Gasoline Taxes and Consumer Behavior
Gasoline Taxes and Consumer Behavior
Citation Li, Shanjun, Joshua Linn, and Erich J. Muehlegger. 2012. Gasoline
Taxes and Consumer Behavior. HKS Faculty Research Working
Paper Series RWP12-006, John F. Kennedy School of Government,
Harvard University.
Terms of Use This article was downloaded from Harvard University’s DASH
repository, and is made available under the terms and conditions
applicable to Other Posted Material, as set forth at http://
nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-
use#LAA
Gasoline Taxes and Consumer
Behavior
Faculty Research Working Paper Series
Shanjun Li
Cornell University
Joshua Linn
Resources for the Future
Erich Muehlegger
Harvard Kennedy School
February 2012
RWP12-006
The views expressed in the HKS Faculty Research Working Paper Series are those of
the author(s) and do not necessarily reflect those of the John F. Kennedy School of
Government or of Harvard University. Faculty Research Working Papers have not
undergone formal review and approval. Such papers are included in this series to elicit
feedback and to encourage debate on important public policy challenges. Copyright
belongs to the author(s). Papers may be downloaded for personal use only.
www.hks.harvard.edu
Gasoline Taxes and Consumer Behavior∗
Shanjun Li †
Joshua Linn ‡
§
Erich Muehlegger
February 2012
Abstract
Gasoline taxes can be employed to correct externalities associated with automobile use, to
reduce dependency on foreign oil, and to raise government revenue. Our understanding of the
optimal gasoline tax and the efficacy of existing taxes is largely based on empirical analysis of
consumer responses to gasoline price changes. In this paper, we directly examine how gasoline
taxes affect consumer behavior as distinct from tax-exclusive gasoline prices. Our analysis
shows that a 5-cent tax increase reduces gasoline consumption by 1.3 percent in the short-run,
much larger than that from a 5-cent increase in the tax-exclusive gasoline price. This differ-
ence suggests that traditional analysis could significantly underestimate policy impacts of tax
changes. We further investigate the differential effect from gasoline taxes and tax-exclusive
gasoline prices on both the intensive and extensive margins of gasoline consumption. We dis-
cuss implications of our findings for the estimation of the implicit discount rate for vehicle
purchases and for the fiscal benefits of raising taxes.
∗
We thank Antonio Bento, Larry Goulder, Ken Small, Chad Syverson, and participants at the Chicago-RFF
energy conference, the Cornell environmental and energy economics workshop, the UC Berkeley Energy Camp,
the AERE Summer Conference, and the Northwestern, Stanford, UC-Davis, and Maryland seminars for excellent
comments and suggestions. Ken Small also shared with us the state-level data, on which part of our analysis is
based. Adam Stern and Pei Zhu provided excellent research assistance. Funding for the project was provided by
the Belfer Center for Science and International Affairs at Harvard Kennedy School. All errors are our own.
†
Dyson School of Applied Economics and Management, Cornell University, 424 Warren Hall, Ithaca, NY 14853.
Email: [email protected].
‡
Resources for the Future, 1616 P St. NW, Washington DC, 20036. Email: [email protected].
§
Harvard Kennedy School and NBER, Taubman 342, 79 John F. Kennedy Street, Cambridge, MA 02138. Email:
Erich [email protected]
1 Introduction
The gasoline tax is an important policy tool to control externalities associated with automobile
use, to reduce dependency on oil imports, and to raise government revenue. Automobile use
imposes externalities including local air pollution, carbon dioxide emissions, traffic accidents, and
traffic congestion (Parry, Walls, and Harrington (2007)). Although the gasoline tax is not the
theoretically optimal tax for all of these externalities, a single tax avoids the need for multiple
instruments (e.g., distance-based taxes and real time congestion pricing) and offers an administra-
tively simple way to control these externalities at the same time. Besides correcting environmental
externalities, the gasoline tax can reduce gasoline consumption and may mitigate concerns about
the sensitivity of the U.S. economy to oil price volatility, constraints on foreign policy, and other
military and geopolitical costs. Moreover, gasoline taxes at the federal and state levels are major
funding sources for building and maintaining transportation infrastructure. Federal fuel taxes
provide the majority of revenue for the Highway Trust Fund, which is used to finance highway
and transit programs. Past increases in federal gasoline taxes have been used to generate revenue
for such programs, but the federal gasoline tax has stayed constant since 1993. Greater infras-
tructure investment needs and declining fuel tax revenues due to the recent economic downtown
have led the Highway Trust Fund to be insolvent since 2008 and required Congress to provide
funding from the General Fund.1
Growing concerns of climate change, air pollution, energy security, the national budget deficit,
and insolvency of the Highway Trust Fund have brought renewed interests in increasing state and
federal gasoline taxes. Understanding how gasoline tax changes affect automobile use and gasoline
consumption is crucial in effectively leveraging this instrument to achieve these policy goals. An
underlying assumption used in previous policy analysis on the effectiveness of higher gasoline
taxes and the optimal gasoline tax is that consumers react to gasoline tax changes similarly
to gasoline price changes. Consequently, the consumer response to oil-price induced changes in
gasoline prices is often used as a proxy for the response to a commensurate change in the gasoline
tax. The recent economics literature finds that consumers respond little to rising gasoline prices
at least in the short run.2 Together with the maintained assumption, these estimates suggest that
a large increase in the gasoline tax would be required to significantly reduce fuel consumption.
Not only may this exacerbate the perceived political cost of increasing gasoline taxes, but it may
partially explain why U.S. policy makers have tended to favor less-salient fuel economy standards
over gasoline taxes, despite the broad conclusion of a long literature examining the Corporate
Average Fuel Economy (CAFE) Standards that gasoline taxes are more cost-effective in achieving
1
In federal fiscal year 2010, $51 billion of spending was committed from the Highway Trust Fund while the total
revenue into the fund was just $35 billion.
2
A partial list includes Small and Van Dender (2007), Hughes, Knittel, and Sperling (2008), Li, Timmins, and
von Haefen (2009), and Klier and Linn (2010). These studies often use variations in gasoline prices driven primarily
by supply and demand shocks.
2
targeted fuel reductions.3
The purpose of our paper is to test the maintained assumption that consumers respond to
gasoline tax and tax-exclusive price changes in the same way. In contrast to the literature, our
analysis directly estimates consumer responses to gasoline taxes by decomposing retail gasoline
prices into tax and tax-exclusive components. We use three outcomes to examine consumer behav-
ior over short time horizons: gasoline consumption, vehicle miles traveled (VMT), and vehicle fuel
economy (miles per gallon, MPG). Gasoline consumption and VMT represent the intensive margin
and MPG represents the extensive margin. Two separate data sets are employed in our analysis:
aggregate state-level data that allow us examine gasoline consumption, and household-level data
that allow us to examine VMT and MPG. We find that rising gasoline taxes are associated with
much larger reductions in gasoline consumption than comparable increases in gasoline prices.
The results from the baseline specification suggest that a 5-cent increase in the gasoline tax re-
duces gasoline consumption by 1.3 percent in the short-run while an equivalent change in the
tax-exclusive price reduces gasoline consumption by 0.16 percent. Dissecting the intensive and
extensive margins, we find a significant differential effect in household MPG, especially among
newer vehicles. Although we focus on short-term responses, the large effect of taxes on MPG
suggests that the long run response to taxes may also be greater than the long run response to
tax-exclusive gasoline prices. Our analysis also shows that the gasoline tax has a stronger effect
on VMT than the tax-exclusive price, but the difference is not precisely estimated.
There are at least two possible (and not mutually exclusive) explanations for the larger re-
sponse to gasoline taxes than to tax-exclusive prices. First, legislation and proposals to change
gasoline taxes are often subject to intensive public debate and attract a large amount of media
coverage. Therefore, changes in gasoline taxes may be more salient than equal-sized changes in
tax-exclusive prices (e.g., due to oil price shocks). As a result, consumers may respond more to
a tax increase than a commensurate increase in the tax-exclusive price. Recent empirical studies
have shown that consumers are more responsive to salient price or tax changes (Busse, Silva-Risso,
and Zettelmeyer (2006), Chetty, Looney, and Kroft (2009), and Finkelstein (2009)).4 Second, the
durable goods nature of automobiles implies that a change in fuel prices depends on consumer
expectations of future fuel costs. If consumers consider tax changes to be more persistent than
gasoline price changes due to other factors, a larger response to gasoline taxes than prices could
arise through vehicle choice in both the short and long run. Although the short-run response
of VMT to gasoline price changes is unlikely to depend on the persistence of price changes, as
our analysis suggests, the long-run response to persistent changes could be greater than to less
persistent changes because of transaction costs involved in travel mode and intensity decisions
3
See, for example, Goldberg (1998), Congressional Budget Office (2003), Austin and Dinan (2005), Fischer,
Harrington, and Parry (2007), Jacobsen (2010), and Anderson and Sallee (2011).
