WHO PAYS THE GASOLINE TAX? HOWARD CHERNICK * & ANDREW RESCHOVSKY **



Similar documents
CONGRESS OF THE UNITED STATES CONGRESSIONAL BUDGET OFFICE CBO. The Distribution of Household Income and Federal Taxes, 2008 and 2009

Social Security Eligibility and the Labor Supply of Elderly Immigrants. George J. Borjas Harvard University and National Bureau of Economic Research

THE DISTRIBUTIONAL EFFECTS OF AN INCREASE IN SELECTED FEDERAL EXCISE TAXES. Staff Working Paper. January 198? The Congress of the United States

The Life-Cycle Motive and Money Demand: Further Evidence. Abstract

Effective Federal Income Tax Rates Faced By Small Businesses in the United States

Executive Summary. 204 N. First St., Suite C PO Box 7 Silverton, OR fax

CBO STAFF DISTRIBUTIONAL EFFECTS OF SUBSTITUTING A FLAT-RATE INCOME TAX AND AVALUE-ADDEDTAXFORCURRENT FEDERAL INCOME, PAYROLL, AND EXCISE TAXES

Jobs and Growth Effects of Tax Rate Reductions in Ohio

MACROECONOMIC ANALYSIS OF VARIOUS PROPOSALS TO PROVIDE $500 BILLION IN TAX RELIEF

Poverty and income growth: Measuring pro-poor growth in the case of Romania

In Search of a Elusive Truth How Much do Americans Spend on their Health Care?

Over the last 40 years, the U.S. federal tax system has undergone three

APPENDIX A The PSID Sample and Family Income

This PDF is a selection from a published volume from the National Bureau of Economic Research

THE U.S. HOUSEHOLD SAVINGS RATE AND CONSUMPTION 3 March 2009, additions 4 March see

Utah Family Energy Costs as Percentage of After-Tax Income

The Risk of Losing Health Insurance Over a Decade: New Findings from Longitudinal Data. Executive Summary

After a severe slowdown in the early 1990s, General Fund

Remarks by Dr. N. Gregory Mankiw Chairman Council of Economic Advisers at the National Bureau of Economic Research Tax Policy and the Economy Meeting

Health Care, Health Insurance, and the Distribution of American Incomes

The Case for a Tax Cut

SOCIAL SECURITY REFORM: work incentives ISSUE BRIEF NO. 6. Issue Brief No. 6. work incentives

Analysis of School Finance Equity and Local Wealth Measures in Maryland

Volume URL: Chapter Title: Individual Retirement Accounts and Saving

IMPACT OF INCREASES IN ELECTRICITY RATES ON LOW AND NON LOW INCOME HOUSEHOLDS IN MANITOBA

WORKING P A P E R. Unequal Giving. Monetary Gifts to Children Across Countries and Over Time JULIE ZISSIMOPOULOS JAMES P. SMITH WR-723.

Chapter 5: Analysis of The National Education Longitudinal Study (NELS:88)

Review of Varshney/Tootelian Report Cost Of AB 32 On California Small Businesses Summary Report Of Findings

Recent reports of corporate downsizing,

Responses to Misleading and Inaccurate Beer Industry Propaganda on Excise Taxes

RE D U C I N G U S G A S O L I N E C O N S U M P T I O N M I G H T S E E M A

The primary purpose of a tax system is to support public goods and services. State and local taxes

Saving and Life Insurance Holdings at Boston University A Unique Case Study

Private Employer-Sponsored Health Insurance

Forecasting Business Investment Using the Capital Expenditure Survey

4. Work and retirement

The Elasticity of Taxable Income: A Non-Technical Summary

Executive Summary Findings Policy Options I. Introduction II. Total State and Local Tax Systems III. Income Taxes...

Finnish Centre for Pensions, Reports 2010:3 EXECUTIVE SUMMARY. Juha Rantala and Ilpo Suoniemi

How Much Do Americans Pay in Federal Taxes? April 15, 2014

13 EXPENDITURE MULTIPLIERS: THE KEYNESIAN MODEL* Chapter. Key Concepts

Forecasts of Macroeconomic Developments, State Revenues from Taxes and Revenue from Other Sources,

PROJECTION OF THE FISCAL BALANCE AND PUBLIC DEBT ( ) - SUMMARY

Life Cycle Asset Allocation A Suitable Approach for Defined Contribution Pension Plans

ICI RESEARCH PERSPECTIVE

Employers costs for total benefits grew

Keywords: Overlapping Generations Model, Tax Reform, Turkey

HANDOUTS Property Taxation Review Committee

En h a n c ed Ea r n i n g s a n d Ta x Revenues from a

Statistical Bulletin. The Effects of Taxes and Benefits on Household Income, 2011/12. Key points

HELP Interest Rate Options: Equity and Costs

An Evaluation of the Possible

New Estimates of the Private Rate of Return to University Education in Australia*

Average Federal Income Tax Rates for Median-Income Four-Person Families

How Much Equity Does the Government Hold?

Quantity of trips supplied (millions)

The 2004 Report of the Social Security Trustees: Social Security Shortfalls, Social Security Reform and Higher Education

Economic Review, April 2012

The Analysis of Health Care Coverage through Transition Matrices Using a One Factor Model

The Decline of the U.S. Labor Share. by Michael Elsby (University of Edinburgh), Bart Hobijn (FRB SF), and Aysegul Sahin (FRB NY)

For Better or For Worse: Default Effects and 401(k) Savings Behavior

Factors Impacting Dairy Profitability: An Analysis of Kansas Farm Management Association Dairy Enterprise Data

How Much Should I Save for Retirement? By Massi De Santis, PhD and Marlena Lee, PhD Research

Inequality, Mobility and Income Distribution Comparisons

Medical Bills and Bankruptcy Filings. Aparna Mathur 1. Abstract. Using PSID data, we estimate the extent to which consumer bankruptcy filings are

READING 11: TAXES AND PRIVATE WEALTH MANAGEMENT IN A GLOBAL CONTEXT

Energy Cost Impacts on Colorado Families

Economic inequality and educational attainment across a generation

SUBMISSION TO THE SOUTH AUSTRALIAN STATE TAX REVIEW

ELASTICITY OF LONG DISTANCE TRAVELLING

Meeting with Analysts

Minnesota Workers' Compensation. System Report, minnesota department of. labor & industry. research and statistics

Unemployment and Economic Recovery

HAS THE PRODUCTIVITY BOOM FINALLY ARRIVED?

Meeting with Analysts

How Equal Pay for Working Women would Reduce Poverty and Grow the American Economy*

CH 10 - REVIEW QUESTIONS

THE IRA PROPOSAL CONTAINED IN S. 1682: EFFECTS ON LONG-TERM REVENUES AND ON INCENTIVES FOR SAVING. Staff Memorandum November 2, 1989

ICI RESEARCH PERSPECTIVE

TAX EVASION AND SELF-EMPLOYMENT OVER TIME: EVIDENCE FROM SWEDEN

Household Energy Expenditure: Measures es off Hardship & Changes in Income

Do Commodity Price Spikes Cause Long-Term Inflation?

