Studies in Regional & Urban Planning

Issue 9, December 2001

FACTORS AFFECTING SHOPPING TRIP GENERATION RATES

IN METROPOLITAN AREAS

K. MERT CUBUKCU
Austin E. Knowlton School of Architecture
City and Regional Planning
The Ohio State University
Revised: August 1, 2002

ABSTRACT

This paper attempts to answer two specific questions on trip generation: (1) What are the
factors that affect the total number of shopping trips in North American metropolitan
areas, and (2) Did the demand for technology-related products, and telecommunication
technologies, and emerging interest in on-line shopping, have any observable effect on
personal shopping trips generated in these areas in the early stages of the Internet? The
estimated model is linear in the continuous independent variables and linear in the
logarithms of the continuous dependent variable. The dependent variable is the total
annual number of shopping trips. The explanatory variables include (1) shopping related
characteristics of the metropolitan areas (total number of retail establishments, population
density, general climatic conditions), (2) socio-economic characteristics of the trip makers
(age, income), and (3) technology-related trip maker characteristics (computer ownership,
modem ownership). The empirical findings, based on OLS estimation of 1995 data for 49
metropolitan areas with population over 1 million indicate that total number of retail
establishments, percentage of the population between ages 34-54, percentage of days
with convenient temperatures for shopping, and computer ownership are positively
related to shopping trip generation rate. Population density and modem ownership are
negatively related. The relations are statistically significant at the .01 level for all the
variables but temperature, which is significant at the .05 level. The model estimates that
the net number of annual shopping trips decrease by between 1.47 and 21.92 millions for
the metropolitan areas in the sample, when both computer ownership and modem
ownership are increased by one percent. Thus, findings indicate a negative net of these
technology-related factors shopping trip generation in the mid 1990’s, when the Internet
revolution has just started. However, the current model does not explain whether these
trips are eliminated or replaced by other trips.

51
 1.INTRODUCTION

Urban transportation covers the movement of both people and goods within an urban area. At the individual
level, urban transportation can be characterized by a trip. However, at the metropolitan area level, millions of
these individual trips define urban transportation (Barber, 1995). A trip is as a journey or a part of a journey
made by an individual between two different points. Each trip is performed using one or multiple transportation
modes for a defined purpose at a given time (Hobbs, 1979).

Trip generation analysis, as Meyer (1974) puts it, seeks to estimate the volume of trips that is made by
individuals to work, shopping, school, and so forth, but not the flows between points within the whole system.
The functioning of metropolitan cities is highly dependent on the movement of people, goods and information
(Muller, 1995) and trip generation studies are a vital part of transportation planning, due to the recursive nature
of urban transportation modeling procedure (Bruton, 1986; Badoe and Steuart, 1997).

Personal trips are commonly classified based on their main purposes (Barber, 1995); work trips, shopping trips,
social trips, recreational trips, school trips, home trips and business trips. A sequence of multi-purpose trips or
single-purpose multi-trips is called a journey (Hensher, 1976). This paper focuses on shopping trips, and the
factors that determine the aggregate number of shopping trips generated in metropolitan areas. A shopping trip
can be defined as a trip from any origin to any retail establishment, irrespective of the size and type of the store
or shop, and whether a purchase is made or not (Barber, 1995). Thus, going to a store, or a shop by any means
of transportation at any time period of the day is a sufficient condition for an individual to perform a shopping trip.
Followed by shopping trips (Vickerman and Barmby, 1984), work trips are the most numerous among all trip
purposes, counting approximately for 40% of all trips generated in North American metropolitan areas (Barber,
1995).

This paper attempts to answer two specific questions. The first one is: “What are the factors that affect the total
number of personal shopping trips generated in North American metropolitan areas?” The second one is an
extension to the first question, and rather historical: “Did the demand for technology -related products, and
telecommunication technologies, and emerging interest in on-line shopping, have any observable effect on
personal shopping trips generated in these areas in the early stages of the Internet? The paper tries to answer
these questions via an empirical model based on cross-sectional data at the metropolitan level for the year 1995.

There are several reasons for selecting the year 1995 for this study. The latest available personal trip generation
data disaggregated by purpose covering the whole U.S. is for the year 1995 at the time of the research. Also, the
mid 1990’s were particularly important in terms of the increasing popularity and availability of technologically
advanced products in the market and the Internet. In 1995, the National Science Foundation (NSF) announced
that it contracted with companies that would be providers of access to the NSF backbone. These companies
would then sell connections to groups, organizations, and companies. Thus, the main U.S. backbone traffic was
routed through interconnected network providers. The Internet was left in commercial hands, and the traditional

52
on-line systems (CompuServe, America Online, Prodigy) started to provide dial-up access. The first on-line
shopping mall had already been opened in 1994. In the year 1995, 100,000 web sites were present on the
Internet with 6.5 million hosts (Zakon, 2001). Parallel to the growing interest in technologically advanced devices
and the Internet, the Bureau of the Census started to conduct ‘Computer Ownership/Uses Supplement’ surveys
in the mid 1990’s. Availability of data from multiple sources makes 1995 a very appropriate choice to measure
the early effects of the Internet and demand for technologically advanced devices as well as the other factors
affecting shopping trip generation. It is aimed that this study will function as a base study and allow further
empirical studies focusing on the relation between telecommunications and shopping trip generation to compare
their cases to the era when the Internet revolution has just accelerated.

The remainder of the paper is organized as follows. Section 2 consists in a literature review. The modeling
methodology is presented in section 3. The data are described in section 4, and the empirical results are
presented and analyzed in section 5. Section 6 concludes the paper and outlines areas for further research.