4
In addition, Finkelstein (2009) finds that the salience of a tax system has a negative impact on equilibrium
tax rates in the context of highway tolls. This leads to the argument that the salient nature of gasoline taxes may
contribute to the low taxes rates in the United States.
3
(such as setting up carpooling or changing where to live and work).
Our findings have several implications. First, they suggest that the gasoline tax would be
more effective than the previous empirical literature has suggested at addressing climate change,
air pollution, and energy security. Several recent proposals have called for higher gasoline taxes
for either fiscal motives (see, e.g., the proposal of the Deficit Reduction Committee), to maintain
the solvency of the Highway Trust Fund, or to internalize the cost of greenhouse gas emissions.
By focusing on the effects of gas taxes, our paper speaks directly to the effectiveness of these
proposals.
Second, separating gasoline taxes from tax-exclusive prices offers a strategy to address a chal-
lenging identification problem in environmental and energy economics. Energy efficiency-related
policies such as CAFE are often advocated because consumers are widely believed to use a high
implicit discount rate to value future energy savings. Beginning with Hausman (1979) and Du-
bin and McFadden (1984), a long literature estimates implicit discount rates consumers use to
evaluate durable good purchases. The identification problem arises because the econometrician
does not observe a consumer’s expectation of future energy costs. Consequently, it is impossible
to estimate implicit discount rates without making assumptions on consumers’ expectations of
future energy prices. In some cases, assumptions of future expectations are innocuous (e.g., for
regulated retail electricity markets) but in others, such as gasoline prices, which are subject to
influences from numerous domestic and international factors, modeling consumer expectations is
not straightforward. Nevertheless, as Section 5.1 illustrates, under assumptions regarding con-
sumer perceptions on state and federal taxes, the implicit discount rate could be identified without
making assumptions regarding consumer expectations of gasoline prices.
Finally, the results have implications for the literature on the optimal gasoline tax (e.g., Parry
and Small (2005)). The literature estimates the optimal tax based partly on empirical estimates
of the elasticity of gasoline consumption to gasoline prices, under assumptions that the gasoline
tax and gasoline price elasticities of demand are the same. Our analysis focuses on short-term
responses to gasoline taxes and tax-exclusive prices. Although we find evidence of differential
responses in gasoline consumption from lagged tax and price changes, we leave the estimation of
long-run responses for future research.
The paper proceeds as follows. In section 2, we present some background on U.S. gasoline
prices and taxes. We present our analysis of the aggregate state-level data in section 3 and present
our analysis of the household-level data in section 4. In section 5, we discuss the implications of
our results for the estimation of implicit discount rates and the elasticity of fiscal revenue. Section
6 concludes.
4
2 Background on U.S. Gasoline Prices and Taxes
Our empirical analysis employs changes in state gasoline taxes and tax-exclusive prices to inves-
tigate the effects of taxes on gasoline consumption, vehicles miles traveled, and vehicle choices.
In this section, we discuss variation of U.S. gasoline prices and taxes.
Taxes make up a substantial portion of U.S. retail gasoline prices. As an illustration, we
decompose gasoline prices into oil prices and excise taxes. We regress the tax-inclusive price
for state s and year t on crude oil prices, federal and state excise taxes, state fixed effects, and
state-specific linear time trends:
RetailP ricest is the retail price, OilP ricet is the crude oil price, and τ is the sum of federal and
state excise taxes. The state fixed effects, αs , capture time-invariant differences in gasoline prices
that arise from differences in transportation costs. The linear time trends allow the retail prices
in each location to adjust at a different linear rate over time. The coefficient on taxes is 1.03
and is statistically indistinguishable from 1, suggesting that gasoline taxes are heavily borne by
consumers. This is consistent with the result in Marion and Muehlegger (2011), which finds that,
under typical supply and demand conditions, state and federal gasoline taxes are passed fully
on to consumers and are incorporated fully into the tax-inclusive price in the month of the tax
change.
Based on these estimates, Figure 1 decomposes the average U.S. retail gasoline price (dashed
line) into an oil component, a tax component, and the state fixed effects and time trends. Although
much of the intertemporal variation in national gasoline prices is correlated with changes in oil
prices, taxes constitute a significant portion of the tax-inclusive gasoline prices for much of the
period. Table 1 reports the average nominal gasoline price, state gasoline tax, and federal gasoline
tax, in cents per gallon for five-year intervals beginning in 1966 and ending in 2008. In addition,
for each period the table reports the percentage of gasoline price changes explained by changes
in gasoline taxes. The percentage varies substantially over time, rising with the federal gasoline
tax (from 4 to 9 cpg in 1983, to 14.1 cpg in 1991, and then to 18.4 cpg in 1994) and state taxes,
and falling during periods of volatile oil prices.
National averages obscure substantial cross-state variation in gasoline taxes. Figure 2 displays
snapshots of per-gallon state gasoline taxes in 1966 and 2006. Figure 3 maps changes in state gas
taxes from 1966 to 1987 and 1987 to 2008. Figure 4 presents the mean, maximum and minimum
state tax as well as the federal gasoline tax over the period. Although the mean state gas tax
rises slowly over time, state taxes rise more quickly in some locations than in others. In 1966,
the difference between the states with the highest and lowest gas taxes was 2.5 cpg. In 2008, the
difference was 30 cpg; Georgia’s excise tax was 7.5 cpg while Washington’s excise tax was 37.5
5
cpg.
States vary substantially in the frequency and magnitude with which they increase gasoline
excise taxes. From 1966 to 2008, state per-gallon taxes changed in approximately 26 percent of
the state-years.5 Gasoline taxes rose in 488 state-years and fell in 44 state-years, out of 2,064
total observations. Nebraska, North Carolina, and Wisconsin changed taxes most often, in 29,
24, and 24 years, respectively.6 Georgia only changed the gasoline excise tax twice.
Figure 5 graphs the proportion of the tax-inclusive retail price made up by excise taxes. At the
median, taxes make up approximately 26 percent of the after-tax price. This varies substantially
over time and across states; the proportion is greatest during the late 1960s and late 1990s when
oil prices were relatively low and taxes were relatively high. The proportion is lowest during the
early 1980s and after 2005, when oil prices rose substantially. At the peak in 1999, the proportion
varies from a low of 25 to 30 percent (at the 5th percentile) to a high of over 40 percent (at the
95th percentile).
Although gasoline taxes constitute a large proportion of after-tax fuel prices, relatively little
research examines political and economic factors that drive state and national fuel taxes.7 The
previous literature identifies a number of political and economic factors that correlate with fuel
tax changes, such as road revenues from other sources (Goel and Nelson (1999)), environmental
regulation and trucking industry employment (Decker and Wohar (2007)), and government debt as
a percent of GDP (Hammar et al. (2004). Finally, Doyle and Samphantharak (2008) use gasoline
tax moratoria that were granted in Illinois and Indiana in 2000 to estimate the incidence of
gasoline taxes. Although in this case taxes were waived in direct response to high gasoline prices,
gas tax moratoria are very rare and constitute a negligible fraction of the observed variation.
Overall, the past literature identifies political and economic factors correlated with tax changes,
but the variables considered explain only a small fraction of total variation. We provide similar
findings in our analysis below.
6
and the intensive margin (vehicle usage).8 In this section, we present our empirical strategy, data,
and results using the aggregate data.
τ τ
ǫp = α − β ; ǫτ = β . (3)
p+τ p+τ
Similarly, we can derive the semi-elasticities, which are defined as the percent change associated
with a unit increase in either the tax-exclusive price or gasoline tax:
∂ln(q) 1 τ ∂ln(q) 1
= α−β ; =β . (4)
∂p p p+τ ∂τ p+τ
This approach provides a direct test of whether taxes are more strongly correlated with behav-
ior than are tax-exclusive gasoline prices. If consumers respond equally to changes in the gasoline
tax and tax-exclusive price (of the same size), α is equal to β. The two semi-elasticities derived
above would be the same and equation (2) reduces to a regression of quantity on the tax-inclusive
gasoline price. If, on the other hand, consumers respond more to a change in taxes than to a
8
Although state-level VMT measures are available, we do not use them to examine the intensive margin because
of their well-known measurement errors.
