Content uploaded by Peng Huang
Author content
All content in this area was uploaded by Peng Huang on Sep 21, 2020
Content may be subject to copyright.
Electronic copy available at: http://ssrn.com/abstract=1266432
1
NO LONGER A TAX HAVEN? THE IMPACT OF TAXES ON
INTERNET PURCHASE BEHAVIOR*
Peng Huang
College of Management, Georgia Institute of Technology, Atlanta, GA 30308, peng.huang@mgt.gatech.edu
Nicholas H. Lurie
College of Management, Georgia Institute of Technology, Atlanta, GA 30308, lurie@gatech.edu
Using the online transaction data of 88,814 U.S. households in 2006, we analyze how local tax rates affect
online purchasing behavior. Although earlier survey-based research has found that consumers who live in
high-tax localities are more likely to shop online, our transaction-based data show the opposite. We find
that higher local tax rates are associated with lower online expenditures, reduced transaction frequency, and
a lower probability of making an online purchase. A disaggregate analysis shows that increased sales tax
does not significantly boost demand from tax avoiding retailers but significantly lowers demand for online
retailers that collect tax. In addition online shoppers are more than twice as sensitive to tax as traditional
store shoppers. Finally, we document that tax losses from Internet sales are more moderate than previously
estimated.
Key words: Internet, e-commerce, tax, public policy
*The authors thank comScore Media Metrix, Inc., and SIMPOVA Web Commerce, for providing the data
used here, as well as Chris Forman, Jerry Thursby, and David Sjoquist for their helpful comments.
Electronic copy available at: http://ssrn.com/abstract=1266432
2
1. Introduction
With retail e-commerce sales in the United States reaching $109 billion in 2006, an increase of 23.5 per-
cent from the previous year (U.S. Census Bureau 2007), state and local governments are concerned about
falling revenues from online evasion of sales tax (Bruce and Fox 2000, Newman 1995) and traditional
retailers are concerned that Internet retailers have an unfair advantage (Anderson et al. 2008). Most re-
cently, New York State has enacted a law that requires online sellers to collect sales tax even if they
maintain no physical operations in the state (McGregor 2008, Yang 2008). However, federal policy mak-
ers are cautious about taxing e-commerce discriminately, in fear that such a tax would discourage people
from conducting business online, thwart entrepreneurial activity, and hurt the development of a vibrant
branch of the U.S. economy (Andal 1997).
In support of the idea that consumers use the Internet to avoid paying local sales taxes, earlier studies,
using 1997 and 2001 surveys of consumers’ Internet purchase behavior, found that people living in high
tax areas are more likely to shop online (Alm and Melnik 2005, Ballard and Lee 2007, Goolsbee 2000).
However, the environment has dramatically changed since this early research. In particular, large retail-
ers, such as such as Toys “R” Us, Wal-Mart, and Target, have begun to collect taxes from consumers who
live in states where these retailers have a physical presence.1 In addition, Internet users more closely re-
semble the U.S. population as a whole (Pew Internet 2008) and 94% of U.S. Internet users have made at
least one online purchase (The Nielson Company 2008) compared to the 20% in earlier research (e.g.,
Goolsbee 2000). Beyond these environmental changes, and their implications for consumer behavior, the
use of survey data has limited prior research to examining the association between tax rates and whether
or not consumers make online purchases. From a public-policy standpoint, the more critical question is
1 Although the U.S. Supreme Court ruled that retailers with a physical presence, or nexus, in a state can be required to collect
sales taxes for the state from where a consumer makes an online purchase, it did not clearly define what constitutes physical pres-
ence (Hellerstein 1997). Prior to 2003, many large retailers avoided collecting taxes from online customers by establishing a sep-
arate legal subsidiary to handle Internet business (Cornia et al. 2004). In February 2003, a consent decree was reached between
38 states and the District of Columbia and large retailers, including Toys “R” Us, Wal-Mart, and Target, that agreed to begin col-
lecting sales taxes from customers living in states with physical stores in exchange for being absolved from liability for back
sales taxes (CNET 2003). Another 30 retailers began collecting taxes by the end of 2004 in return for amnesty on taxes owed on
previous online sales (Krebs and Krim 2003).
3
the link between tax rates and dollar expenditures since this is what determines the revenue that state and
local governments potentially lose from tax-avoiding Internet purchases.
In this article, we use household transaction data to examine the relationship between local tax rates and
purchase behavior on the Internet over a year-long period. Specifically, we study the relationship between
local sales tax rates, online shopping expenditures, purchase frequency, and purchase probability in the
year 2006. To the best of our knowledge, this is the first empirical study to use transaction data on con-
sumers’ complete online shopping baskets to examine the effect of sales tax on Internet purchase beha-
vior. Although prior research has focused on consumer incentives to avoid tax by making online purchas-
es (Alm and Melnik 2005, Ballard and Lee 2007, Goolsbee 2000), we point out that sales tax potentially
has two competing effects on consumer demand for online products: (1) For products for which tax
avoidance is possible, for example by buying from out-of-state sellers, an increase in the local tax rate
may increase online demand; (2) However, for products that are only available from a tax-collecting sel-
ler, an increase in local tax rates reduces online demand through income and product substitution effects.
The overall impact of sales tax on consumers’ purchase behavior depends on the relative size of these
demand-reducing and demand-increasing effects.
We assess the effects of tax on online demand by tracking the Internet purchase behavior of a repre-
sentative sample of over 88,000 U.S. households during the year 2006 and matching their behavior to
state and local tax rates. We control for individual household, regional, and metropolitan characteristics as
well as potential border effects (Asplund et al. 2007, Ballard and Lee 2007, Trandel 1992). We first assess
the aggregate effects of sales tax on online purchase behavior. Then, to separate the demand-increasing
and demand-reducing effects of sales tax, we compare the effects of sales tax on purchases by consumers
who live in states where retailers have no physical presence with those by consumers who live within
states with a physical presence—and whose purchases are subject to tax.2
2 We do not seek to disentangle income from substitution effects in reducing demand; rather our focus is on comparing the de-
mand-reducing and demand-increasing effects of tax.
4
We find, contrary to earlier research based on survey data, that people living in high sales tax locations
have lower total online expenditures, make fewer purchases, and are less likely to make an online pur-
chase during the year-long study period. These results suggest that the effect of sales tax in reducing on-
line demand for taxable items exceeds its effect in encouraging consumers to avoid taxes by purchasing
from out-of-state retailers. Disentangling these effects shows that, although an increase in sales tax signif-
icantly reduces the online demand of consumers who live in states where a retailer has a presence, it does
not increase demand from those living in states without a retailer presence. To better understand the poli-
cy implications of taxing e-commerce we: (1) Compare the tax elasticities of online and offline purchases;
and (2) Estimate the amount of tax revenue lost through online sales. We find that the demand elasticity
of sales tax is much larger for online than traditional retail store purchases (-3.76 vs. -1.35). In addition,
we estimate that in 2006 tax was avoided on about 14.9 percent of online transactions, equivalent to tax
revenue losses of $1.1 billion, or one percent of $108.7 billion in total e-commerce retail sales for that
year.
We contribute to research on the policy implications of Internet taxation in several ways. First, our re-
sults indicate that tax avoidance does not typify current online purchase behavior. The fact that the major-
ity of purchases are from retailers that collect sales tax means that an increase in tax rates reduces demand
for online as well as offline goods. Second, results showing that the demand elasticity with respect to tax
is more than twice as high for online than offline purchases supports arguments that e-commerce should
receive preferential tax treatment from an efficiency standpoint (Goolsbee and Zittrain 1999). Third, our
results suggest that tax revenue losses from e-commerce are more moderate than those suggested by earli-
er studies (Bruce and Fox 2000).
The next section briefly reviews current policy and prior research on Internet taxation. Section 3
presents the theoretical model and propositions. Section 4 describes the data and empirical model. Section
5 presents the aggregate effect of sales tax on e-commerce sales. In section 6, we disentangle demand-
reducing and tax avoidance effect of sales tax. Section 7 compares online and offline sales tax elasticities.
5
In section 8, we estimate total sales tax revenue losses from online purchases for the year 2006. Section 9
concludes.
2. Internet Taxation
Current e-commerce taxation rules are similar to those governing out-of-state mail-catalog sellers, which
require an online merchant to collect and remit sales tax if it has employees or establishes a physical pres-
ence, also known as nexus, in a state. The Internet Tax Freedom Act (ITFA) bars federal, state and local
governments from taxing Internet access and from imposing discriminatory Internet-only taxes such as bit
taxes, bandwidth taxes, and email taxes but does not repeal any existing state sales tax. Although out-of-
state purchases are generally still subject to “use” tax, voluntary compliance with use tax by individual
consumers is extremely rare except for certain products (such as automobiles and boats that must be regis-
tered).
