Professional Documents
Culture Documents
Barry L. Bayus
Uniuersiry of North Carolina, Chapel Hill, NC 27599, USA
Sachin Gupta
Cornell Vniuersi@, Ithaca, NY 14853, USA
Final version
received
April
1992
Correspondence
to: B.L.
School, University of North
USA
Bayus,
Kenan-Flagler
Business
Carolina, Chapel Hill, NC 27599,
* This research
was conducted
while the first author was a
faculty member at Cornell University. Funding for the data
collection
was provided
by the Johnson
Schools Institute
for Research
in Marketing
at Cornell University.
Special
thanks are extended
to Peter Dickson, Dick Wittink, the
editor and two reviewers for their comments
on an earlier
draft.
Intern. J. of Research
North-Holland
in Marketing
9 (1992) 257-267
0167-8116/92/$05.00
0 1992 - Elsevier
Science
Publishers
reserved
258
replacements,
and purchases of additional
units (e.g., Pickering, 1981; Winer, 1985a),
few empirical studies have considered replacement behavior. Attention has tended to
focus on the influence of household characteristics (e.g., income, working wife) on purchase, while product characteristics
(e.g.,
condition of a currently owned unit) are generally ignored. By and large, matching empirical data with econometric models has resulted in disappointing results in terms of
overall statistical fits (e.g., Winer, 1985b) and
predictive ability (e.g., McNeil, 1974). An
exception is Bayus et al. (1989) in which very
good forecasting results for color televisions
were achieved by modeling the separate demand components of sales.
In this paper, we (1) develop a descriptive
model for replacement intentions based on
variables representing product and household characteristics, and (2) use cross sectional data for a set of home appliances to
empirically estimate this replacement model.
Our results indicate that the perceived condition of the currently owned unit and its age
are significantly related to replacement intentions. Whether or not a spouse is working
and the expected future household financial
situation are also significant explanatory
variables. Importantly, we find no intrinsic
product specific effects (i.e., the product specific intercepts in a logit model are not statistically significant).
In the next section, the related literature
for consumer durable demand is reviewed. A
model of durable replacement intentions is
then developed. The sample and data collected are next described, and the dependent
and independent variables are defined. Results based on multivariate logistic regression
analyses are then discussed. Finally, implications of these findings and suggestions for
further research are discussed.
2. Related literature
Major research thrusts in the consumer
durables area have included the following
topics: (1) information search and decision
making (e.g., see the review in Beatty and
Smith, 19871, (2) planning of purchases and
the acquisition sequence of durables (e.g.,
Kasulis et al., 1979; Dickson et al., 1983;
Bayus and Rao, 1989), and (3) post-purchase
behavior (the disposition of durables-e.g.,
Jacoby et al., 1977; DeBell and Dardis, 1979;
consumer dissatisfaction and complaint behavior-e.g.,
Tse and Wilton, 1988). None of
these efforts, however, explicitly studies consumer replacement behavior.
Two recent efforts investigate the timing
of replacement purchases for single durable
products. Antonides (1990) empirically studies the replacement behavior (conditional on
a failure) for washers and finds that income
and household size are positively related to
replacements (as opposed to repairing the
item). Bayus (1991a) reports that demographic characteristics, attitudes and perceptions, and search behavior of consumers
trading in an automobile are significant explanatory variables of the timing of automobile replacements.
Replacement demand is included in the
general conceptual
model for consumer
durable demand
proposed
by Pickering
(1981) and revised by Winer (1985a). Based
on a review of the major efforts concerned
with predicting durable demand, Pickering
(1981) proposes a behavioral model in which
demand is a function of purchase expectations (i.e., intentions). Purchase expectations
in turn are modeled as a function of personal
financial circumstances
and expectations
(which is related to consumer confidence),
personal
circumstances
(e.g., household
move, marriage), orders of acquisition of new
259
find that durable purchase intentions are linearly related to purchase behavior.
We can specify the functional relationship
for replacement intentions as
R(X) =g(X)
+e,
(1)
where g(X) is the deterministic component
of replacement intentions and is dependent
on the set of explanatory variables X. Here,
E is the error term (stochastic component) of
intentions. The set of variables in X include
characteristics
of the product
currently
owned and household characteristics. Following the conceptual model proposed by
Pickering (1981), possible effects due to the
stock of durables owned on replacement intentions for a particular product are not considered (i.e., we assume the indirect utilities
of product replacement are separable). Relaxing this assumption is left for future research.