2. LITERATURE REVIEW

Trip generation models relate trip generation rates to land-use and household characteristics. More often than
not, the focus of research is the number of trips generated and their geographic distribution (Salomon and
Koppelman, 1988). The accuracy in trip generation models is particularly important, since subsequent stages of
transport modeling are highly dependent on the trip generation output (Badoe and Steuart, 1997).

There are at least two approaches in terms of data aggregation in trip generation models: aggregate trip
generation models and disaggregate trip generation models. In aggregate models, data at a given geographic
level, such as neighborhood or city, are used. In disaggregate models, data at the household or individual levels
are used (Koppelman and Pas, 1984). Linear regression and categorical analysis techniques are widely used in
estimating the aggregate models (FHWA, 1975; Hobbs, 1979; Koppelman and Pas, 1984; Bruton, 1986;
Sheppard, 1995). Discrete choice models are used for disaggregate models (Vickerman and Barmby, 1985).

According to Levinson (1976) and Bruton (1986), trip making is a function of the following basic factors: the
socio-economic characteristics of the trip makers residing in the area, and the land -use pattern and
developments in the study area (or the physical characteristics of the area). Two sets of variables det ermining
trip generation rates in an area appear in the literature: socio-economic characteristics of the trip makers, and
physical and demographic characteristics of the area.

The socio-economic characteristics of the trip makers are assumed to be significant determinants of travel
behavior (Koppelman and Pas, 1984; Pas, 1986), and include; income (Bruton, 1986; Koppelman and Pas,
1984; Pas; 1986), age (Boarnet and Crane, 2001), gender, employment status (Koppelman and Pas; 1984; Pas;
1986), auto ownership (Martin et al., 1961; Levinson, 1976; Bruton; 1986), and household size (Levinson, 1976;

53
Bruton; 1986). The physical and demographic characteristics of the area include; population, density (Martin et
al., 1961; Levinson, 1976; FHWA, 1985; Bruton, 1986; Boarnet and Crane, 2001; Hobbs, 1979), and
employment (Hobbs, 1979).

While there is a vast literature on the factors affecting the total volume of trips generated (see Koppelman and
Pas; 1984; Pas, 1986; Osula, 1991), there are much fewer studies focusing on the factors determining the
number of shopping trips directly. That is partly because the factors determining the number of all types of trips
are assumed to be the same as those for shopping trips (Keefer, 1966). However, it is reasonable to believe that
there are shopping related metropolitan area characteristics and trip maker characteristics that solely impact
shopping trip generation rates. The size and structure of the retail sector are such characteristics. Variables that
measure these characteristics include the number of retail establishments, total floor space and accessibility to
retail establishments (Keefer, 1966; Vickerman and Barmby, 1984; 1985).

Keefer (1966) examines the effects of some shopping center and trip maker characteristics on shopping trips
made to these centers. Two models are estimated with linear regression, using data for 23 major shopping
centers in the US. The final model includes such variables as number of parking spaces, total number of person-
work trips, distance from the competitor shopping center, and floor space of the shopping center (Keefer, 1966).
At the household level, Robinson and Vickerman (1976) estimate a pure trip generation model by using only
socioeconomic variables; income, car ownership and household size. The model was not able to explain the
variation in the trip generation rates. Also at the household level, Vickerman and Barmby (1984) estimate the
number of weekly shopping trips and total weekly shopping expenditure by using simultaneous equations with
diary data. The explanatory variables include household size, income, auto ownership, and two indices;
shopping attraction index and shopping travel cost index. However, the levels of explanation of the models are
not satisfactory. In a follow up study, Vickerman and Barmby (1985) apply a binary logistic regression model to
the same diary data. The results point to evidence on the insignificance of car ownership. The effect of
household size is significant, since the number of trips per household naturally increases with increase in
household size. However, the parameters for the attraction index and the shopping trip cost index are found to
be more stable across different trip making levels. The attraction index is calculated by using physical
characteristics of the shopping centers, and the trip cost index is calculated by using the costs associated with
the completion of a trip (Vickerman and Barmby, 1985). Finally, Badoe and Steuart (1997) estimate the total
number of housing shopping trips by using such variables as household size, number of workers, number of
licensed persons, and number of vehicles. Regression is used with 1964 and 1986 data for the Greater Toronto
Metropolitan area. The different model specifications provide little explanation of the variation in household
shopping trips. Household size, which is one of the most common variables in previous studies, is found to have
much less explanatory power, compared to previous studies (Badoe and Steuart, 1997). Badoe and Steuart
(1997) conclude that different approaches are needed to explain the variation in non-work trips, including
shopping trips.

54
Developments in technology and expansion of the market for technologically advanced products stimulate an
interest in the relation between telecommunications and transportation. However, none of the above-mentioned
studies has tried to include variables related to demand for technologically advanced products or on-line
shopping in their models. A vast literature has been developed on the possible effects of such improvements on
trip-making patterns, although empirical research remained limited. Plaut (1997) reviews the
substitution/complementarity debate between travel and telecommunication. In the 1970s and early 1980’s, the
predominant approach was the substitution hypothesis, whereby telecommunications will eventually replace and
reduce travel, or trips, as communication gets cheaper, and more accessible. As Salomon (1985, pp. 225) points
out, “an underlying assumption of trip generation models is a need to perform an activity at a distant location.”
Thus, in an extreme case, trips might be eliminated if all needs are satisfied at a single location (Salomon, 1986).
The different versions of this hypothesis include alteration in timing, frequency, duration, and purpose of the
trips. The complementarity-enhancement hypothesis suggests that developments in telecommunications
technology will increase the efficiency in both public and private transportation, causing additional trip generation
(Plaut, 1997). The empirical studies that provide clear evidence for any of the two different hypotheses are
scarce (Salomon and Koppelman, 1988).