7
change in the tax-exclusive price, β > α. Because we use state fixed effects and annual data, we
interpret the results as short-run effects.
3.2 Sources
We use a panel of data on gasoline consumption and federal, gasoline prices, and state gasoline
taxes by state-year from 1966 to 2008. The data are taken from annual issues of Highway Statistics
Annual, published by the Federal Highway Administration. Tax-inclusive retail gasoline prices
are from the Energy Information Administration State Energy Price Reports. The data contain
demographic variables, including population and average family size from the Current Population
Survey, Bureau of Economic Analysis (BEA) and the Census; and per capita income, gross state
product, and fraction of the population living in metropolitan statistical areas (MSAs) from
BEA. The fraction of the population located in metro areas with rail transit is calculated from
the Statistical Abstract of the United States. There are several additional vehicle-related variables
from the Highway Statistics reports: the number of licensed drivers, number of registered cars
and trucks, and miles of public roads. Except for the federal gasoline tax, all variables vary by
state and year.
8
increase has a much larger effect on gasoline demand than does an equal-sized increase in the
tax-exclusive price.
For comparison with the literature, Table 3 also reports the implied tax and price elasticities
of demand. Similar to the results for the semi-elasticities, the tax elasticity is larger (-0.069) than
the tax-exclusive price elasticity (-0.030). Because of the differences in the scale of the gasoline
tax and the tax-exclusive price, and the fact that it is more natural to compare the effects of a
given monetary change in taxes or tax-exclusive prices, the remainder of the paper focuses on
semi-elasticities.
Columns 3-6 in Table 2 show that the results are robust to adding additional controls and
estimating the same specifications without the regression weights. For comparison with the main
results, we also report the estimates using ordinary least squares (OLS). The coefficient estimates
are much larger with OLS than with FGLS. Table 3 reports the elasticity and semi-elasticity
estimates for the corresponding specifications.
We take two approaches to examine the exogeneity of the gas tax rate. First, we compare the
demographics of high tax and low tax states. We classify states as high tax by comparing the
state tax rate to the weighted average national tax rate in a given year.9 We present the mean
and standard deviations of the demographic variables for the high tax and low tax states in Table
4. In addition, we calculate the difference between the mean of the demographic variables in
the high and low tax states and report whether the means are statistically distinguishable. Tax-
exclusive prices are indistinguishable statistically, which is consistent with consumers bearing the
majority of gasoline taxes. In addition, we do not find significant differences in per capita income,
educational attainment, family size, vehicles per capita, urban population share, unemployment
rates or state budgetary health measured as a state’s budget surplus (or deficit) as a percent of
9
Only five states are exclusively classified as above or below mean in all years. Gasoline taxes in Nebraska,
Washington and West Virginia are above the national average in all years. Gasoline taxes for Missouri and Wyoming
are below the national average in all years. Across all years, states in the 25th percentile report tax rates above
the national average in nine or fewer years and states in the 75th percentile report gas taxes above the national
average in 32 of the 43 years.
9
total expenditures. We do find that high tax states have slightly fewer drivers per capita and a
slightly lower fraction of the population living in a metro area with a rail transit system, although
even these differences are small in absolute value. Furthermore, states with above average gasoline
prices tend to generate a slightly smaller share of Gross State Product (GSP) from manufacturing
and mining.
States with higher than average gasoline taxes tend to elect U.S. senators and U.S. represen-
tatives who receive higher ratings from the League of Conservation Voters and elect a greater
fraction of Democrats to the State senate. We do not find statistically significant differences
between the fractions of Democrats elected to the State House of Representatives in high and low
tax states, nor do we find that high tax states are more or less likely to elect a Democrat as a
Governor.
Second, we test whether changes in economic or political variables predict state gasoline
tax changes. We regress year-to-year changes in nominal state gasoline taxes (per gallon) on
first-differenced economic and political variables. We include the socioeconomic variables (e.g.
GSP per capita, unemployment, urbanization, and educational attainment), political variables
(e.g. League of Conservation Voters scores for a state’s congressional delegation, a dummy for
a governor belonging to the Democratic Party, the fraction of a state’s house and senate seats
occupied by Democrats, and a state’s budget surplus as a fraction of revenues), and industry
variables (e.g. manufacturing and mining GSP shares) from Table 4. Table 5 presents the results.
Specifications (1) through (3) successively add explanatory variables. Specifications (4) through
(6) regress the change in the state gasoline tax on lagged changes in the explanatory variables.
The only significant, consistent correlation we find is that changes (both contemporaneous and
lagged) in state GSP are negatively correlated with gasoline tax changes.10 Collectively, the
political and economic variables explain little of the variation in gasoline prices beyond what is
explained by year fixed effects. The political and economic variables are jointly significant in only
one of the six specifications.
Next, we turn to the exogeniety of the tax-exclusive price. We use crude oil prices to instrument
and correct for the potential endogeneity of the tax-exclusive price. We construct two instruments:
(1) the interaction of the tax-exclusive price in 1966 with the average annual price of imported
crude oil, and (2) one plus the gasoline tax divided by the annual average price of imported
crude oil.11 We present the instrumental variables (IV) results in Table 6. The first two columns
10
As an alternative specification, we also regressed a dummy of whether or not state gasoline taxes rise year-on-
year on the first-differenced explanatory variables. We do not find consistent significant correlations between the
explanatory variables and the timing of state gasoline tax increases.
11
We use the average price of imported crude oil rather than the more commonly used WTI or Brent crude spot
price because the imported price series begins in 1968. Between 1985 and 2008, during which we observe all three
series, the correlation coefficient between the imported crude oil price and the WTI and Brent crude oil spot prices
10
replicate the OLS and FGLS specifications in columns 6 and 2 of Table 2 using the slightly shorter
sample for which our instruments are available (1968 to 2008). Columns 3 and 4 present the IV
results for the same specifications, with column 4 correcting for autocorrelation in the error term
using the IV-GLS method. After instrumenting, the point estimate for the coefficient on the log of
the tax-exclusive price falls slightly, while the point estimate for the coefficient on ln(1+ taxratio)
rises. In all four specifications, the coefficient on ln(1 + taxratio) is significantly greater than the
coefficient on the log of the tax-exclusive gasoline price. Moreover, the parameter estimates from
IV-GLS are very close to those from FGLS, suggesting that the endogeneity of the tax-exclusive
price may not be a serious problem.
To further investigate this issue, we conduct two tests for omitted variables that may be
correlated with both state tax rates and our variables of interest, and may consequently drive a
spurious difference between the tax rate and tax-exclusive gasoline prices. Of particular concern
are unobserved trending variables, i.e., omitted demographic trends affecting vehicle ownership
or driving intensity that are correlated with the state gasoline tax.
First, we examine a shorter state-level panel with monthly gasoline taxes, prices, and con-
sumption from 1983 to 2008. To test for omitted variables, we estimate a first-differenced version
of (2) using the monthly data. First-differencing the higher frequency data makes it less likely the
coefficients will be biased. An omitted variable must change in the same month as the state excise
tax to bias the coefficients of the first-differenced monthly specification. We present the results
in Table 7. As a point of comparison, columns 1 and 2 re-create the earlier levels regressions
from Table 2 using the shorter monthly panel.12 The estimated coefficients from the regressions
in levels are similar to the earlier estimates using the longer, annual, panel. Columns 3 and 4
regress gasoline consumption on the tax-exclusive price and tax rate, after first-differencing. As
in the levels regression, we find a significant difference between the coefficients on tax-exclusive
price and the tax rate.
One drawback of using first-differenced monthly data arises if consumers shift consumption in
response to anticipated changes in gasoline prices or taxes. In this case, first-differenced gasoline
consumption may appear to be more responsive than in our levels regressions. As an additional
check, we aggregate the data up to the season before first-differencing in columns 5 and 6. At
the seasonal level, intertemporal substitution is unlikely to be a problem. Although the size of
both coefficients declines, we continue to find a statistically significant difference between the
coefficients on the tax-exclusive price and the tax rate. This suggests that the results in columns
3 and 4 are not being driven entirely by the strategic timing of gasoline purchases around tax
changes.
Figure 6 presents the second test for omitted variables bias, which plots changes in gasoline
is 0.9988 and 0.9989, respectively.
