National Tax Journal, March 2011, 64 (1), 5984

ON ESTIMATING MARGINAL TAX RATES FOR U.S. STATES

W. Robert Reed, Cynthia L. Rogers, and Mark Skidmore

This paper presents a procedure for generating state-specific time-varying estimates

of marginal tax rates (MTRs). Most estimates of MTRs follow a procedure developed
by Koester and Kormendi (1989) (K&K). Unfortunately, the time-invariant nature of
the K&K estimates precludes their use as explanatory variables in panel data studies with fixed effects. Furthermore, the associated MTR estimates are not explicitly
linked to statutory tax parameters. Our approach addresses both shortcomings.
Using comprehensive tax policy data, we estimate state-specific, time-varying MTRs
for all 50 states over the years 19772004. The inclusion of statutory tax variables
has a significant impact on MTR estimates.
Keywords: state tax revenues, marginal tax rates, tax burden, tax progressivity,
economic growth
JEL Codes: H71, H24, H25

The central role of the government in the economy and the associated high
marginal tax rates mean that the problems of taxing and spending will continue to provide challenging opportunities for research in public economics.
Feldstein (2002, p. 325)
I. INTRODUCTION
his paper generalizes a commonly-used approach for estimating marginal tax rates
(MTRs). MTRs are of particular interest to fiscal policy-makers because it is generally believed that they are influential drivers of economic activity. An important niche in
the voluminous literature on taxes and economic growth is concerned with distinguishing
the effects of marginal from average tax rates. Examples from the international growth
literature include Garrison and Lee (1992), Easterly and Rebelo (1993), Padovano and
Galli (2002), and Myles (2009). Examples from the literature on the growth of U.S.
states include Mullen and Williams (1994), Becsi (1996), Yamarik (2000), and Poulson
and Kaplan (2008).

W. Robert Reed: Department of Economics and Finance, University of Canterbury, Christchurch,

New Zealand (bobreednz@yahoo.com)
Cynthia L. Rogers: Department of Economics, University of Oklahoma, Norman, OK, USA (crogers@ou.edu)
Mark Skidmore: Department of Agriculture, Food, and Resource Economics and Department of Economics,
Michigan State University, East Lansing, MI, USA (mskidmor@msu.edu)

60

National Tax Journal

All of these studies employ a procedure for estimating MTRs that was developed by
Koester and Kormendi (1989, henceforth K&K) in their seminal study of cross-country
growth. K&K use annual observations of income and tax revenues to calculate a single
MTR value for a particular jurisdiction s and time period t = 1,2,T by estimating the
following regression specification,
s st = 0,s + 1,s Incomest + st ; 1, 2,
2, ..., T .

(1)

K&K, and subsequent studies, interpret the slope coefficient on the income term, 1,s,
as a measure of that jurisdictions marginal tax rate (MTR) over all tax bases for the
respective time period,
(2)

MTR
T s=

Tax
a Revenues
esst
= 1,s .
IIIncomeest

Commonly, these MTR estimates are then used as explanatory variables in subsequent
growth regressions.
A secondary application of K&Ks MTR regression is to estimate the income elasticity
of tax revenues. If a jurisdictions MTR is greater than its average tax rate (ATR), tax
revenues are elastic with respect to income.1 Such reasoning has led researchers such
as Becsi (1996) and Crain (2003) to conclude that most state tax systems are income
elastic.2 Other studies, using the same methodology but different time periods, conclude
that state tax systems are moderately inelastic (e.g., Mullen and Williams, 1994).
There are three well-recognized deficiencies of the K&K approach for estimating
MTRs. First, the K&K approach produces a single MTR value for an entire time
period. Thus, it cannot be used as an explanatory variable in panel studies of economic
growth with fixed effects. Second, the K&K approach assumes that the tax structures
of the respective jurisdictions remain constant over the respective time period. This
assumption is demonstrably false, though the practical significance of its violation is
unknown. Third, the associated MTR estimates are not explicitly related to a jurisdictions statutory tax parameters.
This paper presents a simple generalization of the K&K approach that consists of
adding statutory tax variables to (1). It is straightforward to show that this addresses all
three deficiencies above. Our approach allows us to take advantage of available statutory tax data, is easy to implement, and includes the K&K procedure as a nested case.
While it in principle applies equally to country and state studies, data limitations will
make it most applicable to the latter. Accordingly, we use data from U.S. states from
19772004 to estimate state-specific, time-varying MTRs. We demonstrate that the
inclusion of statutory tax parameters results in MTR estimates substantially different
1
2

In the context of (1), B0,s implies MTRs > ATRst for all income levels.
Specifically, Becsi (1996) and Crain (2003) use this condition to argue that state tax systems are progressive. In their context, this is identical to an income elasticity greater than one.

On Estimating Marginal Tax Rates for U.S. States

61

from those generated by the K&K procedure. In addition, we address the possibility
that revenues and income are nonstationary, something ignored in previous research
employing the K&K approach.
Our study proceeds as follows. Section II discusses the relationship between K&Ktype MTR regressions and the literature on the income elasticity of tax revenues. Section
III presents our procedure for estimating MTRs. It highlights some of the details associated with its implementation, including how nonstationarity affects the interpretation of
MTR estimates. Section IV summarizes our data and discusses the statutory tax variables
employed in our analysis. Section V presents our MTR estimates. Section VI concludes.
II. RELATED LITERATURE
The K&K approach to estimating MTRs is related to the extensive literature on the
income elasticity of tax revenues. Like K&K, this literature is concerned with estimating a relationship between revenues and income. Unlike the empirical specification
of (1), this literature typically estimates that relationship using a constant-elasticity
specification,
(3)

Ln(Tax
a Revenues
esst ) = b0,s + b1,s Ln(Incomest ) + est ; t 1, 2, ...T .

Of course, the linear specification of (1) can also be used to construct an income-varying
measure of the income elasticity of tax revenues, st = MTRs/ATRst .3
There are several reasons why one would want to estimate this kind of relationship.
Tax elasticities are useful for forecasting the responses of revenues to expected income
changes (Holcombe and Sobel, 1997). These responses can be focused on particular
types of tax revenues (e.g., personal income taxes, sales taxes) or on total tax revenues.4
The response of total tax revenues to income is particularly pertinent to the literature
on tax structure and the growth of government. If overall tax collections are income
elastic, and expenditures are driven by tax collections, then the public sector will disproportionately expand with income growth (Feenberg and Rosen, 1987; Payne, 2003).
Like K&K, most of these studies are forced to assume a constant tax structure over the
estimation period, though efforts are sometimes made to adjust elasticities for changes
in tax parameters (Sen, 2002).
Although the tax revenue income elasticity and MTR literatures have much in common, they differ in one important respect. The ultimate goal of the income elasticity
literature is the estimation of the revenue-income relationship, while the MTR literature
primarily views this relationship as an input towards subsequent estimation of the effect
of taxes on economic activity.
3

Creedy and Gemmell (2004) report income elasticities constructed in this fashion. Greytak and Thursby
(1979) estimate alternative functional relationships between revenues and income that can be used to
construct estimates of the income elasticity of state tax revenues.
See Dye (2004) for an insightful discussion of issues associated with the former.

