1
Description of a Northern California Shopping Survey
Data Collection Effort
David T. Ory
Department of Civil and Environmental Engineering
and
Institute of Transportation Studies
University of California, Davis
Davis, CA 95616
voice: ( 415) 378- 9102
fax: ( 530) 752- 6572
e- mail: dtory@ ucdavis. edu
and
Patricia L. Mokhtarian
Department of Civil and Environmental Engineering
and
Institute of Transportation Studies
University of California, Davis
Davis, CA 95616
voice: ( 530) 752- 7062
fax: ( 530) 752- 7872
e- mail: plmokhtarian@ ucdavis. edu
March 2007
This research is funded by the University of California Transportation Center.
i
TABLE OF CONTENTS
LIST OF TABLES......................................................................................................................... .............. ii
LIST OF FIGURES ............................................................................................................................... ..... iii
ACKNOWLEDGEMENTS............................................................................................................... ......... iii
1. INTRODUCTION ............................................................................................................................... 1
2. SAMPLING PLAN........................................................................................................................... ... 2
3. SAMPLING....................................................................................................................... .................. 6
4. SURVEY DESIGN......................................................................................................................... ..... 7
5. SURVEY INSTRUMENT DEVELOPMENT ................................................................................... 17
6. DATA COLLECTION ....................................................................................................................... 21
7. INITIAL DATA CLEANING ............................................................................................................ 26
8. SUMMARY AND NEXT STEPS ...................................................................................................... 32
REFERENCES ............................................................................................................................... ........... 32
ii
LIST OF TABLES
Table 1: Target Sampling Plan ..................................................................................................................... 5
Table 2: Media Income, Structure Age, and Tenure by Census Tract .......................................................... 5
Table 3: Delivered Addresses by Census Tract ............................................................................................ 7
Table 4: Distance Measures for Off- Target Addresses................................................................................. 8
Table 5: General Shopping- Related Attitudinal Statements, First Nine Categories ................................... 12
Table 6: General Shopping- Related Attitudinal Statements, Final Ten Categories.................................... 13
Table 7: Design- stage Mode- Specific Attitudinal Statements, First Five Categories ( Internet Mode) ...... 14
Table 8: Design- stage Mode- Specific Attitudinal Statements, Last Nine Categories ( Internet Mode) ...... 15
Table 9: Final Mode- Specific Attitudinal Statements ( Store Mode) .......................................................... 16
Table 10: Response Rate by City................................................................................................................ 25
Table 11: Part A Missing Data Summary ................................................................................................... 27
Table 12: Geographic Imputation Groups .................................................................................................. 27
Table 13: Part D, Section 1 Missing Data Summary .................................................................................. 28
Table 14: Part D, Section 2 Missing Data Summary .................................................................................. 28
Table 15: Sociodemographic Missing Data Summary................................................................................ 29
Table 16: Part A Missing ( Imputed) Data Summary for Final Sample ...................................................... 30
Table 17: Part D, Section 1 Missing ( Imputed) Data Summary for Final Sample ..................................... 31
Table 18: Part D, Section 2 Missing Data Summary for Final Sample ...................................................... 31
Table 19: Sociodemographic Missing Data Summary for Final Sample.................................................... 31
Table 20: Target Status of Working File Cases .......................................................................................... 31
iii
LIST OF FIGURES
Figure 1: Aerial Photos of Davis ( top) and Santa Clara ( bottom), California .............................................. 4
Figure 2: Address Locations and Desired Tracts .......................................................................................... 6
Figure 3: Internet Instrument Flow Chart ................................................................................................... 20
Figure 4: Davis Recruitment Letter ............................................................................................................ 23
Figure 5: First Postcard Reminder .............................................................................................................. 24
Figure 6: Second Postcard Reminder.......................................................................................................... 24
Figure 7: Completed Surveys over Time .................................................................................................... 26
Figure 8: Data Cleaning Process................................................................................................................. 30
ACKNOWLEDGEMENTS
Survey design and data collection was funded by the University of California Transportation Center.
Early phases of the survey design were conducted by Xinyu Cao; we greatly appreciate his thoughtful and
extensive input to the final design. Susan Handy and numerous pretesters also offered many useful
suggestions for improvement. Tara Puzin provided able assistance in the identification and analysis of
prospective survey census tracts.
1
1. INTRODUCTION
Applications of new information and communication technologies ( ICTs) are changing how and where
we work, shop, play, travel, and in other ways live our lives. Yet because ICT development and use is in
such a volatile state, many of those changes and impacts are poorly understood. This report summarizes
the development and deployment of a survey instrument intended to gather information to allow better
understanding of the transportation impacts of business- to- consumer ( B2C) e- commerce. Although the
business- to- business ( B2B) segment dominates e- commerce in terms of the dollar value of transactions
made, B2C remains important for its potential impacts on urban travel and land use patterns, including
potential redistributions of retail land uses, and substantial increases of package delivery trips into
residential neighborhoods.
We see an analysis of the transportation impacts of B2C e- commerce as having two components: ( 1)
assessing the transportation impacts of a given level or pattern of adoption of B2C e- commerce, and ( 2)
investigating the adoption of B2C e- commerce ( who, under what circumstances, in what form). While
transportation planners ultimately need to forecast transportation impacts, to do so accurately they need to
understand adoption processes and trends. Thus, the data collection described here is intended to lead to
modeling the adoption of B2C e- commerce, among other shopping “ modes” ( specifically store and
catalog shopping). Because we take the consumer perspective, we will refer to the use of the internet for
B2C e- commerce as “ e- shopping”. We consider e- shopping to be a subset of “ teleshopping”, which also
includes catalog shopping ( whether placing the order by phone or mail) and shopping from a television
channel ( generally by phone). To frame a manageable project, and because most TV shopping appears to
be impulsive ( Handy and Yantis, 1997), we do not include TV shopping in the survey.
Though the survey instrument collects data on the pre- purchase browsing mode( s) as well as the
transaction mode choice, our definition of shopping requires a purchase to occur, not just information-gathering
or “ window shopping”. Because ( 1) the nature of the shopping process is presumed to be quite
heterogeneous depending on the type of good being considered; ( 2) due to resource constraints our
sample will number in the ( high) hundreds rather than many thousands; and ( 3) attempting to survey
respondents in- depth with respect to many types of goods would pose too great a burden; we focus on two
specific product types. Product classes can be characterized along a number of dimensions, including
frequency of purchase, cost, search area ( local, regional, broader), tangibility ( whether a service, a digital
good, or a material good), perishability, “ differentiability” ( the extent to which retailers can distinguish
their offering of the product from others’), product information complexity, and as a “ search good”
( having features “ that can be evaluated from externally provided information”) versus an “ experience
good” ( needing “ to be personally inspected or tried”; Peterson, et al., 1997).
To yield a sufficient number of purchase occasions in our sample, we focus on low- cost, medium-/ high-frequency-
of- purchase goods. Specifically, we take the tangible product classes of books/ physical
CDs/ DVDs/ videotapes and clothing/ shoes to be analyzed in depth. Books/ CDs/ DVDs/ videotapes per se
are basically search goods with low intrinsic differentiability, though the bookstore experience ( browsing
contents, having coffee) may offer some advantages to some customers. Clothing and shoes are basically
experience goods, although decades of catalog shopping history show that some customers are willing to
buy without trying. The survey also asks a few questions about internet shopping activity of all kinds,
especially internet- based purchases.
The main research questions to be addressed by this exploratory project build on a number of prior
studies: ( 1) For the product class in question, what are the advantages ( motivators and facilitators for
choosing) and disadvantages of ( constraints on choosing) each shopping mode? A number of authors have
identified the potential advantages of e- shopping and/ or store shopping ( e. g. Brynjolfsson and Smith,
2001; Salomon and Koppelman, 1988; Tauber, 1972). Store shopping is so far still very different from e-
2
shopping in terms of such attributes as the information provided, the sensual stimulation, the ability to
compare prices and to attain immediate ownership. Beyond the functional attributes related to shopping
and purchasing, store shopping also offers numerous other experiences, which to varying degrees are less
amenable to electronic platforms. These include, for example, the ability to interact with real salespeople
and to bargain, the opportunity to be outside the home or work environment, some physical exercise, and
so on. Shopping, under many circumstances, is a combined maintenance - leisure activity. Thus, the
choice between store shopping and e- shopping is not unambiguous ( Handy and Yantis, 1997).