Assuming that the error term and is iid
according to a Type I extreme value distribution, (1) can be transformed into the familiar
logistic function. Letting P denote the probability of replacing a durable, the model is
P=l/.[l+exp(-CY-X/3)],
3. A model of durable
replacement
inten-
tions
(2)
where X is the vector of explanatory variables, j3 is the coefficient vector, and a is an
intercept term. We note that the marginal
impact on the replacement probability due to
changes in an explanatory variable (say xi) is
pjP(l - PI. Because P is a function of several explanatory variables, the marginal impact of a single variable thus depends on the
values of the other variables. In other words,
the interactive effects of several explanatory
variables are implicitly included in (2). Since
the dependent variable we will analyze is
binary (likely or not likely to replace),
As part of the empirical
replacement
purchase
study described
data
for
variables
refrigerators
for replacement
and
coffee
data. Re-
and pur-
260
intentions
replacement
distributions
for several
durables, Weibull distributions with y close
to 2 provide very good fits (Bayus, 1988;
1991a). Thus, the Rayleigh distribution (i.e.,
a Weibull with y = 2) is used to model
durable hazard rates. For the Rayleigh, h(t)
= 2et.
3.2 Household characteristics
Although there is some disagreement over
the empirical significance of specific measures, previous studies indicate that the general variables of stage in the family life
cycle and need for convenience are important determinants of durable purchases.
Strober and Weinberg (1977) and Weinberg
and Winer (19831, for example, find that
younger households are more likely to be in
the durable acquisition stage (since they tend
to own fewer durables). Winer (1985b) finds
that age of household head exhibits a nonlinear relationship (i.e., U-shaped) to purchase
probability. His results suggest that replacement demand is associated with older households. Households in the later life cycle
stage are more likely to need a replacement
due to cumulated usage over time and/or
may be better able to afford a replacement
(since young children are generally not present in the household). Contrary to Strober
and Weinberg (19771, Weinberg and Winer
(19831, and Winer (1985b), Kim (1989) finds
that wifes employment is significantly related to durable ownership, even after controlling for income and life cycle effects.
Finally, there is general agreement that a
recent move is associated with durable purchases (e.g., Winer, 1985b; Wilkie and Dickson, 1985). Based on this literature, we hypothesize that older households, households
with working wives, and households that have
recently moved are more likely to have positive replacement
intentions for currently
owned durables.
/ Consumer
durable replacement
Table 1
Explanatory
261
intentions
variables
and hypotheses
Variable
Definition
Hypothesized
direction
of influence
CONDITION
reported condition of
unit currently owned
negative
HAZARD
calculated
hazard rate
of unit currently owned
positive
SPOUSE
positive
HHAGE
reported age of
household head
negative
HHAGE
squared age of
household head
positive
MOVED
positive
INCOME
positive
EXP_
FINANCES
expected future
household financial
situation
positive
262
Table 2
Sample profile
Respondents
Nonrespondents
56 years
59%
53 years
57%
64%
31%
5%
76%
30%
407
62%
31%
7%
80%
40%
154
Total
55 years
58%
64%
31%
5%
77%
33%
561
cleaner, and coffee maker. This set of appliances was selected to represent a range of
product categories. Prior research (Pickering
et al., 1973; Bayus and Carlstrom, 1990) indicates that these appliances can be separated
into three groups based on perceptual measures: major home appliances (stove, refrigerator, washer, vacuum cleaner), housewares (coffee maker), and entertainment
items (color TV, VCR).
Information on the age (in years) and condition (three-point scale: good, fair, poor) of
currently owned units was also collected. Demographic data such as age of household
head (in years), income (eleven-point
ordered scale), and whether a spouse was employed (if married) were also available. Finally, information on the length of residence
at the current address and expected future
household financial situation was collected
(three-point scale: better, same, worse than
now). *
Completed questionnaires were received
from 407 households owning their home or
condominium, representing a response rate
of over 70%. The percentage of households
owning these eight home appliances was generally over 70%. Demographic profiles of the
entire panel and the samples of respondents
and nonrespondents
are given in Table 2.
Generally speaking, the sample of respondents is a little older and has a smaller
percentage of households with a working
spouse than the nonrespondents.
No statistically significant differences exist in terms of
household income, education of the household head, and the percentage of married
households.