Asymmetry and the Cost of Capital

INVESTMENT COMPANY INSTITUTE

Table of Contents. A. Aggregate Jobs Effects...3. B. Jobs Effects of the Components of the Recovery Package...5. C. The Timing of Job Creation...

Taxes and Income: Where Does Kentucky Stand?

This content downloaded on Tue, 19 Feb :28:43 PM All use subject to JSTOR Terms and Conditions

Income Distribution Database (

Chapter 27: Taxation. 27.1: Introduction. 27.2: The Two Prices with a Tax. 27.2: The Pre-Tax Position

THE U.S. CURRENT ACCOUNT: THE IMPACT OF HOUSEHOLD WEALTH

4. Answer c. The index of nominal wages for 1996 is the nominal wage in 1996 expressed as a percentage of the nominal wage in the base year.

The Tax Benefits and Revenue Costs of Tax Deferral

The Property Tax in New York State. Condition Report Prepared for the Education Finance Research Consortium December 2008

How To Find Out How Effective Stimulus Is

The Economic Impact of Texas State University

Chapter 4: Key Conclusions from the Evaluation of the Current Washington Tax Structure

Hard Choices: Revenue-Raising Options for Alaska. Citizens for Tax Justice Alaska Common Ground April 2004

Unemployment Insurance Savings Accounts

VEHICLE SURVIVABILITY AND TRAVEL MILEAGE SCHEDULES

Chapter 6. Inequality Measures

Transcription:

WHO PAYS THE GASOLINE TAX? HOWARD CHERNICK * & ANDREW RESCHOVSKY ** Abstract - The regressivity of the gasoline tax and other consumption taxes has been challenged on the grounds that the use of annual as opposed to lifetime income and consumption data leads to a substantial overestimate of regressivity. Rather than rely on proxies for lifetime income, in this paper panel data on gasoline consumption and income are used to measure incidence over an intermediate time period. When people are grouped into 11-year average income deciles, average gasoline tax burdens are only slightly less regressive than annual burdens. The main reason for the similarity of annual and intermediaterun burdens is the limited degree of income mobility over an 11-year period. INTRODUCTION The taxation of gasoline is a highly contentious issue in the United States. Despite the fact that the combined federal and state tax rate on gasoline in * Department of Economics, Hunter College and the Graduate Center, City University of New York, New York, NY 10021. ** Robert M. La Follette Institute of Public Affairs and Department of Agricultural and Applied Economics, University of Wisconsin Madison, Madison, WI 53706. the United States is approximately onefifth of the average rate in Western Europe, any attempt to increase the tax rate is met with strong opposition. In 1993, a grueling political battle had to be fought in order to raise the federal gasoline excise tax rate by four cents per gallon. The fierce opposition to gas taxes exists despite the view held by many economists and policy analysts that the expanded use of gasoline excise taxes is an appropriate response to automobile congestion and a suitable tool for reducing air pollution and alleviating global warming attributable to vehicular carbon emissions. The opposition to the gas tax on the part of politicians and the general public appears to arise from the widely held perception that the tax is unfair because it is regressive, imposing a greater economic burden on the poor than on higher-income families, and horizontally inequitable, unduly penalizing some regions of the country over others. The evidence for the regressivity of excise taxes comes primarily from crosssectional surveys, which show that lowincome families spend a larger proportion of their annual income on gasoline than do high-income families. Recently, 233

NATIONAL TAX JOURNAL VOL. L NO. 2 the regressivity of the gasoline tax has been challenged by a number of economists who have argued that it is inappropriate to determine tax incidence on the basis of income data from a single year. 1 Their argument, which is based on Friedman s (1957) permanent income theory of consumption and the companion life-cycle model of saving (Ando and Modigliani, 1963), is that, if most people with low incomes are only temporarily poor and if gasoline consumption decisions tend to be made on the basis of lifetime incomes, calculating tax burdens based on data from a single year will yield tax burdens for low-income people that are substantially higher than burdens calculated on the basis of lifetime or permanent income. Although the argument that the regressivity of consumption taxes is overstated when incidence calculations are based on annual income is widely accepted, until recently it has come under little empirical scrutiny. There have been only a handful of studies that have employed alternative measures of income in calculating the incidence of consumption taxes. Almost all of these studies have approached the question of tax incidence from a lifetime perspective, employing various approaches for the calculation of lifetime income. One approach, employed by Davies, St.-Hilaire, and Whalley (1991), uses cross-sectional data and imposes assumptions about income mobility to simulate lifetime income. A second approach, exemplified by Lyon and Schwab (1995), Caspersen and Metcalf (1994), and Rogers (1995), uses limited panel data to estimate lifetime income. Lyon and Schwab combine these estimates with panel data on alcohol and cigarette consumption, in order to calculate long-run excise tax burdens. Caspersen and Metcalf, and Rogers, use consumption data for a single year to calculate annual tax burdens relative to lifetime income, for the value-added tax and for gasoline, alcohol, and tobacco taxes, respectively. Yet another approach, taken by Poterba (1989, 1991a), Metcalf (1994), and the U.S. Congressional Budget Office (1990), is based on the argument that, as long as consumption is a constant fraction of lifetime income, annual consumption expenditures provide a good proxy for lifetime income. These studies, consequently, use total consumption data in a single year as a basis for their tax incidence calculations. Finally, Fullerton and Rogers (1991, 1993) calculate lifetime tax burdens by estimating lifetime income in the context of a large overlapping-generations computable general-equilibrium model. A general conclusion that arises from these empirical studies is that consumption-based taxes are less regressive when incidence calculations are based on lifetime as opposed to annual income. In his studies of the gasoline tax, Poterba (1989, 1991a) concludes that the gasoline tax is actually slightly progressive over the bottom half of the income distribution. Rogers (1995) finds that relative to lifetime income, gasoline tax burdens rise over the bottom three lifetime income quintiles and fall thereafter. Finally, the U.S. Congressional Budget Office (1990) finds that a tax on motor fuels, while regressive relative to annual income, is generally proportional relative to total expenditures. We can thus summarize the literature to date as concluding that the use of annual income in the calculation of gasoline tax burdens creates a substantial annual income bias in the direction of increased regressivity. 234

