this catalog allowed us to consider a modi?cation to
our model speci?cation. The retailer classi?es its items into three categories:
footwear, ‘‘new’’ items and ‘‘established’’ items. The new and established items are
all clothing items (not footwear). The new items were generally introduced in the
most recent season, while established items were offered in previous seasons. The 211
items in this catalog included 55 established items, 128 new items and 28 footwear
101
EFFECTS OF $9 PRICE ENDINGS ON RETAIL SALES
Table 4. Summary of model results in Study 2.
Model 1
$9 ending
$9 ending on
footwear
$9 ending on
established items
$9 ending on
New items
Price (b1)
 0.304
(0.450)
422
 423
Sample size
Log likelihood
Model 2
Model 3
0.143**
(0.039)
0.152**
(0.041)
422
 419
 0.592
(0.486)
422
 419
Model 4
 0.397
(0.247)
0.098
(0.069)
0.218**
(0.059)
 0.480
(0.509)
422
 417
Notes. Item and version-speci?c effects are omitted. Asymptotic standard errors are in parentheses.
**
Signi?cantly different from zero ðp < 0.01Þ.
items. Of the 120 test items, 35 were established, 67 were new and 18 were footwear.
This balance in the distribution of test items is also re?ected in the distribution of $9
endings across the items. Amongst the established items, 55.5% of the established
item prices ended in $9, compared to 55.9% of the new items and 55.4% of the
footwear items. To investigate how the $9 ending effect varied across the product
categories, we interacted the Ninei; j variable with dummy variables identifying each
category. This led to the following model speci?cation:
J1
I1
X
X


ln li; j ¼ a0 þ
aj Versionj þ
wi Itemi þ b1 ln Pricei; j
j¼1
þ
X
i¼1
bc2 Categoryc Ninei; j ;
ð4Þ
c
where Categoryc refers to the category dummies. We label this last speci?cation as
Model 4 and the results of each model are shown in Table 4.
The ?ndings replicate and strengthen many of the ?ndings in Study 1. In Models 2
and 3 we see that $9 endings are again associated with a signi?cant ðp < 0.01Þ
increase in demand. To aid interpretation we calculated the marginal increase in
demand attributable to a $9 ending. On average each version sold 8.7 units of each
item. This increased by approximately 1.3 units (15%) when an item had a $9 ending.
The ?ndings from Model 4 indicate that the $9 ending effect was stronger for new
items than for established items. A $9 ending yielded an estimated increase of 1.8
units or 22% for new items compared to 10% for established items. For the footwear
items the result was dif?cult to interpret; the coef?cient was negative but not
102
ANDERSON AND SIMESTER
Table 5. Summary of model results in Study 2.
Sale cue only
$9 ending only
$9 ending and sale cue
Neither
Total
Control
Version A
Version B
11
176
27
94
308
13
91
16
188
308
14
90
16
188
308
signi?cant. The relatively large standard error for this coef?cient re?ects the small
number of footwear items sold.
Finally, the price coef?cients are again all insigni?cant. This offers further
evidence that customers are more sensitive to the $9 ending cues than to actual price
differences.
Study 3
Study 3 had several goals. First we sought to replicate the $9 ending effect observed
in the two previous studies. Second, we also hoped to replicate the ?ndings in Study
2 revealing that the size of the effect is moderated by whether an item is an
established or new item. Finally, we investigated whether the effect is also moderated
by the presence or absence of a ‘‘Sale’’ cue claiming that a price is discounted.
In Study 3 three different versions of a Sandi’s catalog containing 308 items were
produced. The three catalog versions were distributed to separate, randomly chosen
customer samples, with 90,000 customers receiving each version. The price
descriptions for 38 of the items in the Control version included an explicit claim
that the item was discounted. The prices of these items were presented as follows
‘‘Reg $X SALE $Y’’ compared to the standard description ‘‘$X’’. In addition a total
of 203 items had $9 endings. In the two Treatment versions we varied prices and
removed sale claims from some of these items. Removing a sale cue from an item
results in an unadvertised discount, and catalog managers deemed this potentially
costly but worthy of investigation. In contrast, adding a sale sign to an item without
lowering the price was viewed as a risky strategy that may breach customer
protection laws. For this reason we were unable to add sale cues to regularly priced
items, so that the test contained less variance in the use of sale claims than we would
have liked. The ?nal distribution of sale claims and $9 endings is summarized in
Table 5.