Most of the variables in Table 1 have
natural definitions based on the survey quesintentions
were ditions. Replacement
chotomized into positive (1 = definitely or
likely) and negative (0 = not likely or definitely not) intentions by combining response
categories. 3 Since unit condition was reported using a three-point scale, two dummy
variables are used to represent perceived unit
condition (GOOD = 1 if unit condition is
good, 0 otherwise; POOR = 1 if unit condition is poor, 0 otherwise). Similarly, expected
future household financial situation is represented by two dummy variables (BETTER =
and Gianotten
(1990) use a fiveto much worse than now) to
financial situation. In their analyto have equal interval properties.
such a scale has not been estabordered scale, but only assume
Since replacement
intentions
were collected
using a fourpoint scale, estimation
procedures
such as multinomial
logit
or probit, or ordered
logit could also be used. However,
other analyses (using an ordered logit model) not reported
in this paper indicate
that the main conclusions
remain
unchanged.
Thus, for ease of interpretation
we report results based on a binary dependent
variable.
age (years)
intentions
intentions
size
a From Appliance,
Of all households
tion for each of the eight products considered individually; and (2) the product characteristics summarized in Table 3 suggest that
the expected relationship between unit age,
unit condition, and intentions are similar for
each separate product. Although the model
developed in Section 3 does not consider
differing effects by product category, we allow for intrinsic product specific effects in
the logistic regression model (i.e., product
specific constants; see Chintagunta, 1992) by
defining six appropriate
dummy variables:
STOKE (1 if stove, 0 otherwise), FRlDGE (1
if refrigerator, 0 otherwise), WASHER (1 if
clothes washer, 0 otherwise), CW (1 if color
TV, 0 otherwise), VCR (1 if VCR, 0 otherwise), and VACUUM (1 if vacuum cleaner, 0
otherwise). Here, coffee maker is assumed to
be the base product (i.e., if all product
dummy variables equal zero). Everything else
being equal, if a particular appliance has
some inherent replacement
importance or
priority in the household, then we expect
the dummy variable for that product to be
statistically significant.
5. Resdts
5.1 Model fit
studied
Stove
Mean replacement
263
intentions
Refrigerator
Washer
Color
TV
VCR
Vacuum
cleaner
Coffee
maker
15
13
12
11
15.8
11.0
14.5
9.0
11.5
8.0
9.6
5.7
4.5
2.9
11.2
7.2
4.8
3.8
4b
64
2
30
3
100
5
80
3
100
4
77
0
100
360
368
373
259
366
35
371
September
1984.
owning a stove in good condition.
4% had positive
replacement
intentions.
264
/ Consumer
durable replacement
Table 5
Estimation
infentions
results
Coefficient
*-statistic
5.88
7.78
8.44
7.81
7.90
8.08
8.07
0.07
0.12
0.14
0.12
0.12
0.13
0.13
CY
STOVE
FRIDGE
WASHER
CTV
VCR
VACUUM
CONDITION
GOOD
POOR
HAZARD
SPOUSE
HHAGE
HhYGE
=
MOVED
INCOME
EXP_ FINANCES
BETTER
WORSE
Model chi-square
(d.f.)
- 2.64
1.68
3.71
0.52
0.00
0.00
0.30
0.02
569.04 *
13.35 *
5.94 *
0.00
0.15
1.33
0.20
a
0.56
-0.31
609.04 *
(16)
12.73 *
a Chi-square
statistic calculated
as the difference in model chi
square with and without the two dummy variables
representing this variable (e.g., see Hosmer and Lemeshow, 1989).
* Significant at 0.01 level.
Table 4
Overall model
fit
Actual
replacement
intentions
Total
Positive
Negative
Positive
Negative
111
44
169
1808
280
1852
Total
155
1977
2132
265
show a significant income effect for replacement intentions. However, this result may be
due to the sample analyzed (i.e., respondents
that owned their home) and the inclusion of
the spouses working status. Finally, contrary
to our hypothesis, whether a household has
recently moved is not significantly related to
replacement intentions. We note that this
finding may be due to the fact that very few
households in the sample analyzed reported
a recent move.
5.3 Elasticities
Table 6
Elasticity
estimates
for positive
Variable
HAZARD
(for 1 year increase in unit age)
CONDITION
(from good or fair to poor)
SPOUSE
(from not working to working)
EXP_ FINANCES
(from poor or fair to better)
replacement
intentions
Elasticity
2.91
471.89
18.95
22.70
6. Discussion
and conclusions
266
267