PUBLIC

Holcombe,
10.1177/1091142104264303
FINANCE Lacombe
ARTICLE REVIEW/ EFFECT OF STATE INCOME TAXATION

THE EFFECT OF STATE INCOME TAXATION

ON PER CAPITA INCOME GROWTH

RANDALL G. HOLCOMBE
Florida State University

DONALD J. LACOMBE
Ohio University

This study examines the impact of changes in marginal state income tax rates on per ca-
pita income by comparing income growth in counties on state borders with income
growth in adjacent counties across the state border. Compared to a standard cross-sec-
tional analysis, this border-matching technique is a better way to hold constant many fac-
tors that can vary for geographical reasons, such as climate, culture, and proximity to
markets. The results show that over the 30-year period from 1960 to 1990, states that
raised their income tax rates more than their neighbors had slower income growth and,
on average, a 3.4% reduction in per capita income.

Keywords: state income taxation; per capita income; income growth; state tax poli-
cies; state borders

As state governments grew during the 1960s and 1970s, many

states instituted income taxation for the first time. In 1960, state and
local government expenditures were 10.1% of gross domestic product
(GDP), and by 1976, they had grown to 14.5% of GDP. During that pe-
riod, 10 states adopted personal income taxes. Since then, only 1 state
has adopted state income taxation, but it has been considered in other
states and remains a controversial issue in states that do not have an in-
come tax.1 After Connecticut adopted its personal income tax in 1991
despite much dissention, Texas passed a constitutional amendment
prohibiting personal income taxation. Florida also has a constitutional
prohibition, and in Tennessee, which does not, the legislature has reg-
AUTHORS’NOTE: The authors gratefully acknowledge helpful comments from Bruce Benson,
Thomas Zuehlke, and an anonymous reviewer for this journal. Partial funding for this study came
from the DeVoe Moore Center at Florida State University.
PUBLIC FINANCE REVIEW, Vol. 32 No. 3, May 2004 292-312
DOI: 10.1177/1091142104264303
© 2004 Sage Publications
292
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 293

ularly considered adopting income taxation amid much controversy.

Many arguments can be given on both sides of the issue, and one argu-
ment often given by opponents of state income taxation is that it has an
adverse impact on state income growth. This article presents empiri-
cal evidence to suggest that state income taxes do, indeed, slow the
growth of state income.
The conclusion that state income taxes negatively affect income
growth closely conforms with tax theory. Any tax creates a disincen-
tive to the taxed activity, so an income tax clearly creates a disincen-
tive to earning taxable income. However, given state spending levels,
one might argue that the excess burden of other tax bases is equally
large and that a broader tax base that includes income taxation may
have a lower total excess burden than a narrow tax base that excludes
income taxation. Furthermore, because state taxes are small compared
to federal taxes, and because federal policy creates so much unifor-
mity among states, state tax policies by themselves may not have any
measurable effect on state economic performance. On the other side
of the argument, because any state income tax is added to federal
taxes, the marginal impact of state income taxes may be greater
(Browning 1976), and when two governments tax the same tax base,
the combined tax rate tends to be inefficiently high (Sobel 1997). De-
spite the clear theoretical argument that income taxation should have a
negative impact on income, there is good reason to investigate the
magnitude of the effect.
The next section discusses some of the empirical literature on the
subject. This study adds to that empirical literature by examining the
impact of state income taxes on counties along state borders. There
may be many reasons why income varies from state to state, including
reasons that are related to geography. For example, since the wide-
spread use of air conditioning, one might expect incomes to have risen
faster in southern states such as Texas and Florida, which do not have
income taxes. To control for geographically related effects, this study
employs a border county technique that compares counties on state
borders directly with the counties they are adjacent to across the state
border. Thus, for example, counties in Texas are paired with and com-
pared to adjacent counties in Oklahoma and Louisiana to look for in-
come differences. This technique is a way of holding constant factors
294 PUBLIC FINANCE REVIEW

that may vary from one geographic region to another without using
dummy variables, variables for climate, distance to markets, culture,
or other factors that can affect cross-sectional studies in which obser-
vations may have geographically related differences. This border
county technique provides evidence that state income taxes have a
negative effect on state income growth.

THE LITERATURE ON THE

EFFECT OF INCOME TAXES
Most previous studies on the effect of state taxes on income and
growth have examined the effect of tax levels in general, rather than
specifically looking at income taxes, and most but not all of the studies
suggest that taxes have a negative effect on various measures of eco-
nomic performance.2 Although the typical study in this literature
looks at average overall tax burdens, several studies do examine in-
come taxes specifically. Romans and Subrahmanyam (1979) found
that the level of income taxes did not appear to affect income growth,
but the degree of progression did, with more progressive tax systems
resulting in lower growth. In contrast, after controlling for other fac-
tors, Dye (1980) found that income taxes had no impact on economic
growth rates. Mullen and Williams (1994) looked at the impact of the
marginal state income tax rate and found that it has a negative impact
on state income. Besci (1996) also found a negative impact associated
with higher income tax rates. Dye and Feiock (1995) looked specifi-
cally at the impact of adopting state income taxes and found that after
controlling for other factors, the use of income taxes by itself has a
negative impact on state economic performance. As a whole, past
studies seem to find weak or no effects to average tax levels on income
but find that higher marginal income tax rates negatively affect in-
come. This is consistent with the idea that economic effects occur
because people adjust at the margin to the prices they face.
One problem with all cross-sectional studies examining the effect
of taxes on state economic performance is that it may be difficult to
control for geographically related differences among states. Regional
dummy variables or variables reflecting differences in climate (aver-
age temperature, rainfall, etc.) are sometimes used, but only imper-
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 295