The focus in the substitution/complementarity debate has been telecommuting until the late 1980’s. This is
naturally so, since a larger portion of all trips generated is work-related, and a shift towards telecommuting will
effect the current trip generation rates most. However, most of the work published in the telecommunications-
transportation field is speculative, with little empirical research to test the hypotheses. Most of these limited
empirical works focus on telecommuting (Henderson and Mokhtarian, 1996; Plaut, 1997).

Teleshopping has attracted much less attention (Salomon and Koppelman, 1988). The impact of teleshopping or
demand for technologically advanced products on trip making behavior had not been the subject of any empirical
research until the mid 1980’s (Salomon, 1986). However, in the 1990’s, the popularity of teleshopping has
increased, particularly owing to the spread of Internet shopping. In 1996, about $300 million worth of products
were sold through the Internet (Gould and Golob, 1998). Salomon (1986) argues that teleshopping may
substitute for a portion of customer trips to stores. Since the number of delivery trips will be less than the number
of customer trips, the net impact of teleshopping on shopping trip generation will be negative. The fact that not
all shopping trips result in purchases (Gould and Golob, 1998), but deliveries are made only for purchases,
further supports the idea. Thus, even if shopping trips are to be replaced by other trips, the net number of
shopping trips generated will decline. According to Rosenberg and Hirschman (1980), women’s participation in
the labor force and time limitations are the main motives for teleshopping. Price comparability is definitely
another motive. Other advantages include availability of wider selection of goods and instantaneous information
on current stock, and avoidance of traffic and parking (Gould and Golob, 1998). However, there are more than a
single kind of shopping activity, and teleshopping will have different effects on each. Several authors suggest
that shopping for maintenance of the household (food, cleaning supplies, etc.) is the kind that is more likely to be

55
replaced by teleshopping. This would be less the case for recreational store shopping (Salomon, 1986; Gould
and Golob, 1998). People are also expected to obtain goods that are not locally available through teleshopping
avoiding longer distance trips. However, these ideas have not been tested through empirical research.

3. MODELING METHODOLOGY

Trips are the basic unit for aggregate trip generation models, and trip characteristics such as direction, length, or
duration are ignored (FHWA, 1975). The number of trips made in a pre-identified zone during a given time unit is
called the trip generation rate (Hobbs, 1979). In this paper, the analysis of the factors determining the shopping
trip generation rates in metropolitan areas is approached within the general trip generation rate modeling
framework. Let Ai be the vector of shopping related physical and demographic characteristics of area i, and X i
the vector of socio-economic characteristics of the trip makers, both produc ing the trip generation rate Yi. The
trip generation rate model has the general form (Hobbs, 1979; Bruton, 1986):

Y i = f (A i, Xi)

The equation can be estimated for different trip purposes individually (Hobbs, 1979). Since we are interested in
shopping trips, we can write the equation as:

Y si = f (Asi, Xsi)

where Ysi is the total number of shopping trips generated in area i; Asi is the shopping-related physical and
demographic characteristics vector for area i; and X si is the shopping-related characteristics of people living in
area i.

For the vector As, the variables considered include total number of retail sector establishments, number of retail
centers, population, population density, and percentage of days with preferred average temperatures. For the
vector X s, the variables considered include income, expenditure, age, race, unemployment, and household size.
In the selected model, the metropolitan area characteristics vector, As, includes:

ESTBLMNT: total number of retail sector establishments;

DENSITY: population density;

TEMP: percentage of days with preferred average temperature.

The trip maker characteristics vector, Xs includes:

MIDAGE: percentage of middle aged (35-54) people in the population;

INCOME: median household income.

56
ESTBLMNT, or the total number of retail sector establishments variable, measures the effect of the retail sector
size on the total shopping trip generation rate. These are the establishments that are organized to sell
1
merchandise in small quantities to the general public . The inclusion of the total number of retail establishments
as a variable is natural: more establishments produce more shopping trips providing a larger variety of products
and services (Vickerman and Barmby, 1985). Existence of more retail establishments may also cause more
economic activities owing to agglomeration economies, causing more shopping trips in turn (Tabuchi, 1997).

DENSITY, or the population density variable, is simply population per square mile. The density variable
measures the intensity of population and activities, and it is regarded as a determinant of the shopping trip
generation rate (Bruton, 1986). In denser areas, residential developments are more likely to be in walking
distances to many services, including shopping places. Thus, elimination of auto trips is more probable in denser
areas (Levinson, 1976; FHWA, 1985; Bruton, 1986; Boarnet and Crane, 2001). The argument is perfectly valid
for shopping trips at the metropolitan area level.

TEMP, or the annual percentage of days with preferred temperatures for shopping based on daily average
temperatures, is included in the model to test whether general climatic conditions have any observable effect on
shopping trips. Although, there is not a standard range of temperatures preferred for shopping, it wouldn’t be a
mistake to believe that the ideal temperature for outdoor activities is around 75 °F for most people. With 20 °F
margins, it is assumed that temperatures between 55 °F and 95 °F are preferred temperatures for shopping. It is
expected that higher percentages with convenient temperatures produce more shopping trips, since extreme
temperatures are a good reason to avoid certain types of trips, including a portion of shopping trips. Other
general climatic factors such as wind speed, cloudiness and humidity have significant effects on perceived
temperature; however, these factors are not concerned in this study. It is also assumed that people have the
same range of preferred temperatures throughout the year, although there probably exists seasonal variations.
Nevertheless, the annual percentage of days with preferred temperatures provide sufficient information on the
general climatic variation among the metropolitan areas in the sample.