( 2) Can market segments with different propensities to use alternative modes be identified? A composite
of several studies ( Cairns, 1996; Koppelman, et al., 1991; Tacken, 1990; Gould and Golob, 1997; Gould
et al., 1998; Burke, 1998; Farag et al., 2006; Ferrell, 2005; Ren and Kwan, 2005) identifies four segments
of the population that are likely to be early adopters of e- shopping: the mobility- limited, time- starved,
technophilic, and shopping- phobic. These four segments may well be generalizable to many e- shopping
contexts. In reality, however, they probably represent ( and in general we will treat them as) four
continuous dimensions, with individuals separately falling somewhere along each of them, rather than
four mutually- exclusive and collectively- exhaustive group membership indicators.
( 3) To what extent do customers perceive there to be viable alternative modes for a given shopping
occasion? Previous studies have largely neglected this question, implicitly or explicitly assuming that
each shopping activity involves a true choice among competing alternatives. It is obviously important to
test that assumption, and identify variables associated with perceived mode captivity versus choice.
( 4) Are the various shopping modes substitutes, or complements? For example, do people who do a lot of
e- shopping tend to do less store shopping, or do high amounts of one tend to be associated with high
amounts of the other? Do former catalog shoppers replace the catalog with the internet, or supplement it;
conversely, do internet shoppers become new catalog shoppers as retailers engage in cross- channel
marketing? Are there different relationships for different segments of the population? For example,
shopping modes may be complementary for innate “ shopaholics”, but substitutionary for the time-pressured.
Although to keep the survey at a reasonable length it is not possible to obtain the data for a
rigorous analysis of transportation impacts, the answers to these questions will have direct implications
for the likely transportation impacts of e- shopping.
The remainder of this report focuses on the administration of a survey instrument that will facilitate the
research directions discussed above. The organization of the document is as follows. Section 2 discusses
the design of the sampling plan for the data collection. The following section then describes the actual
sampling process. The fourth section briefly presents the survey design. Section 5 discusses the
development of the survey instruments and Section 6 outlines the deployment of the survey. Next, a
section is devoted to the steps taken to build the final sample through data cleaning. A concluding section
ends the report.
2. SAMPLING PLAN
Obtaining a strictly representative sample is not essential to the success of this study. This is because our
purpose is not primarily to report descriptive statistics for the sample distributions of various measures of
interest and expect them to faithfully represent their population counterparts, but rather to identify
relationships among the variables we measure. Those relationships can be reliably captured, even when
raw univariate distributions are not representative. For example, the sample may over- represent some
income groups and under- represent others, but ( assuming all income groups have a numerically
substantial representation in the sample) the distribution of other responses given a certain income level
can still be properly represented ( see Brownstone, 1998 for a discussion of this issue in the discrete choice
modeling context, and Babbie, 1998 for a general comment).
3
What is critical, then, is to have sufficient diversity with respect to the variables of interest. In particular,
it is important to have a substantial number of e- shopping occasions in the sample. At the same time,
there is currently not a convenient way to systematically sample internet users, let alone e- shoppers. Thus,
we conducted a regular mail recruitment of the sample, but for a setting in which a random sample of
residents is expected to net a high level of internet literacy and shopping activity. Specifically, the
sampling plan segments our population by city, neighborhood type, and census tract. With respect to city,
we split the sample between our university town of Davis, California ( pop. 60,000), which is on the fringe
of the medium- size Sacramento Metropolitan Statistical Area ( MSA; pop. about 2 million) but separated
from the rest of the area by agricultural land and flood control plains, and Santa Clara ( pop. 100,000), an
affluent, computer- literate community in Silicon Valley ( embedded within the large San Francisco MSA,
pop. about 7 million) that also contains a private university ( enrollment 8,000). Figure 1 portrays the two
study areas from about 45,000 feet up. Although the college- town sample will contain a higher- than-average
proportion of younger adults, that actually has some advantages, since those respondents will
tend to be harbingers of future adoption patterns. That is, the ICT behavior of today’s over- 50 adults is apt
to tell us less about the future than that of today’s 20- somethings. Together, the two subsamples will
allow for some limited testing of the role of urban context in shopping mode choice, a factor that has been
found important in the research of Farag, et al. ( 2006) and Krizek et al. ( 2005).
Within those two cities, we targeted “ traditional” neighborhoods, meaning those with closely- spaced
perpendicular streets and nearby retail stores, and “ suburban” neighborhoods, which are typical of more
recent developments and contain curved streets, cul- de- sacs, and major- intersection- located retail strip
malls. Though both Davis and Santa Clara are largely suburban communities and such segmentation may,
in the end, be somewhat arbitrary, we wanted to have at least the possibility of measuring the impact of
different neighborhood types within the two cities.
Due to budget constraints, a recruitment sample size of 8,000 was established with the hope ( based on the
prior experience of the second author with several paper surveys of similar complexity, albeit shorter than
this one) of receiving on the order of one to two thousand valid responses. The sample was first
distributed evenly among city and neighborhood type ( see Table 1), targeting 2,000 residents in each
neighborhood type within each city. To operationalize the administration of the survey, census tracts
within each desired city and neighborhood were selected through map inspection and site visits. Within
each neighborhood type, the number of households in each targeted census tract, per the 2000 census, was
used to allocate the sample by tract. As shown in Table 1, the sampling rate for each tract ranged from
0.382 to 0.582.
As discussed previously, the goal of the sampling plan was not to represent the population of the United
States, or any other general population for that matter. Rather, given limited resources, the goal was to
sample a few typical areas offering a diversity of residential neighborhood type and urban context, and
use the data to investigate a variety of variables’ s on shopping and travel behavior. Table 2 presents a
brief summary of characteristics for the targeted census tracts. We can see that the tracts have a rather
wide range of median incomes, structure age, and owner/ renter mix. It should be mentioned that while the
median incomes in the “ traditional” Davis neighborhoods appear to be rather low, many residents in these
areas are university students who may have spending powers beyond what their nominal income would
suggest.
4
Figure 1: Aerial Photos of Davis ( top) and Santa Clara ( bottom), California
5
Table 1: Target Sampling Plan
City Neighborhood
Type
Census
Tract Households N Sampling Rate
5050.01 2,465 1,432 0.581
Suburb
5050.07 976
2,000
568 0.582
5056 1,246 708 0.568
Santa Clara
Traditional
5057 2,271
4,000
2,000
1,292 0.569
105.07 3,220 1,231 0.382
Suburb 105.08 1,055 403 0.382
106.05 956
2,000
366 0.383
106.02 2,393 1,058 0.442
Davis
Traditional
107.01 2,129
4,000
2,000
942 0.442
Table 2: Media Income, Structure Age, and Tenure by Census Tract
Structure Age Tenure
City Neighborhood
Type
Census
Tract
Median Income
( 1999) Before 1970 After 1970 Owner Renter
5050.01 $ 92,613 19.5% 80.5% 50.2% 49.8%
Suburb
5050.07 $ 74,911 76.8% 23.2% 66.2% 33.8%
5056 $ 42,625 56.2% 43.8% 22.4% 77.6%
Santa Clara
Traditional
5057 $ 53,657 80.8% 19.2% 31.1% 68.9%
105.07 $ 42,813 10.1% 89.9% 42.4% 57.6%
Suburb 105.08 $ 62,537 8.3% 91.7% 61.6% 38.4%
106.05 $ 80,238 0.0% 100.0% 78.1% 21.9%
106.02 $ 25,803 47.0% 53.0% 18.9% 81.1%
Davis
Traditional
107.01 $ 24,204 74.8% 25.2% 24.8% 75.2%
6
3. SAMPLING
To implement the sampling plan, 15,000 names and addresses were purchased from the firm Martin
Worldwide, Incorporated ( http:// www. martinworldwide. net/). We requested that the addresses be located
as summarized in Table 1. Martin Worldwide indicated the request could be satisfied, with a total of
7,500 ( rather than 4,000) addresses to be delivered in each city ( each tract’s sample size is then increased
proportionally to the number of households in the tract – see Table 3).
After receiving the list of 15,000 names and addresses from the vendor, the list was pruned as follows.
First, because we wanted to locate each address in one of the target census tracts, we removed post office
box addresses from the list ( 1,484 of the 15,000). Next, duplicate names, with differing addresses, were
removed ( 1,023). The remaining addresses were then geo- coded using the website
http:// www. batchgeocode. com/. This website allows for a batch of addresses to be entered and plotted on
an interactive map. Twenty- two of the remaining addresses could not be geo- coded. The resulting number
of “ good” addresses was 12,471.
Using geographic information system ( GIS) software, the remaining 12,471 were mapped against the
target census tracts. Figure 2 presents the shaded census tracts along with dots representing each of the
addresses for Santa Clara ( on the left) and Davis ( on the right). As the figure demonstrates, the addresses
are in the same general area as the census tracts, but they are not entirely contained within them.