Prior research, largely using analytical models, has focused on whether, and to what extent, Internet
commerce should be taxed (Bruce et al. 2003, Goolsbee and Zittrain 1999, Zodrow 2006b). Arguments in
favor of no or lower taxes for Internet purchases have been made on the grounds of administrative and
compliance costs (Kaplow 1990), the need to protect an infant industry (Stephenson and Zeisser 1998),
network externalities (Goolsbee and Klenow 2002), and potential differences in the tax elasticities of In-
ternet and retail consumers (Auerbach and Hines 2002). For example, based on research showing that
those living on the borders of different tax jurisdictions tend to have a greater sensitivity to tax rates (As-
plund et al. 2007, Braid 1987, Fox 1986, Mikesell 1970, Mintz and Tulkens 1986, Trandel 1992, 1994,
Walsh and Jones 1988), some argue that Internet consumers are more likely than retail consumers to
change their behavior in response to tax and, from an efficiency standpoint, it is less distortive to apply
lower tax rates to Internet sales (Goolsbee 2000, Goolsbee and Zittrain 1999).
Others have concluded that most of these reasons do not survive close scrutiny. For example, Zodrow
(2003) analyzed three types of network externalities (direct, indirect, and learning) and concluded that the
case for preferential treatment is weak. Using an optimal taxation model, Zodrow (2006a) showed that
6
exempting e-commerce from tax is unlikely to be even close to optimal. Arguments based on equity con-
cerns actually call for higher tax rates on electronic commerce sales since individuals who shop on the
Internet earn approximately $22,000 more in annual income and have nearly two more years of education
than traditional shoppers (Bruce et al. 2003, Goolsbee and Zittrain 1999).
Empirical work on the effects of tax rates on e-commerce, however, has been limited. Goolsbee (2000),
using survey data from 1997, found that higher local tax rates increased the likelihood that consumers
shop online. Later studies by Alm and Melnik (2005) and Ballard and Lee (2007) found similar results
using 2001 survey data. Goolsbee (2000) argued that applying existing sales taxes to Internet commerce
might reduce the number of online buyers by up to 24 percent and, in a subsequent paper (Goolsbee
2001), estimated that tax avoidance on Internet sales led to losses of about $612 million in the year 2000,
or .3 percent of $203 billion in total sales tax revenue during that year.
Research on consumer sensitivity to tax versus price differences is mixed. Brynjolfsson and Smith
(2000) found that consumers who use an Internet shop-bot are twice as sensitive to differences in taxes as
they are to differences in prices. However, Ellison and Ellison (2003) found that consumers pay less at-
tention to sales tax differences than price differences and prefer to buy from in-state firms.
3. Theoretical Framework
Although prior research has focused on incentives for consumers to avoid tax by purchasing on the Inter-
net (Alm and Melnik 2005, Ballard and Lee 2007, Goolsbee 2000), increased collection of local sales tax
by Internet retailers means that a large fraction of online sales are no longer tax-exempt (Krebs and Krim
2003). Given this, it seems important to account for tax-liable as well as tax-avoiding Internet purchases.
Accordingly, we model consumer purchases as coming from four categories—determined by whether a
product is purchased offline or online and whether or not the product is subject to tax.
Consider a representative consumer, with an initial endowment of w, who chooses a consumption bun-
dle to maximize her utility. Let there be four types of normal (i.e., not inferior) consumer goods in the
economy. Type 1 goods are purchased in offline retail stores and are exempt from sales tax either for dis-
7
tributional (i.e., goods consumed in disproportionately large quantities by the poor), social (such as health
care and education), or administrative reasons (e.g., services such as haircuts and car repair), or a combi-
nation of the above (Zodrow 2006a). Type 2 goods are purchased through offline retail stores and subject
to sales tax. Type 3 goods are purchased online and taxed, either because they are purchased from a store
that has a nexus in the state where the consumer resides, or because the type of good (such as hotel stays
or airline tickets) makes avoiding sales tax impossible. Type 4 goods are purchased online but no sales tax
is collected because they are obtained from out-of-state sellers. Let ()u
⋅
be a continuous utility function
representing a locally non-satiated preference relation defined on the consumption set
;4
X
R
+
=. Let t
producer’s price vector be q=d the retail price vector be
he
an
T
1234
[, , , ]
T
qqqq
1 2 3
,(1 ) ,(1 ) ,
T
1234 4
[ , , , ] [ ]
p
pppp q==+q tq tq+, where t is the local sales tax rate. Let (, )
x
pw de-
note the vector of Marshallian demand, and denote the vector of Hicksian demand. The effect of
the local tax rate on Marshallian demand can be expressed as
(,)u hp
(1)
23
(, ) (, ) (, )[0, , ,0]
T
tptp
Dxpw Dxpw Dp Dxpw q q=⋅=⋅
By the Slutsky equation, we have
(2)
23
(, ) [ (,) (, ) (, )][0, , ,0]
TT
tpw
Dxpw Dhpu Dxpw xpw q q=−⋅⋅
Or, in expanded form,
(3)
21 2 2 1 31 3 3 1
1
22 2 2 2 32 33 2
2
23 223 33 333
3
24 2 2 4 34 3
4
(/ /) (/ /)
/
(/ /) (/ /)
/
(/ /) (/ /)
/
(/ /) (/
/
qhpxxwqhpxxw
dx dt
qhpxxwqhpxxw
dx dt
qh px x wqh pxx w
dx dt
qh px x wqh px
dx dt
∂∂−⋅∂∂+∂∂−⋅∂∂
⎡⎤
⎢⎥∂∂−⋅∂∂+∂∂−⋅∂∂
⎢⎥
=
⎢⎥∂∂−⋅∂∂+∂∂−⋅∂∂
⎢⎥∂∂−⋅∂∂+∂∂−
⎣⎦ 34
/)
x
w
⎡⎤
⎢⎥
⎢⎥
⎢⎥
⎢⎥
⋅∂ ∂
⎣⎦
Prior research has shown that own-price elasticities are higher for Internet than traditional retail pur-
chases (Ellison and Ellison 2003, Goolsbee 2000). Higher own-price elasticities are often cited as a rea-
son for preferential tax treatment of e-commerce (Zodrow 2006b). However, cross-elasticities of online
purchases with respect to offline retail prices are usually low (Zodrow 2006a). Under these assumptions,
increases in local tax rates reduce consumer demand for taxed online purchases (i.e., type 3 goods).
8
PROPOSITION 1. Let ij
ε
be the compensated cross-price elasticity of good i with respect to the price
of good j. If the compensated own-price elasticity of type 3 goods (Internet goods on which sales tax is
paid) is larger (in absolute value) than the cross-price elasticity of type 3 goods with respect to the price
of type 2 goods (traditional retail goods on which tax is paid; i.e., 32 33
ε
ε
≤
−), then demand for type 3
goods decreases as the tax rate increases, or
(4)
3/0dx dt <
Proof. We have 333
2
32 33
0
hph
p
hp hp
∂∂
+
∂∂
≤. Multiplying both sides by 3
1
ht
+
, we get
33
23
23
0
hh
qq
pp
∂∂
+
∂∂
≤. Observe that 3
22 33
()
x
qx qx w
∂
+
∂
is strictly positive (i.e., 3
x
is a normal good),
therefore, 33 3
23 2233
23
()0
3/0dx dt
hh x
qxqx
pp w
∂∂ ∂
+−+ <
∂∂ ∂
qq , or
<
.
Although it may reduce demand for taxed online purchases, an increase in the sales tax rate may in-
crease demand for online products for which tax can be avoided (i.e., type 4 goods). The rationale is that,
as local sales tax rates increase, consumers will substitute a fraction of their demand for taxed retail and
Internet products with online products on which tax can be avoided. If the magnitude of substitution ef-
fects exceeds income effects, tax increases will increase demand for type 4 goods.
PROPOSITION 2. Assuming that type 4 goods (Internet goods on which sales tax is avoided) are gross
substitutes for type 2 and type 3 goods (retail and Internet goods on which sales tax is paid), demand for
type 4 goods increases as the tax rate increases, or
(5)
4/0dx dt >
Proof. By gross substitution of goods 4 and 2, we have 4
2
0
x
p
∂
>
∂. By the Slutsky equation we have
. Therefore
4224 4
//hpxxwxp∂∂−⋅∂∂=∂∂
2
/24 2 2 4
(/ /)0qh px x w
∂
∂− ⋅∂ ∂ >. Similarly, by gross subs-
9
titution of goods 4 and 3, we have . Adding these two inequalities we get
.
34 334
(/ /)0qh pxx w∂∂−∂∂>
4/0dx dt >
From these propositions we observe that sales tax has countervailing effects on consumer demand. Al-
though an increase in the local tax rate may lead consumers to buy from out-of-state online sellers, to the
extent that consumers are unable to avoid paying taxes on certain products or choose to purchase products
from in-state sellers—perhaps because of lower search costs, higher levels of trust, and more convenient
returns, an increase in local tax rate reduces consumer demand for online goods. The aggregate impact of
sales tax on Internet purchases depends on the interplay of these countervailing effects. In the following
sections we empirically investigate the overall effect of sales tax on consumer purchase behavior, and
then disentangle the competing effects on consumer demand for online products illustrated in our proposi-
tions.