62

National Tax Journal

III. A PROCEDURE FOR ESTIMATING STATE-SPECIFIC, TIME-VARYING MTRS

A. The Basic K&K Specification
The following discussion focuses on estimating MTRs for U.S. states. The K&K
approach to estimating MTRs can be represented by the regression specification:

a Revenues
esst = 0
(4) Tax

50

s= 2

0 ,s

Ds + 1

50

Ds

1,s

s= 2

IIncomestt + est ,

where Ds is a dummy variable that equals one for state s and zero otherwise, and tax
revenues and income refer to total state and local tax revenues and some measure of
state income (e.g., personal income). State-specific MTRs are measured by the respective income slope coefficients.
To this basic specification we add statutory tax variables, X,
(5)

Tax
a Revenues
esst = 0

0,ss Ds + 1
s= 2
50

+ 2,k X k ,st
k =1
1

r =1

3,r

(X

50

1,s

s= 2

r,st

Ds Incomest

IIncomest + est ,

where R K depending on whether the revenues associated with the respective tax
parameters are affected by income. The MTRs can be derived as follows,
(6)

MTR
T st =

Tax
a Revenues
esst
IIIncomeest

50

s= 2

D
1,s s

r =1

3,r

X r ,st .

The subsequent discussion explores how this MTR equation can be further refined. It
also addresses a number of associated specification and estimation issues.
B. Selection of Variables
1. Statutory Tax Variables

It is useful to divide the sources of tax revenues into personal income (PIT), corporate
income (CIT), general sales (ST), property (PT), and other (OT) tax revenues. There are
obvious difficulties in characterizing each of these taxes with one or just a few statutory parameters. For example, while sales taxes generally consist of a single rate for all
expenditures, definitions of the associated tax base are subject to various exemptions
(e.g., food, medical, services) and these differ across states.
Personal income taxes pose a particular challenge because deductions, credits, tax rate
schedules, and definitions of taxable income vary widely across states. In this respect,
we are greatly aided by the NBER TAXSIM program (Feenberg and Coutts, 1993). In

On Estimating Marginal Tax Rates for U.S. States

63

addition to reporting selected maximum rates, the TAXSIM program allows calculation of average marginal tax rates for a number of different income sources (e.g.,
wage income, interest income, capital gains income, etc.). These provide a convenient
mapping from a complex personal income tax structure to a relatively small number
of tax parameters.
Our empirical analysis investigates a large number and variety of statutory tax variables. We discuss these in greater detail below.
2. Other Control Variables

One can include additional control variables in the specification of (5). Possible
candidates are time trend/dummy, population, and income inequality variables. If these
are interacted with income, they will also appear in the specification of (6). Previous
studies that have used the K&K approach have overwhelmingly excluded other control
variables. In the work that we report, we use the basic K&K type regression specification to facilitate comparison with that work.5
C. Pooled Regression versus Separate State Regressions
Equations (4) through (6) assume an estimation strategy that pools observations across
states. Alternatively, one could estimate separate regressions for each state. Insufficient
degrees of freedom, however, restrict the number of statutory tax variables one can
include as explanatory variables. Pooling mitigates this problem. Further, many of the
advantages of separate state regressions can be captured by including state-specific
slope and dummy variables.
This has implications for the interpretation of the pooled regression coefficients.
Unlike the state-specific Income coefficients (represented by the 1,ss), the statutory tax
parameter coefficients (the 2s and 3s) represent average effects across all states. This
averaging serves a useful purpose. The combination of a limited number of observations per state and infrequent changes in some of the tax parameters over the sample
period can produce spurious correlations. For example, it is possible that a state makes
a small change in its sales tax rate at the same time that it experiences a very large
increase in sales tax revenues, producing a large but spurious estimate of the effect of
that tax variable on revenues.
Accordingly, our approach pools observations across states while allowing a substantial amount of state-specific interaction terms. This enables us to incorporate potentially
important changes in tax policies while at the same time minimizing the risk of spurious
correlations from individual states. Underlying this pooling strategy is the assumption
that the experience of all states provides information about the relationship between
the respective tax parameters and tax revenues for any one state.
5

We note that we did attempt to include annual time dummies in our analysis, but these were jointly
insignificant.

64

National Tax Journal

D. Separate Regressions for Each Revenue Source versus Combined Regression

An advantage of estimating separate regressions for each revenue source is that it
allows one to identify the relative contributions to the overall MTR of the different tax
types.6 Another issue is whether one should include all statutory tax parameters in each
of the revenue source equations. For example, if corporate income taxes increase, some
businesses may choose to re-organize themselves as partnerships or Subchapter S corporations. This would shift business income from the corporate to the personal income
tax base. Indeed, such shifting is reported to have occurred as a result of the Tax Reform
Act of 1986 (Saez, Slemrod, and Giertz, 2009). However, including a large number of
statutory tax variables in each of the individual revenue source equations creates the risk
of estimating spurious correlations. This raises challenging variable selection issues.7 The
results we report below choose to ignore these fiscal externalities, but this can easily be
addressed in future work by including additional tax variables in the respective equations.8
Estimating separate regressions for each revenue source produces the following
revenue equation for tax type i, state s, and fiscal year t:
(7)

Tax
a Revenues
essti = 0i

Ki

i
0 ,ss

+ 2i ,k X ki ,sstt
k =1
1

Ds + 1i

Ri

r =1

i
3,r

(X

i
1,s

i
r,st

Ds

IIncomestt

IIncomest + esti ,

where i = PIT, CIT, ST, PT, and OT, identifying the respective tax type. The corresponding MTR for tax type i is given by:
(8)

MTR
T sti = 1i

i
1,s

Ds + 3i ,r X ri ,st .
r =1

The overall MTR for state s at time t is obtained by summing across all five tax types:
(9)

MTR
T st = M
MTR
TRstPITI + MTR
T stCITI + MTR
MTRstST + MTR
T stPT

MTRstOT .

E. Nonstationarity
Another concern is that revenues and income may be nonstationary, which could
cause conventional regression specifications to measure spurious relationships.9 The
6

An additional benefit is that it allows the individual MTR components to be included as separate explanatory variables in economic growth equations (Yamarik, 2000; Gemmell, Kneller, and Sanz, 2008).
It is a well-known principle in forecasting that the inclusion of too many variables can result in poor
forecasts. Model selection criteria such as adjusted R-squared, Akaike Information Criterion (AIC), and
Schwarz Information Criterion (SIC) address this problem by penalizing the inclusion of additional variables via a penalty function (Reed, 2009).
For example, corporate income tax parameters could be included in the equation for personal income tax
revenues. Likewise, personal income tax parameters could be included in the corporate income tax equation.
We thank William Gentry for emphasizing the implication of nonstationarity for our estimates.