The graphical data of Figure 2 are summarized in Table 3. For Santa Clara, Martin Worldwide promised
to deliver 7,500 addresses in the four target census tracts and delivered 1,398 on- target addresses ( well
short of our requested 4,000). In Davis, 2,057 on- target addresses were delivered.
Santa Clara Davis
Figure 2: Address Locations and Desired Tracts
Combining the “ on- target” addresses from Santa Clara and Davis, a total of 3,455 addresses in the desired
census tracts were delivered by Martin Worldwide. Not wanting to purchase more addresses ( which could
also be off- target) from another vendor and recognizing that the differences between adjacent neighbor-hoods
within Santa Clara and Davis may not generally be dramatic, it was decided to build a sample of
8,000 addresses from those provided by Martin Worldwide.
7
Table 3: Delivered Addresses by Census Tract
Location Delivered N
City Neighborhood Tract
Promised N Requested N
On- Target Off- Target
5050.01 2,686 1,432 542 ---
Suburb
5050.07 1,064 568 232 ---
5056 1,328 708 165 ---
Traditional
5057 2,422 1,292 459 ---
Santa Clara
Total 7,500 4,000 1,398 5,267
105.07 2,308 1,231 663 ---
Suburb 105.08 756 403 253 ---
106.05 686 366 370 ---
106.02 1,984 1,058 409 ---
Traditional
107.01 1,766 942 362 ---
Davis
Total 7,500 4,000 2,057 3,749
The sample was built by first measuring the distance from each address to the boundary of each census
tract ( using GIS software). Next, the points closest to the census tracts were included in the sample. This
was done sequentially, starting with the traditional tracts, taken as a group, and then the suburban tracts.
For example, in Santa Clara we started with the 624 ( 165 + 459) on- target addresses in the traditional
tracts. We then took the next 1,376 addresses closest to either of the traditional tracts until we reached our
“ quota” of 2,000 addresses. We then took the 774 ( 232 + 542) on- target addresses for the suburban tracts,
and added the 1,226 addresses closest to them. This process was repeated for the Davis addresses. With
the realization that some of the more distant addresses could potentially be assigned to either neighbor-hood
type, assignment to the traditional tracts was given priority under the assumption that in both Davis
and Santa Clara, traditional neighborhoods are scarcer than suburban ones.
Table 4 summarizes a variety of distance measures for the off- target addresses ( the “ on- target” addresses
are not included in the average distance calculations). The table shows that the median distance from the
tracts to the addresses is, in three of four cases, less than half a kilometer. The Davis addresses are, in
general, closer to their targets than the Santa Clara addresses.
4. SURVEY DESIGN
As discussed in the next section, both web- and paper- based survey instruments were developed as part of
the study. A brief description of the survey contents is presented here.
The survey started with a simple Welcome question, intended to be easy and fun to answer: “ If you HAD
to spend an hour or two shopping, where would you prefer to be?” Response options were: downtown
shopping district, shopping mall, bookstore, grocery store, electronics store, hardware/ home improvement
store, and “ other ( please specify)”.
Part A: Shopping and General Shopping- related Attitudes followed the Welcome section and contained
42 statements respondents were able to agree or disagree with on a five- point Likert- type scale (“ strongly
disagree”, “ disagree”, “ neutral”, “ agree”, “ strongly agree”). In the design stage, we identified 21
attitudinal dimensions of interest, based on a thorough review of the literature ( Cao and Mokhtarian,
2005) and our own judgment. Then, drawing from previously published surveys and our own judgment,
we prepared the following 79 statements in the 21 categories:
8
Table 4: Distance Measures for Off- Target Addresses
Santa Clara Davis
Measure
Traditional Suburban Traditional Suburban
On- target addresses 624 775 771 1,286
Off- target addresses 1,376 1,225 1,229 714
Minimum 0.2 6.0 0.1 0.9
Maximum 1,435.7 3,344.7 679.6 431.3
Mean 715.8 1,132.0 335.5 255.7
Standard deviation 446.1 1,148.3 208.8 139.1
Distance
from
selected
tracts for off-target
addresses
( meters)
Median 665.0 277.1 354.2 304.6
􀂾 Technology – General
􀂃 I am often among the first to buy new technological products.
􀂃 I often find high- tech products difficult to operate.
􀂃 I like to track the development of new technology.
􀂃 Technology brings at least as many problems as it does solutions.
􀂃 On the whole, technology makes our lives better.
􀂾 Technology – Computers
􀂃 I enjoy using computers.
􀂃 The internet makes life more interesting.
􀂃 Computers are more frustrating than they are fun.
􀂃 The internet plays only a minor role in my life.
􀂾 Variety seeking – General
􀂃 " Variety is the spice of life."
􀂃 I like a routine.
􀂃 Change is unsettling.
􀂃 Change is refreshing.
􀂾 General innovativeness – General
􀂃 I am generally cautious about accepting new ideas.
􀂃 I prefer to see other people using new products before I consider getting them myself.
􀂃 I like to experiment with new ways of doing things.
9
􀂾 eShopping innovativeness – Shopping specific
􀂃 I know more about shopping over the internet than most of my friends.
􀂃 I am among the last in my circle of friends to purchase something over the internet.
􀂃 I enjoy taking chances in buying over the internet just to enrich my shopping experience.
􀂾 Price consciousness – Shopping specific
􀂃 If I really want something, I'll often buy it even if it costs more than it should.
􀂃 I generally compare prices before buying.
􀂃 I look for sales and special offers when I'm shopping.
􀂃 Sales and special offers encourage unnecessary spending.
􀂃 Generic brands usually offer just as good a quality as more expensive ones.
􀂃 It's important to me to get the lowest prices when I buy things.
􀂾 Time consciousness – Shopping specific
􀂃 I'm too busy to shop as often or as long as I'd like.
􀂃 Being a smart shopper is worth the extra time it takes.
􀂃 I'm often in a hurry to be somewhere else when I'm shopping.
􀂃 Saving time when I shop is important to me.
􀂾 Impulsive buying – Shopping specific
􀂃 I generally stick to my shopping lists.
􀂃 Before buying something, I generally take some time to think it over.
􀂃 I often make unplanned purchases.
􀂾 Shopping enjoyment – Shopping specific
􀂃 For me, shopping can be an important leisure activity.
􀂃 I would often prefer someone else to do my shopping.
􀂃 Under the right circumstances, shopping is fun.
􀂃 I enjoy the social aspects of shopping.
􀂃 Shopping is usually a chore for me.
􀂃 Shopping helps me relax.
􀂾 Risk aversion – General
􀂃 Taking risks fits my personality.
􀂃 Before I make a decision, I need to be sure it's the right one.
􀂃 I'm willing to take a big chance for the possibility of a big payoff.
􀂾 Trustingness – General
􀂃 Most people can be counted on to do what they say they will do.
􀂃 One should be very cautious with strangers.
􀂃 People are generally trustworthy.
10
􀂾 Attitude towards credit cards – General
􀂃 I like to pay for everything with cash.
􀂃 Credit cards encourage you to spend unnecessarily.
􀂃 I need to have a credit card.
􀂾 Travel – Shopping specific
􀂃 Shopping is too physically tiring to be enjoyable.
􀂃 Going shopping is mentally stressful.
􀂃 I like to stroll through shopping areas.
􀂃 Getting to where I usually shop is a hassle.
􀂃 Sometimes for me, shopping is mostly an excuse to get out of the house or workplace.
􀂃 Even if I don't end up buying anything, I still enjoy going to stores and browsing.
􀂾 Travel – General
􀂃 " Getting there is half the fun".
􀂃 The only good thing about traveling is getting to the destination.
􀂃 Travel time is generally wasted time.
􀂃 In my daily travel, I try to make good use of my time.
􀂃 The routine traveling that I need to do interferes with doing other things I like.
􀂾 Physical exercise – General
􀂃 Whenever possible, I prefer to walk or bike rather than drive.
􀂃 I follow a regular physical exercise routine.
􀂃 I should be more physically active than I am.
􀂾 Environmental concerns – General
􀂃 To improve air quality, I am willing to pay a little more to use a hybrid or other clean- fuel
vehicle.
􀂃 We should raise the price of gasoline to reduce congestion and air pollution.
􀂃 We need more public transportation, even if taxes have to pay a lot of the costs.
􀂾 Environmental concerns – Shopping specific
􀂃 Shopping travel does not cause very much air pollution.
􀂃 We should try to reduce shopping travel to benefit the environment.
􀂃 A lot of the packaging used for products is wasteful.
􀂾 Status – General
􀂃 For me, a lot of the fun of having something nice is showing it off.