MIDAGE, or the percentage of people between the ages of 35 and 54, measures the effect of a certain age
group on shopping. This group is the most active group in terms of trip making (Levinson, 1976) and people in
this group are expected to make more shopping trips. A greater portion of people between 35 and 54 is usually
financially independent, and auto ownership is relatively high. People under 35, in contrast, are expected to have
less shopping trips compared to the 35-54 group. Boarnet and Crane (2001) show that older people (over 55)
make fewer non-work auto trips, including shopping trips. Thus, it is expected that metropolitan areas with higher
percentages of people between 35 and 54 are expected to produce more shopping trips.

INCOME, or the mean household income variable, measures the economic status of the trip maker. As a
marginal cost is attached to each additional shopping trip, trip makers with higher incomes are more likely to

1
NAICS sector code: 44-45.

57
produce additional shopping trips (Vickerman and Barmby, 1984). Nonetheless, previous studies revealed
conflicting findings on the relation between income and trip generation. Studies dealing with the effect of income
on general trip generation reveal that income is a positive and significant predictor (Pas, 1984). However, no
significant effect is found in studies where the relation between particularly shopping trips and income are tested
(Robinson and Vickerman, 1976; Vickerman and Barmby, 1984). An alternative approach by Vickerman and
Barmby (1985) suggests that expenditure may is a better predictor of trip generation than income. A different
variable, SALES, or the amount of total purchases made, to measure expenditure was then previously
considered replacing INCOME in the model. However, it was dropped, as SALES variable is found to be highly
correlated with the total number of retail sector establishments variable, ESTBLMT (r=.95, significant at the .01
level). Thus, only the income variable is included in the model, assuming that the effect of expenditure is
measured by the variable, ESTBLMT.

In order to measure the first observable effects of teleshopping and technological advances on shopping trip
generation in the mid 1990’s, a technology variable vector, Tsi, is added to the model:

Y si = f (Asi, Xsi, Tsi)

The technology variable vector includes two variables:

COMPUTER: Percentage of households with a computer at home;

MODEM: Percentage of households with a modem at home.

COMPUTER, or the computer ownership variable, is included in the model to test whether the willingness to buy
technologically-advanced products has any observable effect on shopping trip generation. In 1995, a much
smaller percentage of people had computers at home than today. There is no doubt that the computer products
market, including software and hardware, has expanded significantly in the 1990’s, causing additional shopping
trips to purchase such products. However, during the same era, on-line shopping has become more popular
than ever, with the help of declining computer and connection costs.

A second variable, MODEM, or percentage of modem variable, then measures the connection rate. In the mid
1990’s, less than half of the computers had a modem. In 1994, for example, 24.1% of the households had a
computer in the U.S., whereas, only 11.0% of the households had a modem (NTIA, 1999). As clearly stated in
NTIA (1999), modem ownership was a very common measure of determining the level of Internet access, until
1998. That practice ceased in 1998, since modems have become a standard component for personal
computers. Also, there have been developed several alternatives to modems for Internet access such as DSL,
satellite and cable. Nonetheless, modem ownership serves as the best proxy available for measuring Internet
access for historical research purposes pertaining to pre-1998 (NTIA, 1999). As a proxy for Internet access rate,
modem ownership can be regarded as an essential requirement for on-line shopping, or perhaps a measure of

58
eagerness or readiness to perform on-line activities including on-line shopping. Referring to the substitution-
driven teleshopping literature, the number of shopping trips is expected to decline with higher connection rates
as a result of either avoiding a portion of the trips or replacing them with other trips such as recreational trips.
Note that delivery trips resulting from on-line shopping activities are no longer shopping trips.

4. DATA

The data pertain to 49 Metropolitan Areas with over one million population by the year 1995 in the United States.
Metropolitan Statistical Areas (MSAs) and Primary Metropolitan Statistical Areas (PMSAs) are included in the
dataset, as data for all the considered variables were available at the MSA and PMSA level. Moreover, data for
some variables were only available at the MSA and PMSA level. MSA and PMSA are plausible geographic units
for such an analysis, since both MSAs and PMSAs included in the analysis have urbanized populations large
enough to demonstrate an economic and social unity. This economic and social unity is crucial in terms of
defining trip generation rates in the metropolitan areas. The original dataset included 55 Metropolitan Areas, but
2
six were eliminated due to incomplete data.

The data are drawn from five different sources: (1) the shopping trips and the income data are drawn from the
1995 Nationwide Personal Transportation Survey by the Federal Highway Administration and several
cosponsors; (2) the retail establishments data are drawn from the 1997 Economic Census, by the Bureau of the
Census; (3) population density and age data are drawn from the State and Metropolitan Area Data Book 1997-
1998, by the Bureau of the Census; (4) the weather data are drawn from the National Climatic Data Center; and
(5) the telecommunications data are drawn from the 1994 Computer Ownership/Uses Supplement Survey of the
Bureau of the Census.

Table 1. Descriptive Statistics of the Variables (n=49)

Variable Minimum Maximum Mean Standard Deviation
TRIP * 132 1964 503.82 472.790
ESTBLMNT 3,977 34,081 9,317.96 6389.658
DENSITY 32.10 7,504.20 942.70 1,129.649
TEMP (%) 35.9 98.9 61.2 .17639
MIDAGE (%) 25.00 33.00 29.29 1.607
INCOME 34,928 60,921 43,526 6,549.462
COMPUTER (%) 20.34 55.97 35.78 7.625
MODEM (%) 8.47 35.22 20.07 5.783
* : Dependent Variable (in millions)

Table 1 presents the descriptive statistics of the dependent and independent variables. The shopping trips
(TRIP) data include the total annual person shopping trips. A shopping trip is defined as a one-way trip where
the destination is a shopping related activity. Thus, any single-purpose shopping trip, or a trip which is a part of a

2
See Appendix for the list of all MSAs/PMSAs used in the study.

59
multi-purpose journey with one stop for shopping is considered as a shopping trip. The total annual shopping
trips vary between 132 million (Middlesex-Somerset-Hunterdon) and 1.964 billion (Chicago), with a mean of
472.8 million.