12
When regressing in levels, we include state fixed effects and time fixed effects. In the first-differenced specifica-
tion, we only include time fixed effects.
11
consumption over time following a change in the gasoline tax. The figure shows that gasoline
consumption falls steadily after a tax increase. There is no evidence of a pre-existing trend, which
provides further evidence against omitted variables bias. Gasoline consumption decreases after a
tax decrease, but as noted above, there are very few tax decreases in the data.
Finally, we consider the dynamics in the effect of taxes and tax-exclusive prices on gasoline
consumption. The specifications in Table 2 assume a log-linear and contemporaneous relationship
between the dependent variables and the tax and tax-exclusive price. In the following, we examine
potential asymmetric and lagged reposes to price changes. There is some evidence in the literature
that consumers respond more to gasoline price increases than to decreases. Because there are so
few examples of tax decreases in the data (about one per state on average), it is not possible
to assess statistically whether there is a differential tax response. It is possible to investigate
asymmetric responses to tax-exclusive prices, however, by adding to the main specification the
interaction of the tax-exclusive price with a dummy equal to one if the price increased between
the previous and current years. If consumers respond more to a price increase than to a decrease,
the coefficient would be negative, but in fact the coefficient is positive and statistically significant.
The coefficient is quite small, however, and we do not find an economically meaningful difference
in the response to tax-exclusive price increases.
Regressions in Table 8 investigate lagged responses by including lags of the tax-exclusive price
and the tax. The results show that adding three lags of both variables reduces the point estimates
on the current tax and tax-exclusive variables by almost half. Nevertheless, the differential effect
from gasoline taxes and tax-exclusive prices still exist. In an alternative specification, we add
two-year lags and the parameters estimates are comparable in magnitude to the current variables.
Thus, we find that even if we allow for lagged responses to taxes and tax-exclusive prices, we find
a larger response to taxes than tax-exclusive prices. This suggests that a differential effect could
exist even in long-run responses to gasoline tax and tax-exclusive price changes.
3.5 Interpretation
As discussed in the introduction, there are at least two explanations for the larger effect of gasoline
taxes. First, gasoline tax changes at both the federal and state levels are often subject to public
debates and attract a great deal of attention from the media. This could contribute to the salience
of gasoline tax changes: a 5-cents increase in gasoline taxes could very well receive more attention
from the media and consumers than a gasoline price increase of the same size.13 Several recent
empirical studies find that salience is an important factor in consumer responses to prices and
taxes. Using experimental data, Chetty, Looney, and Kroft (2009) find that including sales tax
13
These public debates may even lead to the misperception that gasoline taxes are higher and changed more
frequently than they actually are. A telephone survey of 800 adults conducted on behalf of Building American’s
Future in 2009 showed that 60 percent of respondents believe that the federal gasoline tax goes up every year while
in realty, it has not been changed since 1993.
12
in the price tag (hence increasing its salience) reduces demand by nearly the same amount as an
equivalent price increase. In addition, using observational data they show that alcohol purchases
are more responsive to the excise tax (which is included in the posted price) than the sales
tax. Finkelstein (2009) finds that driving become less elastic under electric toll collection (ETC)
because tolls are less salient than manual toll collection. As a result, toll-setting behavior becomes
less sensitive to local election cycles and toll rates increases after adoption of an ETC system.
Second, consumers may perceive changes in gasoline taxes to be more long-lasting than gasoline
price changes caused by other factors such as temporary demand and supply shocks. Given that
automobiles are durable goods, the expectation of future gasoline prices affects vehicle purchase
decisions. Therefore, vehicle purchasers may respond more to a gasoline tax change than to a
price change caused by other factors.
Both of these explanations could work in concert with each other and we do not attempt to
disentangle the two. In the following, we provide several pieces of suggestive evidence for their
validity.
First, to examine the persistence of gasoline taxes and tax-exclusive prices, we conduct AR(1)
regressions with these two variables using the state-level panel data and controlling for state and
year fixed effects. Using the dynamic panel data approach in Blundell and Bond (1998), we obtain
an AR(1) coefficient of 0.925 with a robust standard error of 0.018 for gasoline taxes and 0.775
with a robust standard error of 0.018 for tax-exclusive prices. This suggests that gasoline tax
changes are more persistent; of course, consumer perceptions of persistence could be different
from these estimates.
Second, if either or both of the explanations are true, consumers should respond more to tax
changes in states that change their taxes infrequently. We use several alternatives to equation (2)
to investigate this implication and we find supporting evidence.
We calculate the number of times the state changes its tax from 1966-2008. Column 1 of
Table 9 adds to the main specification the interaction of the tax variable with the number of
times the state changes its tax. The coefficient on the tax variable is -0.38 (instead of -0.32 in
the baseline). The coefficient on the interaction term is positive and statistically significant. This
suggests that the tax elasticity is smaller (in magnitude) for states that change their taxes more
frequently and vice versa. For a state that changes its tax one standard deviation less frequently
than the average (five tax changes in 43 years instead of 10 tax changes), the coefficient is -0.40.
In column 2, states are assigned quintiles based on the number of tax changes. Defining higher
quintiles as states that change taxes more often, we expect smaller coefficients (in magnitude) for
higher quintiles. The pattern holds, but most of the variation is for states that change their taxes
very infrequently (the lowest quintile includes states that change taxes six times or fewer).
The first two columns in Table 9 use the total number of tax changes over the entire sam-
ple. For some states there are periods in which taxes change frequently and other periods when
taxes change infrequently. Consequently, the total number of changes may not accurately reflect
13
consumer perceptions about the persistence of taxes throughout the sample. To address this
possibility, columns 3-5 add the interactions of the tax variable with the number of tax changes
during the past 15, 10 and 5 years. Column 6 uses the number of years since the last change. Be-
cause these variables are calculated using recent tax changes, they may better capture consumer
perceptions than the variables in columns 1 and 2, which are calculated over the entire sample.
The results are similar for these specifications, which show that the effect of taxes on gasoline
consumption is larger for states that change their taxes less frequently.
where i denotes a household. MPGi is the average MPG of all the new and used vehicles (cars
and light trucks) purchased during the past 12 months by household i. The key explanatory
variables include the tax-exclusive gasoline price and the tax ratio. Importantly, the tax and
price correspond to the household’s state and the quarter of purchase. We include a large set of
household demographics. We use quadratic functions for the non-categorical variables: household
size, the age of the reference person, the number of adults, the number of workers and the number
of drivers in each household. We include full sets of fixed effects for the categorical variables:
14
Vehicle scrappage is part of the extensive margin but is not examined in this paper due to data limitations.
14
household income, education of the reference person, MSA size, worker density by census tract,
population density by census tract, rail availability, and urban and rural indicator variables. We
also include fixed effects for year, month, and location (census division or state).
To examine the effect of the tax-exclusive gasoline price and the gasoline tax on household
travel behavior, we estimate the following equation:
τi
ln(V M Ti ) = αv ln(pi ) + βv ln 1 + + Xi Θv + ei , (6)
pi
where VMTi is the daily total VMT across all vehicles belonging to household i. The VMT
equation includes the same set of variables as the MPG equation with the exception of month
dummies, which are constructed to match the travel period.
4.2 Sources
Household data from the NHTS provide detailed demographic characteristics about each house-
hold. Each household is categorized into one of eighteen income bins and eight education bins.
The data include the number and age of adults, and the numbers of workers and drivers in the
household. In addition, the data provide detailed information about neighborhood (census tract)
demographics such as rural and urban indicators, population, working population, housing den-
sity, and the availability of rail. Consequently, the NHTS data provide a detailed set of controls
for characteristics that may vary with both a state’s tax rate and the household’s driving or
purchase decisions.
For the MPG analysis, we use the 1995, 2001, and 2009 NHTS. The data include the make
and model of the household’s vehicles, which we match to the EPA fuel economy database to
obtain MPG for each vehicle. Gasoline prices at the time of purchase are based on the gasoline
prices used in the aggregate analysis. Because purchases of newer vehicle may respond more to
price changes than purchases of older vehicles, our analysis is conducted on two separate samples.
The first sample, with 52,128 observations includes households who purchased at least one used
or new vehicle during the 12 months prior to the survey. The second sample focuses on newer
vehicles. It has 30,363 households who purchased at least one vehicle during the past 12 months
and all the vehicles purchased are less than four years old. Table 10 provides summary statistics
for the two samples. The average MPG of vehicles in the two samples is almost the same and
other variables are quite close as well. The households in sample 2 (those who purchased newer
vehicles) have slightly smaller household size, higher income and more education.