􀂃 I am often the one introducing a new trend to my friends.
􀂃 Impressing other people is of little interest to me.
11
􀂾 Materialism – General
􀂃 My lifestyle is relatively simple, in terms of material goods.
􀂃 I would/ do enjoy having a lot of expensive things.
􀂃 As long as I have the basics, material things don't matter to me that much.
􀂾 Spending money enjoyment – General
􀂃 I enjoy spending money.
􀂃 Buying things cheers me up.
􀂃 If I got a lot of money unexpectedly, I would probably save more of it than I spent.
􀂾 Social influence
􀂃 My friends and I share many common interests.
􀂃 When my friends try something new, I like to try it as well.
􀂃 A lot of the fun of trying new things is sharing the experience with friends.
For the survey, the 21 categories were combined and eliminated to form 19 categories, and 42 questions
were selected and modified. In keeping with guidance from the survey design literature ( e. g. Baumgartner
and Steenkamp, 2001; Ellard and Rogers, 1993), we diversified the directionality of the final list of
statements, to reduce the tendency to fall into an automatic response mode. We made an effort to include
at least one positively- oriented and one negatively- oriented statement for each construct, but where we
could not produce satisfactory statements by that guideline, we did not force it. The factor analysis
literature ( e. g. Fabrigar et al., 1999) further advises including 3- 5 items ( statements) for each
hypothesized construct, but in view of the large number of constructs we considered important to our
context, and the interconnectedness of many of them ( thus leading to their merging or overlapping in an
exploratory factor analysis), we limited the number of statements per construct to two in most cases –
again as a design compromise to reduce respondent fatigue. The final list, sorted alphabetically by
category label, is included in Table 5 and Table 6.
Part B: Your Purchasing Experiences asked if the respondent had purchased items in 15 product classes
over the past year via the internet, in a store, through a catalog, or by other means. This section was
originally more elaborate, asking for the frequency ( category) of purchasing each product class by each
mode, but was simplified to reduce the burden on the respondent without entirely sacrificing information
on shopping patterns for items other than the two main product classes of interest.
Part C: A Recent Purchase asked a series of detailed questions about a specific recent purchase in one of
two product classes ( the “ search” goods of books/ CDs/ DVDs/ videotapes, or the “ experience” goods of
clothing/ shoes), and were purchased via one of three modes: over the internet, in a store, or through a
catalog. The questions obtained situational information about the specific purchase occasion, and also
established the means by which the respondent initially became aware of the item, obtained information
about it, and experienced the item before purchasing it.
In Part D: Advantages and Disadvantages of Different Ways of Shopping, respondents were asked to
imagine that they will soon be making a purchase similar to the one discussed in Part C. They were then
invited to evaluate two of the three shopping modes – store, internet, and catalog – with respect to such a
purchase. The decision to present only two modes was again a design concession to reduce respondent
fatigue. The evaluation consisted of 28 statements ( for each mode) that the respondents were again asked
to agree or disagree with on a five- point, Likert- type scale. The first set of statements related to the store
mode for all respondents, as the “ anchor” with which it was presumed they would all be familiar. Since
12
the catalog mode was of secondary interest to the study ( in view of the necessity of reducing the
respondent burden), it was presented in the second set of parallel statements only if it were the chosen
mode for the key purchase; otherwise the second mode was the internet.
Similar to the process for the Part A attitudinal questions, in the design stage, we started with 50
statements in 14 categories. In Table 7 and Table 8, these statements are shown as specific to the internet;
in the survey administration, minor changes were made to the statements to make them specific to store
and catalog shopping in turn. In the final survey, these statements were reduced to 28 in 13 categories, as
shown in Table 9 for the store mode.
Table 5: General Shopping- Related Attitudinal Statements, First Nine Categories
Hypothesized
Construct
* Survey Statement
– Credit cards encourage unnecessary spending.
Credit cards
– I prefer to pay for things by cash rather than credit card.
+ We should raise the price of gasoline to reduce congestion and air pollution.
Environmental–
general + To improve air quality, I am willing to pay a little more to use a hybrid or other clean- fuel vehicle.
– Shopping travel creates only a negligible amount of pollution.
Environmental–
shopping- related + A lot of product packaging is wasteful.
+ Whenever possible, I prefer to walk or bike rather than drive.
Exercise
+ I follow a regular physical exercise routine.
+ When it comes to buying things, I'm pretty spontaneous.
Impulse buying
– I generally stick to my shopping lists.
– I am generally cautious about accepting new ideas.
Innovation
– I prefer to see other people using new products before I consider getting them myself.
+ I would/ do enjoy having a lot of expensive things.
Materialism
– My lifestyle is relatively simple, in terms of material goods.
– It’s too much trouble to find or take advantage of sales and special offers.
Price conscious
+ It’s important to me to get the lowest prices when I buy things.
+ Taking risks fits my personality.
Risk- taking
– “ Better safe than sorry” describes my decision- making style.
* Directionality with respect to construct label.
13
Table 6: General Shopping- Related Attitudinal Statements, Final Ten Categories
Hypothesized
Construct
* Survey Statement
– Shopping is usually a chore for me.
+ I enjoy the social interactions shopping provides.
+ Shopping helps me relax.
Shopping enjoyment
+ Shopping is fun.
+ If I got a lot of money unexpectedly, I would probably spend more of it than I saved.
Spending money
+ Buying things cheers me up.
+ I often introduce new trends to my friends.
Status
+ For me, a lot of the fun of having something nice is showing it off.
+ The internet makes my life more interesting.
Technology–
computer- related – Computers are more frustrating than they are fun.
+ I like to track the development of new technology.
Technology– general
– Technology brings at least as many problems as it does solutions.
+ I'm often in a hurry to be somewhere else when I'm shopping.
Time consciousness
+ I'm too busy to shop as often or as long as I’d like.
+
I am generally doing productive or enjoyable things, such as making phone calls or listening to the
radio, while traveling.
Travel– general
– The only good thing about traveling is getting to the destination.
+ Even if I don't end up buying anything, I still enjoy going to stores and browsing.
– Shopping is too physically tiring to be enjoyable.
+ I like to stroll through shopping areas.
Travel– shopping-related
+ For me, shopping is sometimes an excuse to get out of the house or workplace.
+ People are generally trustworthy.
Trust
– I tend to be cautious with strangers.
– I like a routine.
Variety- seeking
+ “ Variety is the spice of life.”
* Directionality with respect to construct label.
14
Table 7: Design- stage Mode- Specific Attitudinal Statements, First Five Categories ( Internet Mode)
Category Statement
When shopping over the internet, I am confident of getting a desired item within a reasonable amount of time.
If necessary, it is easy to return a product purchased over the internet.
Internet shopping provides poor after- purchase customer service.
Customer service
Internet retailers are generally very receptive to customer feedback.
In my experience, most internet stores keep their commitments.
I am concerned that internet stores will fail to meet my expectations.
I am confident in my ability to determine whether a retailer is trustworthy.
I prefer to shop the internet sites of national chain stores.
Trust
I value the anonymity that shopping on the internet provides.
Internet shopping is easy.
Ease of use The product information I need is easy to find over the internet.
I often find shopping over the internet to be frustrating.
Shopping over the internet makes it easier to obtain certain products that are hard to find elsewhere.
A lot of times, products I want are unavailable over the internet.
When it comes to [ clothing, books], I can find anything I want for sale over the internet.
Availability/ Selection
Certain products I purchase are only available over the internet.
The internet makes it easy to check the availability of products.
The internet makes it simple to compare products.
It is difficult to compare products over the internet.
It is easy to get information from a live person when shopping over the internet.
Search costs ( effort
savings)
It takes too long to obtain product information over the internet.
15
Table 8: Design- stage Mode- Specific Attitudinal Statements, Last Nine Categories ( Internet Mode)
Category Statement
All things considered, buying over the internet saves me time.
Time savings
The internet sites I use allow me to fulfill many of my shopping needs in just one location.
When shopping over the internet, I am confident of getting a desired item within a reasonable amount of time.
Gratification delay
I often have to wait too long to receive a product purchased over the internet.
All things considered, buying over the internet saves me money.
Money savings
Considering shipping costs, [ clothing, books] are usually more expensive when purchased over the internet.
Internet stores often fail to offer enough product information.
It takes too long to find a desired product on the internet.
The product information provided by internet stores is generally up to date.
Information ( broad,
fast, comparison)
Product information on the internet is clear and understandable.
The internet allows me to shop at any time I wish.
Internet shopping is available any time I want it.
Internet shopping is available to me anywhere I would like it to be.
I enjoy being able to shop from home without having to get dressed and go out.
Having to get dressed and go to the store is a hassle.