The total number of retail sector establishments (ESTBLMNT) for the 49 metropolitan areas in the sample varies
between 3,977 (Rochester) and 34,081 million (New York), with a mean of 9,318 establishments by the year
1997.

The population density (DENSITY) is meas ured in persons per square miles for the year 1995, and varies
between 32.1 (Las Vegas) and 7504.2 (New York). These measures reflect the estimates by the Bureau of the
Census based on the decennial census data. Similarly, the percentage of population between the ages of 35
and 54 reflects the estimates for the year 1996, and varies between 25.0% (Sal Lake City-Ogden) and 33.0%
(Denver), with a mean of 29.3%.

The mean household income data in dollars (INCOME) is for the year 1995. It varies between 34,928 (Nashville)
and 60,921 (Middlesex -Somerset -Hunterdon, NJ), with a mean of 43,526 dollars.

The annual percentage of days with preferred temperatures for shopping (55 °F - 95 °F) based on daily average
temperatures (TEMP) varies between 35.9% (Portland) and 98.9 % (Miami), with a mean of 61.2%. The data is
drawn for the year 1995.

The computer ownership (COMPUTER) and modem ownership (MODEM) data are for the year 1994. However,
the survey includes observations on both computer ownership and willingness to buy a computer within one
year. Thus, by aggregating the percentages for these two questions, an estimate for the 1995 computer
ownership is derived for the 49 the metropolitan areas in the sample. Likewise, modem ownership for the year
1995 is obtained. Computer ownership varies between 20.34% and 55.97%, with a mean of 35.78, and modem
ownership varies between 8.47% and 35.22%, with a mean of 20.07%, in 1995.

5. ESTIMATION AND RESULTS

The estimated model is linear in the independent variables (ESTBLMNT, DENSITY, TEMP, MIDAGE, UNEMP,
COMPUTER, and MODEM) and linear in the logarithms of the dependent variable (TRIP). As no trips can be
generated when the explanatory variables are all zero, no constant is included in the model:

ln (TRIP)i = ß1 Asi + ß2X si + ß3Tsi + ei

where TRIP is the total number of shopping trips; Asi the shopping related metropolitan area characteristics
vector; X si is the trip-maker characteristics vector; Tsi is the trip maker technology vector for metropolitan area i;
and ei is the error term. The regression analysis is performed with OLS. The error term, ei is assumed to be a

60
random variable, with an expected mean of 0, a constant variance, and is assumed to have significant and
independent of any explanatory variable.
3
Table 2 presents the estimates of the proposed model . Number of retail establishments, density, percentage of
people aged 35-54, computer, and modem ownership are significant at the .01 level, and annual percentage of
days with preferred temperatures is significant at the .05 level. The income variable is found to be insignificant.
Signs for all the statistically significant variables appear to be consistent with the discussion in section 3.

Table 2. Parameter Estimations*

Variable Estimated Coefficients Standard Error t p-value
Unstandardized ( Pr > | t | )
-4
ESTBLMNT 1.355 * 10 0.000 13.556 0.000
DENSITY -2.501 * 10-4 0.000 -4.204 0.000
TEMP 5.397 * 10-3 0.003 2.129 0.039
MIDAGE .142 0.016 8.750 0.000
-6
INCOME -3.957 * 10 0.000 -.448 0.657
COMPUTER 4.913 * 10-2 0.015 3.238 0.002
MODEM -6.029 * 10-2 0.020 -2.960 0.005
* : Dependent Variable: LN_TRIP (Logarithm of Total Annual Shopping Trips in millions)

The estimated total number annual shopping trips is 362.10 million, at the sample mean of each explanatory
variable. The average population for the sample is around 2.38 million. Thus, the model estimates 152 shopping
trips produced annually per person on average in this hypothetical metropolitan area. That is roughly equally to
three shopping trips per week per person.

The positive signs for the number of retail establishments, the percentage of population between 35 and 54, and
the percentage of days with preferred temperatures indicate that the total number of shopping trips increases
with these variables. On the other hand, increase in population density causes decrease in shopping trip
generation.

As expected more retail establishments produce more shopping trips. In semi-log linear models, the effect of one
unit change in any of the ex planatory variables on the dependent variable, keeping the other variables constant,
is equal to:

∆Y ∂Y
= = βiYi
∆X i ∂ X i

Thus, for our hypothetical metropolitan area with 362.10 million trips produced annually, at the sample
mean of each variable, an additional retail establishment will cause 49,064 more shopping trips
annually. This increase varies between 17,886 and 266,122 shopping trips across the sample. This

3
R-square = .999, Adjusted R-square =.997. For regression through the origin (the no-intercept model), R Square measures
the proportion of the variability in the dependent variable about the origin explained by regression. This cannot be compared
to R-Square for models that include an intercept.

61
finding is parallel with the previous studies that suggest more establishments generate more shopping
trips (Vickerman and Barmby, 1985), and supports the hypothesis that the availability of more
shopping options and more variety of goods generate additional economic activity (Tabuchi, 1997),
which in turn generate more shopping trips.