The prices in the Control version were the standard price for each item. In Version
A the prices of 55 items were raised by $1 to $6 and the prices of 53 items were
lowered by $0.50 to $4. In Version B the prices of 58 items were raised by $1 to $4.50
and the prices of 52 items were lowered by $0.50 to $10. The average prices across the
versions and the distribution of price endings are summarized in Table 6. The
manipulations involving $9.50 price endings were introduced by a catalog manager.
103
EFFECTS OF $9 PRICE ENDINGS ON RETAIL SALES
Table 6. Summary of price endings and prices offered in Study 3.
Test items
Control
Version A
Version B
Other items
Price Endings
$0
$1
$2
$4
$5
$6
$7
$8
$9
$9.50
Total
Average price
Std deviation
8
17
8
1
8
19
7
1
148
7
21
27
43
16
148
7
22
27
42
15
148
160
$58.39
$32.94
$58.32
$32.94
$58.44
$32.90
$49.64
$25.66
1
3
3
2
139
12
63
10
6
5
64
This manager reasoned that a $9 ending strategy would be even more pro?table if it
contributed an additional $0.50 of revenue per item.
To analyze the results of Study 3 we modi?ed equation (2) to introduce explicit
controls for the ‘‘Sale’’ cues.
J 1
I1
X
X


ln li; j ¼ a0 þ
aj Versionj þ
wi Itemi þ b1 ln Pricei; j
j¼1
i¼1
þ ð1  Salei; j Þ Ninei; j
þ Salei; j Ninei; j
X
X
bc2 Categoryc
c
c
b3 Categoryc



þ b4 Salei; j 1  Ninei; j :
ð5Þ
c
The Salei; j variables was de?ned as: 1 if the price description of item i in version j
included the word ‘‘Sale’’ and 0 otherwise. This speci?cation allows us to identify
three mutually exclusive effects. The coef?cient bc2 measures the effect of using a $9
ending with no ‘‘Sale’’ cue; bc3 measures the effect of using a $9 ending with a ‘‘Sale’’
cue; and b4 measures the effect of using a ‘‘Sale’’ cue alone. Preliminary analysis
revealed that customers did not respond to $9.50 endings in the same way that they
respond to $9 endings and so we did not incorporate these price endings into the
Nine variable.
We estimated this model using data describing the demand for all 308 items in the
three versions (924 observations). The ?ndings are reported in Table 7 where we also
report analogous benchmark models to those reported for the earlier studies. We did
104
ANDERSON AND SIMESTER
Table 7. Summary of model results in Study 3.
Model 1
$9 Ending 6 No Sale Cue
$9 Ending on Footwear
6 No Sale Cue
$9 Ending on Established Items
6 No Sale Cue
$9 Ending on New Items
6 No Sale Cue
$9 Ending 6 Sale Cue
Sample size
Log likelihood
Model 3
0.044
(0.028)
0.071*
(0.030)
Model 4
0.077
(0.059)
 0.024
(0.112)
0.085*
(0.035)
0.078
(0.079)
$9 Ending on Established Items
6 Sale Cue
$9 Ending on New Items
6 Sale Cue
No $9 Ending 6 Sale Cue
Price (b1)
Model 2
0.240**
(0.069)
 2.244**
(0.464)
924
 1519
924
 1527
0.117
(0.082)
0.215**
(0.071)
 2.312**
(0.492)
924
 1515
 0.109
(0.185)
0.217**
(0.084)
0.178*
(0.073)
 2.360**
(0.503)
924
 1512
Notes. Item and version-speci?c effects are omitted. Asymptotic standard errors are in parentheses.
** Signi?cantly different from zero ðp < 0.01Þ.
not include a variable measuring the joint effect of a ‘‘Sale’’ cue and a $9 ending on
footwear items as there were no footwear items that had both a ‘‘Sale’’ cue and a $9
ending.