4. Data and Specification
4.1 Data
Our primary data source is the comScore 2006 web behavior data base of the domain-level browsing and
buying history of a representative sample of the U.S. Internet population. This data was collected by a
program, the PCMeter, which ran continuously in the background on each household’s computer. The
data contain details on all transactions executed online by each household in the sample, including the
specific products purchased during a transaction, product quantities, product categories, price paid for
each product (excluding taxes) and basket total price (including tax, shipping and coupon discounts), as
well as the online vendor’s domain name. The data also record which domains were visited, the date and
time of each visit, the duration of the visit, and total number of pages viewed during a session. In addi-
tion, the data has comprehensive household level demographic data as well as the zip code of each house-
hold. The 88,814 households in the dataset come from all 50 states and U.S. territories and reside in 2,909
distinct counties and 18,973 distinct zip codes.
10
To identify the local sales tax rate for each household, we use SIMPOVA’s national sales tax database
from January 2006. This data set includes the zip code, state, county, city, and sales tax at the state, coun-
ty, city, and district (the sum of school district, transit, and all other tax) levels, as well as the overall sales
tax rate.3 As controls we use demographic data from the comScore database, the U.S. Census Bureau’s
2000 County and City Data Book, and the Chamber of Commerce’s 2007 ACCRA database.
4.2 Dependent, Independent, and Control Variables
The comScore 2006 web behavior data base makes it possible to explore dependent variables beyond on-
line purchase probability. Although the probability of shopping online has been of central interest in pre-
vious research (Alm and Melnik 2005, Ballard and Lee 2007, Goolsbee 2000), it may be less diagnostic
of current consumer behavior since 94% of current Internet users are estimated to have made at least one
online purchase (The Nielson Company 2008). Accordingly, we aggregate the online transaction data for
each household in 2006 to examine: 1) Total dollar expenditures; 2) The total number of online purchases
transactions; and 3) Whether a household made an online purchase during the year.
Sales tax rates are determined by matching the household’s zip code in the comScore 2006 database
with sales tax information from the January 2006 SIMPOVA sales tax database. Since zip codes may
span cities, we calculate the average sales tax rate within each zip code. Results are similar for all models
when we use minimum or maximum tax rates instead.
We include a comprehensive set of controls for social and economic factors that may influence pur-
chase behavior. At the household level, we use information from the comScore dataset to control for edu-
cation,4 income, age, household size, racial background, presence of children, Internet connection speed,
and Internet usage intensity—which we operationalize as the average number of minutes spent online
each day. Table A1 of the appendix provides additional details on the coding of these controls. To ac-
count for regional differences in buying behavior, we control for land area, population, population densi-
3 This database contains 81,041 non-unique zip codes (with 42,651 distinct zip codes) with disaggregated levels of tax data for
the entire United States and U.S. territories. Non-unique zip codes in the dataset reflect the fact that zip codes may span multiple
cities; therefore tax rate may vary within a single zip code.
4 Information about household highest education is missing for over 75% of the panelists. To check for sampling bias, we com-
pare the summary statistics of households that report education level and those that do not, and found shows they are very similar
11
ty, personal income per capita, and retail trade sales per capita using county-level data from the U.S. Cen-
sus Bureau’s 2000 County and City Data Book.5 We also control for potential differences in consumers’
local market options (Choi and Bell 2008), and the level of product variety in physical stores (Bryn-
jolfsson et al. 2003), by including the number of retail sales establishments in the county. We use metro-
politan area indicators to account for potential differences between urban and rural areas.6 To account for
potential differences in local price levels, we include cost of living as a control variable using data from
the Chamber of Commerce’s 2007 ACCRA data base.7
Table 1 presents summary statistics for the independent and dependent variables as well as individual-
and regional-level controls.
[Insert Table 1 about here]
4.3 Specification
If a consumer avoids paying sales tax when she purchase a product online, and local sales tax does not
affect the local retail price level, the relative price ratio (1 ) /
S
PtP
I
+
measures the price advantage of on-
line shopping, where is the retail store price, is the Internet price for the same item, and t is the tax
rate. Following standard assumptions in the literature (Goolsbee 2000), in our base models we assume
that the relative price is constant across locations. We relax this assumption, and account for dif-
ferences in relative price levels across regions, by including cost of living as a control in our extended
models. For purchases from a tax-collecting vendor, the effective price is
S
P
/
S
P
I
P
I
P
(1 ) I
tP
+
, and the demand re-
duction effect is proportional to (1+t). In either case, demand can be expressed as a function of (1+t).
on most measures.
5 Since zip codes may vary across cities but rarely vary across counties, we use county-level controls.
6 The United States Office of Management and Budget (OMB) defines metropolitan and micropolitan statistical areas according
to published standards that are applied to Census Bureau data. The current standard, Core Based Statistical Area (CBSA), is a
collective term for both metro and micro areas. A metro area contains a core urban area of 50,000 or more residents and a micro
area contains an urban core of at least 10,000 (but less than 50,000) residents. According to the current standard (released in
2003), there are 363 metropolitan and 576 micropolitan statistical areas in the United States. An older definition, the Standard
Metropolitan Statistical Area (SMSA), which divides urban areas into metropolitan statistical area (MSA), primary metropolitan
statistical area (PMSA), and consolidated metropolitan statistical area (CMSA), is still widely used.
7 Cost of living index data are available only for metropolitan areas and select micropolitan areas and including this data reduces
the sample size by half.
12
We measure the effect of sales tax on online purchase intensity in three ways: First, we examine how
tax rates affect total online expenditures for each household during the year 2006; second, we assess the
number of online transactions made by each household during the year; third, to compare our results with
those of previous studies, we assess the effect of sales tax on the likelihood that a household made at least
one online purchase during the year.8 Yearly online expenditures are modeled using left-censored Tobit
regressions, since the dependent variable is censored at zero. Following standard practice we use log
transforms to account for non-constant error variance in this data (e.g., log(1 )s
+
, where s is yearly
household online spending). The marginal effect of tax on online demand in a standard Tobit model can
be captured by
(6) (|; 0) {(1 ( / )[ / ( / )]}
j
j
Ey y
x
βλσσλσ
∂>
=− +
∂
xxβxβxβ
where x, y denote the vector of independent and dependent variables, and () ()/ ()
λ
φ
⋅
=⋅Φ⋅ is the inverse
Mills ratio. The number of online purchase transactions during the year is analyzed using negative bi-
nomial models, since we observe over-dispersion when using Poisson regressions (Greene 2002). Pur-
chase likelihood is assessed through binary probit models.
5. Aggregate Analyses of Online Purchase Behavior
5.1 Online Expenditures
From a policy perspective, the most important question is how sales tax affects a household’s total online
expenditures.9 Column (1) of Table 2 shows results of the baseline Tobit model specifying annual house-
hold online spending as a function of the local sales tax rate, demographics, and regional controls. Col-
umn (2) shows results for a model with the same specification which excludes households that reside in a
sales-tax-free state.10 Column (3) presents result of an extended model that includes households in tax-
free states but also controls for cost of living, reducing the sample size by more than half due to missing
8 The comparison is not exact since prior research examines the probability of “ever” making a purchase.
9 Expenditures are net of sales tax and shipping fees.
10 Five states (Alaska, Oregon, Montana, New Hampshire and Delaware) do not levy sales tax (although some local jurisdictions
in Alaska impose a small local sale tax). Since residents of tax-free states have no incentive to avoid sales tax by purchasing on-
13
data on this variable. Consistently across all three models, an increase in the local tax rate decreases ag-
gregate online expenditures. The predicted value of log spending is 4.52 (implying average expenditures
of $91.84 per year), conditional on positive expenditures and controlling for cost of living (i.e., model 3),
and the implied demand elasticity of online expenditures with respect to local tax rate is about -3.33. The
effect of tax rates on online expenditures is higher when the analysis is restricted only to states that have
sales tax (column 2) and lower when controlling for differences in cost of living (column 3). We also re-
ran the three models after excluding purchases of travel (since local tax rates should have no effect on
travel purchases), food, and clothing (since food and clothing are exempt from sales taxes in many states)
and found similar negative effects of tax rate on online expenditures.
[Insert Table 2 about here]
5.2 Number of Transactions
Since over-dispersion is observed in the Poisson model (Mean = 4.00, s.d. = 10.00, χ2(1) = 1.1e+05, p <
.001), we use a Negative Binomial Model to assess the effect of tax on number of transactions during the
year.11 These results are presented in Column (4) of Table 2 and imply, for consumers who make online
purchases, an average online purchase frequency of four times per year and a demand elasticity of online
expenditures with respect to local tax rates of -5.31.12
5.3 Purchase Probability
Contrary to prior research showing that an increase in tax rates increases the likelihood of shopping on-
line (Alm and Melnik 2005, Ballard and Lee 2007, Goolsbee 2000), our results show that in higher sales
tax regions, the probability of making an online purchase during the year is actually lower.13 Column (5)
of Table 2 presents the results of the probit model of purchase likelihood during 2006 as a function of lo-
cal sales tax, household characteristics, and regional controls. Results indicate that a one percent increase
line, including panelists from those states may reduce estimated tax effects.