On Estimating Marginal Tax Rates for U.S. States

65

fact that tax liabilities are generated as a function of income and/or consumption for
example, via personal income tax schedules provides a strong theoretical argument
against spurious regression. Without exception, previous studies estimating K&K-type
MTR regressions have used levels of revenues and income in their regression specifications. However, if one or more of the variables are nonstationary, then the appropriate
specification is an error correction model (ECM). Note that an ECM model may be
appropriate even if the variables are stationary. If the variables are cointegrated, then
(7) represents a long-run (LR) equilibrium relationship. The state-year MTR estimates
are then interpreted as the LR equilibrium MTRs that would exist if the associated tax
parameters stayed constant at their respective state-year values. We find strong evidence
in support of the ECM and thus interpret our MTR estimates as LR MTRs.10,11
F. Choice of an Income Distribution for Estimating Personal Income Tax Parameters
In order to calculate average marginal tax rates, the NBER TAXSIM model uses
specific income distributions. The model allows three options. One option constructs
separate income distributions for each state and year. A second option constructs separate
income distributions for each state (using 1995 as the benchmark year), but holds this
distribution constant over time. A third option constructs a nationally representative
income distribution and uses this for all states and years. The choice of which income
distribution to use depends on the ultimate goal of the research.
A major concern in studies of taxes and economic growth is that taxes are endogenous.
For example, if MTRs are increasing in income, then this will produce a bias towards
finding a positive correlation between MTRs and economic growth. Therefore, if the
ultimate goal is to estimate the relationship between state taxes and economic activity, one should use the third option described above, because the associated income
distribution is largely independent of any one states tax policy. Feenberg and Rosen
(1987) adopt this approach in their study of tax structure and public sector growth, and
our analysis below does the same.
IV. DATA AND VARIABLES
Our data include all 50 states over the years 19772004, excluding 2001 and 2003 since
state and local fiscal data are not available for these years.12 This limits our sample to a
maximum of 1300 total observations. Table 1 reports shares of total tax revenues by type
of tax over the sample period. Property taxes comprise the largest share of total state and
10
11

12

The corresponding empirical analysis is available from the authors.

We note that recent papers on the income elasticity of taxes are mixed with respect to whether they address
nonstationarity. For example, Gupta, et al. (2009) and Garrett and Coughlin (2009) ignore nonstationarity,
while Nichols and Tosun (2008) address it. Sobel and Holcombe (1996) report little difference in estimates
of the LR equilibrium relationship between revenues and income between these two approaches.
The tax and revenue data come from US Census Government Finances Historical Data. These are available
for state and local governments in 1972, and then from 1977 onward.

66

National Tax Journal

local tax revenues (30.3 percent), followed by sales taxes (23.5 percent), other taxes (23.3
percent), personal income taxes (18.5 percent), and corporate income taxes (4.3 percent).
Some of the states do not collect revenues from all sources; this is reflected as minimum values of zero in Table 1. Our estimation procedure omits states that do not use a
particular revenue source because these states do not contribute information about the
relationship between revenues, income, and tax parameters. For example, the seven states
that had no PIT revenue during this period (Alaska, Florida, Nevada, South Dakota,
Texas, Washington, and Wyoming) were excluded when estimating the PIT revenue
equation.13 A similar procedure was followed in estimating the CIT and ST regressions.14
For state income we use personal income data based on fiscal year as opposed to
calendar year to appropriately match income with tax revenue data. Most studies
use annual data on state personal income (or GSP) in revenue-income regressions. This
causes a mismatch between the left-hand-side variables (tax revenues are calculated
over fiscal years) and right-hand-side variables (personal income is based on calendar
years). We use quarterly Bureau of Economic Analysis data to construct an annual
personal income variable that matches the states fiscal year.
We made exhaustive efforts to assemble data on as many statutory tax parameters as
possible for each tax revenue component. Our efforts were hindered by the diversity
in the statutory tax laws, the need to collect historical data, and the lack of centralized
collection location for the various types of taxes and tax structures. Nevertheless, we did
succeed in collecting data on most of the pertinent tax parameters for each type of tax
for each state and year in our analysis. These are listed and briefly described in Table 2.
Table 1
State and Local Tax Revenue Shares by Type, 19772004
Revenue Type
Personal
Income Tax

Corporate
Income Tax

Sales Tax

Property
Tax

Other

Mean (%)
Maximum (%)

18.5
43.5

4.3
34.81

23.5
49.3

30.3
70.0

23.3
69.7

Minimum (%)

11.1

9.9

Notes: There are only 1,300 state-year observations because state and local fiscal data are not reported
for 2001 and 2003.
(1) Alaska has a much larger corporate income tax share than other states over this time period (average
share is 15.1 percent). This maximum value is from 1981.
Source: The tax and revenue data come from U.S. Census Government Finances Historical Data.

13

14

Alaska had a personal income tax for the first two years of the sample (1977 and 1978). We chose to omit
these observations in the estimation of the PIT equation.
Nevada, Texas, Washington, and Wyoming do not have corporate income taxes. Alaska, Delaware, Montana, New Hampshire, and Oregon do not have sales taxes. Prior to 2008, Texas had a Franchise Tax,
the revenues from which were mainly drawn from a formula based on the federal CIT (Texas Comptroller,
2002). We thank an anonymous reviewer for this clarification.

MAX_CRATE
NUMBER_CBRACKETS

II. Corporate Income Tax (CIT)

MTR_PENSION
MAX_WAGES
MAX_CAPGAINS

MTR_MORTGAGE

I. Personal Income Tax (PIT)

MTR_WAGES
MTR_INTEREST
MTR_DIVIDENDS
MTR_CAPGAINS

Variable

Maximum statutory tax rate on corporate profits

Number of corporate income tax brackets

Average Marginal Tax Rate on mortgage interest paid

(subsidies shown as negative tax rates)
Average Marginal Tax Rate on pension income
Maximum statutory tax rate on wage income
Maximum statutory tax rate on long-term capital
gains income

Average Marginal Tax Rate on wage income

Average Marginal Tax Rate on interest income
Average Marginal Tax Rate on dividend income
Average Marginal Tax Rate on capital gains income

Description

Statutory Tax Variables

Table 2

Compiled by the Federation of Tax Administrators from various sources. Authors

reconciled data with OTPR-University of
Michigan, World Tax Database and Council
of State Governments Book of States
series.

