Escolar Documentos
Profissional Documentos
Cultura Documentos
V
iN[s
;
is a measure of the heterogeneity of the L ) 1 alternatives in {N}.
244 M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260
also use a spline representation). We start from the assumption that the ref-
erence price is determined exogenous to our model.
To develop the spline representation, the total price range is partitioned
into K + 1 price ranges dened by the K psychological prices. Utility is as-
sumed to depend linearly on price, in between the K psychological prices.
The sequence of knots that forms the basis of the spline function, represent-
ing the psychological prices, is dened as t
0
; t
1
; . . . ; t
K
; t
K1
; t
0
denotes the
lowest, t
K1
the highest price oered, and t
k
(k = 1; . . . ; K) are the K psycho-
logical prices. For consumer i, the spline function contains one additional
knot at p
i
r
, his/her reference price level.
We introduce the function S
q
(.):
S
q
(Z) = Z
q
if Z > 0
= 0 if Z 60;
where the exponent q is a nonnegative integer. This function is commonly
used to formulate splines (see Smith, 1979). A simple example follows. Two
intervals of a variable x, say [), t
1
); [t
1
, ] are distinguished, where t
1
denotes a knot. Now, S
1
(x ) t
1
) =x ) t
1
, for x > t
1
. The spline function
is dened in this case as: y =b
1
x + b
2
S
1
(x ) t
1
). This function has regres-
sion coecients b
1
for x < t
1
, and b
1
+ b
2
for x > t
1
. The line segments
are connected at t
1
. In the presence of more intervals, similarly, the coe-
cients are cumulated (b
1
+ b
2
+ b
3
, etc.) across the preceding intervals, in
the calculation of the eect of x in a certain interval. We will use q =0,1,
implying linear splines. The deterministic part of the utility function for a
consumer of market segment s is written as the sum of three eects (each
comprising two terms): a linear price eect + reference price eect + psy-
chological price eect (the interpretation of the terms in the model will
be further explained below):
V
ij[s
= b
00s
b
10s
p
ij
b
2s
p
r
i
b
3s
S
1
(p
ij
p
r
i
)
X
K
k=1
b
0ks
S
0
(p
ij
t
k
)
X
K
k=1
b
1ks
S
1
(p
ij
t
k
): (2)
For an individual consumer in market segment s:, b
00s
denotes the intercept
(INTCPT), b
10s
represents the linear price eect (PRICE0), b
2s
represents
the overall eect of the reference price (REFERP), b
3s
represents the addi-
tional eect of price `losses', where a price loss is dened as the dierence
between the actual/observed price and the reference price if the actual price
M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260 245
is higher than the reference price (PRLOSS), b
0ks
represents the eect of
the kth psychological price level (PSYPRk), b
1ks
represents the additional
linear price eect in the kth price interval, relative to the (k ) 1)th interval
(PRICEk).
The intercept term in Eq. (2) represents that part of the deterministic part
of the utility which is not related to price. It represents the `overall' probabil-
ity of choosing the brand (l =1) by consumers in market segment s. The term
PRICE0 captures the overall linear price eect. The term PRLOSS represents
the additional eect of a price `loss', relative to price `gains' (see Kahneman
and Tversky, 1979; Tversky and Kahneman, 1991; Hardie et al., 1993). Note
that S
1
(p
ij
) p
i
r
) =0 for p
ij
< p
i
r
, and S
1
(p
ij
) p
i
r
) =(p
ij
) p
i
r
), otherwise.
For example, the partial eect of a price gain (p
ij
6p
i
r
) equals b
10s
p
ij
. The
partial eect of a price loss (p
ij
> p
i
r
), can then be represented as
(b
10s
+ b
3s
)p
ij
. Thus b
3s
represents the additional eect of price losses (on
the price p
ij
). The formulation used enables a test of loss aversion: if
b
3s
=0, price eects above and below the reference price level are symmetric.
The term REFERP represents the absolute eect of the reference price on
the choice probabilities. Compare in this respect Tversky and Simonson
(1993), who have shown that in addition to relative evaluations, absolute
evaluations play a role in consumer choice processes.