fectly, to capture geographic differences. The approach used in this

study compares counties on state borders with the counties they bor-
der in adjacent states. A similar methodology was used by Holmes
(1998) to examine the effect of right-to-work laws on manufacturing
employment, but this study advances on Holmes’s methodology by
directly comparing each border county only with its neighbors.
Other researchers have used various matching ideas to determine
the effects of state policy. For example, Card and Krueger (1994) em-
pirically investigated the impact of minimum wage laws by compar-
ing its effects on similar restaurants in New Jersey and eastern Penn-
sylvania. Fox (1986, 387) found that increases in the state and local
sales tax rate reduced the level of retail activity in two of the three met-
ropolitan areas he studied. Isserman and Rephann (1995) compared
counties within Appalachia with a “twin” county outside Appalachia
and determined that between 1969 and 1991, the counties of Appala-
chia grew faster than their control group “twins.” Bronars and Lott
(1998) examined the effect of concealed-weapons laws and con-
cluded that allowing people to carry concealed handguns deters crimi-
nals. Although those studies typically look at one or a few local areas
and compare them with areas that are otherwise similar (Bronars and
Lott’s study is an exception), this study looks at every border county in
the contiguous 48 states and compares them directly with the county
or counties they touch in adjacent states. This minimizes any differ-
ences caused by geography and allows a much cleaner comparison of
differences in state policies. Thus, this study not only contributes in-
formation about the impact of state income taxes on income but also
demonstrates a methodology that can be applied to other state policy
issues.

THE MATCHING TECHNIQUE

The rationale for comparing border counties in one state with adja-
cent counties across the state border is that the technique holds con-
stant many geographic factors that could affect income. Because the
counties are physically adjacent, they should share the same climate
and culture, be similar distances to major markets, and be similar in
other ways that may vary geographically. Plaut and Pluta (1983) note
296 PUBLIC FINANCE REVIEW

that climate and environmental factors may play a role in state indus-
trial growth by affecting the cost of doing business or by attracting a
more productive labor force. Comparing adjacent counties across
state borders adjusts for factors that sometimes are accounted for by
regional dummies, variables for average temperatures, distances from
markets, and so forth. Because the matching technique controls for
these types of differences, any differences between adjacent counties
in different states are more likely to be the result of state government
policies. Holmes (1998, 668), in his study of the effects of right-to-
work laws on manufacturing activity, makes a similar argument con-
cerning the effects of state characteristics unrelated to policy by not-
ing, “If state policies are an important determinant of the location of
manufacturing, one should find an abrupt change in manufacturing
activity when one crosses a border at which policy changes, because
state characteristics unrelated to policy are the same on both sides of
the border.”
This study makes a more direct comparison of border counties than
Holmes (1998) because observations for a county are expressed as a
fraction of that same variable in the adjacent county or counties across
the state border. For example, to calculate the income growth variable
for a county, Mi, using this matching technique, that county’s growth
in income is calculated; then the county’s income growth is used as the
numerator in a fraction, and the denominator is the average for that
same variable in adjacent counties in the other state, following the
formula
countyi
Mi = n
,
1
n
∑ county j
j =1

where countyi is the value of per capita income growth on one side of
the policy border, and countyj is the value of per capita income growth
in the contiguous county or counties on the other side of the policy
border.3 Observations calculated in this way will be referred to as
matched values. The formula calculates the matched value for per ca-
pita income growth as the per capita income growth in a county, ex-
pressed as a percentage of the per capita income growth in the adjacent
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 297

county or counties in the bordering state. Matched income growth,

calculated this way, is a county’s income growth, expressed as a frac-
tion of the income growth in the adjacent county or counties across the
state border.
To get a better idea of how the matching process works, consider
the calculation of Mi for Columbus County, North Carolina, which
borders Horry County, South Carolina. Columbus County’s per capita
income grew by 965.0% from 1960 to 1990, compared to a 1,180.7%
growth in income for neighboring Horry County.4 The matched value
for Columbus County is 965/1180.7 = .8174. Horry County borders
not only Columbus County, North Carolina, but also Brunswick
County, North Carolina, which had a per capita income growth of
1,381.3%. The matched value for income growth for Horry County is
1180.7/((1/2)(965 + 1381.3) = 1.0064. Brunswick County borders
only Horry County, so its matched value for income growth is 1381.3/
1180.7 = 1.1700.
One can see from the formula that the values of the matched de-
pendent variables will be close to 1 and will be greater than 1 for coun-
ties that grew more than their neighbors across the state border and
less than 1 for counties that grew less. Brunswick County, which had
greater per capita income growth than the border county across the
state line, has a matched value for per capita income growth of 1.17.
Horry County grew about as fast as the average of the two counties it
borders, so it has a matched value of per capita income growth of
1.0064—almost exactly 1. Columbus County grew more slowly than
its neighboring county across the state border and has a value of .8174.
All counties have a mean for matched income growth of 1.023, with a
standard deviation of 0.207.
Two different measures of income growth from 1960 to 1990 are
used as dependent variables in the regression models that follow. In-
come growth is measured as the percentage growth in income from
1960 to 1990 in some regressions and as the dollar growth in income
in others. Results are similar, regardless of which measure of income
growth is used. In total, there are 1,129 counties on state borders in the
United States. Some counties border two states (e.g., Houston County,
Alabama, which borders Florida and Georgia), and those counties are
298 PUBLIC FINANCE REVIEW