The coef?cients estimated for the $9 ending variables replicate and extend the
?ndings from the two previous studies. In the absence of a sale sign, a $9 ending is
associated with a signi?cant increase in demand (Model 3), although the signi?cance
of the coef?cient depends upon the inclusion of the price variable (Model 2). On
average the catalog sold 16.9 units of each item in each version and this increased by
1.2 units when an item had a $9 ending (7%). The ?ndings also replicate the evidence
in Study 2 that the $9 ending effect is stronger for new items than for established
items or footwear (Model 4).
The ‘‘Sale’’ cue also led to a positive increase in demand. This is consistent with
?ndings reported elsewhere (see, for example, Inman et al., 1990; Inman and
McAlister, 1993; Anderson and Simester, 2001). The demand increase associated
with the ‘‘Sale’’ cue is larger than that estimated for the $9 ending (see Model 3),
suggesting that customers are more sensitive to this cue than the price-ending cue.
The comparison of the ?ndings when using both a $9 ending and a ‘‘Sale’’ cue on
new items is particularly interesting. Adding a $9 ending to a new item that already
has a ‘‘Sale’’ cue led to a 3.9% demand increase (the difference between 21.7% and
17.8%). In contrast, adding a $9-ending to a new item that does not have a ‘‘Sale’’
EFFECTS OF $9 PRICE ENDINGS ON RETAIL SALES
105
cue yielded an 8.5% increase in demand. One interpretation of this comparison is
that $9 endings are less effective when an item already has a ‘‘Sale’’ cue. However, we
caution that the joint effect of a ‘‘Sale’’ cue and a $9 ending on new items is estimated
using relatively few items. This comparison is perhaps best interpreted as a tentative
?nding worthy of further investigation.
There is an important difference in the ?ndings of Study 3 compared to the
?ndings in the two previous studies. The coef?cients estimated for the price variable
in Study 3 are negative and consistently signi?cant ðp < 0.01Þ. We offer two
explanations for this difference. First, there was considerably more variation in the
prices across catalog versions in Study 3 compared to the previous two studies. A
total of 218 prices were varied in Study 3 (across the two treatment versions),
compared to 43 prices in Study 1 and 120 prices in Study 2. This increased variation
provided more information with which to identify the price coef?cient. Second,
90,000 customers received each version of the catalog in this study, compared to
31,250 in Study 2 and 20,000 in Study 1. This resulted in average demand from each
version of almost 17 units per item compared to less than 9 units per item in Study 2
and just over 2 units per item in Study 1. The increased demand for each item in
Study 3 increased the statistical power of the study, reducing the standard errors.
4. Discussion
The three studies reveal that $9 price endings increase demand but that the effect is
context dependent. In this section, we review different explanations for the $9 ending
effect and evaluate the extent to which these explanations can be reconciled with the
?ndings.
The literature contains a range of explanations for price ending effects. These
explanations fall into two broad categories.2 Explanations in the ?rst category posit
various reasons for why customers ‘‘drop-off ’’ the right-most digits and therefore
overweight the left-most digits. The most common of these explanations is that
customers round prices down (Gabor and Granger, 1964; Lambert, 1975; Schindler
and Kibarian, 1993) and essentially ignore the right-most digits. For example $59.99
might be coded as $59 or, in an extreme case, as $50. A limitation of these theories is
that they do not explain why prices are rounded down rather than up. A related
theory posits that customers engage in left-to-right processing of digits. Thus, when
comparing $55 with $43 a customer that looked at only the left-most digit would
process this as $50 vs. $40 or a $10 price difference (rather than $12). Raising the
price from $55 to $59 would have no impact on this customer’s perceptions of the
price difference. However, lowering the price from $43 to $39 would create the
perception of a $20 price difference (rather than $16). A third class of theories argues
that because of the cognitive costs of processing price information, customers may
2 The authors thank Robert Schindler for his thoughtful suggestions on this taxonomy.
106
ANDERSON AND SIMESTER
not encode some of the right-most digits (Schindler and Wiman, 1989; Schindler and
Kirby, 1997). A common-theme in these theories is that customers imperfectly
process price information, for either rational (cognitive cost) or irrational reasons.