The coecients b
0ks
of the terms PSYPRk (for k = 1; . . . ; K) represent the
changes in intercept at the psychological price levels, and thus the discontinu-
ities in utility at these price levels. The coecients b
1ks
of the terms PRICEk
(k = 1; . . . ; K) represent the changes in the slope of price at the psychological
price levels, relative to the slope in the next lower price range. These coe-
cients are used to investigate the shape of the price-response function.
At this point, we formulate an expression for the conditional choice prob-
abilities P
ij[s
. We use the standard assumption of a logistic distribution of the
dierence
ij[s
)
iN[s
. The following expression for the probability that sub-
ject i, coming from market segment s, wishes to buy the brand at price level
p
ij
, is obtained:
P
ij[s
=
1
1 exp(V
iN[s
V
ij[s
)
=
1
1 exp(V
+
ij[s
)
: (3)
Since the prices of the other brands and thus V
iN[s
is assumed to be con-
stant in segment s during the experiment, this term may be absorbed in the
constant b
+
00s
in V
+
ij[s
. A given consumer i has the probability h
s
that it belongs
to segment s. The unconditional probability that subject i chooses the brand
at price j is:
246 M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260
P
ij
=
X
S
s=1
h
s
P
ij[s
=
X
S
s=1
h
s
1
1 exp(V
+
ij[s
)
: (4)
4.3. Estimation
The parameters of the model are estimated by the method of maximum
likelihood. Using the binomial distribution for the choice probabilities, the
likelihood can be formulated as:
l =
Y
n
i=1
X
S
s=1
h
s
Y
J
j=1
(P
ij[s
)
y
ij
(1 P
ij[s
)
(1y
ij
)
: (5)
The likelihood, or equivalently, the log-likelihood is maximized using an
EM-algorithm (see Wedel and DeSarbo, 1994) to obtain estimates of the pa-
rameters of the model. The EM algorithm has the advantages of being easy to
implement, while convergence of the iterative procedure is ensured.
Once the parameters have been estimated, the posterior probabilities that
subject i comes from segment s, h
is
, can be calculated using Bayes' rule (see
Wedel and DeSarbo, 1994). The asymptotic variances of the estimated
spline-function coecients are obtained from the Fisher information matrix.
These variances allow for signicance testing of the coecients.
In practical applications, the number of segments S is unknown. We use
the ICOMP criterion proposed by Bozdogan (1993) to determine the number
of segments, where the number of segments that yields the minimum value of
ICOMP is selected (the usual likelihood-ratio tests are invalid, because cer-
tain regularity conditions are not satised). ICOMP is an information theo-
retic measure, that improves upon the traditional Akaike (and Consistent
Akaike) Information criterion by adding a correction for model complexity,
and thus controlling for the risks of over-parameterising the model. ICOMP
is dened for our model as:
ICOMP = 2 ln l Q ln (trace(R)) det(R); (6)
where Q is the number of parameters estimated and R is the estimated co-
variance matrix of the parameters. Additionally, the percentage of variance
explained, R
2
, and the entropy E
s
, are used to evaluate the models. R
2
is
dened as 1 minus the ratio of the likelihoods of the current model and
the null-model, where the latter model includes only an intercept (V
ij
=b
0
).
E
s
is a measure that indicates the separation of classes (E
s
=1
M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260 247
indicates complete separation; E
s
=0 indicates complete overlap), and is
dened as:
E
s
= 1
X
n
i=1
X
S
s=1
h
is
ln h
is
ln log(S): (7)
5. Application: Parameterization
5.1. Data collection
In this section we will provide an application of our model to a Gabor
Granger study for a brand in a category of products for personal care. The
brand name cannot be revealed because of the condential nature of the study.
In the study, data were collected from a mall-intercept sample of 377 fe-
male shoppers in The Netherlands. The interview procedure was as follows.
Subjects were not informed beforehand about the purpose of the study. They
were told that some questions were to be posed on their preferences with res-
pect to brands in the category of products for personal care. Nine dierent
price levels, varying from D 2.19 to 4.80 were called out to each of the re-
spondents. In order to obtain substantial price variation while minimizing the
number of prices oered to each respondent, six dierent sequences of nine
prices were used. Respondents were randomly assigned to one of those six se-
quences. The lowest prices in these six sequences were, respectively, D 2.19,
2.20, 2.29, 2.30, 2.39 and 2.40. The prices in a sequence were obtained by suc-
cessively adding D 0.30 to the lowest price. The rst price called out to a
respondent was in the middle of the price sequence, while subsequent prices
were presented in random order to prevent order eects. Monroe (1990),
p. 125, describes such randomization as the appropriate procedure to over-
come order eects in price-research methods. At each price called out, re-
spondents were required to respond with a statement whether they intend
to buy the brand (1) at that price or not (0). The question asking consumers
for their purchase intent was conditioned on a purchase of the category being
made in the subsequent month. Respondents could indicate not to buy the
brand at any of the prices called out.