NBER TAXSIM (AMTR-N95) Model

Source

On Estimating Marginal Tax Rates for U.S. States

67

EXTENT_MACHINERY2

EXTENT_MACHINERY1

III. Sales Tax (ST)

RATE_SALES
RATE_FOOD
EXTENT_SERVICESTAXED

Variable

Dummy variable indicating partial taxation of

machinery
Dummy variable indicating full taxation of
machinery

State-level sales tax rate

State-level tax rate on food
Number of service categories taxed (out of 164)

Description

Statutory Tax Variables

Table 2 (continued)

(19771981 and 19942004) Taxation of

machinery compiled from state statutes, Tax
Foundation-Tax Review (various years), and
Council of State Governments, Book of the
States

(19821993) Taxation of machinery compiled by Due and Mikesell (1983,1994);

(19922004) Taxed services compiled by

the Federation of Tax Administrators from
various sources;

(19771991) Authors search in state statutes

of changes in status of service taxation;

(20032004) Compiled by the Federation of

Tax Administrators from various sources;

(19772002) OTPR- University of

Michigan, World Tax Database;

Source

68
National Tax Journal

Property tax rate per assessed value Assessment

ratio (largest city in state)
Dummy variable taking the value 1 if there is a limit
on the growth rate of assessed value
Dummy variable taking the value 1 if there is
are specific property tax limits
Dummy variable taking the value 1 if there is a
limit on the growth of property taxes
(19962004) Authors search in state
statutes of changes in status of property
taxation

(19962004) Nathan Anderson, University

of Illinois-Chicago;

(19771995) ACIR Report, Tax and

Expenditure Limits on Local Governments;

(19772004) DC Office of Revenue

Analysis1;

Notes: (1) The data were collected from the DC Office of the Chief Financial Officer. Documentation of the methodology is available at http://cof.dc.gov/.
The residential property tax rates were collected from local assessors and state equalization boards. Prior to 1981, residential property tax rates were collected
from the 30 largest cities in the United States. Data for 1983 and 1989 were not available.

V. Other Tax (OT)

No variables

LIMIT_REVGROWTH

LIMIT_SPECIFICRATE

LIMIT_ASSESSGROWTH

RATE_PROPERTY

IV. Property Tax (PT)

On Estimating Marginal Tax Rates for U.S. States

69

70

National Tax Journal

As discussed above, the statutory tax parameters associated with the personal
income tax (PIT) are derived from the NBER TAXSIM model (Feenberg and Coutts,
1993) and consist of two sets of variables: (1) MTRs averaged over the distribution of
taxpayers (MTR_WAGES, MTR_INTEREST, MTR_DIVIDENDS, MTR_CAPGAINS,
MTR_MORTGAGE, and MTR_PENSION), and (2) maximum tax rates (MAX_WAGES,
and MAX_CAPGAINS).
Our data include two statutory tax parameters for the corporate income tax: the maximum rate (MAX_CRATE) and the number of tax brackets (NUMBER_CBRACKETS).
Assuming a positive tax rate elasticity, we expect that the number of tax brackets will
be negatively related to CIT revenues when the maximum rate is held constant.
We collected five statutory tax parameters for the sales tax (ST). These consist of the
state general sales tax rate (RATE_SALES), the state tax rate on food (RATE_FOOD),15
and three variables that characterize the sales tax base. EXTENT_SERVICESTAXED is
a count variable that tracks the number of service good categories included in the state
sales tax base.16 EXTENT_MACHINERY1 and EXTENT_MACHINERY2 are dummy
variables that identify whether machinery is taxed broadly or narrowly, respectively.17
The statutory treatment of property taxes (PT) is characterized by four parameters.
We collected state effective property rates (RATE_PROPERTY) from a yearly survey of
the largest cities in each state conducted by the District of Columbia Office of Revenue
Analysis. The effective rate is calculated as the product of the nominal rate and the
assessment level expressed per $100. Given the infeasibility of collecting effective tax
rates for all localities, these are collected only for the largest city in each state.18 The rates
do not measure the state-wide average nor do they necessarily reflect the property tax
rate for the median household within or across states. They also do not incorporate the
variety of exemptions and credits that affect the taxable property tax base.19 Although
there are obvious, trade-offs associated with using the DC nationwide comparisons,
they offer the advantages of being available on an annual basis, and of incorporating
assessment values rather than income or population as measures of the tax base. Fur15

16

17

18

19

Seven of the states (Connecticut, Massachusetts, Minnesota, New Jersey, Pennsylvania, Rhode Island, and
Vermont) that exempt food also exempt clothing. The RATE_FOOD variable also captures the clothing
exemption.
Table 2 contains the source and information about the 164 service categories. The 164 services are organized into eight groups: utilities; personal services; business services; computer services; admissions and
amusements; professional services; fabrication; repair and installation; and other services.
The coefficient on narrow machinery taxation is expected to be positive (higher marginal tax rate than
no taxation of machinery the omitted category), but smaller than the coefficient on broad taxation of
machinery. See Merriman and Skidmore (2000) for a discussion of these variables and a more extensive
discussion of the distortionary nature of sales tax parameters.
According to Bell and Kirschner (2009), only thirteen states provide information on effective property
rates (Table 9, p. 130).
Bell and Kirschner (2009) provide a thorough discussion of the available measures of effective property
tax rates. The only other comprehensive series available is by the Minnesota Tax Association, which calculates effective rates for median priced homes and accounts for property tax relief. The MTA estimates,
however, are only available periodically (1995, 1998, 2000, 2002, 2004 and 2005).

On Estimating Marginal Tax Rates for U.S. States

71

thermore, as demonstrated by the Bell and Kirshner (2009) comparisons (Table 11, p.
133) the DC estimates are very close or identical for five states and reasonably close
for the other three states that provided estimates of effective tax rates.
We also include variables indicating whether a state limits growth in assessment rates
(LIMIT_ASSESSGROWTH), limits property tax rates by specific entities such as school
districts (LIMIT_SPECIFICRATE), and limits the growth rate of property tax liabilities
(LIMIT_REVGROWTH). LIMIT_ASSESSGROWTH and LIMIT_SPECIFICRATE are
interacted in the regression specification.20
V. DISCUSSION OF MTR RESULTS
A. Core Results for Tax Revenue Regressions
Table 3 presents a summary of our final regression equations for each revenue source,
as specified in (7).21 Given that we find strong evidence in favour of an error correction
specification, these equations should be interpreted as the LR equilibrium relationships
between the variables; and the associated MTRs should be interpreted as LR MTRs.
Variables with insignificant coefficients were dropped from the specification unless
there was a particular interest in certain kinds of tax parameters.22,23 However, as we
discuss below, the only two tax variables that appear to have a substantial impact on
overall MTRs are the average marginal tax rate on wage income (MTR_WAGES) and
the overall sales tax rate (RATE_SALES). Note that each regression equation uses a
different number of observations, because some states do not implement the particular
revenue source and because of missing data.24 For the sake of brevity, the table only
reports the estimates of the statutory tax rate-income interaction variables, since these
are the only tax variables that matter for generating the MTRs. Also included in these
regressions are state dummy and state dummy-income interaction variables, along with
statutory, non-rate tax variables.25