The estimated model suggests that an increase in the percentage of people between the ages 35 and 54 causes
an increase in shopping trips. This result is also consistent with the previous work (Levinson, 1976; Boarnet and
Crane, 2001). An increase of one percent in this age group increases the annual shopping trips between 18.74
and 278.89 million for the sample. For our hypothetical metropolitan area, one additional percentage increases
annual shopping trips by more than 51 million. This finding becomes particularly important when the current
trend in demographic structure in the U.S. is examined. According to the census data, the percentage of people
4
in this age group living in metropolitan areas increased from 25.7% to 44.2% between the years 1990 and 2000 .
According to the model, approximately over 100 million additional shopping trips were then produced in our
hypothetical metropolitan area annually between the years 1995 and 2000, due to the increasing share of the 35
- 54 age group in population.

The negative sign for the density variable indicates that, as population density (and thus activity density)
increases, the shopping trip generation rate decreases. This result is also as expected, since in denser areas the
elimination of auto trips is more probable, due to the fact that service and shopping centers are more likely to be
within walking distances (Levinson, 1976; FHWA, 1985; Bruton, 1986; Boarnet and Crane, 2001). For our
hypothetical metropolitan area, the total number of shopping trips decreases by 90,560 with an increase of one
person per mile square in population density. This decrease varies between 33,013 and 491,196 across the
sample.

The estimated model suggests that more days with preferred daily average temperatures (55 °F - 95 °F)
generate more shopping trips. An additional one percent of preferred daily average temperatures in a year is
expected to increase the number of annual shopping trips by 1.95 million. This increase varies between 0.71 and
10.60 million trips across the sample. The hypothesis that better climatic conditions generate more shopping
trips is supported. However, this variable is less significant compared to the other ones. There is a very plausible
explanation to that: Most of the shopping establishments, if not all, are climate-controlled environments, as well
as the vehicles in which people travel. Thus, the effect of climate on shopping trip generation is limited.
Nonetheless, as the model suggests general climate conditions have still observable effects on shopping trip
generation indicating the possibility that extreme temperatures make people avoid a portion of shopping trips.

As hypothesized, the number of shopping trips increases with an increase in the demand for technologically
advanced products measured by the computer ownership variable. For our hypothetical metropolitan area, a one
percent increase in computer ownership generates 17.79 million additional shopping trips annually. This

62
increase varies between 6.49 and 96.50 millions across the sample. This result may be interpreted in several
ways. One plausible explanation is that computer ownership generates additional economic activity, including
hardware and software purchases, producing extra shopping trips. Equally plausible, computer ownership may
be reflecting the consumer profile in each metropolitan area, perhaps, measuring interest in buying other
technologically advanced products leading to more shopping trips in turn.

The second technology variable, modem ownership, aims to measure the connection rate as a requirement to
shop on-line, or an indicator of the willingness and readiness to shop on -line. The modem ownership rate
appears to be negatively related to shopping trip generation. The total number of shopping trips decreases by
21.83 million with an increase of one percent in modem ownership, in our hypothetical metropolitan area. This
decrease varies between 7.96 and 118.41 millions across the sample. This result may be explained by the
replacement of some shopping trips by on-line shopping. These trips may include shopping for daily low-level
goods like food and cleaning supplies, or high-level luxury goods that are not locally available and require longer
distance trips. In either case, a single on-line shopping activity eliminates one or more shopping trips. Note that
delivery trips are no longer shopping trips, and not all shopping trips end up with a purchase (Gould and Golob,
1998). Thus, it is reasonable to expect a decrease in total shopping trips with an increase in on-line shopping,
which is more likely to happen with higher connection rates, or higher willingness and readiness to shop on-line
measured by modem ownership.

The findings for these separately measured factors provide important information on the relation between
telecommunications an d shopping trip generation in an era where the Internet revolution has accelerated
considerably. The net effect of these two technology -related factors, namely the demand for technologically
advanced products (measured with computer ownership), and willingness and readiness to shop on-line
(measured with modem ownership), shed light on the first observable effects of the Internet on daily life in the
mid 1990’s. The aggregate effect of these two technology variables becomes particularly important, as the
model estimates that an increase in computer ownership causes an increase, and an increase in modem
ownership causes a decrease in shopping trip generation. For our hypothetical metropolitan area with 362.10
million annual shopping trips per year, a one percent increase in both computer ownership and modem
ownership causes a net decrease of 4.04 million shopping trips annually. The decrease in shopping trips
generated varies between 1.47 and 21.92 millions across the sample. However, it is impossible to determine
whether these trips are eliminated or replaced by other trips with the current analysis.

The only variable that is found insignificant is the mean income. This finding contradicts with previous work
focusing on trip generation in general. Pas (1986) shows that income is significant variable in trip generation.
Contrarily, the finding is consistent with research focusing on particularly shopping trip generation. Robinson and
Vickerman (1976) are not able to explain the variation in the number of shopping trips with socio-economic
variables, including income. Similarly, Vickerman and Barmby (1984) report very little variation in shopping trip

4
Author’s own calculation based on decennial census data provided by the Bureau of the Census.

63
making levels between households with different income levels. One explanation to that is the marginal cost of
an additional shopping trip being insignificant and limited to the travel cost, as a purchase is not necessary to
perform a shopping trip. A different approach by Vickerman and Barmby (1985) suggests that expenditure is a
much better predictor of trip generation compared to income. As mentioned before, a different variable, SALES,
or the total amount of purchases, was considered in the model to measure expenditure. However, it was
dropped due to its high correlation to ESTBLMNT, the total number of retail sector establishments variable
(r=.95, significant at the .01 level). One may conclude that income is a determinant of trip generation in general,
but not for shopping trip generation in particular. Expenditure, which is measured by the number of
establishments variable (ESTBLMNT) in this research, appears to be a better predictor. Findings suggest then
more establishments, thus more expenditure indicate more shopping and more shopping trips in turn.