A second class of theories posits that price endings provide information about
relative price levels and/or product quality (Schindler, 1991). In these theories,
customers pay more attention to the right-most digits because of the information
that they convey. This contrasts with the customer’s emphasis on the left-most digits
in the ‘‘dropping off ’’ theories. Researchers have suggested that one inference
customers may draw from $9-endings is that a price is low, discounted, or on ‘‘Sale’’
(Schindler and Warren, 1988). For example, Salmon and Ortmeyer (1993) describe a
department store that uses a 0-cent ending for regularly priced items and 98-cent
endings for clearance items. Similarly, Randall’s Department Store uses 95-cent
endings on all ‘‘value’’ priced merchandize, which is ‘‘meant to indicate exceptional
value to the customer’’ (Salmon and Ortmeyer, 1992).3 Other researchers have
suggested that customers may infer that $0-endings (and perhaps $5 endings) imply
high quality and $9-endings imply low quality (Stiving, 2000). Stiving and Winer
(1997) argued that the opposing implications of price and quality inferences might
help to explain their contradictory empirical ?ndings.
The dichotomy of ‘‘dropping off ’’ effects and ‘‘information’’ explanations
captures many theories of price-endings, but there are some notable exceptions.
The ‘‘change making’’ theory suggests that odd-prices force employees to open the
cash register to make change and this reduces employee theft. An alternative theory
suggests that retailers use price endings to achieve even after-tax prices and reduce
transaction costs. Neither of these explanations appears to apply in retail
environments in which transactions are conducted using credit cards (such as
catalog retailers).
The distinction between whether customers focus on the left or right hand digits
offers an opportunity to discriminate between the ‘‘dropping off ’’ and ‘‘information’’ explanations. This test was proposed by Stiving and Winer (1997) in their
study of demand for yogurt and canned tuna. The prices of yogurt and canned tuna
are all less than a dollar, and so they decomposed these prices into dimes and
pennies. When limiting attention to the two largest brands in each category they ?nd
strong evidence that customers place less weight on the pennies digit relative to the
dimes digit, which is consistent with rounding down theories. However, they ?nd
that the weight on the pennies digits increases when the dimes digits on the two
brands are equal, which is consistent with left-to-right processing of digits. In sum,
their evidence offers support for ‘‘dropping off ’’ theories.
3 Inspection of prices at other retailers revealed many examples that were consistent with this pattern.
For example, Eddie Bauer’s September 1996 catalog used a $X price ending on regular prices and a
$0.99 price ending on discounted items. Similarly, at www.jcrew.com on September 14 2000 all 548
clearance items in 55 categories had 99-cent endings and all regularly priced items had $X endings.
EFFECTS OF $9 PRICE ENDINGS ON RETAIL SALES
107
An equivalent decomposition to Stiving and Winer’s in our setting is to
decompose prices into tens of dollars and dollars. That is, $54 is treated as $50
(tens of dollars) and $4 (dollars). Study 3, in which the variation in price endings was
greatest, provides the richest test of this theory. We modi?ed our Poisson model as
follows:
J 1
I1
X
X


ln li; j ¼ a0 þ
aj Versionj þ
wi Itemi þ b1 Tensi; j þ b2 Dollarsi; j
j¼1
i¼1



þ b3 ð1  Salei; j Þ Ninei; j þ b4 Salei; j Ninei; j þ b5 Salei; j 1  Ninei; j : ð6Þ
The coef?cient b1 describes how variations in the left-hand digit affected demand,
while the effect of the right-hand digit is measured by both b1 and the coef?cients for
the Nine variable. If customers ‘‘round down’’ or place less weight on the dollars
digits, then we would expect to observe jb2 j < jb1 j. This is a relatively weak test as
the variation in demand explained by the right-hand digit is at least partially
explained by the Nine variables. We ?nd that b1 ¼  0:0631 and b2 ¼  0:0651. Both
coef?cients are signi?cantly different from zero ðp < 0.01Þ but the difference between
them is not signi?cant. After controlling for the effect of the $9 price ending there is
seemingly no difference in the weight that customers give to the left and right hand
digits. We conclude that this analysis does not support the ‘‘dropping-off ’’
explanations but it does provide support for an information explanation. The b3
and b5 coef?cients remained positive and signi?cant, which is consistent with
customers forming favorable price inferences when observing a $9 ending.