20

21
22

23

24

25

During periods of housing price increases, property tax rate limits alone are ineffective at constraining
property tax revenue growth. However, rate limits coupled with assessment growth limits can be very
restrictive. Accordingly, we interact the assessment growth limit variable with the rate limit variable since
both are essential for determining property tax liabilities over time. See Skidmore (1999) for a more detailed
discussion. An additional variable characterizing limitations on overall property rates had to be dropped
because of perfect multicollinearity with other included variables.
The OT regression equation is not reported since it does not include any statutory tax variables.
Variables with insignificant coefficients were retained in some cases to allow comparison of the relative
impacts on MTRs of the different variables (e.g., see discussion on impact of MTR_WAGES compared to
MTR_CAPGAINS and MTR_MORTGAGE below).
The literature on variable selection is voluminous and, unfortunately, there is no consensus on the best
way to select variables (see Reed (2009), and the references therein).
Property tax rates were not available for some of the states from 19771980, and state and local tax revenue
data were unavailable for all states for 2001 and 2003.
The set of regression results are available from the authors upon request.

72

National Tax Journal

Table 3
Summary of Regression Results for Individual Tax Revenue Equations
Variable

Coefficient

T-Stat

Prob

MTR_WAGES INCOME

0.003946

5.98

0.000

MTR_CAPGAINS INCOME

0.000145

0.47

0.638

MTR_MORTGAGE INCOME
R-squared = 0.994
Adjusted R-squared = 0.993
Total observations = 1,118

0.000271

0.64

0.521

NUMBER_CBRACKETS INCOME

0.000410

2.03

0.042

MAX_CRATE INCOME
R-squared = 0.976
Adjusted R-squared = 0.974
Total observations = 1,196

0.000800

2.19

0.029

RATE_SALES INCOME

0.003008

5.28

0.000

RATE_FOOD INCOME
R-squared = 0.997
Adjusted R-squared = 0.997
Total observations = 1,170

0.000060

0.18

0.854

0.000904

1.16

0.244

Personal Income Tax Equation

Corporate Income Tax Equation

Sales Tax Equation

Property Tax Equation

RATE_PROPERTY INCOME
R-squared = 0.992
Adjusted R-squared = 0.991
Total observations = 1,130

Notes: This table summarizes the results from estimating Equation (7) in the text. We only report the TAX
VARIABLE INCOME interaction variables because these are the only tax variables that matter for the
calculation of MTRs. All regression specifications also include state dummy, state dummy INCOME
interactions, and other control variables. Regressions are estimated using Weighted Least Squares to address heteroscedasticity, with weights set equal to the reciprocal of the square root of INCOME. Standard
errors are robust to serial correlation. The R-squared and Adjusted R-squared values are those from the
unweighted regression equations.

On Estimating Marginal Tax Rates for U.S. States

73

In assessing the reliability of these regression results, we note that the regression
equations generally produce good fits, with R2 values 97 percent or higher. Even without
the fixed effects, the overall R2 values are quite high.26 The product of the estimated
coefficient and the associated tax variable should be interpreted as the change in the
respective tax revenues resulting from a dollar increase in state personal income. For
example, if we evaluate the coefficient on MTR_WAGES INCOME at the sample
average for MTR_WAGES, a dollar increase in state personal income is estimated to
increase LR PIT revenues by approximately 1.7 cents (0.003946 4.20 = 0.017) via
taxation on wage income.
The relative sizes of the coefficients are in line with expectations: one would expect
a percentage point increase in the tax rate on wages (MTR_WAGES) to have a larger
impact on PIT revenues than a percentage point increase in the tax rate on capital gains or
a percentage point decrease in the tax subsidy for mortgage interest (MTR_CAPGAINS,
MTR_MORTGAGE). Likewise, one would expect a percentage point increase in the
general sales tax (RATE_SALES) to have a larger impact on total ST revenues than the
same increase on food (RATE_FOOD).
A potentially serious econometric issue is that the policy variables included in our
analyses are endogenously determined. For example, states experiencing rapid growth
in property tax revenues are likely to be the same states that support legislation to
restrain property taxes. Similarly, states experiencing disappointing growth in sales
tax revenues may respond by increasing rates and/or expanding the coverage of their
sales tax regimes. In addition, states with large increases in taxable revenues are also
more likely to cut tax rates (Poterba, 1994). The issue of endogeneity is difficult to
overcome. In our case we are hamstrung by a lack of good instruments to determine
whether endogeneity is present, and if so, to use in appropriate econometric methods.27
The presence of endogeneity is likely to cause our estimates in Table 3 to be conservative estimates of the effects of statutory tax variables on LR MTRs. We expect the
bias to be in the opposite direction of the true tax effects. For example, the endogeneity
bias associated with MTR_WAGES should be negative if states with increasing taxable
incomes are more likely to lower tax rates (Poterba, 1994). This would make the true
value of the MTR_WAGES coefficient larger than the estimated value reported in Table
3. This is also true for the other statutory tax rate coefficients reported in Table 3. In this
sense, our results can be thought of representing lower bound estimates of the effects
of tax policy parameters on LR MTRs: correctly-signed coefficients are biased towards
zero. Future research may be able to fashion a better solution to the endogeneity problem.
In the meantime, we believe our methodology represents a substantial improvement
on the K&K approach, which implicitly sets all statutory tax coefficients equal to zero.
26

27

R2 values for the Ordinary Least Squares models with income and the statutory tax variables, but without
fixed effects, ranged from 85 percent (for the corporate income tax equation) to 96 percent (for the personal
income tax equation).
For a good survey of some of the econometric issues, see Stock, Wright, and Yogo (2002).

74

National Tax Journal

B. State- and Year-specific MTR Estimates

We use (8) and (9) to calculate state- and year-specific MTRs.28 Table 4 summarizes
the distribution of estimates. The mean and median values of the overall MTR are 10.45
percent and 10.28 percent, respectively. The individual values range from a minimum
of 7.93 percent to a maximum of 15.23 percent. We discuss the state-by-state results
in more detail below.
Panel B of Table 4 decomposes the level of MTRs by revenue source. Overall, the
respective MTR shares are similar to their shares of overall tax revenues (Table 1).
However, these MTR shares mask much of the heterogeneity across states. For example,
MTRs for PIT and ST are as low as zero in states that do not use these revenue sources,
and as high as 5.6 percent (New York) and 5.1 percent (Washington), respectively. MTRPT
ranges from a low of 1.3 percent (Alabama) to a high of 6.2 percent (Maine), and MTROT
ranges from 1.0 percent (Massachusetts) to 5.2 percent (Delaware).
A different story emerges if we decompose changes in MTRs. Panel C of Table 4
reports mean and median values of the annual change of the respective MTR components
which have time-varying statutory components.29 Changes in MTRPIT contribute the
largest share of annual movement in overall MTRs. The average of the absolute value
of the annual change in MTRPIT is 0.051 percent which is approximately three times
larger than the next largest contributor, the sales tax. The difference is striking given
that both tax types make roughly equal contributions to the level of MTRs (Panel B).
The reason for this discrepancy is that state sales tax policies change much less
frequently than the personal income tax parameters. The median of the absolute value
of annual changes in MTRST is zero; i.e., no change (Panel C). A similar story holds for
property taxes. Thus, while all the tax types except corporate income taxes contribute
substantially to differences in the estimated MTRs across states, most of the movement
of MTRs within states over time is due to changes in PIT parameters.
C. Comparison of MTR estimates with K&K
The inclusion of statutory tax variables makes a substantial difference in LR MTR
estimates. For each state over all years, we calculate the average value of the absolute difference between our MTR estimates and those derived from the K&K procedure. For 27 (out of 50) states, this average difference is larger than the standard
deviation of estimated MTR values.30 For 12 states, it is larger than two standard
deviations.
28