While, on theoretical grounds, the use of lifetime incidence is highly appealing, there are a number of quite serious practical and conceptual problems with this approach to tax incidence. These problems diminish the usefulness of the lifetime approach and suggest that policy conclusions based on the analysis of lifetime tax burdens should be treated with great caution. The basic problem is that we cannot directly observe an individual s lifetime income. Any estimation of lifetime income is thus, by necessity, difficult to carry out and requires a large number of restrictive assumptions. To illustrate, the foundation for Fullerton and Rogers measure of lifetime income (Fullerton and Rogers, 1993) is an estimated ageearnings profile for a sample of households from the Panel Study of Income Dynamics (PSID). One problem with this approach is that forecasting lifetime income based on incomplete earnings experience can lead to substantial errors. As discussed by Moss (1978), any observed age-earnings profile represents a combination of a cohort effect, an aggregate growth effect, and an individual age-earnings profile. Forecast errors for any of these three components can lead to a substantial error in predicting lifetime income. For example, Barthold 1 (1993) points out that if ageearning profiles had been based on data from the 1960s, one would have projected rapidly rising real wages for young 1 workers throughout the 1970s and 1980s, while in fact, real wages grew very slowly during this period. Another problem with existing estimates of lifetime income is that they tend to rely on data from quite restricted sets of households. For example, both Fullerton and Rogers (1991, 1993) and Lyon and Schwab (1995) base their income estimates on a sample of individuals from the PSID who were continuously heads of households over 18- and 20- year periods, respectively. By excluding female-headed households, in the case of Lyon and Schwab, and maritally unstable households, in the case of Fullerton and Rogers, these sample selection rules result in samples that are not representative of the U.S. population. This is not a minor problem, because nearly 40 percent of the women in the first wave of the PSID lived in a female-headed household in one or more years of the sample period, and the average income of these women was about one-half of the income of the women who were in male-headed families. Moreover, divorce or separation is a primary route into poverty 2 for many women and children. Thus, this type of sample selection rule may bias estimates of lifetime income, because 2 individuals included in the sample are likely to have steeper ageearnings profiles than those who are excluded. The use of total annual expenditures as a proxy for lifetime income is also seriously flawed. Like annual income, consumption expenditures vary over time for transitory reasons. Just as a negative transitory component of income could bias upward the measure of tax burdens relative to annual income, the presence of a positive transitory component of consumption could bias tax burdens downward. For example, an unexpected illness that requires substantial out-of-pocket medical expenses will inappropriately be reflected in the measure of lifetime income, resulting in a downward bias in measured tax burdens. Liquidity constraints are another factor that may weaken the link between annual consumption expenditures and lifetime income. For those who are unable to borrow against future income, changes in consumption are more likely to track 235

NATIONAL TAX JOURNAL VOL. L NO. 2 changes in annual income than changes in lifetime income. Zeldes (1989) identifies those with low levels of liquid assets as most likely to be liquidity constrained, and finds that this group tends to consume more food in response to increases in income than would be predicted by the permanent income life-cycle model. Wilcox (1989) presents aggregate evidence that even well-publicized increases in social security benefits lead to greater increases in consumption at the time they are received than one would expect from the life-cycle model. Conceptually, those with borrowing constraints could still smooth consumption out of savings. However, the identification of liquidityconstrained consumption behavior with the lack of fungible wealth suggests that such individuals have been unable to save out of past income. If liquidity constraints affected only a small portion of the population, they might not be important to the consumption-lifetime income issue. However, the low level of wealth in at least the bottom two quintiles of the income distribution suggests that substantial numbers of people are likely to face such constraints (Wolff, 1994). In sum, for a substantial fraction of the population, annual consumption is likely to be a poor proxy for lifetime income. The use of lifetime income as a foundation for tax incidence analysis implicitly minimizes the difficulties some individuals face in consumption smoothing. On the other hand, the use of annual income as the basis for tax incidence calculations tends to overstate the tax burdens faced by some taxpayers with low annual incomes, because it fails to account for the ability of many taxpayers to augment current income by dissaving or by borrowing. One reason why policymakers may reject measures of lifetime tax incidence is that they recognize that many taxpayers, especially those with low and moderate incomes, face substantial liquidity constraints, and thus find it impossible to use expected higher incomes as a basis for borrowing money to pay current taxes. 3 Finally, the assertion that total consumption is a good proxy for lifetime income, and hence a good measure of ability to pay, is predicated on the assumption that, over the course of a lifetime, all income is consumed. This assumption will hold, however, only if individuals make no bequests. Although research on bequests is limited, a careful study by Menchik and David (1982) provides evidence that bequests are a rising share of lifetime income as income rises. Thus, the failure to categorize bequests and gifts as consumption expenditure will overstate the measured progressivity of consumption taxes. In this paper, we propose an alternative measure of tax incidence that provides a practical compromise between two extremes the use of annual income and lifetime income. Our approach recognizes both the potential problems inherent in the use of data from a single year and the difficulties in developing accurate estimates of lifetime income. We calculate intermediate-run tax burdens based on 11 years of data on both income and gasoline consumption. The choice of 11, instead of an alternative number of years, is attributable primarily to the fact that 11 years is the maximum period over which highquality longitudinal data are available on both income and gasoline consumption. Nevertheless, we shall argue that this intermediate-run period provides a practical compromise between annual and lifetime tax incidence measures. Individuals who have low incomes in any given year fall into one of three 236

groups: those with persistently low incomes over a substantial number of years; those who have temporarily low incomes for transitory reasons, such as unemployment or illness; and those who have temporarily low incomes for life-cycle reasons, for example, because they are students or retired elderly. As a consequence, classification on the basis of 11 years of income will provide a more accurate picture of long-term income for some groups than for others. Only for the first group, the persistently poor, will tax burdens calculated using annual income and consumption data provide a reasonably accurate indicator of longer-run burdens. For all those who do not have persistently low incomes, tax burdens based on annual income and consumption data will be biased in the direction of regressivity. Calculating intermediate-run tax burdens should eliminate any annual income bias in the burden calculations for those who have low incomes because of transitory variations in income. Whether the use of 11 years of data is sufficient to eliminate bias in the calculation of tax burdens for individuals whose income varies over time for lifecycle reasons is open to question. Although 11 years will span about a quarter of an individual s working life, this period may not be sufficient to fully capture changes in economic status resulting from movements along some individuals age-earnings profiles. However, recent evidence on growing wage inequality and stagnant real wage growth for many Americans suggests that, for many individuals, especially those with a low level of education and training, age-earning profiles may be quite flat. For example, the median real earnings of all men who were 30 years old in 1972 rose by only 11 percent in the following 20 years, with most of that growth occurring in the first decade (Levy, 1995). Thus, for an increasingly large number of individuals who have relatively flat age-earnings profiles, 11 years of income data may well provide a good estimate of their longer-run income. However, for those with high lifetime earnings, age-earnings trajectories have been rather steep (Fullerton and Rogers, 1993). For some, especially those with graduate and professional degrees, the slope of the age-income profile may have actually increased. While a decade or so of income information will provide an appropriate foundation for calculating intermediate-run tax burdens, for these individuals, a longer period of income data would be required to provide an accurate estimate of lifetime income. This discussion suggests that income mobility, broadly defined to include both temporary changes in economic position and longer-run movements along ageincome profiles, will be crucial to assessing the bias toward regressivity occurring because of the use of annual income data. In this paper, we explore the relationship between income mobility and tax incidence. We examine both the proportion of individuals who are economically mobile and the differences in tax burdens calculated using annual and 11-year income data, for both mobile and immobile individuals. The next section of the paper describes the data and methodology used in this study. We then present the gasoline tax incidence results for alternate income measures and provide an interpretation of the results, focusing on the importance of income mobility in determining tax incidence. We also briefly address the issue of compensating low-income families for gasoline tax increases. 237