The evidence that the $9 ending effect is stronger on new items is also consistent
with an information explanation. The explanation requires that customers lack
information about relative price levels, otherwise there is no need to use the price
endings as a cue to infer this information. We would expect that on average
customers would have more information about the relative price levels of items that
they have seen more frequently in the past, suggesting that they will have less relative
price information about new items compared to established items. A similar
argument might also explain why customers appear to be more sensitive to the priceending cue than to variations in the actual price. If customers are unsure about
relative price levels, small price differences may not help customers evaluate whether
an item is expensive or inexpensive. In the absence of other cues, the price ending
may provide the only source of information that customers can use to help resolve
their uncertainty.
Proponents of the information explanations may ?nd further support for their
arguments from the manipulations involving ‘‘Sale’’ cues. The claim that an item
is discounted is more explicit in a ‘‘Sale’’ cue than in a $9 price ending. This
may explain both why ‘‘Sale’’ cues are more effective at increasing demand, and
the tentative ?nding that price endings are less effective on items that already
have sale signs. Having already informed customers that an item is discounted
108
ANDERSON AND SIMESTER
through a ‘‘Sale’’ cue, little additional information is revealed by the price
ending.
We conclude that these context effects can be reconciled with an information
explanation. However, it is dif?cult to reconcile them with a ‘‘dropping-off’’
explanation. It is not clear why the distinction between new and established items
should affect the extent to which customers imperfectly process right-hand digits. Of
perhaps even greater concern for the dropping off explanations is why the $9 ending
effect does not also extend to $9.50 endings. The evidence suggests that, if anything,
$9.50 endings lead to reduced demand in Study 3.
A criticism of the ‘‘information’’ explanations is that they focus on customer
behavior but fail to consider the seller’s perspective. In particular, the theoretical
concern is that the signal may be equally costly for all sellers to implement. If so,
this violates a necessary condition of information models, as it must be the case that
the low price ?rm can signal at lower cost than the high price ?rm. In such a
situation, an optimal response for every seller is to use $9 endings on each item. At
least two explanations have been offered to suggest why this may not be an optimal
strategy. First, Anderson and Simester (1998) provide an equilibrium explanation
for how retailers may credibly convey information via ‘‘Sale’’ signs and a similar
argument may apply to price endings. They argue that customers have expectations
as to the frequency of a promotional cue prior to arriving at a store and believe
these cues convey information about relative price. Upon arriving, customers
observe how frequently the cues are used (the number of ‘‘Sale’’ signs) and use this
to assess the likelihood that an item with a ‘‘Sale’’ sign is indeed low priced. If a
retailer overuses a cue, it becomes less effective as customers are less likely to believe
the item is indeed low priced. If this theory applies to price endings, retailers should
selectively use $9 price endings. A second theory is that retailers have reputations. If
a retailer uses a $9 price ending on a relatively high priced item then the retailer’s
reputation may be damaged. This may prompt retailers to restrict their use of $9
endings.
Both theories offer a rational explanation for customer reliance on price endings
and retailers’ (limited) use of price endings. However, even these two theories are
incomplete. Neither theory explains why the favorable signal is associated with the
digit 9. The information explanations could equally explain an $8 price ending effect.
Similarly, the theories do not explain how customers learn to use this cue or whether
the cue is speci?c to a product category or a store. The answers to both questions are
important for retailers seeking to implement a price ending policy.
5. Conclusion
We have presented three ?eld experiments demonstrating that $9 price endings
increase demand but that the effect is context dependent. The effect is stronger for
new items that customers have seen less frequently in the past. We evaluated two
types of explanations for the ?ndings. Under the information explanation customers
EFFECTS OF $9 PRICE ENDINGS ON RETAIL SALES
109
use the price ending to infer whether the item is expensive or inexpensive relative to
market price levels. The ‘‘dropping-off ’’ explanations argue that customers
imperfectly process price information. The ?ndings appear to offer greater support
for an information explanation than for a ‘‘dropping off ’’ explanation.