Whereas specic theory as to exactly what constitutes a psychological price
level is currently lacking, we dened prices ending in D 0.99 as potential
psychological prices (cf. Friedman, 1968). Therefore, the prices called out
to the respondents included two psychological prices: D 2.99 and D 3.99.
248 M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260
In a small audit, conducted among 31 supermarkets in the region where
the GaborGranger study was done, we investigated the occurrence of price
setting at the two psychological price levels mentioned. More than 50% of the
(482) items in the personal care products were priced at levels of D 2.99 and
D 3.99. These results illustrate the frequency with which psychological price
setting is used in practice.
Reference price levels were operationalized in this study as the last price
paid for a brand within the product class, assessed through direct question-
ing.
3
A number of demographic, socio-economic and usage characteristics of
the respondents were collected: age in years, monthly income in D 1000, us-
age frequency of the product in times per week, and whether respondents had
previously bought the brand.
5.2. Expected eects
The expectations of the direction of most of the eects captured in the
model depend upon whether consumers see price as a monetary sacrice,
or as an indicator of quality, since in the presence of price-seeking behavior,
the price-response relation is reversed. Table 2 presents a summary of the ex-
pected eects.
The coecients b
1ks
(PRICEk, k = 0; . . . ; K), presenting the price-eects in
the dierent price ranges, are hypothesized to be negative, if price is perceived
as a monetary sacrice. If price is perceived as an indicator of quality, how-
ever, these coecients are positive, indicating that utility increases with price.
3
The asymmetric eect relative to the price last paid could not be estimated for those consumers who
did not remember the price last paid, or for whom the remembered price last paid was outside of the price
range in the study.
Table 2
Expectations of model coecients
Term Coecient Price as monetary sacrice Price as quality indicator
PRICEk b
1ks
) +
REFERP b
2s
+ )
PRLOSS b
3s
) +
PSYPRk b
0ks
)
M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260 249
Manufacturers benet from high reference or expected price levels. High
reference prices (REFERP) ensure that price decreases appear more attrac-
tive to consumers, and regular prices do not seem so unattractive (cf. Blatt-
berg and Neslin, 1990, p. 41). We therefore expect the sign for b
2s
to be
positive. If price is used as an indicator of quality, however, manufacturers
benet from low reference price levels. Here, such low price-expectations will
lead to higher prices being perceived as indicating higher quality across the
entire price range. The sign for b
2s
is therefore expected to be negative in such
situations.
According to the Tversky and Kahneman (1991) prospect theory, consum-
ers should react more strongly to price losses, which gives us reason to expect
b
3s
, the additional eect of a price loss (PRLOSS) to be negative. If consum-
ers use price as an indicator of quality (cf. Hardie et al., 1993), loss-aversion
may occur as well for perceived quality. Consequently, it may be expected
that for prices above the reference price (price losses) a gain in quality is per-
ceived. Therefore the b
3s
are expected to be positive in situations where price
is used as an indicator of quality.
The terms PSYPRk (for k = 1; . . . ; K) the discontinuity in utility at the
psychological price levels. If the kth psychological price causes such discon-
tinuities, b
0ks
is hypothesized to be negative, otherwise b
0ks
is zero. It is not a
priori clear what the eects of psychological prices are in the presence of a
price-perceived quality relationship.
5.3. Results
Our model was applied to the above GaborGranger data of 377 consum-
ers. We specied from S=1 to S=6 segments. Two knots were included in
the spline function, at the psychological price levels of D 2.99 and D 3.99,
respectively. Thus we have two knots and three price ranges: price 62.99;
3.00 < price < 3.99 and price P4.00. Table 3 shows the number of itera-
tions, the log-likelihood, ICOMP, R
2
and entropy (E
s
) criteria for each of
the S=1 to S=6 solutions. The ICOMP criterion indicated the S=5 seg-
ment solution to be optimal. However, the dierences between the S=4
and S=5 segment solutions were small. In the latter solution one segment
was split into two segments with similar interpretation and managerial impli-
cations. We report the S=4 segment solution, because it is more parsimoni-
ous. The four segments were proled by regressing the logit-transformed
posterior probabilities on the consumer descriptor variables.