29
30

While these are LR MTRs, they are year-specific in the sense that they represent the LR MTRs associated with the set of statutory tax parameters in place for that year.
Note that the change in MTROT is always zero, since it employs no time-varying, statutory tax information.
We calculated standard deviation as the standard error of a regression in which the dependent variable was
the MTR estimates and the explanatory variables consisted of state dummy variables.

On Estimating Marginal Tax Rates for U.S. States

75

Table 4
A. Histogram and Associated Summary Statistics of Estimated MTRs

B. Decomposition of Estimated MTRs by Revenue Source

Mean
(in percent)

Share of Total MTR

(in percent)

Personal Income Tax

Corporate Income Tax

2.30
0.32

21
3

Sales Tax

2.43

23

Property Tax
Other Tax

3.18
2.22

30
22

Revenue Source

C. Absolute Value of Annual Changes in Estimated MTRs by Revenue Source

Revenue Source

|Personal Income Tax|

|Corporate Income Tax|
|Sales Tax|
|Property Tax|
|Other Tax|

Mean
(in percent)

Median
(in percent)

0.051
0.006
0.018
0.015
0.000

0.017
0.000
0.000
0.006
0.000

76

National Tax Journal

The bold, solid line in Figure 1 reports our MTR estimates for selected states. MTR
estimates from the K&K procedure are represented by a bold, dotted line; a two-standard
deviation band is indicated around the K&K estimates with light, dotted lines. Note that
some states have missing MTR values for the years 19771980 due to the unavailability
of property tax rates for these years and states.31
The first two graphs in Figure 1 display MTRs for New Hampshire and South Dakota.
These states had the smallest average, absolute difference between the two MTR estimates. As the figures show, the two sets of estimates are virtually indistinguishable.
They represent one end of the spectrum.
Figure 1
Estimated MTRs for Selected States

31

NEW HAMPSHIRE

SOUTH DAKOTA

IOWA

WYOMING

For graphing purposes, we interpolate missing data for the years 1982, 2001, and 2003.

On Estimating Marginal Tax Rates for U.S. States

77

Figure 1 (continued)
Estimated MTRs for Selected States
MAINE

TENNESSEE

CONNECTICUT

ILLINOIS

KENTUCKY

NORTH CAROLINA

78

National Tax Journal

At the other end of the spectrum are Iowa and Wyoming. In the case, of Iowa, almost
all of our MTR estimates lie two standard deviations below the MTR (K&K) estimates.
Wyoming represents the extreme case. The average value of our MTR estimate for
Wyoming is over seven standard deviations below the K&K estimate.
The comparison between K&Ks and our estimates of Wyomings MTRs is illuminating. K&K obtain a high MTR for Wyoming because their MTR is driven by the overall
high correlation between total tax revenues and income. We suspect that oil prices play
an important role here (when oil prices increase, the economy booms and extractive
tax revenues increase). We obtain a much lower MTR than K&K because our approach
does not allow income to affect the MTR on personal or corporate income, as Wyoming
does not have a personal income or corporate income tax. We believe this highlights
one of the advantages of our approach: in the case of the oil industries, the increased
revenues derived from oil production does not reflect a corresponding increase in the
tax rate, but rather higher prices in the national market.32 Therefore, it would be misleading to infer a high MTR on income for Wyoming taxpayers just because income
and revenues are highly correlated.
To better appreciate the impact in the middle of the spectrum, we sort states by average
absolute difference between MTR and MTR (K&K) estimates. The associated median
states are Maine and Tennessee. In the case of Maine, the MTR estimates generally lie
above the K&K estimate. In the case of Tennessee, they generally lie below it.
Much of the within-state movement in MTRs is driven by changes in two tax policy
parameters: (1) the TAXSIM-generated variable, MTR_WAGES, working through
MTRPIT, and (2) the state sales tax variable, RATE_SALES, working through MTRST.
The next four figures in Figure 1 demonstrate how changes in statutory tax parameters
map into changes in estimated MTR estimates.
For example, the sharp increase in Connecticuts MTRs in the early 1990s can be
traced to changes in its income tax code. Connecticuts income tax was broadened in
1991 to include wages, at an initial tax rate of 1.5 percent. This rate tripled the following
year to 4.5 percent. In contrast, Illinois provides an example where an abrupt change
in its MTR is related to its sales tax. In 1990, Illinois increased its sales tax rate from
5.0 percent to 6.25 percent, effective January 1, 1991.
A large increase in Kentuckys MTR from 1990 to 1991 can be attributed to simultaneous changes in both its income and sales tax rates. The deduction of federal income taxes
on state income tax returns was eliminated in 1991. At the same time, the sales tax was
increased from 5 to 6 percent in 1991. Finally, North Carolina provides an example of
a number of tax changes that combined to increase its MTR from the late 1980s to the
early 1990s. North Carolina had major tax reform in 1989. The federal income tax was
adopted as a starting point to calculate North Carolina obligations. Changes to the tax
brackets amounted to a tax increase and the rate structure was streamlined to include
just two rates: 6 percent and 7 percent. In 1991, the state faced budget difficulties and
32

We thank an anonymous referee for pointing out the complexity of tax incidence for extractive industries.
For a detailed analysis of tax incidence associated with energy resources, see Morgan and Mutti (1981).