TABLE 1: Descriptive Statistics for Data Used in Regressions

Standard
Mean Maximum Minimum Deviation

Matched variables (used in Table 2)

Per capita income growth
1960 to 1990: County match 1.023 2.254 0.414 0.207
Per capita income difference
1960 to 1990: County match 1.020 3.438 0.317 0.209
Highest state marginal tax rate
difference –0.050 13.250 –13.250 3.878
Average tax rate difference –3.67E-05 0.039 –0.039 0.014
State and local per capita
government expenditures –24.358 2121.725 –2121.725 606.231
Fantus ranking –0.320 37 –37 13.994
Manufacturing employment (%) 0.088 13.400 –13.400 6.048
Median age –0.018 7 –7 1.792
Mineral production value (millions) –11.765 2,570 –2,570 703
Population density 0.474 776.900 –776.900 152.364
Urban population –0.578 52.100 –52.100 16.269
Per capita income 1960 1.017 2.476 0.407 0.231
Unmatched variables (used in Table 3)
Per capita income growth
1960 to 1990 747.594 2045.024 315.864 197.849
Per capita income difference
1960 to 1990 9679.118 23367.200 2750.333 2354.216
Highest state marginal tax rate
difference 1.156 8.500 –7.350 2.758
Average tax rate difference 0.015 0.033 –0.020 0.012
State and local per capita
government expenditures 3043.260 4999.055 2276.478 498.346
Fantus ranking 20.669 48 1 13.016
Manufacturing employment (%) 17.141 27.300 4 5.481
Median age 32.732 36.200 26.200 1.341
Mineral production value 0.012 0.123 0.000746 0.020
Population density 123.350 1,042 4.700 161.823
Urban population 65.870 92.600 32.200 12.819
Per capita income 1960 1369.201 3252.601 411.184 423.321

included a second time in the data set, producing a sample size of

1,319 border matches.
Table 1 shows some descriptive statistics for the variables used in
the regression models presented below and illustrates how the match-
ing technique affects the variables. The top of the table shows descrip-
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 299

tive statistics for the matched variables, and the bottom shows descrip-
tive statistics for the unmatched variables, which are the actual
magnitude of the variables, with no matching. The first matched vari-
able, per capita income growth, has a mean of 1.023, showing that per
capita income growth in all border counties is, on average, about equal
to the per capita income growth of the counties they adjoin just across
the state border. The maximum of 2.254 shows that the county that
grew fastest compared to its neighbor across the state border had
225.4% higher per capita income growth, whereas the county that per-
formed most poorly compared to its neighbor across the state border
had per capita income growth only 41.4% as fast. Further down the ta-
ble, under “Unmatched Variables,” per capita income growth from
1960 to 1990 averaged $747.59 for all counties, and the county with
the highest per capita income growth had an increase of $2,045.02,
compared to $315.86 for the county with the lowest per capita income
growth.
Some of the variables are calculated as the differences across bor-
der counties, and in these cases, the means will be about zero. Con-
sider, for examples, the highest marginal tax rate difference and popu-
lation density. Colorado had a top marginal income tax bracket of 9%
in 1960, which was reduced to 1.65% by 1990, for a change of 7.35.
Nebraska, which borders Colorado, had no income tax in 1960 and a
top bracket of 5.9% in 1990, for a change of –5.9. The difference be-
tween Colorado’s change in marginal tax rate and Nebraska’s was
7.35 – (–5.9) = 13.25, so counties on the Colorado-Nebraska border
had the highest (Colorado) and lowest (Nebraska) values for these
matched variables. The mean is close to zero because positive values
on one side of a state border are offset by negative values on the other.
Population density is calculated as the population density relative to
the population density of adjacent counties across the border, so its
mean also is approximately zero, and its minimum is the negative of
its maximum. Note that although population density has a mean of
close to zero in the matched data, in the unmatched data, its mean is
about 123 people per square mile. The main results that follow use the
matched data, but the results are repeated with the unmatched data to
show what difference it makes to use the matching technique.
300 PUBLIC FINANCE REVIEW