250 M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260
R
2
is 0.435 for the S=4 segment solution. The value of E
4
of 0.913 indi-
cates that the segments were very well separated. The estimates of the param-
eters of the four segment solution are contained in Table 4. The parameters
of the four segments are well interpretable, and conform to our expectations
to a large extent, as will be detailed below. Fig. 1 displays the form of the
price-utility function in the four segments. (The gure was calculated from
the PRICE# and PSYPR# coecients, holding the other eects constant).
Segment 1 contains 30.6% of the sample and displays a negative price re-
sponse function (Fig. 1). In the lower price range demand is inelastic
(PRICE0). Above D 2.99 utility decreases faster with price (PRICE1,
p < 0.10). This decrease is enhanced in the highest price range (PRICE2,
p < 0.10). To this segment of consumers, price represents the amount of
money that must be sacriced in order to obtain the brand, and therefore
Table 4
Parameter estimates for the S=4 solution
Parameter /Segment 1 2 3 4
INTCPT 7.240
b
3.880
b
)20.960
b
)11.190
b
PRICE0 )0.012 )0.020
b
0.068
b
0.056
b
PSYPR1 (2.99) )0.022 1.987
b
0.794 0.520
PRICE1 )0.024
a
)0.032
b
)0.058
b
)0.028
PSYPR2 (3.99) )0.783
a
1.390 0.075 )3.710
b
PRICE2 )0.020
a
)0.009 )0.014
a
)0.072
b
PRLOSS )0.042
b
)0.025
b
0.143
b
0.035
b
REFERP 0.009
b
0.015
b
)0.011
b
)0.017
b
h
s
(relative size of segment s) 0.306 0.314 0.205 0.175
a
p < 0.10.
b
p < 0.05.
Table 3
Statistics of the S=1 to S=6 solutions
No. of Classes No of Iterations Log L ICOMP R
2
Entropy E
s
1 2 )2184.320 4447.256 0.094 )
2 14 )1801.777 3778.856 0.277 0.903
3 49 )1549.788 3354.552 0.376 0.910
4 28 )1383.251 3130.700
b
0.435 0.913
5 19 )1290.894 3052.293
a
0.465 0.911
6 33 )1242.535 3071.440 0.480 0.912
a
Denotes minimum ICOMP.
b
Denotes the solution selected.
M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260 251
prices aect purchase probabilities negatively. At D 2.99 there is no signi-
cant psychological price eect. Apparently at prices upto D 3.00 consumers
hardly perceive price-changes to be important. At D 3.99, there is a signi-
cant psychological price eect. Apparently, D 3.99 is perceived as a reduced
price relative to a possible fair price of D 4.00. At prices above D 4.00,
consumers are more sensitive to price changes, since the perceived sacrice
increases sharply. The eect of the reference price (REFERP) is positive, in-
dicating that higher expected prices lead to higher intentions of buying the
brand. This was expected to occur in the presence of negative linear price ef-
fects (PRICEk). If the reference price is higher, the whole price-sequence of-
Fig. 1. The price-utility function in segments 14, personal care brand.
252 M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260
fered appears to be less of a monetary sacrice. The coecient of PRLOSS is
negative, which indicates that prices above the reference price more negative-
ly aect purchase probabilities than prices below the reference price. This was
expected on the basis of prospect theory, according to which consumers react
more strongly to price losses (higher prices) than to price gains. Membership
in segment 1 is associated with lower income, and not previously having pur-
chased the brand.
In Segment 2, comprising 31.4% of the sample, utility also decreases as a
function of price. The decrease is not monotonic: in the second price range
(PRICE1) it is stronger than in the rst price range (PRICE0). The price
response function decreases at the same rate in the third price range
(PRICE2). To these consumers price represents a monetary sacrice. For
this segment the psychological price eects are quite dierent from those
in segment 1. A signicant psychological price eect is present at D
2.99 (PSYPR1). There is no psychological price eect at D 3.99
(PSYPR2). Apparently, at D 2.99 price is perceived to be reduced relative
to a fair price of D 3.00. The coecient of REFERP is positive, indicating
that higher reference price leads to a higher probability of indicating a buy,
which is consistent with expectations and the ndings in the rst segment.