On Estimating Marginal Tax Rates for U.S. States

79

added a third rate of 7.75 percent. In 1992, the state sales tax rate was increased from
3 to 4 percent.
D. Evaluating Proxies for Time-varying MTRs
The preceding discussion highlights an advantage of our procedure: it allows changes
in estimated MTRs to be directly linked to changes in underlying statutory tax parameters. Previous studies attempt to capture such movements using proxies. Two standard
proxies for time-varying MTRs are (1) the top, marginal, personal income tax rate, and
(2) the change in tax burden (Mullen and Williams, 1994).33
To assess how well these proxies correlate with our estimated MTR values, we calculate simple correlations for the respective pairs of variables. The associated correlations
between our MTR estimates and the top marginal personal income tax rates are 0.52 and
0.62, depending upon whether state fixed effects are partialled out or not, respectively.
In other words, approximately two-thirds to three-fourths of the variance in our estimated MTRs cannot be explained by changes in the top, marginal, personal income
tax rate.34 Whether this qualifies the latter to be an acceptable proxy is questionable.
In contrast, changes in tax burden are very poorly correlated with our estimated timevarying MTRs. The simple correlations without and with adjustments for fixed effects
are 0.05 and 0.08, respectively. Clearly, this variable should not be used as a proxy for
state-specific MTRs in a panel setting. This supports Reed and Rogers (2006) conclusion that tax burden measures do a poor job of capturing changes in state tax policy.
E. State MTR rankings
Understanding the precise nature of the theoretical and empirical links between tax
policy and economic growth continues to stimulate research efforts (Feldstein, 2002).
State policy makers pay particular attention to their tax burden rankings, which are
readily available. In contrast, accurate measures of MTRs which come closer to
reflecting the rates that affect behavioral responses to policy decisions have been
elusive. Both the absolute and relative ranks of MTRs across states are useful for gaining
such insights. Accordingly, we present state-level MTR estimates from our analysis.
Given the substantial changes in MTRs over time, we focus on the last five years of
our sample, 20002004.
Previous research has claimed a link between MTRs and economic growth (Mullen
and Williams, 1994; Becsi, 1996). For this and other reasons, knowing both the absolute
33

34

Tax burden is typically calculated as the ratio of total state and local tax revenues to income, where income
is based on the calendar year at that ends during the corresponding fiscal year (Reed and Rogers, 2006).
Mullen and Williams (1994) use a modified tax burden measure calculated as a states average tax rate
(ATR) divided by the mean ATR for the sample.
Correlations of 0.52 and 0.62 translate to R-squares of 0.251 and 0.361, respectively. Thus, between 0.749
and 0.649 of the variance is unexplained.

80

National Tax Journal

and relative rankings of MTRs across states is of interest. The five states with the largest
average MTRs over the years 20002004 are: New York (14.11 percent), Maine (14.00
percent), Hawaii (12.50 percent), New Mexico (12.46 percent), and Wisconsin (12.38
percent). The five states with the smallest average MTRs over the 20002004 period
are: Alabama (8.91 percent), South Dakota (8.79 percent), Tennessee (8.55 percent),
Wyoming (8.40 percent), and Alaska (8.01 percent).
For those familiar with such comparisons, this ranking of states by MTRs will
appear similar to rankings of state tax burdens, with two egregious exceptions. Alaska
and Wyoming have the two lowest MTRs, but generally rank highest in terms of tax
burdens. This discrepancy is attributable to severance taxes, the revenue from which is
largely independent of the states own income. When Alaska and Wyoming are omitted,
the simple correlation between average MTRs and average tax burdens across states
during 20002004 in the remaining sample is 0.79.35
V. CONCLUSION
This article develops a procedure for estimating state-specific, time-varying marginal tax rates (MTRs). It generalizes the time-invariant procedure of Koester and
Kormendi (1989) by incorporating state tax policy variables for all fifty states over the
years 19772004. We find that the inclusion of statutory tax variables has a substantial
impact on estimates of MTRs. Further, we find strong evidence that an error correction
model is the appropriate specification when relating tax revenues with its determinants,
so that the associated MTRs should be thought of as the long-run revenue impact of a
dollar increase in state income.
Our procedure produces estimated MTRs that vary widely across states and years,
ranging from a low of 7.9 percent to a high of 15.2 percent. Using data from the last five
years of our panel (20002004), we rank individual states on the basis of their MTRs.
The top five states (in descending order) are New York, Maine, Hawaii, New Mexico,
and Wisconsin. The states with the smallest MTRs are Alaska, Wyoming, Tennessee,
South Dakota, and Alabama.
To obtain MTRs, we estimate revenue regressions for each of five major tax categories.
All of the tax types other than corporate income taxes make a substantial contribution
to the level of state MTRs. In contrast, annual changes in MTRs are primarily driven
by statutory changes in personal income taxes and, to a lesser degree, by sales and
property taxes. A benefit of our procedure is that it allows one to explicitly link timeseries movement in estimated MTRs to actual changes in state tax policy parameters.
Two proxies for state MTRs that have been employed in economic growth studies
are the top marginal income tax bracket and changes in state tax burdens. We find that
the first is moderately correlated with our estimated MTRs, while the latter performs
very poorly as a proxy.
35

Mullen and Williams (1994) find a large (0.635) and highly significant simple correlation between their
measure of tax burden and a states average MTR over the 19691986 period.

On Estimating Marginal Tax Rates for U.S. States

81

There are two straightforward directions in which this research can be extended. First,
our procedure enables MTRs to be employed in panel studies of economic growth. This
should advance the study of the impact of MTRs on aggregate economic activity. Second,
there are a number of refinements, particularly regarding endogeneity and specification
issues, that could lead to more reliable estimates of MTRs.
ACKNOWLEDGEMENTS
We thank Daniel Feenberg for countless clarifications concerning the TAXSIM model
and data tables, and Ed Wyatt of the District of Columbia Office of Revenue Analysis
for providing historical property tax rate documents. Carlos Lamarche, Xin Huang,
and Les Oxley provided helpful comments regarding nonstationarity in a panel data
framework. We also thank two anonymous referees and the editor for thoughtful comments that substantially improved this paper.
REFERENCES
Becsi, Zsolt, 1996. Do State and Local Taxes Affect Relative State Growth? Federal Reserve
Bank of Atlanta Economic Review 81 (2), 1836.
Bell, Michael E., and Charlotte Kirschner, 2009. A Reconnaissance of Alternative Measures of
Effective Property Tax Rates. Public Budgeting and Finance 29 (2), 111136.
Crain, W. Mark, 2003. Volatile States: Institutions, Policy, and the Performance of American
State Economies. The University of Michigan Press, Ann Arbor, MI.
Creedy, John, and Norman Gemmell, 2004. The Income Elasticity of Tax Revenue: Estimates
for Income and Consumption Taxes in the United Kingdom. Fiscal Studies 25 (1), 5577.
Dye, Richard, 2004. State Revenue Cyclicality. National Tax Journal 57 (1), 133145.
Easterly, William, and Sergio Rebelo, 1993. Marginal Income Tax Rates and Economic Growth
in Developing Countries. European Economic Review 37 (23), 409417.
Feenberg, Daniel, and Elisabeth Coutts, 1993. An Introduction to the TAXSIM Model. Journal
of Policy Analysis and Management 12 (1), 189194.
Feenberg, Daniel, and Harvey Rosen, 1987. Tax Structure and Public Sector Growth. Journal
of Public Economics 32 (2), 185201.
Feldstein, Martin, 2002. The Transformation of Public Economics Research: 19702000.
Journal of Public Economics 86 (3), 319326.
Garrett, Thomas A., and Cletus C. Coughlin, 2009. Inter-temporal Differences in the Income
Elasticity of Demand for Lottery Tickets. National Tax Journal 62 (1), 7799.
Garrison, Charles and Feng-Yao Lee, 1992. Taxation, Aggregate Activity and Economic Growth: Further Cross-Country Evidence on Some Supply-Side Hypotheses. Economic Inquiry 30 (1), 172176.