We use a subset of the NHTS data to examine VMT. During the 1995 and 2001 installments,
participants received an initial survey followed by a second survey several months later. In both,
participants were asked to report odometer readings of all of their vehicles. We calculate daily
VMT per vehicle across vehicles owned by a household by comparing the two odometer readings
15
for each vehicle. We also construct the average gasoline price during the odometer reading period
based on the date of the odometer readings and weekly state gasoline prices. Unfortunately, not
all survey participants report the second odometer reading and there are many missing values for
the first odometer reading. We drop approximately two-thirds of the households in the 1995 and
2001 survey waves that have missing data for either of the two odometer readings for any of the
vehicles owned by the household. The final VMT data set contains 28,303 observations. Table 11
reports summary statistics under sample 1.
To our knowledge, whereas the previous literature has used self-reported annual VMT, this is
the first use of VMT data based on two odometer readings from NHTS. We compare the results
using odometer-based VMT with the results using self-reported annual VMT. The data set with
self-reported annual VMT is larger and contains 61,795 observations. Table 11 reports summary
statistics for the self-reported sample under sample 2.
To compare daily VMT based on the two types of VMT estimates, we use 24,528 households
with both values. The (weighted) average daily VMT based on odometer readings is 49.9 with
a standard deviation of 35.3, while the average self-reported daily VMT is 50.1 with a standard
deviation of 45.4. The top graph in Figure 7 plots Kernel densities of the two VMT measures;
the distributions of the two variables are quite similar. Nevertheless, the comparison of the two
distributions masks the differences that exist for a given observation. We find that although the
means of the two variables are quite close, the difference between the two measures (for a given
household) can be quite large: the mean difference is 0.2 but the standard deviation is 38.5.
To further understand the difference, we compare the two VMT measures for two subsamples
that are defined according to whether the odometer-based VMT is above or below the sample
mean of 49.9 (which we refer to as high and low VMT households). The average daily odometer-
based VMT for the two subsamples is 83.0 and 26.0, while the average self-reported daily VMT
is 74.7 and 32.3 for the two subsamples. The middle graph in Figure 7 plots the kernel densities
of the difference between the odometer-based and self-reported VMT for the high and low VMT
subsamples separately, with vertical lines indicating the sample averages. Households who travel
more tend to under-report their travel intensity, and those who travel less tend to over-report.
Given that two odometer readings could happen any time (2-6 months apart in general) during
the year, part of the differences could be caused by seasonality in driving. To check if this is driven
by seasonality, we compare the two measures month by month and find that the pattern still holds
in each of the 12 months as shown in the bottom graph in Figure 7. To the extent that gasoline
prices are negatively correlated with travel, this finding implies that using self-reported VMT in
the regression analysis could attenuate the effect of gasoline prices on travel demand.
Because the observations used for our VMT analysis only constitute about one-third of the full
sample, it is important to know how representative the estimation sample is. Table 12 compares
the characteristics of the subsample of participants who report two odometer readings with the
characteristics of the full 1995 and 2001 samples. We find that the mean tax-exclusive price and
16
gasoline taxes for the VMT subsample and full sample compare quite closely. Households in the
estimation sample are slightly older (mean age 50.66 vs. 49.59) and less likely to live in an MSA
with a subway system (14 percent vs. 16 percent). Overall, however, the mean and the 10th and
90th percentiles of the variables are quite similar for the full sample and the estimation sample.
Figures 8 and 9 show the distributions of the categorical variables for both samples. Similarly to
the other variables, the distributions for these variables are very similar for the full NHTS and the
estimation samples. These comparisons suggest that the estimation sample may be representative
of the full NHTS sample, but we treat the estimation results with caution.
The empirical strategy for gasoline consumption and VMT uses cross-state and time-series
variation in the tax-exclusive prices and gasoline taxes. An important concern is that the tax-
exclusive prices, taxes, and the dependent variables may be correlated with omitted variables.
To investigate this possibility, we examine whether gasoline taxes and tax-exclusive prices are
correlated with the independent variables. In particular, we separate states according to whether
they have high gasoline taxes or tax-exclusive prices. Across the samples, we compare the means
and the 10th and 90th percentiles of the independent variables. In general, we find that the
distributions are quite similar, which is consistent with the assumed exogeneity of tax-exclusive
prices and taxes.
4.3 Results
Table 13 reports key parameter estimates and elasticities for 12 regressions, examining how gaso-
line prices and taxes affect the fuel economy of recently purchased vehicles. Panel A shows six
regressions of the effect of the tax-inclusive gasoline price on average MPG of recently purchased
vehicles. Columns 1 to 3 use sample 1 (all households), and columns 4 to 6 use sample 2 (house-
holds purchasing newer vehicles).
Columns 1 to 3 differ according to whether census division dummies or state dummies are
included. The parameter estimates are very similar and are statistically significant. The elasticity
of MPG with respect to the tax-inclusive gasoline price is 0.065 in the preferred specification, which
includes state dummies. The elasticity estimates from sample 2 (presented in columns 4 to 6) are
similar across the three specifications, and they are only slightly larger than their counterparts
from sample 1. The elasticity estimates are close to those in several recent studies: Small and
Van Dender (2007) estimate a short-run elasticity of 0.044; Gillingham (2010) finds a medium-run
(2-year) fuel economy elasticity of 0.09; Klier and Linn (2010) estimate an elasticity of about 0.12
using monthly data.
Panel B shows regressions that separate the gasoline tax from the tax-exclusive gasoline price;
columns are analogous to those in Panel A. Columns 1 to 3 shows that the gasoline tax has a
larger effect than the tax-exclusive gasoline price. The difference is 0.68 percent and statistically
significant in column 3 when state dummies are included. The differential effect is stronger for
17
newer vehicles: it is 2 percent in column 6. This comparison is intuitive: if the differential effect is
driven by the durable good nature of automobiles, we would expect a stronger effect among newer
vehicles that have a longer remaining lifetime. Supporting this hypothesis, Busse, Knittel, and
Zettelmeyer (2009) find that the adjustment in the new vehicle market to gasoline price changes
is primarily in market shares, while it is primarily in vehicle prices in the used vehicle market.
The gasoline prices are matched to the month of vehicle purchase in the specifications presented
above. We also estimate the same regressions using the 3-month or 12-month averages of the
gasoline price and tax (including the purchase month and months prior to purchase). The results
(not reported) are very similar to those in Table 13, but the parameter estimates are less precise.
This may reflect the fact that the average gasoline price is a noisier predictor of the expected
future gasoline prices.
Table 14 presents key parameter estimates for the VMT analysis, VMT elasticities, and semi-
elasticities. The six regressions in Panel A include total gasoline prices on the right side, while
those in Panel B separate tax-exclusive prices from gasoline taxes. The dependent variable in
columns 1 to 3 is the log of household daily odometer-based VMT (in logarithm), and that in
columns 4 to 6 is the self-reported VMT (sample 2).
The VMT elasticity with respect to gasoline prices from sample 1 ranges from -0.33 to -0.50
in the three specifications. The preferred specification in column 3 with state dummies provides
an estimate of -0.39. When separating gasoline taxes from tax-exclusive gasoline prices in Panel
B, the VMT elasticity with respect to gasoline taxes is not statistically significantly different in
the second and third specifications. This could reflect more limited variations in gasoline taxes
changes than for the MPG analysis in the previous section (recall that the VMT analysis only uses
the 1995 and 2001 NHTS because the second odometer readings were not collected in the 2009
survey). In all specifications, the percent changes in VMT from a 5-cent gasoline tax increase are
larger than those from changes in tax-exclusive gasoline prices of the same magnitude. However,
none of the differences are statistically significant.
We conduct the same analysis in columns 4 to 6 based on self-reported VMT. We use average
gasoline prices over the same period as VMT reporting. The VMT elasticity from gasoline prices
is -0.27 from column 4, while it is smaller in magnitude and statistically insignificant in columns
5 and 6. The regressions in the second panel separate the gasoline tax from the tax-exclusive
gasoline price. There are no statistically significant differential effects on VMT from the tax-
exclusive gasoline price and gasoline tax. Nevertheless, when state dummies are included, the
difference in the VMT effect is quite large in magnitude with equally large standard errors. We
conduct additional regressions using self-reported VMT data based on a larger sample that also
includes data from the 2009 NHTS. The findings, not reported here, are qualitatively the same.