Convenience
The stores I want are conveniently located.
Internet stores often provide misleading product information.
Internet shopping generally enables me to experience products before buying to the extent that I want to.
Products purchased over the internet often fail to meet my expectations.
Product risk
I'm concerned that a product I purchase over the internet will not perform as expected.
It is risky to release credit card information over the internet.
I am generally nervous about providing personal information over the internet.
Potentially having to pay a fee to return an unsatisfactory product is a reasonable risk to take.
The prospect of having to return a product that I've purchased over the internet doesn't really bother me.
Financial risk
I'm concerned that an internet store will fail to deliver a product I've purchased.
I enjoy shopping over the internet.
General enjoyment
Shopping over the internet is boring.
When it comes to [ clothing, books], I have a strong preference for shopping at one or a few particular internet
Store- brand sites.
attachment
With respect to buying [ clothes, books], I am always on the lookout for a new internet site to check out.
16
Table 9: Final Mode- Specific Attitudinal Statements ( Store Mode)
Category * Statement
Availability/ + When it comes to buying books/ CDs/ DVDs/ videotapes, I can find anything I want in stores.
selection – A lot of times, products I want are unavailable in stores.
+ The stores I want/ need to shop at are conveniently located.
Convenience + Getting dressed and going out is an enjoyable aspect of store shopping for me.
+ Stores are open whenever I want to shop.
Customer – Stores typically provide poor after- purchase customer service.
service + If necessary, it is easy to return a product purchased at a store.
+ The product information I need is easy to find in stores.
Ease of use – I often find shopping in stores to be frustrating.
+ It is risky to release credit card information to stores.
Financial risk + I am uncomfortable about providing personal information to stores.
General – Shopping in stores is boring.
enjoyment + I enjoy shopping in stores.
Gratification + I often have to wait too long for a store to obtain the product I want to purchase.
delay – When shopping in stores, I am able to immediately obtain the products I purchase.
– Considering taxes and other costs, books/ CDs/ DVDs/ videotapes are usually more expensive when
Money savings purchased in stores.
+ All things considered, buying in stores saves me money.
+ I'm concerned that a product I purchase in a store will not perform as expected ( e. g. quality, etc.).
Product risk _ When shopping in stores, I am able to experience products before buying, to the extent that I want to.
Search costs _ It is difficult to compare products at stores.
( effort savings) + When shopping in stores, it is easy to check the availability of products.
– With respect to buying books/ CDs/ DVDs/ videotapes, I am always on the lookout for a new store to
Store- brand check out.
attachment + When it comes to books/ CDs/ DVDs/ videotapes, I have a strong preference for shopping at one or a few
particular stores.
+ I value stores that allow me to fulfill many of my shopping needs in just one location.
Time savings + All things considered, buying in stores saves me time.
+ I prefer to shop at independent stores rather than national chains.
Trust – I value the anonymity ( e. g. paying with cash) that shopping in stores provides.
– I am concerned that unfamiliar stores will fail to meet my expectations.
* Directionality with respect to construct label.
17
In Part E: Frequency of Shopping, more general questions are asked about the frequency of shopping by
mode for the key item discussed in Parts C and D. Part F: Your Use of Internet and Communication
Technology continues the move back to the general by asking questions about internet use, as well as the
use of other technologies. Finally, Part G: Some Information About Yourself asks general
sociodemographic questions that will allow our sample to be compared to more general populations.
5. SURVEY INSTRUMENT DEVELOPMENT
The survey was administered primarily over the internet, though three separate paper survey versions
were also developed. This section discusses the development of both instrument types.
The internet version of the survey was developed using the commercial software vendor Zoomerang
( http:// www. zoomerang. com). Zoomerang allows users to develop surveys and then handles the
administration and data collection of the surveys. Though the software does limit the question types and
format of the instrument, the advantages in terms of security and data collection made the choice to use a
commercial vendor superior to collecting the data on our own. The advantages and disadvantages, in the
context of this survey, of using a web- based commercial vendor in general, and Zoomerang in particular,
are as follows:
􀂾 Advantages
􀂃 Data- entry is done automatically and accurately;
􀂃 Issues related to web- security and server/ database management are taken care of;
􀂃 Development time is much faster;
􀂃 Data can easily be downloaded;
􀂃 Relatively low cost;
􀂃 The software is constantly improving.
􀂾 Disadvantages
􀂃 The types of questions that can be asked are limited;
􀂃 Branching logic can only be related to a single response at a time;
􀂃 The software did not permit the respondent to save a partially- completed survey and
automatically return to the same point at the next log- in;
􀂃 Hitting the “ back” button on one’s web browser and changing one’s response on a previous
screen did not overwrite the previous response, so the only way respondents could change
their answers after moving past a page was to start over entirely;
􀂃 Only one type of “ button” is available for each question type and in the case of multiple
choice questions, the difference between the “ clicked” and “ not clicked” images is not great;
􀂃 The font size and style can be changed using html tags, but not on a global level ( i. e. each
statement/ question/ instruction required its own html tag);
18
􀂃 Entered data, such as an identification number, cannot be checked against an existing
database;
􀂃 Although individual responses could be limited to the provided options, multiple responses
could not be checked for internal consistency ( e. g. that number of workers did not exceed
number of household members) in real time;
􀂃 No “ thermometer” showing percent of survey completed was automatically available; we
manually created one at several points in the survey;
􀂃 The order of questions could not be made random;
􀂃 Respondents could not be automatically prompted to complete missing responses without
forcing them to do so – i. e. there was no ability just to remind respondents that a field was left
blank in case it was inadvertent, yet to reduce respondent irritation we wanted to limit the
number of questions for which we required a response;
􀂃 Neither survey pages nor questions could be copied and pasted, which made creating this
survey incredibly tedious and subject to typographical errors;
􀂃 Zoomerang customer service was largely unresponsive to problems with their software,
which, as with all software, did have bugs;
􀂃 The survey is administered on a Zoomerang- hosted website, which made a few respondents
concerned about the legitimacy of our claimed relationship with UC Davis.
The Zoomerang service works by allowing the user to design any number of survey pages. Within each
survey page the user can include any number of survey questions. There are a variety of question types,
including multiple choice, ranking, write- in, etc. Answers to any particular question can then be used to
facilitate skipping logic between pages ( e. g. if the answer to Question 4 is “ yes”, go to page 4). After
developing the survey, a website is then built for survey administration. The Zoomerang software then
takes care of the administration and data storage.
Figure 3 presents a flow chart of the web pages used in the survey administration and this figure is
discussed in detail here. Each box in the figure represents a group of web pages. The top number in the
box refers to the web page numbers included in that section; the next line gives the “ Part” of the survey of
which the pages belong to ( see Section 4); the last line gives a description of the pages.
The first page in the survey and the top box in Figure 3 requires the user to enter a 10- digit identification
number ( see Section 6 of this report for more information on the ID number). The next three boxes in the
figure reference the Welcome, Part A, and Part B portions of the survey, which include web pages 2, 3 to
5, and 6, respectively. All users complete these portions of the survey.
The branching starts in Part C, which focuses on the purchase of a particular item in one of two product
classes – books/ CDs/ DVDs/ videotapes (“ search” goods) or clothing/ shoes (“ experience” goods) –
purchased via one of three modes: over the internet, in a store, or through a catalog. We narrowed the
questioning to specific product classes on the assumption ( supported by other research) that relevant
variables could be weighted differently depending on the nature of the product. We chose relatively low-cost,
high- frequency- of- purchase product types to ensure the presence of a sufficient proportion of
relatively recent purchase occasions in the sample, and we chose these two to represent the difference
between experience goods ( those often needing to be tried in some way before being purchased) and
19
search goods ( those that can often be satisfactorily evaluated on the basis of externally- provided
information alone; Peterson et al., 1997). The product class pertaining to a given respondent will hence-forth
be referred to as the “ key purchase” or “ key item”.
The first branching is based on the purchase mode. We first inquire about a recent ( within six months or
so) internet purchase of one of the key items ( on page 7). If such a purchase was made, the respondent is
directed to page 8 and the survey continues with more detailed questions about the internet purchase. If
such a purchase was not made, we inquire about a recent purchase made in a store ( page 9). Again, if such
a purchase has been made, the survey continues with more detailed questions about the store purchase. If
no recent store purchase has been made, we ask about a catalog purchase. If no recent purchase of a key
item has been made via any of these three modes, the survey inquires about any purchase of a key item
the respondent can recall. If the respondent cannot recall the purchase of any of the key items ( book, CD,
DVD, videotape, clothing, or shoes), they are directed to Part F of the survey.