NATIONAL TAX JOURNAL VOL. L NO. 2 METHODOLOGY The primary source of data for computing the intermediate-run burden of the gasoline tax is the PSID. The PSID is a large longitudinal survey that has followed the members of over 5,000 families since 1968. The high quality of the income data and the availability of some longitudinal consumption data make the PSID the best available data set for examining long-term tax burdens. 2 Our sampling frame consists of all family heads and spouses who were in the PSID in 1982. For most of our analysis, the sample is an 11-year panel constructed by following the 1982 sample backward to 1976 and forward to 1986. The sample consists of 10,906 individuals. Making the individual the unit of analysis avoids severe sample selection problems inherent in the choice of a sample based on headship or on marital status. Because changes in family composition are major determinants of changes in income (Bane and Ellwood, 1986), it is important that inclusion or exclusion from the sample not be conditional on such changes. 3 Our sampling strategy is designed to make our results comparable to annual incidence studies based on crosssectional data sets such as the Current Population Survey. It duplicates the annual approach by choosing a sample at a point in time, computing annual incidence in that year, and then comparing the results to a measure of intermediate-run incidence by following the sample backward and forward over time. This approach contrasts with the strategy followed by a number of other incidence studies, which compare individuals economic positions in an initial period to their economic positions in some future year or years. 4 We chose our approach because, from a taxincidence perspective, we want the best possible measure of longer-run ability to pay for a given annual distribution of income. Particularly for those individuals who are liquidity constrained in their ability to consume, a measure of longerrun income based on expected (or actual) future income does not provide as good an estimate of the ability to pay taxes as a measure that combines past and future income. Although tax burdens are calculated over an entire income distribution, most political discussions of consumption taxes focus on the degree of regressivity at the bottom end of the distribution. If a family is poor in the current year and in the midst of an extended spell of poverty, its economic well-being and its ability to pay taxes are presumed to be lower than the well-being and taxpaying ability of a family that has recently fallen into poverty. The literature on spells of poverty has emphasized the distinction between those who are poor at a given point in time and those who ever become poor (Bane and Ellwood, 1986). While many of those who will ever become poor will experience only a short spell of poverty, of those in poverty in any given year, a substantial fraction (over 50 percent in Bane and Ellwood s work) are in the midst of a long spell of poverty. Censoring of the sample, either from the left (prior to the observation period) or the right, can give a misleading picture of the persistence of poverty. Thus, the permanence of low-income status will be understated if we look only at the length of time in poverty of those who are currently poor, ignoring future income levels, or look only at poverty status in the future, ignoring income levels in previous years. 5 Because the public finance literature does not typically calculate tax burdens according to arbitrary thresholds such as 238

the poverty line, the results of research on spells of poverty cannot be applied directly to quantile divisions (e.g., deciles) of the population. Nonetheless, the general point about sample truncation applies. The closer we come to examining completed periods at any given income level, the more accurate the picture of long-run income status we will get for a given annual distribution. Our approach is thus to compare annual burdens for a centered year to average burdens for periods surrounding that year. The basic unit of analysis for intermediate-term incidence is the individual. 6 We assume that income and expenditures are pooled among all members of the family unit. Therefore, at any point in time, the resources available to an individual equal the total resources of the family unit in which the individual resides, and consumption tax burdens depend on family expenditures relative to family income. Our measure of intermediate-run income is the average income of the family (or families) in which an individual lives over the 11- year period covered by our sample. If the individual stays in the same family unit over the sample period, then family and individual incidence are the same. If the individual passes through different family units, intermediate-run incidence is the average of family incidence at each point in time. Due to missing data, it was not possible to calculate average incomes and tax burdens over the full 11-year period for every individual in the 1982 sample. However, the full 11 years of data are available for a third of the sample, and data are available for eight or more years for fully 83 percent of the sample. The implications of the missing data are discussed in a data appendix available from the authors. On the whole, our tax incidence results remain largely unchanged when burden calculations are based on a subsample of individuals with no missing data. Although the PSID provides considerable data on sources of income, it provides only limited information on consumption expenditures of individuals in the sample. In order to develop an estimate of expenditures on gasoline, we rely on data available on the PSID on the annual number of miles driven by the members of each family. 7 Indirect consumption of gasoline through public transportation and reflected in the shipping cost of goods is not considered in this analysis. Gasoline expenditures are imputed to the sample using a three-step procedure. First, miles driven are divided by an imputed measure of the fuel efficiency (miles per gallon) for each family s vehicles. Our miles-per-gallon estimates are based on methodology developed by Kayser (1994) that combines data from the 1983 Survey of Consumer Finances on the make, model, and year of cars owned by each family with data on the average gasoline mileage of each type of car. 8 These data are used to estimate a regression of the assigned fuel efficiency on a number of household-specific socioeconomic and locational variables. 9 The coefficients from this regression are then used to calculate fuel efficiencies for each individual in the PSID. As the data needed to estimate the regression were only available for 1983, we had to assume that the structural relationship that existed in 1983 between individual characteristics (including income) and fuel efficiency also held true in the other ten years of our sample period. Miles-per-gallon data for 1979, 1985, and 1988 from the Residential Energy Consumption Survey (U.S. Energy Information Administration, 239

NATIONAL TAX JOURNAL VOL. L NO. 2 various years) suggest, however, that the relationship between income and fuel efficiency was weaker in the period prior to 1980 than in the period after 1980. Although this suggests that our use of the 1983 structural relationship between income and fuel efficiency may be overestimating the regressivity of gasoline expenditures, a test of this proposition (described in the data appendix) indicates that the magnitude of this overestimate is quite small. 10 The final step in estimating gasoline expenditures is to multiply the number of gallons consumed by each individual by the price per gallon of gasoline. The data on gasoline prices are inclusive of federal excise taxes and state excise and sales taxes. The gasoline price data, which were provided by the Bureau of Labor Statistics, are disaggregated by region and for a number of major metropolitan areas. Individuals living outside the major metropolitan areas were assigned the average price for the region of the country and the size of the jurisdiction in which they lived. The data indicate that poorer families tend to drive older and less fuel efficient cars than families with higher incomes. This implies that an imputation of gasoline expenditures based on national average fuel efficiency would result in a less regressive pattern of gasoline expenditure burden than we observe using our more elaborate imputation procedure. 11 There was a considerable amount of volatility in the price of gasoline during our sample period. The real price of gasoline was stable from 1976 to 1979, increased by 56 percent between 1980 and 1982, and then fell steadily from 1983 through 1986. In real terms, however, prices at the beginning and the end of our sample period were approximately the same. Consumer responses to price changes can occur along two margins, adjustments in miles driven and adjustments in the fuel efficiency of cars. The latter can be accomplished by replacement of fuelinefficient cars, or, in multicar families, by substituting away from the less efficient car(s). In any given year, the miles driven variable reflects the net effect of price adjustments along both of these margins. Miles driven declined in 1980 and 1981 but started to increase again in 1982. Meanwhile, gallons of gasoline consumed decreased continuously until 1985 (Kayser, 1994). This suggests that both types of price responses did, in fact, occur. If the price elasticity of demand varies by income level, then volatility in gasoline prices may influence our tax incidence results. The fact that, during the 11-year period between 1976 and 1986, real gasoline prices were highest in 1982, the base year of our study, suggests that, if gasoline consumption is at all responsive to the level of gasoline prices, our results may overstate the regressivity of the gasoline tax. Although we cannot precisely quantify the magnitude of any overestimate of regressivity, we have used data from the U.S. Department of Energy on fuel efficiency by income class for the year 1979 to make a rough adjustment for any overestimate of fuel efficiency for high-income individuals. The results of this adjustment to the data suggest that the methodology we employ for estimating fuel efficiency does not result in either a significant increase in the regressivity of gasoline expenditures or a substantial underestimate of the annual income bias. 12 240