Just as in segment 1, we nd a negative coecient for PRLOSS. Note, how-
ever, that loss aversion is less strong in this segment as compared to seg-
ment 1. Consumers that have previously purchased the brand have a
higher probability of belonging to this segment.
Segment 3 is smaller than the segments 1 and 2 and contains 20.5% of
the sample. Contrary to segments 1 and 2, in this segment utility increases
with price in the rst price range (PRICE0). The price responses in the sec-
ond and third price range are almost inelastic. Consumers in this segment
use price as an indicator of quality. Note that consumers apparently
trade-o perceived-quality and perceived-sacrice. The priceperceived
quality relationship is primarily found in the lowest price range, but in
the higher price ranges the monetary sacrice becomes more important,
so that the positive price eect levels o. No signicant psychological price
eects are observed in this segment. When the price-response is positive,
psychological prices may not be perceived as price reductions relative to
a fair price, so that transaction utility is not increased at such price levels.
Consistent with the positive price relation, the coecient of reference price
(REFERP) is negative, which was expected. This indicates that low price
expectations (i.e. a low price last paid) will lead to prices across the entire
oered price sequence being perceived as indicating higher quality, which
M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260 253
results in higher purchase probabilities. The coecient of PRLOSS is pos-
itive, which conrms our expectations on the basis of prospect theory. Since
price is an indicator of quality in this segment, higher prices losses corres-
pond to perceived quality gains, so that the term PRLOSS, representing
higher prices relative to the reference price, indicates perceived quality
gains. The coecient for prices above the reference price (PRLOSS) is pos-
itive, indicating a stronger priceperceived quality relation for these prices.
This nding supports prospect theory, which predicts the eect of a per-
ceived quality loss to be greater than that of a corresponding quality gain.
The large negative intercept indicates that consumers in this segment have a
much lower overall probability of choosing the brand, as compared to seg-
ments 1 and 2. Segment membership does not display strong associations
with consumer descriptors, although consumers in this segment tend to
have a somewhat higher (p < 0.10) income. This corresponds to the nd-
ings of e.g. Steenkamp (1989), p. 205, that consumers with higher incomes
are more quality conscious.
Segment 4 is the smallest segment and contains 17.5 percent of the
sample. Here a clear bell-shaped price response function is observed, as
was reported to occur in the original studies of Gabor and Granger
(1966). At lower price levels consumers' utility signicantly increases with
price (PRICE0). Since PRICE1 is not signicant, the increase appears to
continue in the middle price range. In the highest price range the price-
response is clearly negative. This segment of consumers use price as an
indicator of quality in the lower and middle price range. For prices above
D 3.00 the importance of the perceived sacrice is apparently greater
than that of perceived quality, and the price response function decreases.
At D 2.99, where the price-response function is positive, there is no sig-
nicant psychological price eect. This nding is consistent with the re-
sults for segment 3, where also no psychological price eects were
found for prices that were used as an indicator of quality. At D 3.99
there is a signicant psychological price eect. Note that the psychological
price eect is very large indeed, and that this eect occurs at the onset of
the negative price-response function. Consistent with our expectations the
coecient of REFERP is negative. Just as in segment 3, and consistent
with prospect theory, the coecient for PRLOSS is positive, indicating
that prices below the price last paid are less strong indicators of quality.
Judged by the magnitude of the coecients, the loss aversion is much
greater in segment 3 than in segment 4. Again, as evidenced by the inter-
cept, the overall probability of choosing the brand is low as compared to
254 M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260
segments 1 and 2. Consumers that have not previously purchased the
brand have a higher probability of belonging to this segment. This nding
may indicate that these consumers are less sure of the quality of the
brand, and therefore depend more heavily on price as a quality cue
(Steenkamp, 1989, p. 95).