82

National Tax Journal

Gemmell, Norman, Richard Kneller, and Ismael Sanz, 2008. Fiscal Policy Impacts on Growth in
the OECD: Are They Long-Run? Working Paper. University of Nottingham, Nottingham, UK.
Greytak, David, and Jerry Thursby, 1979. Functional Form in State Income Tax Elasticity Estimation. National Tax Journal 32 (2), 195200.
Gupta, Sanjay, Jared Moore, Jeffrey Gramlich, and Mary Ann Hofmann, 2009. Empirical Evidence on the Revenue Effects of State Corporate Income Tax Policies. National Tax Journal
62 (2), 237267.
Holcombe, Randall G., and Russell S. Sobel, 1997. Growth and Variability in State Tax Revenue:
An Anatomy of State Fiscal Crises. Greenwood Press, Westport, CT.
Koester, Reinhard B., and Roger C. Kormendi, 1989. Taxation, Aggregate Activity and Economic Growth: Cross-Country Evidence on Some Supply-Side Hypotheses. Economic Inquiry
27 (3), 367386.
Merriman, David, and Mark Skidmore, 2000. Did Distortionary Sales Taxation Contribute to
Growth of the Service Sector? National Tax Journal 53, (1), 125142.
Morgan, William E., and John H. Mutti, 1981. Shifting, Incidence, and Inter-State Exportation
of Production Taxes on Energy Resources. Land Economics 57, (3), 422-435.
Mullen, John K. and Martin Williams, 1994. Marginal Tax Rates and State Economic Growth.
Regional Science and Urban Economics 24 (6), 687705.
Myles, Gareth D., 2009. Economic Growth and the Role of Taxation: Aggregate Data. Working Paper No. 714. Organisation for Economic Co-operation and Development, Economics
Department, Paris, France.
Nichols, Mark W., and Mehmet Serkan Tosun, 2008. The Income Elasticity of Gross Casino
Revenues: Short-Run and Long-Run Estimates. National Tax Journal 61 (4, Part 1), 635652.
Padovano, Fabio, and Emma Galli, 2002. Comparing the Growth Effects of Marginal vs. Average Tax Rates and Progressivity. European Journal of Political Economy 18 (3), 529544.
Payne, James E., 2003. A Survey of the International Empirical Evidence on the Tax-Spend
Debate. Public Finance Review 31 (3), 302324.
Poterba, James M., 1994. States Responses to Fiscal Crises: The Effects of Budgetary Institutions and Politics. Journal of Political Economy 102 (4), 799821.
Poulson, Barry W., and Joules Gordon Kaplan, 2008. State Income Taxes and Economic Growth.
Cato Journal 28 (1), 5371.
Reed, W. Robert, 2009. The Determinants of U.S. State Economic Growth: A Less Extreme
Bounds Analysis. Economic Inquiry 47 (4), 685700.
Reed, W. Robert, and Cynthia L. Rogers, 2006. Tax Burden and the Mismeasurement of State
Tax Policy. Public Finance Review 34 (4), 404426.

On Estimating Marginal Tax Rates for U.S. States

83

Saez, Emmanuel, Joel Slemrod, and Seth Giertz, 2009. The Elasticity of Taxable Income with
Respect to Marginal Tax Rates: A Critical Review. NBER Working Paper No. 15012. National
Bureau of Economic Research, Cambridge, MA.
Sen, Pronab, 2002. A Note on Estimating Tax Elasticities. Working Paper No. 6/2002-6.
Perspective Planning Division, Planning Commission, Government of India, New Delhi, India.
Skidmore, Mark, 1999. Tax and Expenditure Limitations and the Fiscal Relationships between
State and Local Governments. Public Choice 99 (12), 77102.
Sobel, Russell S., and Randall G. Holcombe, 1996. Measuring the Growth and Variability of
Tax Bases over the Business Cycle. National Tax Journal 49 (4), 535552.
Stock, James H., Jonathan Wright, and Motohiro Yogo, 2002. A Survey of Weak Instruments
and Weak Identification in Generalized Method of Moments. Journal of Business and Economic
Statistics 20 (4), 518529.
Texas Comptroller, 2002. The Texas Franchise Tax on Corporations (TX 96-114), Window
on State Government, State of Texas. Downloaded on 11/26/30 from www.window.state.tx.us/
taxinfo/taxpubs/tx96_114.html
Yamarik, Steven, 2000. Can Tax Policy Help Explain State-Level Macroeconomic Growth?
Economics Letters 68 (2), 211215.

84

National Tax Journal

APPENDIX

Table A1
Summary Statistics
Variables

OBS

Mean

Minimum

Maximum

STDEV

EXTENT_MACHINERY1

1,300

0.17

0.38

EXTENT_MACHINERY2

1,300

0.25

0.43

EXTENT_SERVICESTAXED

1,300

46.1

160

39.4

LIMIT_ASSESSMENTGROWTH

1,300

0.13

0.34

LIMIT_REVGROWTH

1,300

0.45

0.50

LIMIT_SPECIFICRATE

1,300

0.56

0.50

MAX_CAPGAINS

1,300

4.21

16.37

3.03

MAX_CRATE

1,300

6.47

12.25

2.95

MAX_WAGES

1,300

5.22

19.8

3.41

MTR_CAPGAINS

1,300

3.77

14.87

2.71

MTR_DIVIDENDS

1,300

4.60

11.15

2.54

MTR_INTEREST

1,300

4.15

0.07

10.9

2.34

MTR_MORTGAGE

1,300

3.40

9.45

1.28

2.85

MTR_PENSION

1,300

3.38

0.68

8.87

2.50

MTR_WAGE

1,300

4.20

9.76

2.54

NUMBER_CBRACKETS

1,300

1.78

10

1.85

RATE_FOOD

1,300

1.55

7.00

2.08

RATE_PROPERTY

1,130

1.76

0.3

7.87

0.94

RATE_SALES

1,300

4.25

8.00

1.77

TAXES_CINCOME

1,300

462,979

6,925,916

887,027

TAXES_PINCOME

1,300

2,137,560

39,574,649

3,944,435

TAXES_PROPERTY

1,300

3,107,940

78,221

34,499,304

4,516,354

TAXES_SALES

1,300

2,429,041

34,283,279

3,624,823

TAXES_TOTAL

1,300

10,119,074

389,039

133,893,624

14,540,798

INCOME

1,300

94,966,663

2,892,750

1,223,301,000

129,605,026