What remains largely unexplained by our work is ‘‘Why 9?’’ The data and
explanations are silent on how customers form beliefs that $9 endings convey
favorable information that increases demand for an item. Future work may want to
explore how these associations are formed, and this may help to explain the varied
use of price endings across markets and geographies.
Acknowledgments
We wish to thank seminar participants at Northwestern, MIT and Harvard for their
comments. We would also like to thank the company that provided the data
reported in the paper. Eric Anderson thanks the Kilts Center for Marketing at the
University of Chicago for research support. We thank Robert Schindler for his
comments.
References
Anderson, Eric T. and Duncan I. Simester. (1998). ‘‘The Role of Sale Signs’’, Marketing Science 17(2),
139–155.
Anderson, Eric T. and Duncan I. Simester. (2001). ‘‘Are Sale Signs Less Effective When More Products
Have Them?’’ Marketing Science 20(2), 121–142.
Blattberg, Robert C. and Scott A. Neslin. (1990). Sales Promotion: Concepts, Methods and Strategies.
Englewood Cliffs, NJ: Prentice-Hall.
Daily Mail. (2000), ‘‘Asda Axes The 99p Price Ploy’’, London, May 22, p. 15.
Direct Marketing Association. (2000). State of the Catalog Industry Report, New York: DMA.
Gabor, Andre and Granger, C.W.J. (1964). ‘‘Price Sensitivity of the Consumer’’, Journal of Advertising
Research, 4, 40–44.
Ginzberg, Eli. (1936). ‘‘Customary Prices’’, The American Economic Review 26(2), 296.
Inman, Jeffrey J. and Leigh McAlister. (1993). ‘‘A Retailer Promotion Policy Model Considering
Promotion Signal Sensitivity’’, Marketing Science 12(4), 339–356.
Inman, Jeffrey J., Leigh McAlister, and Wayne D. Hoyer. (1990). ‘‘Promotion Signal: Proxy For A Price
Cut’’, Journal of Consumer Research 17(1), 74–81.
Lambert, Zarrel V. (1975). ‘‘Perceived Prices as Related to Odd and Even Price Endings’’, Journal of
Retailing 51 (Fall), 13–22.
Salmon, Walter J. and Gwendolyn K. Ortmeyer. (1992). Randall’s Department Stores. Harvard Business
School Publishing, 9-593-032, pp. 1–24.
Salmon, Walter J. and Gwendolyn K. Ortmeyer. (1993). Duncan Department Stores. Harvard Business
School Publishing, 9-594-012, pp. 1–36.
Schindler, Robert M. (1991). ‘‘Symbolic Meanings of a Price Ending’’, Advances in Consumer Research
18, 794–801.
Schindler, Robert M. and Thomas Kibarian. (1993). ‘‘Testing for Perceptual Underestimation of 9-ending
Prices’’. In Advances in Customer Research, Vol. 20. Provo, UT: Association for Customer Research,
580–585.
110
ANDERSON AND SIMESTER
Schindler, Robert M. and Thomas Kibarian. (1996). ‘‘Increased Consumer Sales Response Through Use
of 99-Ending Prices’’, Journal of Retailing 72(2), 187–199.
Schindler, Robert M. and Patrick N. Kirby. (1997). ‘‘Patterns of Rightmost Digits Used in Advertised
Prices: Implications for Nine-Ending Effects’’, Journal of Customer Research 24 (September), 192–201.
Schindler, Robert M. and Lori S. Warren. (1988). ‘‘Effect of Odd Pricing on Choice of Items from a
Menu’’, Advances in Consumer Research 15, 348–353.
Schindler, Robert M. and Alan R. Wiman. (1989). ‘‘Effect of Odd Pricing on Price Recall’’, Journal of
Business Research 19 (November), 165–177.
Stiving, Mark. (2000). ‘‘Price-Endings When Prices Signal Quality’’, Management Science 46(12),
December, 1617–1629.
Stiving, Mark and Russel S. Winer. (1997). ‘‘An Empirical Analysis of Price Endings with Scanner Data’’,
Journal of Customer Research June, 57–67.

PLACE THIS ORDER OR A SIMILAR ORDER WITH US TODAY AND GET AN AMAZING DISCOUNT 🙂