6. Validation
In this section the predictive validity of our model is compared to the la-
tent class logit model with linear price eects. The deterministic part of the
utility function of this model has the following, simple structure
V
ij[s
= c
00s
c
10s
p
ij
; (8)
where for an individual consumer in segment s, c
00s
denotes the intercept and
c
10s
represents the linear price eect. This is the model proposed by Kama-
kura and Russell (1989) in a binary context. Two validation studies were car-
ried out, establishing internal and external validity, respectively.
First, as there were no holdout data available, we randomly eliminated
10% of the data of the GaborGranger study, estimated the models on the
remaining data, and predicted choices for the eliminated price levels. To this
end, Eq. (5) was used. This procedure was repeated ve times for each model.
To evaluate the predictive accuracy we calculated the correlation between
predictions and holdout data. This dependent measure was analyzed by uni-
variate Analysis of Variance to test dierences among the methods (the AN-
OVA has 90% power to detect eects that account for approximately 15% of
the variance, Cohen, 1991).
The predictive t of the latent class spline model was signicantly better
than that of the latent class linear model: the validation correlation for the
former was 0.636, the validation correlation for the latter 0.603 (Standard er-
ror of the dierence, SED=0.005, p < 0.001). The results support the inter-
nal validity of our results. The proposed spline model is superior to the model
with linear price eects only, although the dierence is modest. The improve-
ment is about 5% in the present application, and will depend on the extent to
which the price response functions are nonlinear.
Second, the external validity of our model for the eects of psycholog-
ical pricing in GaborGranger price studies was investigated. This implies
the validation of the data collection procedure which measures purchase
intentions and the validation of the model which has been calibrated using
M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260 255
these data. Market shares and prices were obtained for the brand in 12
two-monthly periods for a representative sample of stores in The Nether-
lands, from A.C. Nielsen. The period concerns two years, covering the pe-
riod in which the data for the GaborGranger study were collected. The
market shares were adjusted for price-eects of competing brands using re-
gression methods, and one extreme observation was eliminated from the
data. The two models were used to predict the brand's market shares in
the (remaining) 11 periods in the basis of aggregate level prices for each
segment.
The correlations of observed and predicted shares were 0.500 for the latent
class linear logit model, and 0.517 for the latent class spline model. The cor-
relation of market shares and prices calculated directly from the two-monthly
data was 0.479. The predictive validity of both latent class models thus
outperform the predictive t of a regression model tted on the market shares
themselves. The correlations conrm the somewhat better predictive validity
of the latent class spline model. The improvement of the spline model is of the
same order of magnitude as found above. The results support the external va-
lidity of the GaborGranger procedure in conjunction with the proposed
model as a procedure for assessing price sensitivity.
7. Discussion and conclusions
Below we discuss the substantive ndings derived from our model.
7.1. Priceperceived quality
Our study has yielded supportive evidence of a number of hypotheses and
previous empirical ndings. It has provided additional support for consum-
ers' use of price as an indicator of quality, in two out of four segments of
the market. One of these segments displayed a bell-shaped price response
function. Such a response curve was already observed by Gabor and Granger
(1966) in their original studies (see also Monroe, 1990, p. 114). The bell-
shaped curve can be explained as a situation where at lower price levels the
importance of perceived quality, as inferred from price, is higher than the im-
portance of perceived sacrice, whereas the reverse holds at higher price
levels.
In our application consumers appeared to be very heterogeneous in their
reaction to price. This is in accordance with the ndings of e.g. Zeithaml
256 M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260
(1988), and Dickson and Sawyer (1990). The allocative (price as perceived
monetary sacrice) and informative (price as an indicator of quality) aspects
of price were found to operate simultaneously in two consumer segments,
while the allocative aspects clearly dominated in two other segments. This
nding corresponds with the ndings of Lichtenstein and Burton (1989)
who reported both priceperceived-quality groups of consumers and no-
priceperceived-quality groups of consumers.
7.2. Loss aversion
The application of our model has provided additional support for the ex-
istence of loss aversion with respect to price (see Kahneman and Tversky,
1979). For all four segments the parameter estimates of PRLOSS
(
^
b
3s
; s = 1; . . . ; 4) are signicantly dierent from zero. We have provided
supportive evidence for the ndings of Hardie et al. (1993), that consumers
display loss aversion with respect to quality as well. In the segments with a
positive price response function, emanating from the use of price as an indi-
cator of quality, such loss aversion with respect to quality was inferred. Con-
sumers in these segments relied less on prices below the reference price
(indicating quality loss) as indicators of quality. Research is needed to further
substantiate these ndings (see Bell and Lattin, 1993). Segments of consum-
ers diered in the extent to which they displayed such loss aversion, both with
respect to price and with respect to quality.