6. CONCLUSION

This paper aimed at answering two questions: (1) What are the factors that affect the total number of personal
shopping trips in North American metropolitan areas? and (2) Did the demand for technology-related products,
and telecommunication technologies, and emerging interest in on -line shopping, have any observable effect on
personal shopping trips generated in these areas in the early stages of the Internet? For the first question, the
estimated model shows that shopping related metropolitan area characteristics such as total number of retail
establishments and percentage of days with preferred temperatures have positive, and population density has
negative effects. Regarding trip maker characteristics, the percentage of people in the age group 35-54 is found
to have a positive effect, while, income is found to have no observable effect.

For the second question, shopping trips are found to be increasing with the demand for technology -related
products, as measured by computer ownership, and decreasing with on-line shopping, as measured by
connection rate, or modem ownership. The combined net effect of these two technology-related factors
pertaining to the mid 1990’s is negative, indicating a net decrease in shopping trips when both variables are
increased by same percentage. For our hypothetical metropolitan area with 362.10 million annual shopping trips
per year, one percent increase in both computer ownership and modem ownership causes a net decrease of
4.04 million trips annually. The decrease in shopping trips varies between 1.47 and 21.92 millions across the
sample. This is as expected as a single on-line shopping activity is likely to eliminate one or more shopping trips:
Although an on-line shopping activity ends with a purchase and an a delivery trip, not all shopping trips end with
a purchase, and an individual may make multiple shopping trips to make a single purchase. Moreover, delivery
trips are counted as work trips, thus no shopping trips are generated as a result of an on-line shopping activity.

Several areas for further research can be delineated. First, the current model cannot explain whether the
avoided shopping trips are totally eliminated or replaced by other trips. An expanded model might consider trips
for different purposes simultaneously to answer this question. Second, the variables included in the model might

64
interact in complex ways in trip generation (Bruton, 1986), and testing the effect of the interactions between the
variables would be informative. Finally, this research aimed to see if any observable effect of demand for
technologically advanced products and on-line shopping were present in the mid 1990’s, the early stages of the
Internet. The estimations are based on 1995 data indicating significant net negative effect of the two technology-
related factors on shopping trip generation. However, as mentioned, modem ownership was a very common
measure of determining the level of Internet access until 1998 (NTIA, 1999). That practice ceased in 1998, since
modems have become a standard component for personal computers. Research examining the current relation
between the Internet and shopping trip generation has to measure these two technology-related factors in a
different way. There is no question that neither computer ownership can measure the demand for technology
related prod ucts, nor modem ownership can measure connectivity rate today (2001). For such research,
alternatives to computer ownership include expenditure on technologically advanced devices; and alternatives to
modem ownership include actual connectivity rates, and the volume of on-line sales. Nonetheless, the computer
ownership and modem ownership were able to capture interest in technologically advanced products and
eagerness or readiness to perform on-line activities separately for the era that the study focuses on. Findings
provide evidence that the net effect of these factors were negative on shopping trip generation in metropolitan
areas in the mid 1990’s, when the Internet revolution has just started.

ACKNOWLEDGEMENT

I would like to thank Jean-Michel Guldmann and Philip A. Viton, for their help in the formation and writing of this
article and Ebru Cubukcu, for her helpful comments. However, any mistakes that remain are my own.

REFERENCES

BADOE, D. A., and G. N. STEUART (1997) Urban and Travel Changes in the Greater Toronto Area and the
Transferability of Trip-generation Models, Transportation Planning and Technology, vol. 20, pp. 267-290.
BARBER, G. (1995) Aggregate Characteristics of Urban Travel, in The Geography of Urban Transportation, ed.
by Susan Hanson, The Guilford Press, New York.
BOARNET G. B., and R. CRANE (2001) Travel by Design: The Influence of Urban Form on Travel, Oxford
University Press, New York.
BRUTON, M. J. (1986) Introduction to Transportation Planning, Hutchinson, London.
th
BUREAU OF CENSUS, U.S. DEPARTMENT OF COMMERCE, State and Metropolitan Area Data Book (5
Edition), Washington D.C., 1998.
FEDERAL HIGHWAY ADMINISTRATION (FHWA) (1975) Trip generation Analysis, U.S. Government Printing
Office, Washington, D.C.
GOULD, J. and T. F. GOLOB (1998) Will Electronic Home Shopping Reduce Travel? Access, vol. 12, spring, pp.
26-31.
HENDERSON, D. K. and P. L. MOKHTARIAN (1996) Impacts of Center-based Telecommuting on Travel and
Emissions: Analysis of the Puget Sound Demonstration Projects, Transportation Research D, vol. 1, no. 1,
pp. 29 -45.
HENSHER, D. A. (1976) The Structure of Journeys and Nature of Travel Patterns, Environment and Planning A,
vol. 8, pp. 655-672.