Although we find that vehicle purchase decisions (as reflected in average MPG) respond more
strongly to gasoline tax changes than commensurate tax-exclusive price changes, there is no
statistically significant evidence of a differential effect for VMT. This contrast could be viewed
18
from two rather different angles. First, the limited variations in gasoline taxes in the VMT analysis
could prevent us from precisely estimating a differential effect. Second, the findings could reflect
different natures of MPG and VMT decisions. As discussed above, if consumers view gasoline
tax changes as more persistent than changes in other components of gasoline prices, their vehicle
purchase decisions could react more strongly to gasoline tax changes due to the durable good
nature of automobiles. Regarding VMT, although gasoline prices could affect consumers’ day-to-
day travel decisions, the underlying cause of price changes (e.g., whether they are from gasoline
tax changes or oil price shocks) should not matter for their travel decisions (e.g., how to go to
work) in the short run.15
5 Implications
Our central and robust finding of differential responses to gasoline taxes and tax-exclusive prices
has important implications for the effectiveness of using gasoline taxes to address climate change,
air pollution, and energy security. In particular, gasoline taxes would be more effective than
suggested by previous empirical estimates of the effect of gasoline price changes on gasoline con-
sumption. The results also have implications for the implicit discount rate and the tax revenue
from gasoline taxes, which we discuss in this section.
19
face in the empirical analysis because neither consumers’ discount rates nor their expectation of
future gasoline prices is observed. Consequently, it is not possible to estimate implicit discount
rates without making assumptions on consumer expectations of future gasoline prices.
In the following, we use a canonical vehicle demand model to illustrate how separating gasoline
taxes from tax-exclusive gasoline prices can aid the identification of the implicit discount rate.
The vehicle demand model is a linear model that can be estimated using market-level sales data.
This model can be derived from a multinomial logit model at the consumer level:
where j is the index for a model (e.g., a 1999 Toyota Camry) and 0 indexes the choice of not
purchasing a new vehicle (denoted as the outside good). sj is the market share of model j. cj
is the total expected cost during the vehicle’s lifetime. X is a vector of vehicle attributes that
capture consumer utility from the vehicle. The present value of the total expected cost of owning
vehicle j is equal to:
T T
X 1 pet V M Tt X 1 (τte + epet )V M Tt
cj = vpj + = vp j + , (8)
t=1
(1 + r)t M P Gj t=1
(1 + r)t M P Gj
where vpj is the vehicle price. T is the vehicle’s lifetime and r is the discount rate. pet is the
expected gasoline price at year t, which is the sum of the expected gasoline tax (τ ) and tax-
exclusive gasoline price (ep). In order to estimate parameters in equation (7), one needs to make
assumptions on the discount rate and expected future gasoline prices. If consumers view the
gasoline tax as being (more) permanent in nature, the identification of the implicit discount rate
is possible by assuming τte = τ , without making an assumption on future tax-exclusive gasoline
prices epet .
Decomposing cj into tax and tax-exclusive components, equation (7) can be written as the
following:
T T
h τ X V M Tt ep X V M Tt epet i
ln(sj /s0 ) = α vpj + + + Xj β + ej
M P Gj (1 + r)t M P Gj (1 + r)t ep
t=1 t=1
τ ep
= αvpj + γ1 + γ2 + Xj β + ej , (9)
M P Gj M P Gj
PT V M Tt
where γ1 is equal to α t=1 (1+r)t assuming that the VMT profile is the same across vehicles.
τ ep
M P Gj is tax dollars per mile for vehicle j and M P Gj is the tax-exclusive dollars per mile, both of
which can be easily constructed from available data. Equation (9) can be estimated in a linear
framework. With estimates for α and γ1 , we can recover the implicit discount rate with a given
VMT profile over the vehicle’s lifetime. Relaxing the assumption of vehicle-invariant VMT profile
20
is straightforward and would entail nonlinear estimation to recover the implicit discount rate
simultaneously with other model parameters.
We estimate the above model using sales data at the vehicle model level in 22 MSAs from
1999 to 2006 (Li, Timmins, and von Haefen 2009). To save space, we do not report the full results
and they are available upon request. In the estimation, we control for vehicle price endogeneity
using product characteristics of other vehicles models produced by the same firm and other firms,
which are standard instruments in the vehicle demand literature (Berry, Levinsohn and Pakes
1995). To control for possible endogeneity of gasoline prices, we use similar instruments as in the
state-level analysis: the crude oil price interacted with gasoline prices prior to the sample period.
When using overall gasoline prices to identify the discount rate by assuming a random walk
process, the estimates suggest a large implied discount rate (0.42), implying significant under-
valuation of future fuel costs. When estimating the model by separating gasoline taxes from
tax-exclusive prices, we find that vehicle choices are more responsive to gasoline taxes, consistent
with our findings in the household analysis. The parameter estimates imply a much smaller dis-
count rates (0.11), which is more in line with market interest rates (e.g., for auto financing). A
more flexible demand model (e.g., taking into account consumer heterogeneity) would be desired
for a full evaluation of the implicit discount rate and is left for future work. Our purpose here is
to illustrate the point that consumers’ differential responses to gasoline taxes and tax-exclusive
prices can be exploited to identify the implicit discount rate.
21
6 Conclusion
Despite multiple policy goals that the gasoline tax can help to achieve, the United States taxes
gasoline at the lowest rate among industrialized countries. In 2009, average state and federal
gasoline taxes were 46 cents per gallon, compared to $3.40 per gallon in the United Kingdom.
Heightened environmental and energy concerns, a record national budget deficit, and an insolvent
Highway Trust Fund have brought about renewed interest in raising the gasoline tax in the United
States. Estimates of the effects of higher gasoline taxes often rely on the estimated gasoline
demand elasticity with respect to gasoline prices, with an implicit assumption that consumers
respond to a change in gasoline taxes in the same way as they respond to a commensurate change
in tax-exclusive gasoline prices.
This paper investigates this underlying assumption by separately estimating consumer re-
sponses to gasoline taxes and the tax-exclusive gasoline price. We examine the short-run impacts
of changes in these two components on gasoline consumption, vehicle miles traveled, and vehi-
cle choices using both state-level and household-level data. We find strong and robust evidence
that gasoline tax changes are associated with larger changes in gasoline consumption and vehicle
choices than are commensurate changes in the tax-exclusive gasoline price.
The finding that not all variations in gasoline prices are created equal has important implica-
tions for transportation and tax policies. First and foremost, our work indicates that fuel taxes
may be a more effective measure of reducing gasoline consumption or inducing consumers to adopt
more fuel efficient vehicles than previously thought. Second, our research shows that gasoline tax
changes could provide a useful source to identity the implied discount rate and to quantify the
extent of an energy paradox in automobile demand. Third, our estimates suggest that traditional
analysis on gasoline taxes may slightly overestimate the fiscal benefits of a gasoline tax.
Our research points to three questions that warrant further investigation. First, recent studies
have estimated that the optimal gasoline tax in the United States is more than twice as large as
the current level (Parry and Small, 2005, and West and Williams, 2007). These studies employ the
long-run consumer response to gasoline prices as one of the key inputs for analysis. Our analysis
provides evidence of a differential effect of lagged price changes. Nevertheless, the precise long-run
estimates and their implications for the optimal gasoline tax are unknown. Second, we conjecture
two potential sources for the differential effect: the more long-lasting nature of tax changes, and
the more salient nature of tax increases. Further work is needed to disentangle the importance of
these two factors. Perhaps most importantly is the question of generality. Our finding suggests
that it may be important to consider the source of the price variation when estimating demand
elasticities and conducting policy analysis for other goods and services.
22
References
[1] H. Allcott and N. Wozny (2010). Gasoline Prices, Fuel Economy and the Energy Paradox.
Working paper.
[2] S. T. Anderson and J. M. Sallee (2011), Using Loopholes to Reveal the Marginal Cost of
Regulation: The Case of Fuel-Economy Standards. American Economics Review 101 (4):
1375-1409.
[3] D. Austin and T. Dinan (2005). Clearing the Air: The Costs and Consequences of Higher
CAFE Standards and Increases in Gasoline Taxes. Journal of Environmental Economics and
Management 50(3): 562-582.
[4] S. Berry, J. Levinsohn, and A. Pakes (1995). Automobile Prices in Market Equilibrium.
Econometrica, 63: 841-890.
[5] R. Blundell and S. Bond (1998). Initial conditions and moment restrictions in dynamic panel
data models. Journal of Econometrics 87: 115-143.