7.3. Psychological price eects
Finally, and most importantly, the results have demonstrated that psycho-
logical price setting may cause discontinuities in demand (see Monroe, 1973).
Whereas Blattberg and Wisniewski (1987) already demonstrated such eects
on the aggregate level using scanner data, our study includes psychological
prices in a consumer choice model and demonstrates the eects at the disag-
gregate level. Although the existence of discontinuous eects on demand is
supported by our results, it appears from our study that they do not occur
at all psychological price levels and for all market segments. As has been es-
tablished in the literature on deal-proneness (cf. Blattberg and Neslin, 1990,
pp. 7781), consumers may vary in the extent to which they are sensitive to
price reductions. Segments exist in which the perceived price reduction at the
psychological price level does not lead to an increase in transaction utility.
Also, it appeared that consumers may perceive a price reduction at one
M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260 257
psychological price level, but not at another. From these ndings, it may be
concluded that the importance of psychological price setting may be overes-
timated in pricing decisions in the market place.
7.4. Limitations and future research
We conclude that the approach proposed here is a valuable tool in the inv-
estigation of consumers' price sensitivity and we argue that the benets of our
model accrue from its use in conjunction with the GaborGranger method
and related methods used in pricing research (see Monroe, 1990, pp. 106
137). A limitation of the GaborGranger procedure is that the eects of pric-
es of competitive brands are not considered explicitly. Our model can in fu-
ture research be extended to multibrand situations. The external validity of
the proposed procedure was supported by our validation study on store-level
market shares, although the gain in predictive validity over competing proce-
dures was modest. Future research should corroborate these ndings. It is
clear that the external validity of the proposed procedure can yet be further
enhanced by designing in-store experiments in which actual choices are ob-
served in response to price changes, e.g. through check-out scanning devices.
The analyses of such experimental scanner data may provide evidence of the
existence of bell-shaped price-response curves, loss aversion, and psycholog-
ical pricing eects in actual market situations. We leave this for future re-
search.
Acknowledgements
We wish to acknowledge A.C. Nielsen, Netherlands for providing
aggregate-level sales data. We also thank the reviewers and especially
Alan J. Mac Fadyen for helpful suggestions on an earlier version of this
paper.
References
Bell, D.R., Lattin, J.M., 1993. Loss aversion and heterogeneity in price sensitivity. Research paper no.
1259, Graduate School of Business, Stanford University.
Ben-Akiva, M., Lerman, S.R., 1985. Discrete Choice Analysis. MIT Press, London.
Blattberg, R.C., Buesing, T., Peacock, P., Sen, S.K., 1978. Identifying the deal prone segment. Journal of
Marketing Research 25, 369377.
258 M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260
Blattberg, R.C., Neslin, S.A., 1990. Sales Promotion, Concepts, Methods and Strategies. Prentice-Hall,
Englewood Clis, NJ.
Blattberg, R.C., Wisniewski, K.J., 1987. How retail price promotions work: Empirical results. Working
Paper 42, University of Chicago.
Bozdogan, H., 1993. Mixture-model cluster analysis using model selection criteria and a new
informational measure of complexity. In: Bozdogan, H. (Ed.), Multivariate Statistical Modelling 2.
Kluwer Academic Publishers, Dordrecht, The Netherlands.
Brenner, G.A., Brenner, R., 1982. Memory and markets, or why are you paying $ 2.99 for a widget?
Journal of Business 55 (1), 147158.
Cohen, J., 1991. Statistical Power Analysis for the Behavioral Sciences. Friedman Erlbaum, Hillsdale, NJ.
Dickson, P.R., Sawyer, A.G., 1990. The price knowledge and search of supermarket shoppers. Journal of
Marketing 54 (July), 4253.
Elrod, T., Winer, R.S., 1982. An empirical evaluation of aggregation approaches for developing market
segments. Journal of Marketing 46 (Fall), 6574.
Emery, F., 1970. Some psychological aspects of price. In: Taylor, B., Wills, G. (Eds.), Pricing Strategy.