TAX INCIDENCE RESULTS Gasoline Expenditures as a Percentage of Income To allow easier comparison to other studies and because the incidence of gasoline taxes is quite similar to the incidence of gasoline expenditures, most of our analysis will focus on gasoline expenditures rather than gasoline taxes. We begin with the traditional approach used in tax incidence analysis, namely, comparing gasoline consumption in a single year with income in that year. In the left panel of Table 1, we order our sample of individuals by deciles of 1982 income 13 and calculate average gasoline expenditure burdens by decile (defined as gasoline expenditures by 1982 annual income). The data indicate that, as expected, annual gasoline expenditure burdens are regressive, with burdens falling from 7.3 percent of income in the second decile of 1982 annual income to 3.1 percent in the top decile. To be sure that 1982 was not an atypical year, we also computed annual gasoline expenditure burdens for each year between 1976 and 1986. Average burdens are lower in nine of the ten other years, reflecting lower real prices of gasoline relative to 1982. The incidence of expenditure burdens, however, is similar to the pattern observed in 1982. We also averaged the annual expenditure burdens for each decile over the 11 years and found the resulting distribution of burdens to be similar to, although slightly more regressive than, the annual incidence pattern in 1982. Our finding that gasoline expenditures are regressive when measured using annual data on income and expenditures is consistent with the findings of other recent studies. Poterba (1991a) reports a similar regressive pattern for gasoline and motor oil expenditures relative to annual income. Using 1985 expenditure and income data from the Consumer Expenditure Survey, he finds that the average burden falls from 6.5 percent in decile 2 to 2.4 percent in decile 10. The U.S. Congressional Budget Office (1990) reports that gasoline expenditures as a percent of post-tax family income decline from 6.9 percent in the bottom annual income quintile to 1.5 percent in the top quintile. Rogers (1995) finds an even more highly regressive pattern of annual gasoline expenditure burdens. Her measure of annual burden falls from 1982 Family Income Decile Gasoline Expenditures as a Percentage of Annual Income: 1982 Data from PSID TABLE 1 GASOLINE EXPENDITURE BURDENS 1976 86 Average Family Income Decile 1 (lowest) 6.4 % 1 (lowest) 3.9 % 2 7.3 2 5.3 3 6.7 3 5.1 4 5.8 4 5.0 5 5.7 5 4.6 6 5.2 6 4.3 7 5.0 7 4.0 8 4.6 8 3.8 9 3.9 9 3.4 10 (highest) 3.1 10 (highest) 2.5 Source: Authors tablulations based on data from the PSID. Average Family Gasoline Expenditures During the Period from 1976 to 1986 as a Percentage of Average Family Income During the Period from 1976 to 1986 241

NATIONAL TAX JOURNAL VOL. L NO. 2 0.19 in the bottom annual income quintile to 0.02 in the top quintile. Calculating average burdens for the bottom decile is complicated by the fact that a number of individuals report negative, zero, or very small incomes. The standard response to this problem is to exclude individuals with very low incomes from the analysis. In their wellknown study of the distribution of tax burdens, Pechman and Okner (1974) excluded those in the bottom five percent of the income distribution from the bottom decile. In this paper, we have excluded all individuals in the lowest one percent of the income distribution from the average burden calculations for the lowest decile. Although we have excluded as few individuals as possible, any exclusion rule is, by definition, arbitrary. A close look at the pattern of income over time of the excluded individuals shows very large year-to-year income swings, suggesting that many excluded individuals are self-employed and comfortably middle class, yet suffer occasional business losses. Similar conclusions are reached by Slemrod (1992), who, drawing on a panel of tax return data, reports that those with negative adjusted gross incomes in a given year have relatively high seven-year average incomes. The numbers reported in the tables for the lowest decile also reflect the fact that a substantial portion of the individuals in the lowest decile do not have access to a car and, hence, have a zero burden. If we exclude nondrivers from our calculations, the distribution of burdens is regressive starting with the lowest decile. We expect that the use of annual income in calculating expenditure or tax burdens will result in a more regressive pattern of burdens than if income is measured over a period of years (up to a lifetime). Because measured annual income contains a transitory component that by assumption, is weakly correlated with total consumption, the elasticity of gasoline consumption with respect to annual income should also be smaller than the elasticity with respect to longer-run income. Since the fraction of measured income, which is transitory, should decline as the time period is lengthened, expenditures on any given item, in this case, gasoline, should be less regressive when compared to longrun income than when compared to annual income. Consumption as well as income has a long-run component and a transitory component. Therefore, basing calculations of gasoline tax burdens on data from a single year of gasoline consumption may give a misleading picture of the long-run burden of the gasoline tax. Assuming that the ratio of transitory to long-run consumption declines as the measurement period increases, average gasoline expenditures over 11 years will provide a better approximation of longrun expenditure patterns. Therefore, a more precise measure of the intermediate-run burden of gasoline taxes is given by comparing the average amount of money spent on gasoline during each year of the 1976 to1986 period with the average amount of real income earned during each year of that same period. The right panel of Table 1 presents our intermediate-run incidence results. As we believe that intermediate-run income provides a better measure of taxpayer ability to pay than annual income, we rank individuals by their 11-year average income and assign them to averageincome deciles. Expenditure burdens are defined as average family gasoline expenditures over the 11-year period from 1976 to 1986 divided by average 242