[6] M. Busse, J. Silva-Risso and F. Zettelmeyer (2006). $1000 Cash-Back: the Pass-through of
Auto Manufacturing Promotions, American Economic Review 96(4): 1253-1270.
[7] M. Busse, C. Knittel and F. Zettelmeyer (2009), Pain at the Pump: The Differential Effect
of Gasoline Prices on New and Used Automobiles, NBER working paper 15590.
[8] R. Chetty, A. Looney and K. Kroft (2009). Salience and Taxation: Theory and Evidence.
American Economic Review 99(4): 1145-1177.
[9] Congressional Budget Office (2003). The Economic Costs of Fuel Economy Standards Versus
a Gasoline Tax. U.S. Congress: Washington, DC.
[10] M. Davis and J. Hamilton (2004). Why Are Prices Sticky? The Dynamics of Wholesale
Gasoline Prices. Journal of Money, Credit, and Banking 36: 17-37.
[11] L. Davis and L. Kilian (2011) Estimating the Effect of a Gasoline Tax on Carbon Emissions.
Journal of Applied Econometrics 26(7): 1187-1214
[12] C. Decker and M. Wohar (2007). Determinants of State Diesel Fuel Excise Tax Rates: The
Political Economy of Fuel Taxation in the United States. The Annals of Regional Science 41:
171-188.
[13] Doyle, J. J. and K. Samphantharak (2008), $2.00 Gas! Studying the Effects of a Gas Tax
Moratorium. Journal of Public Economics 92: 869-884.
[14] J. Dubin, and D. McFadden (1984). An Econometric Analysis of Residential Electric Appli-
ance Holdings and Consumption. Econometrica 52: 345-362.
[15] A. Finkelstein (2009). E-Ztax: Tax Salience and Tax Rates. Quarterly Journal of Economics,
124(3): 969-1010
23
[16] C. Fischer, W. Harrington, and I. Parry (2007), Do Market Failures Justify Tightening
Corporate Average Fuel Economy (CAFE) Standards? Energy Journal 28 (4): 1-30.
[17] K. Gallagher and E. Muehlegger (2011). Giving Green to Get Green? Incentives and Con-
sumer Adoption of Hybrid Vehicle Technology, Journal of Environmental Economics and
Management 61: 1-15.
[18] K. Gillingham (2010). How Do Consumers Respond to Gasoline Price Shocks? Heterogeneity
in Vehicle Choice and Driving Behavior. Working paper.
[19] R. Goel and M. Nelson (1999). The Political Economy of Motor-Fuel Taxation. Energy Jour-
nal 20: 43-59.
[20] P. K. Goldberg (1998). The Effects of the Corporate Average Fuel Efficiency Standards in
the US. Journal of Industrial Economics 46: 1-33.
[21] H. Hammar, A. Lofgren and T. Sterner (2004). Political Economy Obstacles to Fuel Taxation.
Energy Journal 25: 1-17.
[22] J. Hausman (1979). Individual Discount Rates and the Purchase and Utilization of Energy-
Using Durables, Bell Journal of Economics 10: 33-54.
[23] G. Helfand, and A. Wolverton (2010). Evaluating the Consumer Response to Fuel Economy:
A Review of the Literature. U.S. Environmental Protection Agency.
[24] J. Hughes, C. Knittel and D. Sperling (2008). Evidence of a Shift in the Short-Run Price
Elasticity of Gasoline Demand. Energy Journal 29(1), January 2008.
[25] M. R. Jacobsen (2010). Evaluating U.S. Fuel Economy Standards in a Model with Producer
and Household Heterogeneity. Working paper.
[26] T. Klier and J. Linn (2010). The Price of Gasoline and New Vehicle Fuel Economy: Evidence
from Monthly Sales Data. American Economic Journal: Economic Policy 2(3): 134-53.
[27] S. Li, C. Timmins, and R. von Haefen (2009). How Do Gasoline Prices Affect Fleet Fuel
Economy? American Economic Journal: Economic Policy 1(2): 113-37.
[28] J. Marion and E. Muehlegger (2011). Fuel Tax Incidence and Supply Conditions. NBER
working paper w16863.
[29] I. Parry and K. Small (2005). Does Britain or the United States Have the Right Gasoline
Tax? American Economic Review 95(4): 1276-1289.
[30] I. Parry, M. Walls, and W. Harrington (2007). Automobile Externalities and Policies. Journal
of Economic Literature 45(2): 373-399.
[31] J. Sallee, S. West, and W. Fan (2010). Do Consumers Recognize the Value of Fuel Economy?
Evidence from Used Car Prices and Gasoline Price Fluctuations. Working paper.
[32] K. Small and K. Van Dender (2007). Fuel Efficiency and Motor Vehicle Travel: The Declining
Rebound Effect. Energy Journal 28(1): 25-51.
24
[33] S. West and R. Williams (2007). Optimal taxation and Cross-price Effects on Labor Supply:
Estimates of the Optimal Gas Tax. Journal of Public Economics 91: 593-617.
25
Figure 1: Gasoline Price Decomposition
26
Figure 2: State Gasoline Tax Rates
7.5 6
6
6
7
6 6 7 6
6 6.5
7
5 6
7 6.5
7.5 6 7
7
5 7 6
6
6 6 7 7
6 7
5 5
7 7 7
6.5 7 7
7.5
7 6
7
7 7 6.5
5
7
(30,35]
(25,30] 7
(20,25]
(15,20]
(10,15]
[5,10]
34 23
27.75
20
26.8
25 22 29.9 19
24 19
19.6
14 23.95
20.7 21
27.1 25 30
31.2
19 28 10.5
18
24 24.5 23.523
22 27
24 17
18 19.7 17.5
17 20 29.9
21.5
18 18.875
16
18.4 18 7.5
20
20
(30,35]
(25,30] 14.9
(20,25]
(15,20]
(10,15]
[5,10]
27
Figure 3: State Gasoline Tax Rates
10.5 7
14
11
7
8.5 7 13 9
6 6.5
7
3 2
9 4.5
10.4 13 8
5
8 7.7 2
8
10 13 11.5 9
12 8.300001
6 6
2 8 10.5
9.5 10 8.5
6
9 8
8
8 6 1
10
9
(25,30]
(20,25] 2.7
(15,20]
(10,15]
(5,10]
[0,5]
19.5 10
7.8
5.5
14.4
10.5 9 10.9 4
12 8
5.6
6 16.5
5 10
8.1 6 15
18
6 13.3 2.5
4
8 5.5 5 7
4 16.9
13 6
9 7.5 0
1 3 14.6
8
2 4.9
1
3.4 5 0
5
4
(25,30]
(20,25] 5.9
(15,20]
(10,15]
(5,10]
[0,5]
28
Figure 4: Distribution of Gasoline Taxes, by year
200
.2
100
.1
29
Figure 6: Change in Gas Consumption, pre- and post-tax change
Gasoline Consumption
30
Figure 7: Densities of Daily VMT from Odometer Readings and Self-reported Annual VMT
.015
VMT based on odometer readings
Self−reported VMT
.01
Density
.005
0
High−VMT households
Low−VMT households
.03
Density
.02 .01
0
High−VMT households
Low−VMT households
−5 0 −10
0 5 10 15
31Month
Figure 8: Summary Statistics for NHTS Categorical Variables
App Fig 3.1: Income Group Shares App Fig 3.2: Income Group Shares
by Sample (unweighted) by Sample (weighted)
0.12 0.12
0.1 0.1
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0 0
1 3 5 7 9 11 13 15 17 1 3 5 7 9 11 13 15 17
NHTS Estimation NHTS Estimation
App Fig 3.3: Education Group Shares App Fig 3.4: Education Group
by Sample (unweighted) Shares by Sample (weighted)
0.35 0.35
0.3 0.3
0.25 0.25
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0 0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
NHTS Estimation NHTS Estimation
App Fig 3.5: Housing Density Group App Fig 3.6: Housing Density
Shares by Sample (unweighted) Group Shares by Sample
(weighted)
0.35 0.35
0.3 0.3
0.25 0.25
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0 0
1 2 3 4 5 6 1 2 3 4 5 6
NHTS Estimation NHTS Estimation
32
Figure 9: Summary Statistics for NHTS Categorical Variables (Cont.)