Brandon/System Press, Princeton, NJ, pp. 98111.
Erickson, G.M., Johansson, J.K., 1985. The role of price in multi-attribute product evaluations. Journal of
Consumer Research 12, 195199.
Friedman, L., 1968. Psychological pricing in the food industry. In: Phillips, A., Williamson, O.E.
(Eds.), Prices, Issues in Theory, Practice and Public Policy. University of Pennsylvania Press,
Philadelphia.
Gabor, A., Granger, C.W.J., 1961. On price consciousness of consumers. Applied Statistics 10, 170188.
Gabor, A., Granger, C.W.J., 1966. Price as an indicator of quality: Report on an inquiry. Economica 32,
4370.
G on ul, F., Srinivasan, K., 1993. Modelling multiple sources of heterogeneity in multinomial logit models:
Methodological and managerial issues. Marketing Science 12, 213229.
Hanna, N., Dodge, H.R., 1995. Pricing, Policies and Procedures. MacMillan Business, London.
Hardie, B.G.S., Johnson, E.J., Fader, P.S., 1993. Modelling loss aversion and reference dependence eects
on brand choice. Marketing Science 12, 378394.
Helson, H., 1964. Adaptation Level Theory. Harper & Row, New York.
Kahneman, D., Tversky, A., 1979. Prospect theory: An analysis of decision under risk. Econometrica 47
(March), 263291.
Kalwani, M.U., Yim, Ch.K., 1992. Consumer price and promotion expectations: An experimental study.
Journal of Marketing Research 29, 90100.
Kalwani, M.U., Yim, Ch.K., Rinne, H.J., Sugita, Y., 1990. A price expectations model of customer brand
choice. Journal of Marketing Research 27, 251262.
Kamakura, W.A., Russell, G.J., 1989. A probabilistic choice model for market segmentation and elasticity
structure. Journal of Marketing Research 26, 379390.
Lichtenstein, D.R., Burton, S., 1989. The relationship between perceived and objective price quality.
Journal of Marketing Research 26, 429443.
Lichtenstein, D.R., Ridgway, N.M., Netemeyer, R.G., 1993. Price perceptions and consumer shopping
behavior: A eld study. Journal of Marketing Research 30, 234245.
Liefeld, J., Heslop, L.A., 1985. Reference prices and deception in newspaper advertising. Journal of
Consumer Research 11, 868876.
Monroe, K.B., 1973. Buyers' subjective perceptions of price. Journal of Marketing Research 10, 7080.
Monroe, K.B., 1990. Pricing: Making Protable Decisions. McGraw-Hill, New York.
Rao, V.R., 1993. Pricing models in marketing. In: Eliashberg, J., Lilien, G.L. (Eds.), Handbooks in
Operations Research and Management Science. Vol. 5. Marketing. North-Holland, Amsterdam, pp.
517549.
Schindler, R.M., 1991. Symbolic meaning of a price ending. Advances in Consumer Research 18, 794801.
M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260 259
Smith, P.L., 1979. Splines as a useful and convenient statistical tool. The American Statistician 33, 5762.
Steenkamp, J.B.E.M., 1989. Product Quality. Van Gorcum, Assen.
Thaler, R., 1985. Mental accounting and consumer choice. Marketing Science 4, 199214.
Tversky, A., Kahneman, D., 1991. Loss aversion and riskless choice: A reference dependent choice model.
Quarterly Journal of Economics 106, 10391061.
Tversky, A., Simonson, I., 1993. Context-dependent preferences. Management Science 32, 645659.
Wedel, M., DeSarbo, W.S., 1994. A review of recent developments in latent class regression methods. In:
Bagozzi, R.P. (Ed.), Advanced Methods of Marketing Research, pp. 352388.
Wildt, A.R., McCann, J.M., 1980. A regression model for market segmentation studies. Journal of
Marketing Research 17, 335340.
Winer, R.S., 1986. Reference price models of brand choice for frequently purchased products. Journal of
Consumer Research 13, 250256.
Zeithaml, V.A., 1988. Consumer perceptions of price, quality and value: A means end model and synthesis
of evidence. Journal of Marketing 52 (July), 222.
260 M. Wedel, P.S.H. Leeang / Journal of Economic Psychology 19 (1998) 237260