11 Poison-probit hurdle and Negative Binomial-probit hurdle models (Mullahy 1986) show similar results.
12 To see if tax rates affect average expenditures per transaction, in addition to number of transactions, we ran Tobit models on the
effect of tax on the average purchase amount. No significant effects of tax were found, suggesting that effects of tax on total ex-
penditures are driven primarily through number of transactions.
13 A two-part hurdle model, using a binary probit for online market participation and a log-normal model for online spending
(Cragg 1971, Duan et al. 1983), shows similar results.
14
in the tax rate leads to a reduction in purchase probability of -.021, which implies an elasticity of -1.14
based on the predicted mean purchase probability of .68, in contrast to the positive elasticity of 2.30 found
by Goolsbee (2000).
5.4 Border Effects
One possible pitfall for these results is that we have overestimated the effective local tax rate for some
consumers. If the measurement error of the effective tax rate is correlated with the observed tax rate, un-
der classical error-in-variables (CEV) assumptions, the estimated effects of tax will be biased and incon-
sistent (Wooldridge 2002). Specifically people living on geographic borders tend to make purchases on
the side of the border with the lowest taxes, particularly if they live on state borders, since states often
have large differences in sale tax rates (Fox 1986). Similarly, one might argue that people living in areas
that span more than one zip code will likely conduct most of their retail purchases in the zip code with the
lowest sale tax rate. To address these issues, we determined the latitude and longitude of the centroid of
each zip code in our dataset. Then, for each household, we determined the minimum tax rate among all
zip codes (including the zip code of residence) within a 10-mile radius of the centroid of the zip code of
residence and reran the models using this rate. Table 3 shows the results for annual expenditures, number
of transactions, and purchase probability using this minimum tax rate. Results are consistent with the pre-
vious models, and magnitudes of tax elasticities are quite similar, indicating that our conclusions are ro-
bust to potential border effects.
[Insert Table 3 about here]
6. Tax Avoidance vs. Demand Reduction
The aggregate-level analyses show, contrary to earlier research using surveys of online purchase beha-
vior, that the overall effect of increased tax rates is to reduce consumer demand for online goods. Howev-
er, the aggregate analysis masks the countervailing effects of sales tax on consumer demand. In this sec-
tion we try to separate the demand-increasing and demand-reducing impact of sales tax on online pur-
chases.
15
Propositions 1 and 2 suggest that increased tax rates can simultaneously encourage and discourage on-
line shopping, by increasing demand for goods from (out-of-state) retailers that do not charge taxes and
by reducing demand for goods for which tax cannot be avoided (or for which the consumer chooses not to
avoid tax). To get a better sense of the relative size of the demand reducing and increasing effects of sales
tax on online purchases, taxed online transactions need to be separated from those that avoid tax. Since
the comScore data does not indicate whether tax is paid on a transaction, we limit our analysis to purchas-
es from leading retailers for which we can identify the states in which they have a physical presence (i.e.,
nexus). This allows us to determine which transactions are likely to be subject to sales tax. Two analyses
are conducted. The first assesses the demand-reducing effect of taxes by examining purchase behavior
from top Internet sellers with a presence in almost every state, and for which the vast majority of transac-
tions are taxed. The second compares the demand-reducing to the demand-increasing effect of sales tax
by examining purchase behavior from retailers who have a nexus in only a few states and contrasting the
shopping behavior of consumers that live in these states with that of consumers who live in other states.
6.1 The Demand Reducing Effect of Online Tax
To determine the size of the demand reducing effect of taxes, and compare it to the aggregate effect found
in our earlier analyses, we first identified the 50 largest online vendors based on sales revenue in our data-
set. Among these, we eliminated travel websites since taxes on travel services are generally not based on
buyer location.14 We then selected those retailers with a physical presence in almost every state. This left
us with 18 nation-wide “brick-and-click” sellers for which purchasing online offers no tax advantage.15
We visited the website of each of the 18 sellers to confirm that they collect tax from residents of all states
that have a sales tax. Table 4 presents a reanalysis based on purchases from these 18 sellers.
[Insert Table 4 about here]
14 Airline tickets are subject to a uniform 7.5% sales tax. Sales taxes on hotels and car rentals are based on the rate where the
hotel or rental agency is located.
15 The 18 sellers are: dell.com, jcpenney.com, qvc.com, ups.com, officedepot.com, staples.com, walmart.com, quillcorp.com,
victoriassecret.com, costco.com, sears.com, circuitcity.com, target.com, bestbuy.com, quixtar.com, intuit.com, apple.com and
oldnavy.com.
16
Results (column (1) of Table 4) show that an increase in sale tax has a significant negative effect on
yearly household online spending from tax-collecting sellers. Average yearly expenditures at the 18 sel-
lers (conditional on purchase) are $53.52. The tax elasticity of demand is about -3.76, which is slightly
higher (in absolute value) than the aggregate elasticity of -3.33. Similar negative effects of sales taxes are
found for number of transactions as well as probability of purchase, as illustrated in columns (2) and (3)
of Table 4. For these dependent variables, limiting the analysis to the 18 tax-collecting firms leads to elas-
ticities that are larger than those from the aggregate analyses (-8.8 vs. -5.31 for purchase count and -1.63
vs. -1.14 for purchase probability).
6.2 Demand-Reducing Versus Tax Avoiding Effects
Limiting the analysis to firms that have a nexus in all states, and therefore collect taxes from most con-
sumers, is one way to identify the effect of local tax on reducing online demand. An alternative approach
is to limit the analysis to retailers who have a nexus in only a few states and to compare the behavior of
consumers that live in those states with that of consumers from states where the retailer does not have a
nexus.
In order to eliminate potential effects of local retail store presence on online purchase behavior, we re-
strict our sample to pure-play online retailers whose physical presence (nexus) is limited to warehouses or
other administrative operations. Among the top sellers in our sample for which physical presence, or nex-
us, can be determined, we identify five such online retailers: amazon.com, zappos.com, seventhave-
nue.com, newport-news.com, and 1800flowers.com. For these five firms, we compare transactions from
households that reside in states with a nexus with those of households residing in states without a nexus;
this allows us to separate demand-reducing from demand-increasing (tax avoiding) effects of sales tax.
Let Yij be the yearly online spending (in log form) of all orders that i-th household placed from j-th firm,
Zi be a vector of household demographic and regional controls, Ti be the sales tax rate for the zip code
where the household resides, Dj=[d1j,d2j,…djj,…dJj] be a vector of firm dummies, and Nij be an indicator
that j-th firm has a nexus in the state in which the i-th household resides. The corresponding left-censored
Tobit Model is:
17
(7)
*(1
2
|, , (0, )
(0, *)
YZDT
ij i j i ij
ZTD Normal
ij i i j
YMaxY
ij ij
)
α
βγ η ε
εσ
=+ + + + +
=
～
Online purchase frequency and purchase probability can be similarly specified using Negative Binomial
and Probit models.
We conducted separate analyses to compare the demand-increasing (i.e., Nij =0) and demand reducing
(i.e., Nij =1) effects of sales tax. Results are presented Table 5. The first and second columns compare the
online expenditures of the two groups of consumers. For consumers who can avoid taxes, increased sales
tax has a positive, but non-significant, effect on demand. In contrast, for households that pay sales tax on
their purchases, there is a significant negative effect of increased sales tax. For those subject to sales tax,
the results imply a tax elasticity of -6.77. We find similar results for number of purchases and purchase
likelihood as shown in columns (3) to (6). Because this analysis is limited to a few leading online retail-
ers, this large negative elasticity cannot be generalized to the general population of online sellers. Howev-
er, it illustrates the extent to which demand-reducing dominate demand-increasing effects of sales tax for
online purchases. Comparing this tax elasticity with that of the earlier analysis (based on firms with a
presence in almost every state; -6.77 vs. -3.76) suggests that having physical stores, rather than just ware-
house or administrative operations, tempers the effect of tax in reducing online demand.
[Insert Table 5 about here]
The aggregate-level analyses show that consumers who live in high-tax areas spend less money on In-
ternet purchases. The disaggregate-level analyses show that this effect is almost entirely driven by the
demand-reducing effects of tax. Although prior research suggests that higher tax rates can increase de-
mand for online goods (Alm and Melnik 2005, Ballard and Lee 2007, Goolsbee 2000), this effect is not
significant in our more current transaction-based data. To better understand the policy implications of
these results, we conducted two additional analyses: (1) We compare the tax elasticities for online prod-
18
ucts to those for offline products; and (2) We estimate potential tax losses in 2006 from Internet purchases
and compare our results to earlier estimates (e.g., Goolsbee 2001).