RESULTS FROM THE MATCHED SAMPLE

Table 2 shows regression results using as the dependent variable the

two different measures of matched per capita income growth from 1960
to 1990. The comparison was done over a period of three decades for
several reasons. First, the impact of income taxes on income growth is
likely to be a long-run phenomenon that may be overwhelmed by
other factors in the short run. Second, a comparison over several de-
cades can eliminate some other effects on state income that might be
unrelated to state policy but that might affect states differently. Over
this time period, for example, disruptions in energy markets during
the 1970s had a negative impact on income in many states but helped
states that produced a substantial amount of coal and oil. The effect
went the other way in the 1980s when oil prices fell, but effects such as
these on state incomes should be minimized by looking at a three-de-
cade time period. The first and third regressions use as the dependent
variable the percentage growth in nominal per capita income from
1960 to 1990, matched using the above formula with the county or
counties in the neighboring state that county borders. In other words,
representing a county’s per capita income in year i as YPCi, the first
and third regressions measure income growth as (YPC90 – YPC60)/
YPC60 and use this as the county observation that is matched with ad-
jacent counties in the above formula. The second and fourth regres-
sions use the county’s nominal per capita income in 1990 minus its
nominal per capita income in 1960 (YPC90 – YPC60), again matched
with its bordering counties, as a measure of the change in the level of
per capita income over the period. These per capita income growth
measures matched following the above matching formula to calculate
the dependent variable.
The first independent variable is the change in the difference be-
tween neighboring states’highest marginal income tax rate from 1960
to 1990, and this variable is also matched against border counties in
the neighboring state.5 For example, in 1960, the highest marginal in-
come tax rate was 7% in California and 4.5% in Arizona, so on the
California-Arizona border, the difference in the highest marginal in-
come tax rates was 2.5% in 1960. In 1990, California’s highest mar-
ginal income tax rate was 9.3% and Arizona’s was 8%, so the differ-
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 301

TABLE 2: Effect of Tax Rates on Various Measures of Per Capita Income Using
Full Sample of Matched Border Counties (t statistics in parentheses)

Dependent Variable
Per Capita Per Capita Per Capita Per Capita
Income Income Income Income
Growth, Difference, Growth, Difference,
1960 to 1990: 1960 to 1990: 1960 to 1990: 1960 to 1990:
Independent Variable County Match County Match County Match County Match

Constant 1.526770 0.550400 1.527856 0.551386

(70.67) (25.27)*** (70.80)*** (25.34)***
Highest state marginal
tax rate difference –0.003768 –0.003607 –0.002935 –0.002851
(–2.96)*** (–2.81)*** (–2.21)** (–2.13)**
Average tax rate
difference NA NA –0.862868 –0.783500
(–2.18)** (–1.97)**
State and local per
capita government
expenditures 2.90E-05 3.09E-05 3.33E-05 3.49E-05
(2.87)*** (3.04)*** (3.25)*** (3.37)***
Fantus ranking –0.002433 –0.002394 –0.002308 –0.002281
(–6.50)*** (–6.34)*** (–6.10)*** (–5.98)***
Manufacturing
employment (%) –0.000536 –0.000619 –0.000530 –0.000614
(–0.63) (–0.72) (–0.62) (–0.71)
Median age 0.017086 0.019767 0.016824 0.019529
(6.16)*** (7.08)*** (6.07)*** (6.99)***
Mineral production
value 2.13E-11 1.70E-11 2.58E-11 2.11E-11
(2.67)*** (2.11)** (3.13)*** (2.54)**
Population density 7.87E-05 7.20E-05 8.91E-05 8.14E-05
(1.86)* (1.69)* (2.10)** (1.90)*
Urban population –0.000284 –2.68E-05 –0.000520 –0.000241
(–0.60) (–0.06) (–1.08) (–0.50)
Per capita income
1960 –0.495729 0.462042 –0.496737 0.461126
(–23.94)*** (22.14)*** (–24.02)*** (22.11)***
Number of
observations 1,319 1,319 1,319 1,319
F statistic 69.85*** 71.60*** 63.52*** 64.97***
2
Adjusted R 0.319781 0.325270 0.321736 0.326746
NOTE: NA = not applicable.
*Significant at the 10% level. **Significant at the 5% level. ***Significant at the 1% level.
302 PUBLIC FINANCE REVIEW

ence in 1990 was 1.3%. Thus, the gap between California’s and
Arizona’s highest marginal income tax rate narrowed by 1.2%. In the
data for Table 2, Arizona counties that border California would have a
value of 1.2 for the highest state marginal tax rate difference variable,
indicating that relative to its bordering counties in California, Ari-
zona’s income tax rate increased by 1.2%. Conversely, California
counties bordering Arizona would have a value of –1.2 for that same
variable, indicating that their income tax rate was 1.2% lower relative
to Arizona in 1990 compared to 1960. 6
This border county analysis is intended to hold other things con-
stant, but other state-specific factors besides the income tax might also
affect per capita income in a county. One major item is state and local
government spending. The budgets of states and localities tend to be
close to balanced, so the average tax rate serves as a close proxy for
spending out of own-source revenues, but federal revenues to states
can vary considerably. For this reason, state and local per capita ex-
penditures is also included as a dependent variable.7 It is significant
and positive in every specification, indicating that more government
expenditures increase per capita income growth, holding other things
(including taxes) constant.
The Fantus ranking is a ranking of business climate and was used
by Holmes (1998) to adjust for state policies that may favor business.8
It too is significant in every specification. Lower Fantus numbers indi-
cate a more favorable business climate, so the negative sign indicates
that the more favorable the business climate according to this ranking,
the higher will be per capita income growth. Plaut and Pluta (1983)
also use a business climate variable based in part on the Fantus rank-
ing and find that a poor business climate negatively affects capital
stock growth. Manufacturing employment is the percentage of the
state workforce employed in manufacturing, and it is included to cap-
ture any effects that might result from a county being located in a state
with a larger manufacturing base but is not significant.9 Median age of
the state’s population could affect the county if, for example, an older
population created a more productive workforce, resulting in higher
income growth.10 The consistently positive sign and statistical signifi-
cance show that a higher median state population age is associated
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 303

