Professional Documents
Culture Documents
DAVID C. SCHMITTLEIN
University of Pennsylvunia
Abstract
The purpose of this article is to describe shopping trip behavior empirically using shopping trip
data. A better understanding of this aspect of the purchase process can assist in generating testable
hypotheses as to how the shopping trip decision may influence other choice decisions.
1. Introduction
In studying the purchase process, marketing researchers have focused on the con-
sumer’s brand choice decision (Corstgens and Gautschi, 1983; Meyer and Kahn,
1990), the timing of the brand purchase (Morrison and Schmittlein, 1988), the
effects of store knowledge and time available for shopping on purchase behavior
(Park, Iyer, and Smith, 1989), and how various marketing influences (e.g., pro-
motions) affect these elements of the purchase process (Guadagni and Little,
1983; Gupta, 1988). Before a consumer can choose among brands, however, he or
she must first decide to enter the store to shop. If the decision to go shopping and
the brand choice decision are not statistically independent, then it is incorrect to
model these decisions separately. Ignoring the first decision (or the shopping de-
cision) in this hierarchical purchase process can result in inconsistent or inefficient
model parameter estimates at later stages of the process. (For an analogous ex-
ample of such hierarchical effects see Krishnamurthi and Raj, 1988). The purpose
of this paper is to describe shopping trip behavior empirically using IRI shopping
trip data.’ With a better understanding of this aspect of the purchase process, we
can generate testable hypotheses as to how the shopping trip decision may influ-
ence other choice decisions.
Briefly, our analyses yielded the following general conclusions about shopping
trip behavior.
1. The distribution of times between shopping trips and the distribution of times
between purchases of a specific brand are significantly different. Whereas both
interpurchase times and shopping trip histograms show evidence of 7-day (or
multiples of 7 days) cycles, these effects are much stronger for brand interpur-
chase times than for intershopping trip times.
2. There is evidence that these 7-day peaks in the shopping trip data are created
by strong preferences for shopping on a’specific day of the week rather than
based on when the product is used up.
56 BARBARA E. KAHN AND DAVID C. SCHMITTLEIN
In the remainder of the paper we discuss each of these findings in detail, and point
out the implications of the results for models of choice, purchase timing, or pro-
motional-effects.
Perhaps the best way to present a single snapshot of shopping trip behavior is
through an aggregate level histogram of intershopping trip times. The histogram
in figure 1 was created by randomly choosing one time between shopping trips for
each of the panelists in the IRI shopping trip subsample (n = 1442). This histogram
shows marked peaks at 7 days and at 14 days as might be expected, but also shows
a large peak at 24 days.
This intershopping trip times histogram is very different from interpurchase-
times histograms generated by Dunn, Reader, and Wrigley (DRW), 1983, for prod-
uct categories such as toilet tissue, baked beans, and instant coffee. The DRW
interpurchase times histograms show very strong peaks at 7 days, 14 days, 21
days, etc., but show no peak at 2-4 days. These differences between the two types
of histograms are intriguing and suggest research questions, but before drawing
any conclusions about the nature of the differences, we first ascertained that
We believe that there are two plausible explanations for the intershopping trip
histogram to be smoother than the histograms for interpurchase times. First, if
the 7-day peaks occur in the interpurchase-times data because the decision to
purchase a product correlates with what day of the week it is (e.g., Friday is fish
day), and if this day of the week varies across product classes, (e.g., Friday may
be fish day but Saturday is canned goods day), then pooling across product cate-
gories to the shopping trip level will result in a smoother histogram. In this paper,
we are only studying one product category and so cannot test this hypothesis per
se; however, we can investigate whether there is some general relationship be-
tween when consumers choose to go shopping and preferences for a specific day
of the week.