The order of the mode questions heavily influences the “ mode shares” of the key purchases. As such, we
made two versions of the internet survey. The first inquired about internet purchases first ( and store
purchases second), as presented above and as shown in the figure. The second asked about store pur-chases
first ( and internet purchases second). The two surveys were deployed at different times ( out of
view of the respondents) to try and gather a balanced sample in terms of store- versus- internet mode
shares. The internet- first version was active from June 1 - 13, 2006 and collected 439 ( not all unique or
complete) responses; the store- first version was active from June 14 - September 14, 2006 and collected
649 responses ( again, not all unique or complete). As the catalog aspect of the survey was of secondary
interest, the catalog option was always asked last.
After branching based on shopping mode, Part C then branched on item type. Pages 8, 10, 12, and 14 to
16 ask which key item, among books, CDs, DVDs, videotapes ( the search goods) and clothing and shoes
( experience goods) was purchased most recently. This branching resulted in respondents being directed
down one of six Part C tracks representing the item- mode combination of their key purchase, namely:
book- internet ( pages 17 to 27; “ book” is used to represent all the search goods), clothing- internet ( pages
28 to 38; “ clothing” is used to represent all the experience goods), book- store ( 39 to 49), clothing- store
( 50 to 60), book- catalog ( 61 to 71), and clothing- catalog ( 72 to 82). In this portion of the survey, the
respondent answers a series of detailed questions regarding the key purchase.
Part D of the survey asks respondents to compare shopping modes in the context of purchasing the key
item they described in Part C. First, each respondent is asked to respond to 28 statements about
purchasing the key item, either for books/ CDs/ DVDs/ videotapes or clothing/ shoes, in a store. Next, the
respondent does the same for 28 companion statements for either the internet ( if the item was purchased
in the internet or store) or a catalog ( if the key item was purchased in a catalog). Thus, Part D collapses
into the four tracks ( book- store + book- internet; clothing- store + clothing- internet; book- store + book-catalog;
clothing- store + clothing- catalog) shown in Figure 3.
Part E asks only item- specific questions, thus collapsing the tracks into two, book and clothing. Finally,
Parts F and G are asked of each respondent.
20
No No No No
Yes Yes Yes Yes
Part E Part E
Use of technology
106 to 117
Part G
Socio-demographics
103 to 105
Part F
Part D
Book, catalog
99, 100 101, 102
Book purchases Clothing
purchases
Part D
Clothing, catalog
93, 94
95, 96
Part D
Clothing, store
85, 86 97, 98
Part D
Book, internet
89, 90
Part D
Clothing, internet
Clothing, catalog
83, 84
Part D
Book, store
87, 88
Part D
Clothing, store
91, 92
Part D
Book, store
Clothing, store
11
Part C
Recent catalog
12
Part C
Purchase type
61 to 71
Part C
Book, catalog
9
Part C
Recent store
10
Book, internet
13
Part C
Recent
14 to 16
Part C
Product and mode
28 to 38
Part C
Purchase type
17 to 27
Part C
50 to 60
Part C
39 to 49
Part C
72 to 82
Part C
Part C
Recent internet
8
Part C Part C
Purchase type
1
ID Number
xx- xxxxx- xx
2
Welcome
Where would you
like to shop
3 to 5
Part A
Attitudes
6
Part B
Purchases
7
Clothing, internet Book, store
Figure 3: Internet Instrument Flow Chart
As discussed in the next section, paper versions of the survey were made available to those with either a
preference for a paper survey or an inability to complete the internet version. Because of the limiting
nature of a paper survey, the full branching options included in the web survey were not available to those
21
completing the paper surveys. To partially reduce this constraint, the following three paper surveys were
administered:
􀂾 Version 1: Book
􀂃 Captures book/ CD/ DVD/ videotape purchases;
􀂃 Part C: asks first about a recent internet purchase; if none is recalled, asks about a store
purchase;
􀂃 Part D: first set of statements relates to store purchases, second set to internet purchases.
􀂾 Version 2: Clothing- internet
􀂃 Captures clothing/ shoe purchases;
􀂃 Part C: asks first about a recent internet purchase; if none is recalled, asks about a store
purchase;
􀂃 Part D: first set of statements relates to store purchases, second set to internet purchases.
􀂾 Version 3: Clothing- catalog
􀂃 Captures clothing/ shoe purchases;
􀂃 Part C: asks first about a recent catalog purchase; if none is recalled, asks about a store
purchase;
􀂃 Part D: first set of statements relates to store purchases, second set to catalog purchases.
As those requesting a paper survey had to do so by e- mail or phone, screening questions were used to
determine the most appropriate paper survey for each user. The screening questions are as follows:
1. Do you have access to the internet? Yes 􀃆 Question 2; No 􀃆 Version 3;
2. ( If “ yes” to question 1:) Do you shop more often for books or clothing? Books 􀃆 Version 1;
Clothing 􀃆 Version 2.
It should be noted that Versions 1 and 2 ( and the catalog track of Version 3) of the paper surveys are
subsets of the web- based survey and, as such, the results of the paper survey were entered directly into the
web- based survey. A modified web survey was developed for the store track of Version 3 in order to enter
those surveys into the database. A different survey was needed because the Version 3 paper surveys
presented catalog- and store- specific Part D statements, whereas in the web survey, those purchasing
items in a store are presented store- and internet- specific Part D statements. The modified web survey
included the catalog- and store- specific Part D statements for those making a purchase in a store.
6. DATA COLLECTION
The data were collected over a three- month period from June to August, 2006. The first step in the pro-cess
was sending out a recruitment letter to each of the 8,000 selected addresses. We debated several
different ways to address the envelopes and letters, either to the householder by name ( as provided by the
vendor), to “ Current Resident”, or to “[ name] or Current Resident”, and whether to treat the envelopes the
22
same way as the letters. The advantage of the latter two approaches for the envelope is that no mail
should be returned as undeliverable due to the occupant in the vendor’s database having moved; the
obvious disadvantage is that the letter is immediately marked as a mass mailing, possibly “ junk mail”,
and therefore more likely to be discarded. On the other hand, using the resident’s name only was
potentially “ friendlier”, but also potentially more threatening or annoying to some (“ how did you get my
name and know where I live?”), and had the additional disadvantage that if the resident had moved, the
letter would be returned as undeliverable ( a non- trivial factor, as shown below). Ultimately, we chose to
use the resident’s name on the envelope, and “ Dear [ city] resident” as the salutation on the letter – hoping
that the name on the envelope would get it opened, while the anonymity of the salutation would reduce
the threat level. The envelopes were especially printed with the return address of “ Prof. Patricia L.
Mokhtarian”, to pique curiosity and to further distinguish the recruitment from commercial solicitations.
The recruitment letters were identical for Davis and Santa Clara residents, except for ( 1) the two
appearances of the city name, ( 2) the use of two different telephone numbers for David Ory so that calls
to him would be essentially local for residents of either city ( a 530 area code number for the Davis letters,
and a 415 area code number for the Santa Clara letters), and ( 3) the word “( collect)” after the 530 area
code telephone number for Patricia Mokhtarian on the Santa Clara letter. A copy of the letter sent to the
Davis residents is presented in Figure 4.
The letters were printed by the Reprographics Division of the University of California, Davis, and folded,
stuffed, and mailed out on June 5, 2006 by the UCD Bulk Mail Center.
The identification number mentioned in the letter is a unique 10- digit code that contains the respondent’s
census tract, street address, and unique identifying number ( for those in apartment complexes). The code
uses a private formula to disguise the linkage to location from any unauthorized observer: the first two
digits are unique to each of the 35 census tracts in the target sample; digits 3 through 7 contained a
scrambled version of the street address, wherein the address numbers are converted one to one ( 0= 8, 1= 4,
2= 6, 3= 0, 4= 7, 5= 3, 6= 5, 7= 2, 8= 1, 9= 9), and the order is scrambled ( 3rd digit, 1st digit, 4th digit, 2nd digit,
5th digit), with 0s filling in the blanks, and the 0’ s are then converted to 8s, so that 1234 becomes
84607 􀃆 68047. The reason for devising such a code was to preserve useful information ( rather than
being just a random identifying number) in case the correspondence between the code and the addresses
was misplaced or disturbed. The web- based survey instrument required the respondent to enter the 10-
digit code before proceeding with the survey ( though any number, or string of characters, for that matter,
could be entered in this location of the survey). The paper- based surveys had the 10- digit numbers
recorded on them before being mailed out. However, the codes resulted in numerous entry errors on the
part of the internet respondents – most of which could be identified and fixed, but not all. In retrospect,
the benefit of the code probably did not exceed its cost.