with higher per capita income growth in border counties in that state.
Mineral production is the share of state personal income derived from
mineral production.11 Niskanen (1992) suggests that states that enjoy
natural resource endowments may have larger incomes, and the posi-
tive and significant sign on mineral production agrees with
Niskanen’s argument. Population density may affect the cost of deliv-
ering government services in the state12 and is consistently significant
and positive. Urban population is included because more urban states
may find themselves with a larger income tax base but also with a
greater demand for state services.13 However, urban population is not
statistically significant.14
By far the most significant independent variable is per capita in-
come in 1960. When the dependent variable is growth in per capita in-
come, per capita income has a negative effect, implying convergence.
Counties that start with lower incomes in 1960 tend to converge so
have higher income growth, whereas those that start with relatively
high incomes grow slower. When the dependent variable is the differ-
ence in income between 1960 and 1990, the per capita income vari-
able has a positive sign, indicating that even though per capita income
growth is greater in counties that start with low income, the (nominal)
gap in per capita income tended to widen. The significance of the ini-
tial per capita income variable supports the convergence hypothesis
and is consistent with other studies that have examined the issue, such
as Besci (1996) and Rasmussen and Zuehlke (1990).
The first two regressions are identical, with the exception of the de-
pendent variable, which is the matched growth in per capita income in
the first regression and the matched change in the level of per capita
income in the second. In both cases, the change in the highest mar-
ginal income tax rate is negative and highly significant, showing that
the more states raised their highest marginal income tax rate com-
pared to their neighboring states, the slower was their per capita in-
come growth. The second two regressions are identical to the first two,
with the exception that the change in the state’s average tax rate is in-
cluded, and also matched against border counties in the adjacent state.
The average tax rate is calculated as per capita state and local taxes as a
percentage of state and local income, so this variable looks at the
304 PUBLIC FINANCE REVIEW

growth in average taxes, compared to that same change in the

bordering state. It also is negative and significant.
The first two independent variables show that the more states raise
their income tax rates compared to their neighbors, the slower will be
their per capita income growth, and that the more average taxes rise in
a state, the slower will be per capita income growth. Higher taxes
mean slower income growth, and larger increases in marginal income
tax rates have a negative impact on income growth, even holding con-
stant the change in average taxes. To get an idea about the magnitude
of the effect, consider this rough calculation that evaluates the vari-
ables at their means. Looking at the matched data in the third regres-
sion in Table 2, the typical pair of states has a difference in the
matched value of the highest state marginal tax rate difference vari-
able of 13.25.15 Thus, for a typical pair of states, the state that in-
creased its income tax rate less than its neighbor would reduce the de-
pendent variable by 0.039 (which is 13.25 • 0.002935) compared to its
neighbor. The mean of the dependent variable is very close to 1 (a look
at the formula showing its calculation reveals why), so the percentage
growth in per capita income would be 3.9% lower over the 30-year pe-
riod in the state with the larger income tax increase. Looking at the to-
tal percentage growth in per capita income over the period implies a
reduction in state per capita income from this income tax increase of
about $377.49, or 3.4% of per capita income.16 Thus, the impact is not
only statistically significant but economically significant as well.
The results in Table 2 provide strong evidence that increases in a
state’s highest marginal income tax rate result in a lower growth rate in
per capita income. The results are similar whether one looks at the per-
centage growth in income or the dollar amount of income growth, re-
gardless of whether the state’s average tax burden is included in the re-
gression. When the average tax burden is included, it too has a
negative and significant effect on income growth. Not surprisingly,
initial per capita income has the greatest explanatory value in the re-
gression, but even when initial income and other differences are ac-
counted for, changes in the state’s highest marginal tax rate have a sig-
nificant impact. In terms of explanatory power, all of the regression
models explain approximately 32% of the variation in the dependent
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 305

variable. When looking at these results, one must keep in mind the na-
ture of the data. This is not a simple cross-sectional regression; rather,
the dependent variable is per capita income growth in a county as a
percentage of per capita income growth in adjacent counties just
across the state border. This border county matching technique holds
constant many differences that might otherwise have to be accounted
for using independent variables, and it is likely that without this
matching technique, geographic differences will be accounted for
only imperfectly. Thus, the significant results in these regressions
provide strong evidence that raising income tax rates lowers income
growth.