7. Online vs. Offline Effects of Sales Tax
From a policy standpoint, it is interesting to compare the effect of sales tax on taxed online purchases to
taxed offline purchases. Using data from the 3,143 counties in the 2000 U.S. County and City Data Book,
we estimate the average tax elasticity for offline expenditures. We start with aggregate retail sales data for
each county and divide by the number of households in the county to get average household retail sales.16
We specify the log transform of this variable as a function of the county average sales tax while control-
ling for other county characteristics such as land area, population, population density, household income
level, education level of population, male/female ratio, age, racial composition of the population, country
of origin of population, whether the county belongs to a metropolitan area, cost of living, and percentage
of the population living below the poverty line. The results are presented in Table 6. We find that an in-
crease in the county average sales tax has a significant negative effect on county average household retail
(offline) spending. With predicted average household yearly retail sales of $16,155, we find that the elas-
ticity of offline retail demand with respect to sale tax rate is -1.35. Comparing this number with the elas-
ticity calculated earlier for online purchases from leading online sellers who have a presence in almost
every state (-3.76 from Table 4) suggests that, despite higher income levels among Internet users (Bruce
et al. 2003, Goolsbee and Zittrain 1999), online purchases are about 2.8 times as sensitive to tax as offline
purchases. There are several possible reasons for these differences, including the fact that products ac-
quired from online channels are often non-necessities and that taxes are more saliently displayed in online
settings (Chetty et al. 2007). Our results confirm Goolsbee and Zittrain’s (1999) claim that online sales
are more tax elastic, supporting arguments for the preferential tax treatment of e-commerce.
[Insert Table 6 about here]
8. Tax Losses for Online Purchases
16 Assuming that most households purchase products within the county that they reside, this number should be approximately
19
Estimating the magnitude of sales tax revenue losses from e-commerce is difficult for a number of rea-
sons. Beyond the technical difficulties of obtaining accurate data on the tax rates for different types of
goods in different locations, it is unclear that e-commerce actually hurts tax revenues; on the contrary,
online shopping may complement rather than substitute for retail store shopping (Advisory Commission
on Electronic Commerce 2000). Also, if sales taxes were imposed on all online sales (perhaps through a
uniform federal tax; [McLure 1997, 1999]), there is no guaranty that online demand would remain at the
same level. Some demand may shift to tax-exempt shopping channels such as mail catalogs and some
portion of online demand would likely disappear.
Despite the complexities inherent in such estimates, developing a quantitative measure of tax revenue
losses on consumer e-commerce has important implications for public policy. Since sales taxes account
for 46 percent ($346 out of $750 billion) of state government tax revenues (U.S. Census Bureau 2008b), it
is understandable that state and local governments are extremely concerned about the potential for sales
tax evasion through online shopping. Goolsbee (2001) made a rough estimate of tax losses and concluded
that, in the year 1999, taxes were not collected on about $9.7 of $20.3 billion in online sales, implying tax
revenue losses of $612 million. Goolsbee forecasted that tax revenue losses would equal $6.88 billion in
2004, or 2.6 percent of projected sales tax revenue.
To obtain a preliminary understanding of the extent to which consumers avoid taxes by buying from In-
ternet sellers, we identified the top 50 online retailers based on yearly sales in the comScore data set.
These vendors account for approximately 70% of revenues. Notably, the majority of these are travel web-
sites and purchases from these sites are therefore subject to tax. For each of the remaining vendors in the
top 50 list, we identify whether online purchases from them are tax advantaged. Most are large “brick and
click” retailers, with a physical presence, or nexus in almost every state, and are therefore required to col-
lect sales tax for all purchases. There are only six exceptions among the top 50 sellers: Amazon, The
Home Shopping Network, Yahoo, Overstock, Cabelas, and LL Bean (Table A2 in the appendix ranks
these sellers by revenue and identifies whether they are tax advantaged).
equal to county average household retail spending.
20
To gain a more complete understanding of the magnitude of tax losses, we identify which transactions
are tax-avoiding by determining which retailers are tax advantaged. This is in contrast to Goolsbee
(2001), who estimated revenue losses by broadly classifying online products as suffering little, partial, or
full tax revenue losses. First, we retrieve a list of all online sellers (626 in total) and their 2006 sales reve-
nues from households in our data set (which totaled $40.3 million). From this list, we remove travel sites,
such as air and train tickets, hotel reservation, and car rental firms; large brick and click vendors that have
physical presence in almost every state; other vendors whose products are subject to tax, such as event
ticket sellers, telecom and wireless communication vendors, express mail delivery services and the U.S.
Postal Service; as well as vendors whose online sales are delivered almost exclusively through local phys-
ical stores, such as grocery and fast food. The aggregate sales of the remaining firms are $6.67 million, or
16.6 percent of total online sales in our 2006 dataset. We assume that 90% of these firms’ sales are tax-
avoiding (since online retailers must establish a nexus in some states and collect sales tax in those states).
Given a mean sales tax rate in the United States of 6.8 percent, we estimate the sales tax loss at 1.0
(.166*0.9*6.8) percent of total online sales revenue. Total U.S. e-commerce retail sales for 2006 were
estimated at $108.7 billion (U.S. Census Bureau 2007), implying a tax revenue loss of $1.1 billion in the
year 2006,17 or .33 percent of $332.8 billion in total sales tax revenue in 2006 (U.S. Census Bureau
2008a). Even if we make the most conservative estimate that 100% sales from tax-advantaged retailers
are tax-avoiding, the estimated tax loss is only 1.1 percent of total online sales revenue, or equivalent to
$1.2 billion in revenue loss. Reflecting strengthened tax collection efforts by state governments, the pro-
portion of tax-evading transactions is considerably lower than previously estimated (14.9% vs. 47.8% in
Goolsbee 2001); yet the magnitude of revenue loss as a percentage of total sales tax revenue is quite simi-
lar (.33% vs. .30%), due to rapid growth in e-commerce sales from $20.3 billion to $108.7 billion be-
tween 1999 and 2006.
9. Conclusion
17 Compare to Goolsbee’s (2001) estimate of a $.74 billion (in 2006 real dollars) revenue loss in 1999.
21
9.1 Summary of results
In this article we present an empirical analysis of how local taxation affects consumers’ purchase beha-
vior. Contrary to earlier survey-based research, which found that consumers who live in high tax locations
are more likely to shop online, our results, based on a comprehensive analysis of actual consumer transac-
tions, show that the overwhelming effect of increased sales tax for today’s online consumers is a reduc-
tion in online purchases. In particular, Internet users living in high tax areas spend less, conduct fewer
transactions, and are less likely to buy things online. It is difficult to attribute these effects to heterogenei-
ty among individual households; nor can they be explained by regional social-economic factors that are
correlated with tax rates. Differences between our results and those of previous research may be due to
changes in the Internet population, the fact that the majority of Internet users now make online purchases,
as well as increased tax collection by major online retailers. In addition to finding very different results
than earlier researchers, our household-level transaction data allow us to go beyond purchase probability
to examine the relationship between taxes and online expenditures and purchase frequency over a year-
long study period. This is important as consumer online spending and purchase frequency directly deter-
mine the amount of money that local governments lose though online purchases. Moreover, results show-
ing lower tax elasticities for online retailers with physical stores in consumers’ home states than that for
pure-play online retailers with warehouses or other administrative facilities suggests that today’s online
consumers may be less concerned with tax evasion than in interacting with retailers whose local presence
may increase brand awareness and trust in addition to making returns easier (Ellison and Ellison 2003).
Future research could examine this issue in more detail.
To better understand the aggregate effect of sales tax on online purchase behavior, we compare the ex-
tent to which sales tax lowers demand for taxable online products with its effect on increasing demand for
products on which tax can be avoided. Limiting the analysis to major retailers with a nexus in every state,
and for which most online purchases are therefore taxed, shows even stronger negative effects of tax on
purchase behavior. Limiting the analysis to major retailers with a nexus in only a few states shows that
22
increased tax rates significantly reduce purchases from households living in states with a retailer nexus,
but do not significantly boost purchases by households living in states with no physical retail presence.
To understand the public policy implications of these results we conduct two additional analyses: (1)
We compare the elasticities for online and offline purchase; and (2) We estimate tax revenue losses from
Internet purchases in 2006 and compare them to earlier estimates. Comparing online and offline purchas-
es from tax-collecting sellers shows that the demand elasticity of sales tax is much larger for online than
for traditional retail store purchases (-3.76 vs. -1.35). In addition, our estimates suggest that online tax-
avoidance is more moderate than previously suggested since the vast majority of online purchases are
from retailers who collect sales tax in almost every state.
9.2 Policy implications
This research has several important implications for policy. First, because the vast majority of online pur-
chases are now subject to tax, either because they are travel related, because major online retailers have a
presence in most states, or because consumers prefer to buy from retailers with a local presence, sales tax
reduces demand for taxed online purchases but does not significantly increase demand for tax avoiding
purchases. These results hold even when travel, food, and clothing-related expenditures are removed from
the analysis. This indicates that the extent to which local governments are losing revenue to e-commerce,
and local retailers are losing sales to online competitors, may not be as high as earlier estimates suggest.
Second, our results show that online consumers are indeed more sensitive to sales tax than offline con-
sumers. Although it may be that online shoppers have higher levels of income and education, these gaps
are declining (Goolsbee 2001), and in our sample there is no systematic difference in income and educa-
tion between households that participate in the online market and those that do not. This suggests that dis-
tributional equity concerns about current tax policy may diminish over time and support optimal tax ar-
guments in favor of preferential tax treatment for e-commerce. As a whole, these results suggest that initi-
atives to increase collection of tax on online purchases, especially on interstate sales, be approached with
caution.