Second, it is possible that the mode at 2-4 days in the shopping trip histogram
represents shopping trips that are fill-in or quick trips (Frisbie, 1980; Kollat and
58 BARBARA E. KAHN AND DAVID C. SCHMITTLEIN
Willett, 1967; McKay, 1973). Thus consumers may make a major shopping trip
weekly (e.g., Monday) and also a minor fill-in trip weekly (e.g., Friday before the
weekend). Or, these fill-in trips could be unplanned and thus irregularly pat-
terned. The lack of a mode at 2-4 days in the product interpurchase times histo-
gram could exist because:
1. the product categories that have been studied (e.g., toilet tissue, baked beans,
instant coffee, and crackers) are not bought on fill-in trips, or
2. all product categories generate occasional spur-of-the-moment stock-outs ne-
cessitating a quick run to the store, and these stock-outs occur independently
for different categories. The shopping trip then represents a superposition of
all the individual product class spur-of-the-moment trips. Any time a spur-of-
the-moment need arises for any product class, it shows up in the shopping trip
data thus accumulating to a relatively large number of occurrences. Whereas a
spur-of-the-moment trip shows up infrequently for any specific product class.
3. Day-of-the-week effect
Baseline
% of shopping probabil.
Individual patterns trips (s.e.) (simulated) Average amount spent (s.e.)
% of all
shopping trips Average amount % as most likely % as least likely
Day of week (s.e.) spent (s.e.) shopping day shopping day
ply looking at shopping trips by day of the week at the aggregate level, one ob-
serves a slight preference for Saturdays and Fridays but no clear pattern emerges.
We have found that households have different shopping trip loyalties for different
days of the week. It is a question for future empirical study whether purchases
made on favorite-day shopping trips differ from purchases made on nonfavorite-
day shopping trips. (It might also be interesting to test how resistent to change
these favorite-day loyalties are.) It is plausible that there may be differences in
price-sensitivity, choice patterns, response to promotions, etc., on shopping trips
made on different types of days. Models of choice and promotion effects may
therefore be missing an important explanatory variable that is easy to measure
and to include.
These findings also have implications for stochastic analyses of purchase timing
models. Currently, there are no models that explicitly include this strong day-of-
the-week phenomenon (although it is discussed in Dunn, Reader, and Wrigley,
1983; and Kahn and Morrison, 1989). Since some researchers (e.g., Grover and
Rao, 1988) have used interpurchase times to try and understand market structure,
60 BARBARA E. KAHN AND DAVID C. SCHMITTLEIN
NO. OF TRIPS/HOUSEHOLD
As discussed earlier, the aggregate histogram of shopping trip times suggests the
existence of fill-in shopping trips, or quick trips. Similarly to Frisbie (1980), we
distinguish between quick and regular shopping trips by the amount of money
spent on each. However, our cutoff rule is more individually tailored to the habits
of each household. Sample histograms of the amount spent on each shopping trip
across consumers were examined and two dominant patterns were discovered.
These two patterns are illustrated in figures 4 and 5. In order to determine the
number of quick and regular shopping trips for each household, we map each
panelist’s histogram of amounts spent onto one of the two graphs shown in figures
4 and 5 depending upon which is appropriate (i.e., whether a second mode occurs
in the histogram). Thus quick trips are characterized by amounts spent to the left
or equal to the arrow (as shown in the graphs) and regular trips by amounts spent
to the right of the arrow. Table 2 shows the distribution across households of these
cutoff values.
These cutoff values were found to be significantly related to several demo-
graphic variables (namely, income, family size, and age of the adult female in the
household). In general, the lower cutoff values are associated with lower incomes,
small families, and older female heads of the household.3
SHOPPING TRIP BEHAVIOR 61
$5.00 20%
$10.00 16%
$15.00 16%
$20.00 19%
$25.00 30%
62 BARBARA E. KAHN AND DAVID C. SCHMITTLEIN
Using the cutoff rule defined above, we calculated the number of quick and reg-
ular shopping trips for each household. Then we defined the Quicks to be those
households that made more quick purchases than regular purchases, and Regulars
to be those households that made more regular purchases than quick purchases.
This scheme is obviously ad hoc, but will generate some interesting results below.
The reliability of this segmentation scheme was tested by comparing classifi-
cation of each household using the first half of the shopping trips to classification
of the same household using the second half of the shopping trips, for all panelists.
We found that the segmentation scheme was 85% reliable on a test-retest basis.