RESULTS FROM THE

CROSS-SECTIONAL SAMPLE

The border county matching technique used in the previous section

is unusual enough that one might wonder what difference it makes to
use the matched variables rather than undertake a simple cross-sec-
tional analysis of border counties, as is more typically done. Table 3
shows the same specifications as appeared in Table 2 but using un-
matched variables. The dependent variables are simply the growth of
county per capita income from 1960 to 1990 and the change in the
level of income from 1960 to 1990. Similarly, the highest marginal tax
rate is the change in the rate in that county’s state, the Fantus ranking is
that state’s Fantus ranking, and so forth. In all of the specifications, the
marginal tax change variable remains statistically significant at
greater than the 1% level. The average tax variable is significant in one
specification but not in the other. All of the other independent vari-
ables are statistically significant also, in all four regressions, with the
exception of mineral production, which is never significant. However,
mineral production was significant in every specification in the
matched regressions in Table 2.
Taken as a whole, the regression specifications appearing in Table 3
explain almost 60% of the variation in the dependent variable, and the
R2s in Table 3 are nearly double those in Table 2. The only difference
in the regressions in the two tables is that Table 2 uses the matched
306 PUBLIC FINANCE REVIEW

TABLE 3: Effect of Tax Rates on Various Measures of Per Capita Income Using
the Full Sample of Nonmatched Border Counties (t statistics in paren-
theses)

Dependent Variable
Per Capita Per Capita Per Capita Per Capita
Income Income Income Income
Growth, Difference, Growth, Difference,
Independent Variable 1960 to 1990 1960 to 1990 1960 to 1990 1960 to 1990

Constant 692.7189 –572.9924 656.1157 –527.4344

(5.62)*** (–0.40) (5.28)*** (–0.36)
Highest state marginal
tax rate difference –8.650916 –85.75549 –7.272899 –87.47063
(–5.79)*** (–4.89)*** (–4.47)*** (–4.57)***
Average tax difference NA NA –844.1765 1050.700
(–2.12)** (0.22)
State and local per
capita government
expenditures 0.055225 0.646515 0.058279 0.642714
(4.84)*** (4.83)*** (5.08)*** (4.76)***
Fantus ranking –2.731153 –28.94050 –2.528858 –29.19229
(–6.97)*** (–6.29)*** (–6.28)*** (–6.16)***
Manufacturing
employment (%) 5.364541 38.44621 5.457384 38.33065
(6.25)*** (3.81)*** (6.36)*** (3.80)***
Median age 10.92146 140.2610 11.79957 139.1681
(3.25)*** (3.55)*** (3.49)*** (3.50)***
Mineral production
value 282.4712 –1098.420 323.4002 –1149.362
(1.19) (–0.39) (1.36) (–0.41)
Population density 0.392584 5.636742 0.390705 5.639080
(12.20)*** (14.92)*** (12.16)*** (14.92)***
Urban population –2.004047 –27.71954 –1.995661 –27.72998
(–5.21)*** (–6.14)*** (–5.20)*** (–6.14)***
Per capita income
1960 –0.303861 3.561092 –0.301546 3.558212
(–27.82)*** (27.78)*** (–27.52)*** (27.61)***
Number of
observations 1,129 1,129 1,129 1,129
F statistic 178.73*** 187.1308*** 161.8066*** 168.2798***
2
Adjusted R 0.586447 0.597599 0.587729 0.597258
NOTE: NA = not applicable.
**Significant at the 5% level. ***Significant at the 1% level.
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 307

variables, whereas Table 3 uses the unmatched observations. This

lends evidence to the claim that using the matching technique removes
some variation in the data caused by geographical differences because
the independent variables (which may vary due to geographically re-
lated differences in Table 3) explain more of the variation in the de-
pendent variables in the unmatched regressions than in the matched
regressions.17
To get an idea of the magnitudes in the unmatched cross-sectional
regressions in Table 3, the average income tax rate increase in all states
over the time period was 1.5%. Multiplying that by the coefficient on
the marginal tax variable yields 0.11, which implies an impact on per
capita income of $1,064.70, or 9.6% of income. In other words, if the
typical state had not increased its highest marginal income tax rate,
rather than increased it by 1.5%, its 1990 per capita income would
have been 9.6% higher. The impact appears much larger using the
cross-sectional regression without the match, but the reason for doing
the county match methodology was to hold constant factors that might
not otherwise be controlled for in the regression. Of course, these cal-
culations are not directly comparable because the variables are mea-
sured differently in the regressions in Tables 2 and 3. Still, these calcu-
lations provide some idea about the magnitude of the effects and
suggest that the matching technique for holding other factors constant
reduces the measured magnitudes of the effects.
A comparison of the matched and unmatched results shows the
value of undertaking the matching methodology. The results in both
cases point in roughly the same direction, but the magnitudes of the ef-
fects are substantially different, and we believe that the border match
methodology more accurately captures the magnitude of the effect.
The lower R2s in the matched regressions show that the matching tech-
nique does remove some correlation in the data due to geographic dif-
ferences not captured by the independent variables. In addition to the
empirical results themselves, the matched border county methodol-
ogy used in this study illustrates a technique that could be applied to
other questions. This study contributes not only insight into the im-
pact of state income taxes but also offers a methodology that can be
used to analyze other state policy issues.
308 PUBLIC FINANCE REVIEW