9.3 Extensions and future research
23
Future studies could extend our work in several ways. For example, future research could compare the
effect of sales tax on online demand for different product categories. Such studies could help policy mak-
ers decide whether certain online products should receive preferential tax treatment. Given earlier re-
search showing that online consumers prefer to buy from retailers with a local presence (Ellison and Elli-
son 2003), future research could also examine how physical distance affects online store choice, and how
the effects vary for different types of physical presence. Finally, given our results showing that the de-
mand elasticity of sales tax is much higher for online than offline goods, future research could reevaluate
whether e-commerce deserves preferential tax treatment, while incorporating efficiency and equity con-
siderations.
24
References
Advisory Commission on Electronic Commerce. 2000. Report to congress. Retrieved June 15, 2008,
http://www.ecommercecommission.org/acec_report.pdf.
Alm, J., M.I. Melnik. 2005. Sales taxes and the decision to purchase online. Public Finance Rev. 33(2)
184-212.
Andal, D.F. 1997. State and local taxation of electronic commerce: Read my e-mail, no new taxes. Sym-
posium: Multi-Jurisdictional Taxation of Electronic Commerce, International Tax Program and
the Society for Law and Tax Policy, Harvard School of Law, April 5, 1997.
Anderson, E.T., N. Fong, D. Simester , C. Tucker. 2008. How does an obligation to collect sales tax affect
consumer and firm behavior? Journal of Marketing Research forthcoming.
Asplund, M., R. Friberg, F. Wilander. 2007. Demand and distance: Evidence on cross-border shopping.
Journal of Public Economics 91(1-2) 141-157.
Auerbach, A.J., J.J. Hines. 2002. Taxation and economic efficiency. A.J. Auerbach, M. Feldstein, eds.
Handbook of Public Economics, Elsevier, New York, 1347-1421.
Ballard, C.L., J. Lee. 2007. Internet purchases, cross-border shopping, and sales taxes. National Tax
Journal 60(4) 711-725.
Braid, R.M. 1987. The spatial incidence of local retail taxation. Quarterly Journal of Economics 102(4)
881-891.
Bruce, D., W.F. Fox. 2000. E-commerce in the context of declining state sales tax bases. National Tax
Journal. 53(4) 1373-1388.
Bruce, D., W.F. Fox, M. Murray. 2003. To tax or not to tax? The case of electronic commerce. Contem-
porary Economic Policy. 21(1) 25-40.
Brynjolfsson, E., M.D. Smith. 2000. Frictionless commerce? A comparison of Internet and conventional
retailers. Management Science. 46(4) 563-585.
Brynjolfsson, E., H. Yu, M.D. Smith. 2003. Consumer surplus in the digital economy: Estimating the val-
ue of increased product variety at online booksellers. Management Science. 49(11) 1580-1596.
Chetty, R., A. Looney, K. Kroft. 2007. Salience and taxation: Theory and evidence. Working Paper,
NBER, http://www.nber.org/papers/w13330.
Choi, J., D.R. Bell. 2008. Virtual agglomeration of physically distant types: How local market options
affect willingness to shop online. Working Paper, University of Pennsylvania, Wharton School.
CNET. 2003. Sales taxes coming to a site near you. Retrieved June 15, 2008, http://www.news.com/2100-
1017-983636.html.
Cornia, G.C., D.L. Sjoquist, L.C. Walters. 2004. Sales and use tax simplification and voluntary com-
pliance. Public Budgeting & Finance. 24(1) 1-31.
25
Cragg, J.G. 1971. Some statistical models for limited dependent variables with application to the demand
for durable goods. Econometrica. 39(5) 829-844.
Duan, N., W.G. Manning, C.N. Morris, Jr., J.P. Newhouse. 1983. A comparison of alternative models for
the demand for medical care Journal of Business & Economic Statistics. 1(2) 115-126.
Ellison, G., S.F. Ellison. 2003. Tax sensitivity and home state preferences in internet purchasing. Work-
ing Paper, MIT, http://web.mit.edu/gellison/www/tax2.pdf.
Fox, W.F. 1986. Tax structure and the location of economic activity along state borders. National Tax
Journal 39(4) 387-401.
Goolsbee, A. 2000. In a world without borders: The impact of taxes on internet commerce. Quarterly
Journal of Economics 115(2) 561-576.
Goolsbee, A. 2001. The implications of electronic commerce for fiscal policy (and vice versa). Journal of
Economic Perspectives. 15(1) 13-23.
Goolsbee, A., P.J. Klenow. 2002. Evidence on learning and network externalities in the diffusion of home
computers. Journal of Law & Economics. 45(2) 317-343.
Goolsbee, A., J. Zittrain. 1999. Evaluating the costs and benefits of taxing internet commerce. National
Tax Journal. 52(3) 413-428.
Greene, W.H. 2002. Econometric analysis. Prentice Hall, Upper Saddle River, New Jersey.
Hellerstein, W. 1997. Transaction taxes and electronic commerce: Designing state taxes that work in an
interstate environment. National Tax Journal 50(3) 593-606.
Kaplow, L. 1990. Optimal taxation with costly enforcement and evasion. Journal of Public Economics.
43(2) 221-236.
Krebs, B., J. Krim. 2003. Big stores to charge sales taxes online; retailers agree to collect for states.
Washington Post (February 7) A01.
McGregor, J. 2008. Attention, online shoppers: Taxes ahead. BusinessWeek (June 16) 89.
McLure, C.E., Jr. 1997. Electronic commerce, state sales taxation, and intergovernmental fiscal relations.
National Tax Journal. 50(4) 731-749.
McLure, C.E., Jr. 1999. Electronic commerce and the state retail sales tax: A challenge to American fede-
ralism. International Tax and Public Finance. 6(2) 193-224.
Mikesell, J.L. 1970. Central cities and sales tax rate differentials--border city problem. National Tax
Journal. 23(2) 206-213.
Mintz, J., H. Tulkens. 1986. Commodity tax competition between member states of a federation--
equilibrium and efficiency. Journal of Public Economics. 29(2) 133-172.
Mullahy, J. 1986. Specification and testing of some modified count data models. Journal of Econome-
trics. 33(3) 341-365.
26
Newman, N. 1995. Prop 13 meets the Internet: How state and local government finances are becoming
road kill on the information superhighway. Retrieved June 15, 2008,
http://www.nathannewman.org/EDIN/budget/tax-internet.html.
Pew Internet. 2008. Who's online. Retrieved May 20, 2008, http://www.pewinternet.org/trends.asp.
Stephenson, J., M. Zeisser. 1998. Don't tax the Internet--yet. Wall Street Journal (June 6) A22.
Streamlined Sales Tax Project. 2001. Streamlined sales tax system for the 21st century. Retrieved June
15, 2008, http://www.geocities.com/streamlined2000/.
The Nielson Company. 2008. Over 875 million consumers have shopped online--the number of Internet
shoppers up 40% in two years. Retrieved May 20, 2008,
http://www.nielsen.com/media/2008/pr_080128b.html.
Trandel, G.A. 1992. Evading the use tax on cross-border sales--pricing and welfare effects. Journal of
Public Economics. 49(3) 313-331.
Trandel, G.A. 1994. Interstate commodity tax differentials and the distribution of residents. Journal of
Public Economics. 53(3) 435-457.
U.S. Census Bureau. 2007. Quarterly retail e-commerce sales, 4th quarter 2006. Retrieved June 15, 2008,
http://www.census.gov/mrts/www/data/html/06Q4.html.
U.S. Census Bureau. 2008a. State government tax collections: 2006. Retrieved June 15, 2008,
http://www.census.gov/govs/statetax/0600usstax.html.
U.S. Census Bureau. 2008b. State government tax collections: 2007. Retrieved June 15, 2008,
http://www.census.gov/govs/statetax/0700usstax.html.
Walsh, M.J., J.D. Jones. 1988. More evidence on the border tax effect--the case of West-Virginia, 1979-
84. National Tax Journal 41(2) 261-265.
Wooldridge, J. 2002. Introductory econometrics: A modern approach. South-Western College Publishing,
Mason, Ohio.
Yang, J.L. 2008. The end of tax-free e-commerce. Retrieved June 15, 2008,
http://money.cnn.com/2008/05/30/technology/yang_amazon.fortune/.
Zodrow, G.R. 2003. Network externalities and indirect tax preferences for electronic commerce. Interna-
tional Tax and Public Finance. 10(1) 79-97.
Zodrow, G.R. 2006a. Optimal commodity taxation of traditional and electronic commerce. National Tax
Journal 59(1) 7-31.
Zodrow, G.R. 2006b. Should electronic commerce receive preferential tax treatment? State Tax Notes.
41(7) 423-435.