family income over the same 11-year period. 11 The results indicate that intermediaterun burdens decline monotonically from 5.3 percent in the second decile to 2.5 percent in the highest decile. If we characterize individuals in the lowest several deciles as persistently poor, where persistence is defined as low average income over an 11-year period, then an ad valorem tax on gasoline would impose an economic burden on the persistently poor that is nearly twice as high as the burden placed on the persistently rich. Although intermediate-run burdens remain regressive, they appear to be less regressive than annual burdens. To summarize and compare progressivity under different measures of income and consumption, we use the Suits index of tax progressivity, which ranges from 1 (most regressive) to +1 (most progressive). 15 The Suits index for gasoline expenditures using annual data is 0.173. The index drops to 0.153 when 11-year average income and expenditure data are used. A criticism sometimes levied against the Suits index is that it is sensitive to the distribution of income (Kiefer, 1984). For our purposes, however, the fact that the income distribution tends to become more equal when income is measured over longer periods of time is an important factor in explaining the annual income bias. It is, thus, appropriate that the Suits index indicates decreased regressivity as we move to longer time periods. To examine the relationship between the length of the accounting period and measured incidence, we also calculate average gasoline expenditure burdens for the five-year period from 1980 to 1984. The value of the Suits index for five-year average income and expenditure is 0.165. As a percentage of the one-year Suits index, the difference between the one-year and the 11-year index represents an 11.5 percent decline in the regressivity of gasoline expenditure burdens. The decline in regressivity is fairly uniform over the 11-year observation period, with a 4.6 percent change in the first five years. The incidence pattern indicated by the Suits indexes is illustrated in Figure 1, which plots one-year, five-year, and 11- year gasoline expenditure burdens by decile of income. For comparability, individuals are ranked according to their income position over the relevant accounting period. Thus, for 11-year burdens, individuals are ranked by 11- year average income position, while for the five-year period, they are ranked by five-year income deciles. Figure 1 shows that, although the average burden falls as the accounting period is lengthened and regressivity is somewhat reduced by not relying on annual income and consumption data, the overall regressive pattern of the gasoline tax remains regardless of time period. Lyon and Schwab (1995), in a study of alcohol and cigarette tax incidence, arrive at results that are similar to our results for gasoline consumption. They find that taxes on cigarette and alcohol consumption are only slightly less regressive when incidence is measured with respect to both five-year and lifetime incomes than when it is measured with respect to annual income. Our results, however, contrast with those of several other recent studies of gasoline expenditure incidence. Poterba (1991a) calculates expenditure burdens by dividing 1985 gasoline and motor oil expenditures by total 1985 expenditures 243

NATIONAL TAX JOURNAL VOL. L NO. 2 FIGURE 1. Three Measures of Gasoline Expenditure Burdens as a Percentage of Income: Single Year, Five-Year Averages, and 11-Year Averages for families ranked by expenditure deciles. He finds that the pattern of expenditure burdens is progressive through the bottom four deciles before turning regressive over higher deciles. 16 The U.S. Congressional Budget Office (1990) also uses total expenditures as a measure of ability to pay. They calculate motor fuel expenditures relative to total expenditures using annual data, but rank families by their post-tax annual income. 17 They find that spending on motor fuels is proportional to total expenditures over the bottom four income quintiles and regressive over the top income quintile. Rogers (1995) uses gasoline expenditures in a single year and a proxy for lifetime income to calculate gasoline expenditure burdens. Ranking individuals by her lifetime income measure, she finds that gasoline expenditure burdens follow a hump-shaped pattern, with the highest burdens in the middle lifetime income quintile. Gasoline Tax Burdens If the only tax on gasoline was in the form of a national ad valorem excise tax, then the gasoline expenditure incidence displayed in Table 1 would be proportional to the incidence of the gasoline tax. In fact, about two-thirds of the total gasoline excise tax rate faced by the average American consumer is composed of state excise taxes. In 1982, 244

these state taxes ranged from 8 to 22 cents per gallon. In addition, nine states included gasoline consumption in their sales tax base during some or all of the 11-year period between 1976 and 1986. To determine whether the incidence of the gasoline tax is identical to that of gasoline expenditures, we calculated gasoline tax burdens by assuming that gasoline taxes are shifted fully forward to consumers and by applying statespecific total gasoline excise tax rates to gasoline consumption (in gallons) and, where appropriate, state sales tax rates to gasoline expenditures. The results show that the incidence of tax burdens is very similar to the incidence of gasoline expenditure burdens displayed in Table 1. The average intermediate-run gasoline tax burden is 0.90 percent in the second average income decile, 0.77 percent in the fifth decile, and 0.43 percent in the top decile. To determine whether the variation among states in excise tax burdens contributes to the overall regressivity of the tax, we simulated a revenue-neutral uniform national gasoline excise tax rate. The results of the simulation indicate that the variation in rates across the states actually contributes to a slight reduction in regressivity. This occurs because states with a high concentration of residents with relatively low intermediate-run incomes tend to have below-average gasoline tax rates. INTERPRETATION OF THE RESULTS The purpose of this section is to explore the reasons why the intermediate-run incidence of gasoline expenditures (and taxation) is quite similar to the annual incidence. Following Poterba, we refer to the differences between these two measures of incidence as annual income bias. The bias from using annual data to conduct tax incidence analysis occurs for two reasons. First, annual income bias will be larger to the extent that annual income differs from longer-run income. If, over time, most individuals experience limited changes in their real income (for either transitory or life-cycle reasons), the measurement of tax incidence will be quite insensitive to the length of time over which income is measured. Second, given that annual income differs from longer-run income, the annual income bias will be larger to the extent that individuals gasoline consumption depends on their longer-run rather than their annual income. To help us analyze the annual income bias, it is useful to write out a simple model of gasoline consumption behavior. Let us assume that annual gasoline expenditures in year t (G t ) depend on longer-run income (Y l ) and on the difference (Y d ) between annual t income in year t (Y t ) and longer-run income. 18 This relationship can be expressed as 1 G t = α 0 + α 1 Y l + α 2 Y d. t If α 2 = 0, then annual gasoline expenditures depend only on longer-run income. In that case, the consumption model in equation 1 is in the spirit of the permanent income model. If α 2 > 0, then positive deviations of annual income from longer-run income would lead to an increase in annual gasoline expenditures. Defining Y l as equal to Y t 245

NATIONAL TAX JOURNAL VOL. L NO. 2 minus Y d, substituting into equation 1, t and dividing by annual income, we obtain the following expression for the annual gasoline expenditure burden (B t ): 2 B t = Gt = α 0 + α 1 Yt + (α 2 α 1 ) Y t. If Y d equals zero, annual income equals t longer-run income in year t, and the annual burden (the first term in equation 2) is equal to the longer-run burden. If Y d is not equal to zero, then t the first term represents the longer-run component of the measured annual burden. Labeling this term B l, we can define the annual income bias in year t (β t ) as 3 Y t Y t Y β t = B t B l = (α 2 α 1 ) t. Yt From equation 3, we can see that the annual income bias depends on the responsiveness of gasoline consumption to changes in both annual and longerrun income and to income mobility, where mobility is measured as the difference between annual and longer-run income relative to annual income (Y d t /Y t ). d Y t d In this section, we analyze the roles income mobility and gasoline consumption behavior play in determining the incidence of gasoline expenditures. The discussion will help explain our incidence results and will explore the reasons our results differ from those of other studies. As will be explained below, our ability to look separately at the impact of income mobility and gasoline consumption behavior on tax incidence will be enhanced by separating the calculation of intermediate-run incidence into two parts. The first is the calculation of annual, five-year average, and 11-year average expenditure burdens for each individual. The second is the ranking of individuals by alternative measures of their ability to pay taxes, namely, annual, five-year, and 11- year average incomes. Summary statistics generated from this two-step procedure are presented in Table 2, which displays two measures of the incidence of gasoline expenditure burdens using three alternative measures of burden and three alternative income rankings. For each expenditure burden measure, horizontal movements in Table 2 represent alternative rankings of individuals by annual, five-year, and 11-year average incomes. Vertical movements (down the columns) represent the use of different burden measures under a constant income ranking. The Suits indexes presented in the previous section of the paper are displayed in the upper-left to lower-right diagonal cells of Table 2. Thus, in the upper-left cell (which we will call cell a 11 ), burdens are calculated using 1982 annual income and gasoline expenditure data and individuals are ranked by their 1982 annual incomes. In the bottomright cell (cell a 33 ), burdens are calculated using 11-year average gasoline expenditures and income and individuals are ranked by their 11-year average incomes. As discussed previously, the observed decline in the Suits index from 0.173 to 0.153 indicates a modest reduction in regressivity as we move from annual to intermediate-run incidence. As Suits indexes can only be calculated when expenditures (or taxes) and income are ranked according to the 246