CONCLUSION

There is a substantial literature that examines the effect of state

taxes on state economic performance. This study adds to that literature
by using a matched border county methodology to evaluate the effects
of state income taxes on state income. When compared to a straight-
forward cross-sectional analysis, the matched border county method-
ology used here has the advantage that it makes a direct comparison of
conditions in one county with adjacent counties across the state bor-
der, so it provides a good way of controlling for factors that may vary
for geographic reasons. Thus, one can be more confident that the em-
pirical results reflect the impact of state policies rather than other
factors that can vary from one location to another.
The empirical results consistently show that states that increase
their income tax rates more than their neighbors have slower per ca-
pita income growth. Increases in average state tax burdens also have a
negative impact on per capita income growth. The border county
matching methodology holds constant factors that vary geographi-
cally, but other variables that were consistently significant in the
matched sample were state per capita income (suggesting conver-
gence), population density, the value of mineral production as a per-
centage of state income, a ranking of state business climate, and per
capita state and local government expenditures. Holding those factors
constant and using the border county match to hold geographic factors
constant, income tax increases consistently lead to lower income
growth.
State income taxes increased in importance as a source of state rev-
enue in the last half of the 20th century, both because states that did not
have an income tax adopted it and because states raised their rates.
Eleven states have adopted income taxes since 1960, and the highest
marginal income tax rate rose from an average of 4.1% in 1960 to
5.3% in 1990. There is a substantial literature suggesting that income
taxes have negative economic effects. This study adds to that literature
by showing that income tax increases lower per capita income growth.
A rough calculation suggests that the typical state that has increased
its income tax rates more than its neighbor has a per capita income that
is about $377.49, or 3.4%, lower than its neighbor across the border
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 309

that had lower income tax increases over the period from 1960 to
1990.

NOTES

1. States without personal income taxes in 2001 are Alaska, Florida, Nevada, New Hamp-
shire, South Dakota, Tennessee, Texas, Washington, and Wyoming. New Hampshire and Ten-
nessee have limited taxes on interest and dividend income only. Data on state income taxes are
from Holcombe and Sobel (1997, Tables 1.1, 3.3, and 3.5).
2. See, for examples, Plaut and Pluta (1983), Helms (1985), Benson and Johnson (1986),
Canto and Webb (1987), Rasmussen and Zuehlke (1990), Vedder (1990, 1995), Mofidi and Stone
(1990), Berry and Kaserman (1993), and Crain and Lee (1999).
3. Adjacent counties were determined by examining a map of the United States and deter-
mining which counties in adjoining states touched each other. The relationships are not necessar-
ily reciprocal because, for example, County A in State 1 might touch Counties X and Y in State 2,
but County X in State 2 might touch only County A in State 1.
4. These growth figures are in nominal terms, but because the matching process uses ratios,
adjustment to real growth would have the same impact on the numerator and denominator.
5. Marginal state income tax rates for 1960 are available in Facts and Figures on Govern-
ment Finance (Tax Foundation, Inc. 1960-1961). Marginal state income tax rates are from Facts
& Figures on Government Finance (Tax Foundation, Inc. 1991).
6. One might make an argument for creating a more complex measure of a state’s marginal
income tax rate structure, but in the typical state, most taxpayers face the highest marginal rate.
For states that have income taxes, the median income level at which taxpayers face the highest
marginal tax rate is $20,000. There is a good reason for using the marginal rate rather than some
other measure, such as average tax payment, because people adjust to the marginal prices they
face.
7. The state and local per capita government expenditure figures for 1960 are available in
the United States Statistical Abstract (U.S. Bureau of the Census 1962, 423). State and local gov-
ernment expenditure data are from Facts & Figures on Government Finance (Tax Foundation,
Inc. 1993). Population figures for 1990 are available in the United States Statistical Abstract
(U.S. Bureau of the Census 1995, 28). The state and local per capita government figures for 1990
are the author’s calculation (i.e., state and local government expenditures divided by population).
Local expenditures are included because there is a substantial variation across states in the per-
centage of state and local government expenditures provided by localities. However, funding
sources are similar, and state funds make up a substantial share of local government revenues.
8. Holmes (1998) gives the state Fantus rankings and explains the ranking system.
9. Manufacturing employment figures for 1960 are found in the United States Statistical
Abstract (U.S. Bureau of the Census 1965, 225). Manufacturing employment figures for 1990
are found in the Geographic Profile of Employment and Unemployment (U.S. Department of
Labor 1990, 66).
10. The median age figures for 1960 are available in the United States Statistical Abstract
(U.S. Bureau of the Census 1961, 29). Median age figures for 1990 are from the Population Esti-
mates Program, Population Division, U.S. Bureau of the Census, Washington, D.C. The data are
310 PUBLIC FINANCE REVIEW

also available in electronic form from http://www.census.gov/population/estimates/state/