27
Table 1
Summary Statistics
Variable Obs Mean Std. Dev. Min Max
Online buyer 88814 .652 .476 0 1
Number of transactions 88814 3.998 10.000 0 662
Online total spending 88814 453.191 1279.878 0 135771
Internet usage intensitya 88814 191.574 334.554 0 31879
Household highest education 19871 2.531 1.493 0 5
Household size 88814 2.906 1.358 1 6
Household oldest age 88747 6.736 2.425 1 11
Household income 88814 4.339 1.772 1 7
Children 88814 .586 .493 0 1
Connection speedb 88814 .771 .420 0 1
White 88767 .939 .239 0 1
Black 88767 .044 .205 0 1
Asian 88767 .011 .106 0 1
Other race 88767 .005 .074 0 1
Hispanic origin 88814 .212 .409 0 1
Metro area 87473 .818 .386 0 1
Micro area 87473 .111 .314 0 1
MSA 88814 .777 .417 0 1
PMSA 88814 .315 .464 0 1
Population per square milec 86779 1.867 6.302 .000 66.835
Personal income per capitac 86479 26.854 7.372 8.162 72.194
Retail trade establishmentsc 86782 3.090 5.110 0 27.577
Retail trade sales per capitac 86721 9.300 3.006 0 68.085
Land areac 86782 1.318 2.443 .002 145.9
Populationc 86782 917.782 1726.953 .319 9637.494
Cost of living index 39481 106.850 20.191 81.633 172.057
Sales tax rate 87768 .068 .015 0 .107
Notes:
aInternet usage intensity is measured by average daily time online (in minutes).
bConnection speed is 1 if broadband, 0 otherwise.
cIn thousands.
28
Table 2
Effect of Sales Tax on Online Expenditures, Number of Transactions, and Purchase Likelihood
Yearly Online Spending
(4) Number of
Transactions
(5) Purchase
Likelihood
(2) Tax-free states
excluded
(3) With cost of
living Variable (1) Base Model
-8.483***
(2.118)
-11.148***
(3.078)
-5.855*
(3.193)
-2.053**
(.996)
-4.971***
(1.127)
1+tax
.053**
(.021)
.053**
(.022)
.066**
(.031)
.015
(.010)
.001
(.011)
Household highest education
.034
(.029)
.028
(.030)
.006
(.043)
.004
(.013)
.009
(.015)
Household size
-.008
(.014)
-.010
(.014)
-.020
(.020)
-.006
(.006)
-.005
(.007)
Household oldest age
.121***
(.019)
.125***
(.019)
.145***
(.027)
.032***
(.008)
.039***
(.009)
Household income
.441***
(.083)
.447***
(.085)
.467***
(.122)
.132***
(.037)
.048
(.043)
Children
1.909***
(.076)
1.912***
(.078)
2.031***
(.115)
.445***
(.033)
.725***
(.041)
Connection speed
1.213***
(.274)
1.264***
(.278)
.569
(.376)
.114
(.108)
.118
(.131)
White
.043
(.302)
.092
(.306)
-.727*
(.419)
-.135
(.120)
-.396***
(.147)
Black
.377
(.349)
.487
(.353)
-.445
(.483)
-.152
(.141)
-.292*
(.168)
Asian
-.621***
(.082)
-.607***
(.084)
-.648***
(.120)
-.134***
(.036)
-.313***
(.042)
Hispanic origin
.001***
(.000)
.001***
(.000)
.001***
(.000)
.001***
(.000)
.001***
(.000)
Internet usage intensity
.310
(.194)
.315
(.197)
.376
(.338) --a --a Metro area
.094
(.157)
.073
(.161) --a -.109
(.102)
-.171
(.118)
Micro area
-.152
(.155)
-.121
(.157)
-.151
(.235)
-.083
(.071)
-.086
(.083)
MSA
.180**
(.086)
.193**
(.087)
-.083
(.145)
.009
(.045)
-.039
(.051)
PMSA
-.027***
(.007)
-.026***
(.007)
.027
(.043)
-.001
(.014)
.008
(.014)
Population per square mile
.034***
(.008)
.031***
(.008)
.016
(.013)
.002
(.004)
.008*
(.005)
Personal income per capita
-.156**
(.067)
-.148**
(.068)
-.034
(.145)
-.011
(.045)
.018
(.051)
Retail trade establishment
-.001
(.015)
.003
(.015)
.013
(.023)
.009
(.007)
-.013
(.008)
Retail trade sales per capita
.076***
(.016)
.074***
(.017)
.075***
(.019)
.019***
(.006)
.018*
(.007)
Land area
.000**
(.000)
.000**
(.000)
.000
(.000)
.000
(.000)
-.000
(.000)
Population
.011***
(.004)
.001
(.001)
.005***
(.001)
Cost of living index
7.034***
(2.302)
9.844***
(3.323)
3.985
(3.508)
1.682
(1.097)
5.272***
(1.241)
Constant
Observations 19440 18913 8764 8764 8764
Pseudo R-squared .018 .018 .018 .051 .018
LR Chi-square 1521.69 1479.57 683.20 568.25 775.09
Standard errors in parentheses.
***p < .01, **p < .05, *p < .10.
aCost of living data available only for metro and select micro areas.
29
Table 3
Reanalysis Using Minimum Tax Rate in a 10-Mile Radius of the Zip code of Residence
Variable (1) Yearly online spending (2) Number of transactions (3) Purchase likelihood
-7.304**
(3.704)
-4.295***
(1.383)
-1.938*
(1.047)
1+tax
.071**
(.034)
.020
(.013)
.015
(.010)
Household highest education
.014
(.047)
.016
(.017)
.004
(.013)
Household size
-.032
(.022)
-.013
(.008)
-.006
(.006)
Household oldest age
.150***
(.030)
.040***
(.011)
.032***
(.008)
Household income
.496***
(.135)
.087*
(.050)
.132***
(.037)
Children
2.207***
(.127)
.606***
(.047)
.444***
(.033)
Connection speed
.696*
(.416)
.219
(.153)
.114
(.108)
White
-.732
(.464)
-.344**
(.171)
-.134
(.120)
Black
-.455
(.535)
-.220
(.197)
-.154
(.141)
Asian
-.753***
(.133)
-.296***
(.050)
-.135***
(.036)
Hispanic origin
.001***
(.000)
.001***
(.000)
.001***
(.000)
Internet usage intensity
.417
(.373) --a --a Metro area
--a -.129
(.139)
-.106
(.102)
Micro area
-.216
(.260)
-.136
(.096)
-.087
(.071)
MSA
-.103
(.162)
-.103*
(.060)
.006
(.045)
PMSA
.030
(.048)
.007
(.017)
-.001
(.014)
Population per square mile
.015
(.014)
.009
(.005)
.002
(.004)
Personal income per capita
-.062
(.161)
-.080
(.059)
-.014
(.045)
Retail trade establishment
.015
(.025)
-.011
(.009)
.009
(.007)
Retail trade sales per capita
.078***
(.021)
.018**
(.008)
.018***
(.006)
Land area
.000
(.000)
.000
(.000)
.000
(.000)
Population
.013***
(.004)
.006***
(.002)
.002
(.001)
Cost of living index
5.312
(4.008)
5.171***
(1.499)
1.525
(1.134)
Constant
Observations 8765 8765 8765
Pseudo R-squared .017 .010 .051
LR Chi-squared 663.66 542.18 567.20
Standard errors in parentheses.
***p < .01, **p < .05, *p < .10.
a Dropped due to multi-collinearity (cost of living data only available for metro and select micro areas).
30
Table 4
Demand-Reducing Effects of Sales Tax
(1) Yearly online spending (2) Number of transactions (3) Purchase likelihood
-13.865*
(8.407)
-8.268***
(1.993)
-1.236
(.997)
1+tax
.016
(.060)
-.106***
(.021)
.003
(.010)
Household highest education
.163**
(.082)
-.010
(.027)
.033**
(.013)
Household size
-.080**
(.039)
.009
(.013)
-.014**
(.006)
Household oldest age
.285***
(.053)
.047***
(.017)
.045***
(.009)
Household income
.346
(.236)
.066
(.079)
.054
(.038)
Children
2.621***
(.237)
.949***
(.075)
.404***
(.037)
Connection speed
2.080**
(.817)
.413*
(.247)
.321**
(.131)
White
-.182
(.905)
-.338
(.277)
.014
(.145)
Black
1.183
(.999)
-.617*
(.319)
.148
(.162)
Asian
-1.003***
(.236)
-.366***
(.077)
-.147***
(.038)
Hispanic origin
.001***
(.000)
.000***
(.000)
.000***
(.000)
Internet Usage Intensity
1.263*
(.690) --a --a Metro area
--a -.653***
(.212)
-.052
(.105)
Micro area
-.350
(.453)
-.193
(.144)
-.052
(.073)
MSA
-.124
(.281)
-.163*
(.095)
-.038
(.046)
PMSA
-.060
(.082)
.015
(.024)
-.008
(.014)
Population per square mile
-.027
(.025)
-.015*
(.008)
-.002
(.004)
Personal income per capita
.332
(.281)
-.053
(.097)
.032
(.046)
Retail trade establishment
-.046
(.044)
-.004
(.015)
-.008
(.007)
Retail trade sales per capita
.030
(.037)
.010
(.014)
.002
(.006)
Land area
-.001
(.001)
.000
(.000)
-.000
(.000)
Population
.024***
(.008)
.010***
(.003)
.003***
(.001)
Cost of living index
3.478
(9.122)
7.212***
(2.204)
-.279
(1.100)
Constant
Observations 8538 8764 8764
Pseudo R-squared .016 .018 .030
LR Chi-squared 333.73 358.37 309.38
Standard errors in parentheses,
***p < .01, **p < .05, *p < .10.
a Dropped due to multi-collinearity (cost of living data only available for metro and select micro areas).