TABLE 2 DISTRIBUTIONAL INDICES FOR GASOLINE TAX INCIDENCE: SUITS INDEX (IN BOLD) AND RATIO PROGRESSIVITY INDEX * (IN SQUARE BRACKETS) Alternative gasoline expenditure burdens Ranking of Individuals by Alternate Ability to Pay Measures (1) Annual Income (1) 0.173 annual [0.68] (2) Five-Year Average Income [0.72] (3) 11-Year Average Income [0.76] (2) Five-year average [0.75] 0.165 [0.72] [0.73] (3) 11-year average [0.78] [0.73] 0.153 [0.73] * The ratio progressivity index is defined as the average burden in deciles 7 through 9, divided by the average burden in deciles 2 through 4. Source: Authors calculations. same criterion, we must use an alternative summary measure to determine incidence in the off-diagonal cells of Table 2. As an alternative incidence measure, we take the ratio of the average burden in the top part of the income distribution (deciles 7 through 9) to the average burden in the bottom part (deciles 2 through 4). For convenience, we will call this measure the ratio index of tax progressivity. 19 The larger the value of the index, the greater the progressivity of the underlying distribution. A value of one indicates proportionality. The Role of Economic Mobility We consider first the role of economic mobility in explaining the magnitude of the annual income bias. At the outset, it is important to emphasize that income mobility is a particularly elusive concept, and any conclusions about the extent of mobility are highly sensitive to the way in which it is measured. As Fields and Ok (1996) point out,...the very notion of income mobility is rather a multifaceted one, and any attempt to devise a measure that aims to incorporate all aspects of income mobility is therefore destined for failure (p. 4). Results may differ depending on whether economic movement is defined in relative or absolute terms and depending on the time frame for measurement. Defining Mobility For any given year t, we define income mobility in terms of the difference between an individual s annual income decile and his or her 11-year average income decile. Average income is defined in terms of an income period centered on year t. As discussed previously, defining average income in this way, rather than as average income in a period starting in year t, provides a better measure of ability to pay because it considers both past and future income levels and tends to reduce the impact of sample truncation. We refer to individuals whose annual income position in a given year is close to their intermediaterun income position as economically 247

NATIONAL TAX JOURNAL VOL. L NO. 2 immobile and individuals whose annual position differs substantially from their intermediate-run position economically mobile. This definition of mobility emphasizes changes in relative income position and makes no distinction between mobility that occurs for transitory reasons and that which takes place for life-cycle reasons. To calculate the number of individuals in our PSID sample whose relative income position remain unchanged over a period of 11 years, we calculate the proportion of individuals who are in the same or an adjacent decile of 1982 annual income and 11-year average income. 20 A U- shaped pattern that is characteristic of transition matrix measures of income mobility emerges, with a higher proportion immobile in the tails of the distribution than in the middle (Atkinson, Bourguignon, and Morrisson, 1992). Overall, nearly 85 percent of individuals remained in the same or an adjacent average-income decile and, thus, are classified as immobile. Among households whose incomes are in the bottom two deciles in 1982, 87 percent had low 11-year average incomes. 21 And even in the middle-income deciles, where economic mobility is greatest, we classify over 75 percent of households as economically immobile. It should be pointed out that other studies that have reported larger annual income biases than we do, for example, Fullerton and Rogers (1993), Caspersen and Metcalf (1994), and Davies, St.-Hilaire, and Whalley (1991), also tend to find a substantial amount of income mobility. These differences in reported income mobility stem from two sources. The first is that most other studies rely on restricted samples, generally household heads. These individuals tend to experience more economic mobility than the population as a whole. 22 The second is that, for the reasons articulated earlier, we use a centered measure of mobility, whereas most other studies use a prospective measure of mobility comparing income in year t to income in a period following year t. Furthermore, other studies tend to estimate income over a longer time period, generally a lifetime. For those individuals whose real income follows a linear time trend over the observation period, the interior year will be a better predictor of average income than the initial year. Hence, centered measures of mobility are more likely to reflect transitory changes in income while minimizing the impact of life-cycle changes (movements along an age-earnings profile). 23 We are now ready to use the ratio progressivity indices presented in Table 2 to help us look more closely at the question of why annual gasoline expenditure incidence differs from intermediate-run incidence. We start by asking what happens to the distribution of annual expenditure burdens if individuals are ranked by their average income. The answer to this question can be seen by looking at the first row of Table 2 and comparing the value of the ratio progressivity index in cell a 11 (0.68) to the value of the index in cell a 31 (0.76). As burden measures remain unchanged for all horizontal moves within Table 2, any changes in the value of the index must be due entirely to changes in relative income. The change in the ratio index from 0.68 to 0.76 indicates that income reranking reduces regressivity. Recall that reranking of individuals from their annual to their 11- year average income position results in little change in the relative income positions for most individuals. Thus, the reduction in regressivity indicated by a move from cell a 11 to a 13 reflects the income mobility of a relatively small number of individuals. Many of these 248