Approximately two weeks after sending out the recruitment letter, a postcard reminder was supposed to
be sent to everyone who had not completed the survey by the time the information had to be provided to
Bulk Mail ( which turned out to be everyone except the 320 individuals who completed the survey within
the first couple of days). One week later, a second postcard reminder was to be sent out to those who had
not yet completed the survey. Accordingly, we had 7,800 copies of the first postcard printed in advance,
and 7,600 copies of the second. Due to an error in the mail room, however, the second postcard was
labeled for mailing first, and we were notified when they ran out of postcards ( having only 7,600 instead
of the 7,680 needed for the first mailing). Once the mistake was identified, the 7,680 first postcards were
labeled and mailed out. A week later, the 7,600 erroneously labeled second postcards were mailed
( without removing those who had returned the survey, fearing that that could lead to further errors),
leaving the last 80 names on the first- reminder mailing list without the second reminder postcard.
23
Figure 4: Davis Recruitment Letter
24
The first postcard reminder was printed on mint green card stock and is included as Figure 5 below. It was
mailed on June 20, 2006. The second postcard reminder, shown in Figure 6, was printed on bright yellow
card stock and mailed on June 26, 2006.
Figure 5: First Postcard Reminder
Figure 6: Second Postcard Reminder
25
The web surveys were “ active” from June 1 to September 14, collecting a total of 996 unique responses
( complete and partial). Combined with 71 returned paper surveys ( out of 80 mailed out), a total of 1,067
surveys were returned ( note: these numbers and all those in tables below do not include one additional
paper survey returned months later, but still in time to include in future analyses). To compute a response
rate for the surveys, we first subtracted the number of “ bad” addresses provided by the commercial
vendor. These were computed by tallying ( by city) the number of recruitment letters returned as
undeliverable. A total of 1,426 letters were returned to the University, indicating that 17.8% of the
addresses supplied by the vendor were out- of- date in terms of the name of the current resident ( not
surprisingly in view of the mobility of its heavily college- related population, the bad address rate was
higher in Davis than in Santa Clara). Subtracting this number from 8,000 leaves 6,574 households who
( presumably) received our correspondence. Of the 6,574, 1,068 completed the survey, resulting in a
response rate of approximately 16 percent.
Table 10 summarizes the response rate by city. Interestingly, the response rate for Davis residents is
considerably higher ( at near 23%) than for Santa Clara residents ( at near 9%). We attributed this
difference to two factors. First, it is likely that the affection Davis residents have for the University of
California campus in their city, which accounts for the lion’s share of economic activity in the city,
accounted for the majority of the boost in response rate. Also, by coincidence the survey corresponded
with a debate in the city over the proposed construction of a Target Store ( no so- called “ big- box” retail
stores are currently located in Davis). It is likely ( anecdotally corroborated by conversations with several
respondents and by several comments written on the survey) that those with strong opinions towards the
Target store saw the survey as an opportunity to share their views about the issue.
Table 10: Response Rate by City
Quantity Davis Santa Clara Unknown Total
Mailed out 4,000 ( 100%) 4,000 ( 100%) --- 8,000 ( 100%)
Undeliverable mail 754 ( 18.9%) 674 ( 16.9%) --- 1,428 ( 17.9%)
Submitted
( complete and
partial) surveys
760 297 10** 1,067
Response rate* 23.4% 8.9% --- 16.2%
* Percent of delivered letters resulting in a submitted survey. ** ID codes entered incorrectly so location could not be
ascertained, but survey appeared to be legitimate.
The first web- based survey was submitted on June 9; the last survey was submitted on August 22, 2006.
Figure 7 shows the distribution of completed surveys over time. Note that the chart does not include
partial responses, as they are not time- stamped by the website. Paper surveys entered into the web- based
survey, by us, are also omitted from the distribution.
More than 13% of all completed responses were collected on the first day and over half by the end of the
first week. The impact of both postcard reminders is evident in the chart: more than 40% of all responses
were received after the first reminder was mailed, with almost 20% of the total occurring after the second
reminder. However, the timing of the first reminder corresponded with the stated deadline for entering the
raffle for the cash prize, and thus that “ jump” in the response rate could be the result of both effects. The
data collection was essentially complete after one month.
26
Responses over Time
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
0 10 20 30 40 50 60 70
Days after First Response
Cumulative Share of Responses
50% reached on 7th day
14% ( 128) on first day
95% reached after ~ one month
1st postcard
reminder
2nd postcard
reminder
Figure 7: Completed Surveys over Time
7. INITIAL DATA CLEANING
The initial data cleaning activities focused on three aspects of the survey: the attitudinal statements in Part
A and D, and the sociodemographic data in Part G. As the attitudinal statements allow for the most
interesting analyses of the data and the sociodemographics allow for behaviors to be related to general
populations, these portions of the survey are viewed as the most crucial.
Part A contains 42 attitudinal statements to which respondents can either agree or disagree on a five- point
Likert- type scale. The distribution of completed responses is shown in Table 11. A natural cut- off was 3
or fewer missing data items, after which the distribution thins out dramatically. As such, those missing
more than 3 responses were excluded from the working sample. Those missing three or fewer responses
had their missing data imputed with geographically- segmented means. Using census tracts, the sample
was segmented into six imputation groups ( see Table 12). Mean values on a given statement were then
computed for each of these groups, rounded to the nearest valid answer ( the integers between 1 and 5,
inclusive) and imputed accordingly.
As can be deduced from Table 11, only 108 out of 1,033 x 42 = 43,386 responses ( 0.25%) were imputed
in this way. No single question in Part A had more than 8 ( 0.77% of 1,033) responses imputed.
27
Table 11: Part A Missing Data Summary
Missing Valid Frequency Percent Cumulative
Percent
0 42 946 88.7% 88.7%
1 41 72 6.7% 95.4%
2 40 9 0.8% 96.3%
3 39 6 0.6% 96.8%
5 37 2 0.2% 97.0%
7 35 2 0.2% 97.2%
8 34 1 0.1% 97.3%
9 33 1 0.1% 97.4%
14 28 4 0.4% 97.8%
28 14 4 0.4% 98.1%
31 11 4 0.4% 98.5%
42 0 16 1.5% 100.0%
Total 1,067 100.0%
Table 12: Geographic Imputation Groups
Imputation Group N Census Tracts
1 281 105. XX
2 335 106. XX
3 144 107. XX
4 48 5049. XX
5 155 5050. XX to 5053. XX
6 94 5054. XX to 5060. XX
Key: “ XX” represents any number, i. e. 105. XX = 105.01, 105.02, 105.03, etc.
A similar selection and imputation process was performed on the Part D data. Recall from the previous
section that Part D is segmented by product type and shopping mode. Part D, Section 1 contains the
questions regarding store purchases ( for both books ( search) and clothing ( experience)); Section 2
contains questions regarding either internet or catalog purchases – again for both item types. The
distribution of these two sections are shown in Table 13 ( Section 1) and Table 14 ( Section 2). A cut- off of
28
three missing was again implemented. Those missing more than three responses in either section were
excluded from the working sample.
The imputed means were specific to geography ( using the groups of Table 12), item type ( book or
clothing), and shopping mode ( for Section 2 only – catalog or internet). Only 77 of 991x28 = 27,748
responses ( 0.28%) in Section 1 ( store) were imputed in this way, with no single question in Section 1
having more than 6 responses ( 0.61% of 991) imputed. Only 91 of 981x28 = 27,468 responses ( 0.33%)
in Section 2 ( internet or catalog) were imputed in this way, with no single question in Section 2 having
more than 8 responses ( 0.82% of 981) imputed.
Table 13: Part D, Section 1 Missing Data Summary
Missing Valid Frequency Percent Cumulative
Percent
0 28 925 86.7% 86.7%
1 27 57 5.3% 92.0%
2 26 7 0.7% 92.7%
3 25 2 0.2% 92.9%
6 22 2 0.2% 93.1%
7 21 1 0.1% 93.2%
14 14 6 0.6% 93.7%
16 12 1 0.1% 93.8%
28 0 66 6.2% 100.0%
Total 1,067 100.0%
Table 14: Part D, Section 2 Missing Data Summary
Missing Valid Frequency Percent Cumulative
Percent
0 28 906 84.9% 84.9%
1 27 61 5.7% 90.6%
2 26 12 1.1% 91.8%
3 25 2 0.2% 91.9%
5 23 1 0.1% 92.0%
12 16 1 0.1% 92.1%
14 14 4 0.4% 92.5%
15 13 1 0.1% 92.6%
28 0 79 7.4% 100.0%
Total 1,067 100.0%
29
Though numerous demographic measures were captured by the survey, the most important were felt to
be: gender, number of vehicles, number of workers, household size, employment status, and income.