This reliability is significantly higher than any of the standard benchmark proce-
dures (e.g., the criterion generated by assuming that proportions of the segments
are equal in both halves (Mosteller and Bush, 1954): 55%; the proportional chance
criterion (Morrison, 1969); 56%; or the largest group criterion (Lehmann, 1985):
67%).
Reoular Sesment
QuickTrips Q Q Q
Regular Tdps A i R R ; R R R R A
: ,: ;! :
Observed ,i Qd R QR i A RQ R
Quick Trips Q Q Q Q Q Q Q
i j:
Regular Trips R: Ri j R j
/: ;i I
Observed Q Q dQ Q R Q Q d 6
0 7 14 2, 28 35 42
Tab/e 4. (continued)
The segments are also similar in two interesting respects. One is that the average
amount spent on quick and regular trips does not vary across segments. The av-
erage amount spent on a quick trip is a little over $7 for both segments and the
average amount spent on a regular trip ranges between $35-$40 for both segments.
This substantial difference between amounts spent on quick versus regular trips,
and its similarity in both segments, lends additional credibility to the segmentation
scheme.
The other similarity between the two segments is the relative behavior with
regard to day-of-week shopping preferences. As table 3 shows, both segments
shop on their most preferred shopping day approximately 30% of the time. The
most preferred modal shopping day is Saturday and the least preferred shopping
day is Sunday for both segments. As above, the similarity between the two seg-
ments on the day-of-the-week preference phenomenon suggests the robustness of
the variables constructed.
ables (i.e., the nature of the shopping trip and the segment identification of the
consumer) should be included in logit models of brand choice.
Understanding shopping trip behavior can also contribute to our understanding
of whether or not promotions accelerate purchases (Neslin, Henderson, and
Quelch, 1985). There are two components to purchase acceleration: 1) the accel-
eration of the consumer’s purchase of the brand (beneficial to the manufacturer),
and 2) the acceleration of the shopping trip to the store (beneficial to the retailer).
To test for the latter acceleration, one should understand quick and regular shop-
ping trips and Quick and Regular shoppers.
For example, if features are accelerating shopping trips we would expect the
percentage of featured brand purchases made on quick trips to be higher than
the percentage of nonfeatured brand purchases made on quick trips, at least for the
Regular segment. Similarly, by examining figure 6 we would also expect that the
average time to the last shopping trip would be shorter for trips in which featured
brands were purchased than the average time to the last shopping trip for trips in
which featured brands were not purchased (again for the Regular segment). Yet
this natural approach for assessing purchase acceleration is not likely to be effec-
tive for the Quick shoppers, as the figure 6 illustrates. Thus, understanding pur-
chase acceleration requires an appreciation of a household’s baseline inclination
to undertake frequent, small, relatively randomly occurring shopping trips.
Identifying the types of shopping trips made could also help determine whether
these accelerated trips to the store increase overall spending. By comparing the
average amount spent across different patterns of shopping trips (e.g., regular-
quick-regular compared to regular-regular for consumers in the Regular segment),
we could assess whether quick trips to the store increase overall expenditures or
just spread the total expenditure over more trips.
6. Conclusions
shopping trips. The Regular consumers make fewer quick runs to the store and
instead are more apt to have a once-a-week regular shopping day. The robustness
and strength of these phenomena indicate that they may impact brand choice,
purchase time, and the effectiveness of promotional variables, and therefore are
factors worthy of inclusion in future marketing model-building efforts.
Notes
I. Our analyses are based on the Information Resources, Inc. Academic Research Data Base,
Crackers I/02/84-12/29/85. The data are from panelists in Williamsport. PA; Rome, GA; and
Midland, TX.
2. The 30% modal day shopping figure cannot simply be compared with 117 (7 days equally likely)
due to a selection effect. It turns out that the baseline expectation depends on the number of
shopping trips observed for each household. The expectation decreases towards 117 as the num-
ber of shopping trips increases to infinity and, of course, equals 100% for any household ob-
served to make only one shopping trip.
3. This cutoff rule therefore takes into account income, family size, and age of adult female. The
alternative is to use a standard cutoff rule for everyone. We redid the analyses using a standard
$15 cutoff rule. which is the median value (and also may roughly correspond to the maximum
amount spent at an Express Lane) and found no differences in the implications of cur results.