31
Table 5
Demand-Reducing vs. Demand-Increasing Effects of Sales Tax
Yearly Online Spending Number of Transactions Purchase Likelihood
(1) Without
nexus
(2) With
nexus
(3) Without
nexus
(4) With
nexus
(5) Without
nexus
(6) With
nexus
Variable
6.383
(4.265)
-60.748**
(25.499)
1.586
(1.349)
-16.733**
(7.618)
.865
(.589)
-7.328**
(2.987)
1+tax
.081*
(.044)
-.075
(.155)
.034**
(.014)
-.019
(.046)
.012*
(.006)
-.009
(.018)
Household highest education
.088
(.060)
-.070
(.204)
.007
(.019)
-.011
(.061)
.013
(.008)
-.009
(.024)
Household size
-.038
(.028)
.049
(.097)
.011
(.009)
.002
(.030)
-.006
(.004)
.005
(.011)
Household oldest age
.259***
(.039)
.099
(.132)
.085***
(.012)
-.009
(.038)
.035***
(.005)
.012
(.016)
Household income
.075
(.174)
.403
(.593)
.070
(.056)
.077
(.177)
.008
(.024)
.050
(.070)
Children
2.231***
(.177)
3.335***
(.686)
.585***
(.054)
1.267***
(.214)
.303***
(.024)
.388***
(.079)
Connection speed
1.165*
(.641)
.241
(1.879)
-.049
(.186)
.331
(.605)
.169*
(.088)
.016
(.220)
White
-.457
(.706)
-.527
(2.160)
-.430**
(.209)
.204
(.685)
-.059
(.097)
-.080
(.254)
Black
.474
(.779)
-1.783
(2.422)
-.181
(.237)
-.841
(.802)
.068
(.108)
-.216
(.284)
Asian
-1.287***
(.184)
-.971*
(.588)
-.427***
(.059)
-.456**
(.182)
-.175***
(.025)
-.111
(.069)
Hispanic origin
-.011
(.418)
.511
(1.493)
-.064
(.135)
.239
(.449)
-.002
(.058)
.054
(.177)
Metro area
.288
(.336)
-.437
(1.319)
.036
(.107)
-.227
(.405)
.039
(.046)
-.058
(.155)
Micro area
-.354
(.329)
.605
(1.176)
.026
(.108)
.301
(.346)
-.050
(.045)
.063
(.139)
MSA
.484***
(.175)
.144
(.602)
.115**
(.055)
.014
(.181)
.0651***
(.024)
.015
(.071)
PMSA
.002
(.014)
-.096***
(.036)
.006
(.004)
-.027***
(.010)
.000
(.002)
-.011***
(.004)
Population per square mile
.074***
(.015)
.073
(.053)
.020***
(.005)
.013
(.016)
.010***
(.002)
.009
(.006)
Personal income per capita
-.284**
(.138)
.966**
(.399)
-.096 **
(.044)
.318***
(.116)
-.039**
(.019)
.115**
(.047)
Retail trade establishment
-.024
(.030)
-.279**
(.118)
-.006
(.009)
-.128***
(.034)
-.003
(.004)
-.032**
(.014)
Retail trade sales per capita
.052
(.033)
.151
(.097)
.006
(.010)
.047*
(.028)
.007
(.005)
.018
(.011)
Land area
.001**
(.000)
-.002**
(.001)
.000**
(.000)
-.001**
(.000)
.000**
(.000)
-.000**
(.000)
Population
.001***
(.000)
.001**
(.001)
.001***
(.000)
.000**
(.000)
.000***
(.000)
.000**
(.000)
Internet usage intensity
-15..941***
(4.662)
35.772
(27.330)
-4.038
(1.471)
11.310
(8.214)
-2.171***
(.643)
4.439
(3.219)
Constant
Observations 84282 12918 84282 12918 84282 12918
Pseudo R-squared .142 .098 .152 .132 .223 .151
LR Chi-squared 6364.48 424.40 5836.29 477.42 6416.83 436.42
Standard errors in parentheses.
***p < .01, **p < .05, *p < .10.
To save space, coefficients of firm dummies are not reported.
32
Table 6
County-level Offline Retail Sales
Dependent variable Retail sales per household (log form)
1+tax -1.267**
(.551)
MSA .057**
(.023)
PMSA -.043
(.042)
Land area .008**
(.004)
Population .000***
(.000)
Population per square mile -.016***
(.005)
Percent Hispanic .004***
(.001)
Median age -.045***
(.003)
Male per 100 female -.010***
(.001)
Percent white -.002
(.001)
Percent black -.002
(.001)
Percent Asian .012**
(.005)
Persons per household -.756***
(.080)
Percent high school .007***
(.002)
Percent bachelor .000
(.002)
Household median income .013***
(.003)
Percent below poverty .001
(.003)
Constant 14.708***
(.745)
Observations 2890
R-squared .266
Standard errors in parentheses
***p < .01, **p < .05, *p < .10
1
Appendix
A1. Individual Level Controls
Most of the control variables at the individual level come from demographic information in the comScore
dataset. Table A1 provides details on the coding of education, income, age, household size, presence of
children, and Internet connection speed. Racial background dummies indicate if a family is white, black,
Asian, or another race. “Country of origin” indicates whether the family is of Hispanic origin. Because
higher levels of Internet usage are likely to be associated with greater online purchase frequency and
spending, we calculated “Internet usage intensity” as the average daily time (in minutes) spent online by
each household during the month of January 2006.
TABLE A1
Individual Level Control Variables Coding
Most Educated Head of the Household Household Income Age of Eldest Head
of Household
Household
Size
0 Less than a high school diploma 1 Less than 15k 1 18-20 1 1
1 High school diploma or equivalent 2 15k-24.999k 2 21-24 2 2
2 Some college but no degree 3 25k-34.999k 3 25-29 3 3
3 Associate degree 4 35k-49.999k 4 30-34 4 4
4 Bachelor’s degree 5 50k-74.999k 5 35-39 5 5
5 Graduate degree 6 75k-99.999k 6 40-44 6 6+
99 Missing 7 100k+ 7 45-49
8 50-54
Connection Speed Child Present 9 55-59
0 Not broadband 0 No 10 60-64
1 broadband 1 Yes 11 65 and over
A2. Top 50 Retailers
Table A2 lists the top 50 retailers by sales revenue in the comScore 2006 data set. We visited each of their
websites to determine whether sales from these retailers are tax-advantaged.
2
TABLE A2
Top 50 Online Sellers and Sales Tax Advantage
Tax-advantageda Note
Domain Name
expedia.com No travel
southwest.com No travel
dell.com No required to collect tax in all states
orbitz.com No travel
travelocity.com No travel
amazon.com Yes collect tax in KS, KY, ND and WA
ticketmaster.com No required to collect tax in all states
aa.com No travel
jcpenney.com No required to collect tax in all states
cheaptickets.com No travel
yahoo.com Yes tax depends on seller and buyer
qvc.com No required to collect tax in all states
continental.com No travel
ups.com No required to collect tax in all states
delta.com No travel
officedepot.com No required to collect tax in all states
staples.com No required to collect tax in all states
walmart.com No required to collect tax in all states
jetblueairways.com No travel
priceline.com No travel
quillcorp.com No required to collect tax in all states
victoriassecret.com No required to collect tax in all states
itn.net No travel
hilton.com No travel
hsn.com Yes collect tax in CA and FL
airtran.com No travel
nwa.com No travel
marriott.com No travel
usairways.com No travel
overstock.com Yes collect tax in Utah
costco.com No required to collect tax in all states
sears.com No required to collect tax in all states
circuitcity.com No required to collect tax in all states
target.com No required to collect tax in all states
bestbuy.com No required to collect tax in all states
ichotelsgroup.com No travel
alaskaair.com No travel
quixtar.com No required to collect tax in all states
hotels.com No travel
intuit.com No required to collect tax in all states
apple.com No required to collect tax in all states
hertz.com No travel
enterprise.com No travel
cabelas.com Yes nexus not published
oldnavy.com No required to collect tax in all states
3
peapod.com No groceries, local delivery only
llbean.com Yes nexus not published
avis.com No travel
hotwire.com No travel
aol.com No all transactions are travel related
Notes. Sorted in descending order by online sales in the comScore 2006 data. comScore’s terms of use
do not allow us to disclose domain-specific sales or transactions.
aSites are classified as tax-advantaged if they only collect tax in a small number of states.