Taking each of these six questions as an individual item, the distribution of missing items is shown in
Table 15. The seemingly magic number of three was again used as a cut- off: those missing more than
three of these six items were excluded from the working sample. No imputation was done for the missing
values.
Table 15: Sociodemographic Missing Data Summary
Missing Items Valid Items Percent Cumulative
Percent
0 856 80.2% 80.2%
1 118 11.1% 91.3%
2 23 2.2% 93.4%
3 8 0.7% 94.2%
4 1 0.1% 94.3%
5 4 0.4% 94.7%
6 57 5.3% 100.0%
Total 1,067 100.0%
Using the cut- off criteria for Part A, Part D, and the sociodemographics, our original sample of 1,067
responses was reduced to 966 ( 91% of the original total) via a sequential process illustrated in Figure 8.
The 966- observation file will become the “ working file” on which all subsequent work will be based.
Table 16 through Table 19 summarize the missing and/ or imputed data distribution for the final working
file.
30
N = 966
Case discarded if missing more than 2 of the following items:
gender, number of vehicles, number of workers, household
size, employment status, and income
Final sample
N = 1,067
Remaining N = 1,033
Remaining N = 984
Remaining N = 973
Remaining N = 966
Part D, Section 1 cleaning
Case discarded if missing more than 3 responses out of 28
Sociodemographic cleaning
Part D, Section 2 cleaning
Case discarded if missing more than 3 responses out of 28
Raw sample
Part A cleaning
Case discarded if missing more than 3 responses out of 42
Figure 8: Data Cleaning Process
Table 16: Part A Missing ( Imputed) Data Summary for Final Sample
Missing Valid Frequency Percent Cumulative
Percent
0 42 889 92.0% 92.0%
1 41 65 6.7% 98.7%
2 40 8 0.8% 99.6%
3 39 4 0.4% 100.0%
Total 966 100.0%
31
Table 17: Part D, Section 1 Missing ( Imputed) Data Summary for Final Sample
Missing Valid Frequency Percent Cumulative
Percent
0 28 904 93.6% 93.6%
1 27 55 5.7% 99.3%
2 26 6 0.6% 99.9%
3 25 1 0.1% 100.0%
Total 966 100.0%
Table 18: Part D, Section 2 Missing Data Summary for Final Sample
Missing Valid Frequency Percent Cumulative
Percent
0 28 897 92.9% 92.9%
1 27 58 6.0% 98.9%
2 26 9 0.9% 99.8%
3 25 2 0.2% 100.0%
Total 966 100.0%
Table 19: Sociodemographic Missing Data Summary for Final Sample
Missing Items Valid Items Percent Cumulative
Percent
0 828 85.7% 85.7%
1 116 12.0% 97.7%
2 22 2.3% 100.0%
Total 966 100.0%
Table 20: Target Status of Working File Cases
Santa Clara Davis
Measure
Traditional Suburban Traditional Suburban
On- target addresses 41 53 109 237
Off- target addresses 177 346
32
8. SUMMARY AND NEXT STEPS
The survey design and data collection effort can be considered reasonably successful, producing a
relatively clean dataset with enough cases to permit the application of numerous statistical methods. The
dataset is extraordinarily rich with attitudinal variables, as well as having a number of different behavioral
indicators and the conventional sociodemographic traits. In fact, these data offer the most comprehensive
set of shopping- related variables that we have seen empirically measured by a single study. Accordingly,
we believe they will continue to provide useful insights for some time to come, especially with respect to
the role the internet is playing in the shopping behavior of ordinary Americans.
Initial analysis plans call for factor- analyzing the Part A and Part D attitudinal statements, and clustering
cases on factor scores to identify market segments having similar attitudinal profiles. It is also of interest
to examine attitudes differ by mode as well as product type ( book/ CD/ DVD/ videotape, a search good,
versus clothing/ shoes, an experience good). Eventually, discrete choice models will be developed using a
variety of dependent variable formulations ( actual and intended choices for a single purchase, as well as
frequencies and shares for multiple purchases), and beyond that, applications of more sophisticated
methodologies such as latent class and structural equations models await.
REFERENCES
Babbie, Earl ( 1998) The Practice of Social Research, 8th edition. Belmont, CA: Wadsworth Publishing.
Baumgartner, Hans and Jan- Benedict E. M. Steenkamp ( 2001) Response styles in marketing research: A
cross- national investigation. Journal of Marketing Research 38( May) 143- 156.
Brownstone, David ( 1998) Multiple imputation methodology for missing data, non- random response,
and panel attrition. Chapter 18 in Tommy Garling, Thomas Laitila, and Kerstin Westin, eds. Theoretical
Foundations of Travel Choice Modeling. Amsterdam: Elsevier, 421- 449.
Brynjolfsson, Erik and Michael D. Smith ( 2001) The great equalizer? Consumer choice behavior at
Internet shopbots. Working paper, Sloan School of Management, Massachusetts Institute of Technology,
Cambridge, MA, April. Available at ebusiness. mit. edu/ papers/ tge.
Burke, Raymond R. ( 1998) Real shopping in a virtual store. In Sense and Respond: Capturing Value in
the Network Era. Eds. Stephen P. Bradley and Richard L. Nolan. Boston: Harvard Business School.
Cairns, Sally ( 1996) Delivering alternatives: Successes and failures of home delivery services for food
shopping. Transport Policy 3( 4), 155- 176.
Cao, Xinyu and Patricia L. Mokhtarian ( 2005) The Intended and Actual Adoption of Online Purchasing:
A Brief Review of Recent Literature. Research Report UCD- ITS- RR- 05- 23, Institute of Transportation
Studies, University of California, Davis, May.
Available at http:// repositories. cdlib. org/ itsdavis/ UCD- ITS- RR- 05- 07/.
Ellard, John H. and T. B. Rogers ( 1993) Teaching questionnaire construction effectively: The Ten Com-mandments
of question writing. Contemporary Social Psychology 17( 1) ( March), 17- 20.
Fabrigar, Leandre R., Duane T. Wegener, Robert C. MacCallum, and Erin J. Strahan ( 1999) Evaluating
the use of exploratory factor analysis in pyschological research. Psychological Methods 4( 3), 272- 299.
33
Farag, Sendy, Kevin J. Krizek, and Martin Dijst ( 2006) E- shopping and its relationship with in- store
shopping: Empirical evidence from the Netherlands and the USA. Transport Reviews 26, 43- 61.
Ferrell, Christopher E. ( 2005) Home- based teleshopping and shopping travel: Where do people find the
time? Transportation Research Record 1926, 212- 223.
Gould, Jane and Thomas F. Golob ( 1997) Shopping without travel or travel without shopping? An
investigation of electronic home shopping. Transport Reviews 17( 4), 355- 376.
Gould, J., T. Golob, and P. Barwise ( 1998) Why do people drive to shop? Future travel and
telecommunication trade- offs. Presentation at the Annual Meeting of the Transportation Research Board,
Washington, DC, January.
Handy, S. and T. Yantis ( 1997) The Impacts of Telecommunications Technologies on Nonwork Travel
Behavior. Research Report SWUTC/ 97/ 721927- 1F, Southwest Region University Transportation Center,
Center for Transportation Research, The University of Texas at Austin, January.
Krizek, Kevin J., Yi Li, and Susan L. Handy ( 2005) Spatial attributes and patterns of use in household-related
ICT activity. Paper no. 05- 1933 on the conference CD of the 84th Annual Meeting of the
Transportation Research Board, Washington DC, January.
Koppelman, Frank, Ilan Salomon, and Kimon Proussaloglou ( 1991) Teleshopping or store shopping? A
choice model for forecasting the use of new telecommunications- based services. Environment and
Planning B: Planning and Design 18, 473- 489.
Peterson, Robert A., Sridhar Balasubramanian, and Bart J. Bronnenberg ( 1997) Exploring the
implications of the internet for consumer marketing. J. of Academy of Marketing Science 25 ( 4), 329- 346.
Ren, and Mei- Po Kwan ( 2005) The impact of e- shopping on individual temporal constraints and out- of-home
activities. Paper 05- 1647 on the CD- ROM of the 84th Annual Meeting of the Transportation
Research Board, January, Washington DC.
Salomon, Ilan and Frank S. Koppelman ( 1988) A framework for studying teleshopping versus store
shopping. Transportation Research A 22( 4), 247- 255.
Tacken, M. ( 1990) Effects of teleshopping on the use of time and space. Transp. Res. Rec. 1285, 89- 91.
Tauber, E. ( 1972) Why do people shop? Journal of Marketing 36, 46- 49.