4. Although we used the number of quick versus regular trips to determine the segments, the pro-
portion of each type of trip per segment did not have to differ this greatly from 50-50. The greater
the difference in these proportions between the two groups, the more spread apart the segments
are.
References
Corstgens, Marcel L, and David A. Gautschi. “Formal Choice Models in Marketing.” Mrrrketirtg
Scie/7ce 2 (Winter 1983), 19-56.
Dunn, R., S. Reader. and N. Wrigley. “An Investigation of the Assumptions of the NBD Model as
Applied to Purchasing at Individual Stores.” Applied Stutistics 32 (3, 1983). 249-259.
Frisbie, Gil A., Jr. “Ehrenberg’s Negative Binomial Model Applied to Grocery Store Trips.” lo~r-
na/ of Marketing Reseurch XVI? (August 1980), 385-390.
Grover, Rajiv, and Vithala Rao. “Inferring Competitive Market Structure Based on a Model of
Interpurchase Intervals.” Interrrcrfion~I Jor/rno/ of Research in Mnr-keting 5 (I, 1988), 55-72.
Guadagni, Peter M.. and John D. C. Little. “A Logit Model of Brand Choice Calibrated on Scanner
Data.” Marketing Science 2 (Summer 1983). 203-238.
Gupta, Sunil. “An Integrated Model of Inter-purchase Time, Brand Choice and Purchase Quan-
tity.” Journul of Murketir~g Reserrrch XXV (November 1988).
Kollat, David T.. and Ronald P. Willett. “Customer Impulse Purchasing Behavior.” Jorrr-rral of
Murkefing Reseurch IV (February 1967). 21-31.
Kahn, Barbara E., Donald G. Morrison, and Gordon P. Wright. “Aggregating Individual Purchases
to the Household Level.” Murkeriilg Science 5 (Winter 1986). 260-268.
Kahn, Barbara E., and Donald G. Morrison. “A Note on ‘Random’ Purchasing: Additional Insights
from Dunn, Reader and Wrigley.” Applied Stufi~tic~ 38 (1, 1989), 111-l 14.
Krishnamurthi, Lakshman, and S. P. Raj. “A Model of Brand Choice and Purchase Quantity Price
Sensitivities.” Marketing Science 7 (Winter 1988), 1-19.
SHOPPING TRIP BEHAVIOR 69
Lehmann, Donald R. Ma&et Research and Anulysis, 2nd ed. Homewood, IL: Irwin. 1985.
McKay, D. B. “A Spectral Analysis of the Frequency of Supermarket Visits.” Jo~r-nrrl o~MNI.XC~-
ing Research 10 (February 1973). 84-90.
Meyer, Robert .I., and Barbara E. Kahn. “Probabilistic Models of Consumer Choice Behavior.”
Handbook of Consumer Behavior: Theory and Resemch, eds. H. Kassarjian and T. Robertson.
Prentice Hall. 1990.
Morrison, Donald G. “On the Interpretation of Discriminant Analysis.” Jorrrnrrl of‘ Mtrrhetirlg
Reseuuch 6 (May 1969), 156-163.
Morrison, Donald G., and David C. Schmittlein. “Generalizing the NBD Model for Customer
Purchases: What Are the Implications and Is It Worth the Effort. 7" Journul of Business & Eco-
rzomic Statistics 6 (April 1988), 145-159.
Mosteller, Frederich, and Robert R. Bush. “Selective Quantitative Techniques.” Hrrtzdbook of
Social Psychology, vol. I, ed. Gardner Lindzey. Reading, MA: Addison-Wesley. 1954.
Neslin, Scott, Caroline Henderson, and John Quelch. “Consumer Promotions and the Acceleration
of Product Purchases.” Marketing Science 4 (Spring 1985), 147-165.
Park, C. Whan, Easwar S. Iyer, and Daniel C. Smith. “The Effects of Situational Factors on In-
Store Grocery Shopping Behavior: The Role of Store Environment and Time Available for Shop-
ping.” Jo~vlzal of Corzsurrter Research I5 (March 1989), 422-433.