App Fig 3.7: Worker Density Group App Fig 3.8: Worker Density Group
Shares by Sample (unweighted) Shares by Sample (weighted)
0.2 0.25
0.18
0.16 0.2
0.14
0.12 0.15
0.1
0.08 0.1
0.06
0.04 0.05
0.02
0 0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
NHTS Estimation NHTS Estimation
App Fig 3.9: Population Density App Fig 3.10: Population Density
Group Shares by Sample Group Share by Sample (weighted)
(unweighted)
0.25 0.25
0.2 0.2
0.15 0.15
0.1 0.1
0.05
0.05
0
1 2 3 4 5 6 7 8 9 0
1 2 3 4 5 6 7 8 9
NHTS Estimation NHTS Estimation
App Fig 3.11: Urban Group Share App Fig 3.12: Urban Group Share
by Sample (unweighted) by Sample (weighted)
0.3 0.4
0.35
0.25
0.3
0.2
0.25
0.15 0.2
0.15
0.1
0.1
0.05
0.05
0 0
1 2 3 4 5 1 2 3 4 5
NHTS Estimation NHTS Estimation
33
Table 1: Nominal Prices and Taxes (cpg), over time
34
Table 3: Elasticities and Effects of Price Changes on Gasoline Demand
35
Table 4: Comparison of States Above and Below Mean Tax Rate
Notes: States are grouped above or below mean tax relative to the weighted average state
gasoline tax in each year. Standard deviations are in parentheses. *, **, and *** denote
that the difference in means is statistically significant at the 10%, 5%, and 1% levels.
36
League of Conservation Voters scores are available from 1983-2008. GSP shares are available
from 1985-2008.
Table 5: Gasoline Taxes and Political Environment
(1) (2) (3) (4) (5) (6)
GSP per capita (000s) -0.107** -0.0745* -0.0954** -0.164*** -0.131** -0.120*
(0.0430) (0.0434) (0.0472) (0.0564) (0.0589) (0.0603)
Pop. share in MSA with rail 0.195 -1.011 0.176 -0.573 -0.104 0.350
(0.844) (1.793) (1.228) (0.529) (2.851) (3.239)
Fraction of Democrats in State Senate -0.312 -0.594 -0.302 -0.166 0.450 0.567
(0.489) (0.822) (0.927) (0.335) (0.522) (0.512)
Fraction of Democrats in State House 0.839* 0.693 0.668 0.158 0.602 0.528
(0.448) (1.090) (1.153) (0.471) (0.835) (0.817)
Percent State Budget Surplus -0.131 -0.283 -0.283 -0.103 0.291 0.165
(0.261) (0.357) (0.381) (0.204) (0.430) (0.456)
First-differenced
Levels First-differenced Seasonal Data
Variable (1) (2) (3) (4) (5) (6)
38
Table 8: Coefficient Estimates: Lagged Prices and Taxes
39
Table 9: Effect on Gas Consumption of Frequency of Tax Changes
40
Table 10: Summary Statistics: MPG of Recently Purchased Vehicles
Average MPG 21.42 5.13 10.00 49.80 21.40 5.30 10.12 49.25
Tax-excl. gas price 1.59 0.74 0.70 3.94 1.60 0.75 0.70 3.94
Gas tax 0.42 0.07 0.24 0.69 0.42 0.07 0.24 0.69
Household size 2.97 1.43 1 14 2.82 1.35 1 14
Number of drivers 2.10 0.83 0 10 2.06 0.76 0 10
Number of adults 2.11 0.79 1 10 2.07 0.73 1 10
Number of workers 1.60 0.94 0 10 1.55 0.90 0 10
Age of reference person 45.20 14.53 17 92 46.99 14.64 17 92
Household income 10.66 5.21 1 18 12.10 4.93 1 18
Education of ref. person 3.02 1.13 1 5 3.23 1.11 1 5
MSA size 4.12 1.45 1 6 4.16 1.39 1 6
Worker density 1,201 1,446 25 5,000 1,227 1,436 25 5,000
Population density 3,610 4,903 50 30,000 3,636 4,747 50 30,000
With rail 0.21 0.41 0 1 0.24 0.43 0 1
Without rail 0.79 0.41 0 1 0.76 0.43 0 1
Second city 0.19 0.39 0 1 0.18 0.38 0 1
Suburban 0.27 0.45 0 1 0.31 0.46 0 1
Town and country 0.41 0.49 0 1 0.39 0.49 0 1
Urban 0.12 0.33 0 1 0.12 0.33 0 1
Notes: Sample 1, with 52,128 observations, includes households who purchased at least one vehicle
within the past year, from the 1995, 2001, and 2009 NHTS. Sample 2, with 30,363 observations, in-
cludes households who purchased at least one vehicle during the past year such that all the vehicles
purchased are less than four years old. Tax-exclusive gasoline price and gasoline tax correspond to
the purchase month (and are averaged in case of multiple vehicle purchases). Household income,
education of reference person, MSA size, worker density, and population density at the Census
tract level are all categorical variables.
41
Table 11: Summary Statistics: Household VMT
Average daily VMT 50.82 35.71 0.01 347.96 58.94 50.03 0.00 499.09
Tax-excl. gas price 1.07 0.12 0.74 1.69 1.13 0.13 0.87 1.68
Gas tax 0.46 0.06 0.29 0.71 0.45 0.06 0.29 0.68
Household size 2.33 1.32 1 12 2.51 1.38 1 14
Number of drivers 1.69 0.67 0 6 1.79 0.73 -8 10
Number of adults 1.74 0.66 1 8 1.83 0.71 1 9
Number of workers 1.16 0.88 0 6 1.32 0.91 -8 10
Age of reference person 50.98 16.42 17 88 47.85 16.12 17 88
Household income 9.33 4.96 1 18 9.48 4.96 1 18
Education of reference person 3.09 1.18 1 5 3.02 1.15 1 5
MSA size 4.06 1.40 1 6 4.10 1.42 1 6
Worker density at census tract 1,397 1,507 25 5,000 1,396 1,531 25 5,000
Population density at census tract 4,000 4,907 50 30,000 3,890 4,856 50 30,000
With rail 0.22 0.41 0 1 0.21 0.41 0 1
Without rail 0.78 0.41 0 1 0.79 0.41 0 1
Second city 0.21 0.41 0 1 0.21 0.41 0 1
Surbaban 0.32 0.47 0 1 0.29 0.45 0 1
Town and country 0.34 0.47 0 1 0.37 0.48 0 1
Urban 0.13 0.34 0 1 0.13 0.34 0 1
Notes: Sample 1, with 28,303 observations, is from the 1995 and 2001 NHTS and includes households who
reported two odometer readings for each of the vehicles owned. Sample 2, with 61,795 observations, includes
households with self-reported annual VMT for each vehicle owned. In sample 1, the tax-exclusive gasoline
price and gasoline tax are averaged during the period of the two odometer readings. Household income,
education of reference person, MSA size, worker density, and population density at the Census tract level
are all categorical variables.
42
Table 12: Comparison of NHTS and Estimation Subsample
Panel A: Unweighted
MSA has subway 111,850 0.16 0.37 0.00 1.00 34,234 0.14 0.34 0.00 1.00
Panel B: Weighted
Log (tax-excl. gas price) 0.087 0.085 0.105 0.099 0.111 0.134
(0.020) (0.019) (0.029) (0.024) (0.027) (0.037)
Log(1 + tax ratio) 0.151 0.174 0.288 0.201 0.274 0.405
(0.082) (0.073) (0.126) (0.067) (0.077) (0.148)
R-squared 0.053 0.055 0.059 0.082 0.085 0.091
Elasticity w.r.t. tax-excl. gas price 0.051 0.044 0.038 0.052 0.047 0.040
(0.015) (0.017) (0.016) (0.023) (0.023) (0.023)
Elasticity w.r.t. gas tax 0.035 0.041 0.067 0.047 0.064 0.094
(0.019) (0.017) (0.029) (0.016) (0.018) (0.034)
% ∆ in MPG: 5c ↑ tax-excl. price 0.158 0.137 0.117 0.162 0.143 0.123
(0.047) (0.050) (0.050) (0.070) (0.071) (0.072)
% ∆ in MPG: 5c ↑ gas tax 0.418 0.482 0.798 0.552 0.759 1.124
(0.277) (0.202) (0.348) (0.187) (0.213) (0.410)
Difference in % changes 0.260 0.345 0.682 0.396 0.616 1.001
(0.245) (0.224) (0.361) (0.218) (0.236) (0.433)
44
Table 14: Gasoline Taxes, Tax-exclusive Prices, and Household VMT
Sample 1 Sample 2
(1) (2) (3) (4) (5) (6)
45