individuals have high annual gasoline expenditure burdens. The income reranking moves them up the averageincome distribution. The movement of individuals with high burdens from the bottom of the annual income distribution to a higher position in the averageincome distribution results in the observed reduction in measured regressivity. A different story can be told by looking at horizontal movements in the third row of Table 2. Again, we are isolating the impact of income mobility, but here we look at the distribution of average expenditure burdens as individuals move from a ranking by annual income to a ranking by average income. In cell a 31, average expenditure burdens are arrayed by 1982 annual incomes. 24 By reranking these individuals by their average incomes, the ratio progressivity index is reduced from 0.78 to 0.73, indicating an increase in regressivity or, alternatively stated, a reduction in the magnitude of the annual income bias. To understand this change, remember that the very process of calculating intermediate-run burdens has tended to reduce burdens for those with unusually high annual burdens (and low annual incomes) and increase burdens for those with unusually low annual burdens (and high annual incomes). The reordering from annual to intermediate-run income position (represented by the move from cell a 31 to a 33 ) takes mobile individuals with low annual incomes and what are now relatively low average burdens and moves them up the income distribution. Conversely, mobile individuals with high annual incomes and moderate or high average burdens tend to move down the income distribution. The result is an increase in measured regressivity. This suggests that using measures of longrun tax burdens, but failing to rank individuals by the corresponding measure of ability to pay, will overstate the magnitude of the annual income bias. The Role of Gasoline Consumption Behavior The importance of intermediate-run income as opposed to annual income in determining annual gasoline expenditures depends on the magnitude of α 1 relative to α 2 (see equation 3). The annual income bias will be larger to the extent that gasoline consumption decisions are made on the basis of longer-run income rather than on the basis of annual income. To help isolate the relationship between income and gasoline expenditures from the impacts of income mobility, we move down column 1 of Table 2. The movement from cell a 11 (annual burdens ranked by 1982 annual income) to cell a 31 (11- year average burdens for individuals ranked by 1982 annual income) indicates a substantial reduction in regressivity, with the ratio index going from 0.68 to 0.78. In terms of the ratio index of progressivity, this reduction in regressivity is twice as large as the reduction in regressivity that we observe when we calculate intermediate-run burdens and rank individuals by their intermediate-run incomes. The conclusion we draw is that the annual income bias is substantially overstated if we calculate average expenditure burdens but continue to rank individuals by their annual incomes. To help us understand the reasons for this reduction in regressivity, we divide our sample into three groups according to their magnitude and direction of income mobility. Individuals are defined as upwardly mobile if their 11-year average income decile is more than one decile higher than their 1982 annual income decile, downwardly mobile if 249

NATIONAL TAX JOURNAL VOL. L NO. 2 their average income decile is more than one decile lower than their annual income decile, and immobile if they remain in the same or an adjacent average and annual income decile. For each of these three groups, we calculate annual and 11-year average gasoline expenditure burdens. Figure 2 illustrates the results of these calculations for individuals in each group ranked by their 1982 annual incomes. 25 The top panel of Figure 2 demonstrates that, for those who are upwardly mobile, the use of annual data in calculating gasoline tax burdens will substantially overestimate the regressivity of the gasoline tax. For the lowest three deciles of the 1982 annual income distribution, annual burdens are about eight percentage points higher than average burdens. The annual income bias is especially high in the lowest annual income decile, equalling almost ten percent of income. By contrast, for those who are classified as immobile or downwardly mobile, the annual income bias appears to be negligible. These results suggest that, for the relatively small number of individuals who are upwardly mobile, gasoline consumption decisions are primarily influenced by intermediate-run incomes rather than by (temporarily) low annual incomes. Just as income mobility reduces the annual income bias (increases regressivity) once we have replaced annual by intermediate-run burdens (represented by horizontal movement in Table 2 from a 31 to a 33 ), so replacing annual by intermediate-run burdens, after we have taken into account economic mobility (represented by vertical movement from a 13 to a 33 ), also decreases the annual income bias. In this case, some high intermediate-run income individuals with high annual burdens are assigned lower intermediate-run burdens, leading to greater regressivity. To summarize the results of this section, we argue that the magnitude of the annual income bias depends on the extent of economic mobility and the difference in consumption behavior with regard to longer-run income and annual income. When we compare annual to 11-year incidence for gasoline expenditures, we find that the annual income bias is relatively modest. Looking first at mobility, we find limited economic mobility in our 11-year sample. Only a small proportion of the sample have more than one decile difference between their annual and intermediateincome positions. Comparing gasoline expenditure burdens for those who are mobile and those who are not shows that the annual income bias is substantial only for the small percentage of the sample who are upwardly mobile. Though mobility in our sample is limited, the annual income bias is still overstated if we take into account only economic mobility. The bias is similarly overstated if we take into account only the difference between annual and intermediate-run burdens. The bias is correctly measured by comparing intermediate-run incidence to annual incidence, in other words, by substituting intermediate-run for annual expenditure burdens and reranking individuals according to longer-run ability to pay. The discussion in this section should help clarify why a number of other studies of the incidence of the consumption-based taxes conclude that the use of annual data leads to a substantial overestimate of regressivity. As pointed out previously, the U.S. Congressional Budget Office (1990) study of excise taxes calculated expenditure burdens 250

FIGURE 2. Annual and 11-Year Average Gasoline Expenditure Burdens for Individuals Classified by Income Mobility 251

NATIONAL TAX JOURNAL VOL. L NO. 2 using a single year of consumption data and used total consumption expenditures in the same year as a measure of ability to pay. We have argued in the Introduction that annual expenditures are likely to be a poor proxy for the long-run ability to pay taxes. In addition, the Congressional Budget Office draws conclusions about long-run incidence based on the distribution of expenditure burdens for individuals ranked by their annual income. As demonstrated above, the failure to rank individuals by a measure of long-run ability to pay results in an overestimate of the annual income bias. In his study of gasoline tax incidence, Poterba (1991a) also uses annual expenditures as a proxy for lifetime income. He uses annual gasoline expenditures to calculate burdens, and then ranks individuals by decile on the basis of their total annual expenditures. We calculated that Poterba s results imply a ratio progressivity index for gasoline expenditure burdens of 0.84. This number compares with the ratio index values for annual burdens of 0.68 and for 11-year average burdens of 0.73, and, hence, indicates that the use of total expenditure data as a measure of lifetime income results in a substantially larger annual income bias than the bias relative to intermediate-run income. The difference between Poterba and our gasoline incidence results demonstrates that total annual expenditures provide a poor proxy for intermediate-run income. One possible explanation for this result is that, while the 11 years over which intermediate-run income is measured is long enough to eliminate the bias from transitory income variations, the use of annual expenditures may serve to remove the annual income bias attributable to both transitory and life-cycle effects. If this were so, annual expenditures would indeed be a good proxy for lifetime income. It seems unlikely, however, that annual expenditures are devoid of both transitory and life-cycle effects. Annual consumption almost certainly has a transitory component and a permanent component. As suggested by Caspersen and Metcalf (1994), to the extent that transitory consumption expenditures are constant across income classes, the use of annual expenditure data will lead to a downward bias in tax (or expenditures) burdens at lower income levels. In terms of life-cycle measurement, because annual expenditures exclude bequests, and bequests are a larger proportion of lifetime income for individuals with higher lifetime incomes, the use of annual expenditures as a measure of ability to pay will bias tax burdens upward at the top of the income distribution. Davies, St.-Hilaire, and Whalley (1991) report that the lifetime incidence of sales and excise taxes, while remaining regressive over the entire income distribution, is substantially less regressive than the annual incidence. The ratio index of progressivity for sales and excise taxes, calculated from their Table 2, increases from 0.78 for annual incidence to 0.94 for lifetime incidence an increase of 22 percent. This contrasts with our results, which indicate that the ratio index increases by only seven percent as we move from annual to intermediate-run incidence. In their study, Davies, St.-Hilaire, and Whalley use Lillard s (1977) results on earnings mobility to construct their estimates of lifetime income. By parameterizing a mobility probability that is greater than that for the entire population, and that does not reflect differences in mobility across the annual income distribution, Davies, St.-Hilaire, and Whalley impose unrepresentative mobility on the sample. Our longitudinal 252