65
HOBBS, F. D. (1979) Traffic Planning & Engineering, Pergamon Press, London.
KEEFER, L. E. (1966) Urban Travel Patterns for Airports, Shopping Centers, and Industrial Plants, Highway
Research Board, National Academy of Sciences-National Academy of Engineering.
KOPPELMAN, F. S. and E. I. PAS (1984) Estimation of Disaggregate Regression Models of Person Trip
th
Generation with Multiday Data, in Proceedings of the 9 International Symposium on Transportation and
Traffic Theory, ed. by J. Volmuller and R. Hamerslag, Ulrecht, the Nedherlands, pp. 513-529.
LEVINSON, H. S. (1976) Urban Travel Characteristics, in Transportation and Traffic Engineering Handbook ,
edited by J. E. Baerwald, Prentice-Hall, New Jersey.
MARTIN, B. V., MEMMOTT, F. W., and A. J. BONE (1962) Principles and Techniques of Predicting Future
Demand for Urban Area Transportation, Research Report No. 38, MIT, Cambridge.
MEYER, J. R. (1974) “The Future and Implications for Regional Transportation Planing” Perspectives on
Regional Transportation Planning edited by DeSalvo, J. S., Lexington Books.
MULLER, P. O. (1995) “Transportation and Urban Form: Stages in the Spatial Evolution of the American
Metropolises” in The Geography of Urban Transportation edited by Susan Hanson, The Guilford Press,
New York.
NTIA – National Telecommunications and Information Administration (1998) Falling Through the Net II: New
Data on the Digital Divide, US Department of Commerce.
NTIA – National Telecommunications and Information Administration (1999) Falling Through the Net: Defining
the Digital Divide, US Department of Commerce.
PAS, E. I. (1986) Multiday Samples, Parameter Estimation Precisopn, and Data Collection Costs for Least
Squares Regression Trip-generation Models, Environment and Planning A, vol. 18, pp. 73-87.
PINDYCK, R. S. and D. L. RUBINFELD (1976) Econometric Models and Economic Forecast, Mc-Graw Hill, New
York.
PLAUT, P. O. (1997) Transportation-Communications Relationships in Industry, Transportation Research A, vol.
31, no. 6, pp. 419-429.
ROBINSON, R. V. and R. W. VICKERMAN (1976) The Demand for Shopping: A Theoretical and Empirical
Study, Applied Economics, vol. 8, pp. 167-187.
ROSENBERG, L., and E. HIRSCHMAN (1980) Retailing without Stores, Harvard Business Review, Jul-Aug, pp.
103-112.
SALOMON, I. (1985) Telecommunications and Travel Relationships: A Review, Transportation Research A, vol.
20A, no. 3, pp. 223-238.
SALOMON, I., and F. KOPPELMAN (1988) A Framework for Studying Teleshopping versus Store Shopping,
Transportation Research A, vol. 22A, no. 4, pp. 247-255.
SHEPPARD, E. (1995) Modeling and Predicting Aggregate Flows, The Geography of Urban Transportation
edited by Susan Hanson, The Guilford Press, New York.
TABUCHI, T. (1998) Urban Agglomeration and Dispersion: A Synthesis of Alonso and Krugman, Journal of
Urban Economics, vol. 44, pp. 333-351.
VICKERMAN, R. W., and T. A. BARMBY (1984) The Structure of Shopping Travel: Some Developments of the
Trip Generation Model, Journal of Transport Economics and Policy, vol. 18, no. 2, pp. 109-121.
VICKERMAN, R. W., and T. A. BARMBY (1985) Household Trip Generation Choice: Alternative Empirical
Approaches, Transportation Research B, vol. 19, no. 6, pp. 471-479.
ZAKON, Robert (2001) Hobbes Internet Timeline, published electronically: available at
http://www.zakon.org/robert/internet/timeline/ (accessed in August, 2001)

66
APPENDIX: The List of MSAs/PMSAs
MSA/PMSA Metropolitan
Code Area
520 Atlanta, GA MSA
720 Baltimore, MD PMSA
1123 Boston, MA-NH PMSA
1280 Buffalo-Niagara Falls, NY MSA
1520 Charlotte-Gastonia-Rock Hill, NC-SC MSA
1600 Chicago, IL PMSA
1640 Cincinnati, OH-KY-IN PMSA
1680 Cleveland-Lorain-Elyria, OH PMSA
1840 Columbus, OH MSA
1920 Dallas, TX PMSA
2080 Denver, CO PMSA
2160 Detroit, MI PMSA
2680 Fort Lauderdale, FL PMSA
3120 Greensboro—Winston-Salem--High Point, NC MSA
3360 Houston, TX PMSA
3480 Indianapolis, IN MS A
3760 Kansas City, MO-KS MSA
4120 Las Vegas, NV-AZ MSA
4480 Los Angeles -Long Beach, CA PMSA
4920 Memphis, TN-AR-MS MSA
5000 Miami, FL PMSA
5015 Middlesex-Somerset-Hunterdon, NJ PMSA
5080 Milwaukee-Waukesha, WI PMSA
5120 Minneapolis -St Paul, MN -WI MSA
5190 Monmouth-Ocean, NJ PMSA
5360 Nashville, TN MSA
5380 Nassau-Suffolk, NY PMSA
5560 New Orleans, LA MSA
5600 New York, NY PMSA
5640 Newark, NJ PMSA
5720 Norfolk-Virginia Beach-Newport News, VA-NC MSA
5775 Oakland, CA PMSA
5880 Oklahoma City, OK MSA
5960 Orlando, FL MSA
6160 Philadelphia, PA-NJ PMSA
6200 Phoenix-Mesa, AZ MSA
6280 Pittsburgh, PA MSA
6440 Portland-Vancouver, OR -WA PMSA
6840 Rochester, NY MSA
6920 Sacramento, CA PMSA
7040 St Louis, MO -IL MSA
7160 Salt Lake City-Ogden, UT MSA
7240 San Antonio, TX MSA
7320 San Diego, CA MSA
7360 San Francisco, CA PMSA
7400 San Jose, CA PMSA
7600 Seattle-Bellevue-Everett, WA PMSA
8280 Tampa-St Petersburg-Clearwater, FL MSA
8840 Washington, DC-MD-VA-WV PMSA

67