st9820.txt.
11. Mineral production value for 1960 is from the United States Statistical Abstract (U.S.
Bureau of the Census 1965, 710), and personal income statistics for 1960 are from the United
States Statistical Abstract (U.S. Bureau of the Census 1962, 319). Mineral production values for
1990 are from the United States Statistical Abstract (U.S. Bureau of the Census 1995, 710), and
personal income figures are from the United States Statistical Abstract (U.S. Bureau of the Cen-
sus 1992, 438). Personal income figures for 1960 are available in the United States Statistical Ab-
stract (U.S. Bureau of the Census 1962, 319). Personal income figures for 1990 are available in
the United States Statistical Abstract (U.S. Bureau of the Census 1992, 43). For each year, the
mineral production value was divided by the personal income figure to obtain the values for the
independent variable.
12. Population density figures for 1960 are from the United States Statistical Abstract (U.S.
Bureau of the Census 1965, 13), and the population density figures for 1990 are from the United
States Statistical Abstract (U.S. Bureau of the Census 1992, 23).
13. Urban population figures for 1960 are available in the United States Statistical Abstract
(U.S. Bureau of the Census 1965, 16), and urban population figures for 1990 are available in the
United States Statistical Abstract (U.S. Bureau of the Census 1995, 43).
14. The unemployment rate for 1960 and 1990, as well as the percentage of the state’s popu-
lation with a high school diploma for 1960 and 1990, was also used in the fully specified models
but was never significant at standard confidence levels and hence was dropped from the regres-
sion models reported here.
15. The mean of the matched highest marginal income tax variable is approximately zero be-
cause for every county with a positive value, there will be a county in the adjacent state with an
offsetting negative value. The mean would be exactly zero if county borders in states exactly
lined up with borders in adjacent states. The range of the matched highest marginal rate variable
is –13.25 to 13.25, making the midpoint of the absolute value of the range 6.625%. A typical state
with a smaller tax increase than its neighbor will have a value of +6.625, and the neighboring
state will have a value of –6.625, for a difference between them of 13.25.
16. Nominal per capita income in 1960 averaged $1,369.20 and was $11,048.32 in 1990, for
an average increase of $9,679.12, or an increase of 707%. Taking 3.9% of this yields $377.49,
which is the amount by which income in the state with the smaller income tax increase would ex-
ceed income in the state with the larger increase. Taking that difference as a fraction of average
per capita income shows that the negative impact of the higher than average income tax increase
costs 3.4% of income. That is, at the end of the 30-year period from 1960 to 1990, per capita in-
come is 3.4% lower, or $377.49 lower, in states that had higher than average income tax
increases.
17. A referee asked whether these results might be driven by a few outliers with the most ex-
treme changes in the tax rate. To address this issue, all eight regression models were reestimated,
excluding those observations with the most extreme values in the tax variable. For the regressions
in Table 2, the tax variable was restricted to the interval (–10,10), which reduced the sample size
to 1,289 observations. The tax variable became more significant, with t statistic estimates rang-
ing in value from –2.96 to –3.68, and the other variables in the specifications retained their quali-
tative and quantitative characteristics. For the regressions in Table 3, the tax variable was re-
stricted to the interval (–5,5), which reduced the sample to 996 observations. The tax variable in
this case remained about the same in terms of statistical significance, with t values ranging from –
2.78 to –3.70. Again, the specifications retained their qualitative and quantitative characteristics.
 Holcombe, Lacombe / EFFECT OF STATE INCOME TAXATION 311

REFERENCES

Becsi, Zsolt. 1996. Do state and local taxes affect relative state growth? Federal Reserve Bank of
Atlanta Economic Review 81 (2): 18-36.
Benson, Bruce L., and Ronald N. Johnson. 1986. The lagged impact of state and local taxes on
economic activity and political behavior. Economic Inquiry 24 (3): 389-401.
Berry, Dan M., and David L. Kaserman. 1993. A diffusion model of long-run state economic de-
velopment. Atlantic Economic Journal 21 (4): 39-54.
Bronars, Stephen G., and John R. Lott Jr. 1998. Criminal deterrence, geographic spillovers, and
the right to carry concealed handguns. American Economic Review 88 (2): 475-79.
Browning, Edgar K. 1976. The marginal cost of public funds. Journal of Political Economy 84
(2): 283-98.
Canto, Victor A., and Robert I. Webb. 1987. The effect of state fiscal policy on state relative eco-
nomic performance. Southern Economic Journal 54 (1): 186-202.
Card, David, and Alan B. Krueger. 1994. Minimum wages and employment: A case study of the
fast-food industry in New Jersey and Pennsylvania. American Economic Review 84 (4): 772-
93.
Crain, W. Mark, and Katherine J. Lee. 1999. Economic growth regressions for the American
states: A sensitivity analysis. Economic Inquiry 37 (2): 242-57.
Dye, Thomas R. 1980. Taxing, spending, and economic growth in the American states. Journal
of Politics 42 (4): 1085-1107.
Dye, Thomas, and Richard C. Feiock. 1995. State income tax adoption and economic growth.
Social Science Quarterly 76 (3): 648-54.
Fox, William F. 1986. Tax structure and the location of economic activity along state borders.
National Tax Journal 39 (4): 397-401.
Helms, L. Jay. 1985. The effect of state and local taxes on economic growth: A time series-cross
section approach. Review of Economics and Statistics 67 (4): 574-82.
Holcombe, Randall G., and Russell S. Sobel. 1997. Growth and variability in state tax revenue:
An anatomy of state fiscal crises. Westport, CT: Greenwood.
Holmes, Thomas J. 1998. The effect of state policies on the location of manufacturing: Evidence
from state borders. Journal of Political Economy 106 (4): 667-705.
Isserman, Andrew, and Terance Rephann. 1995. The economic effects of the Appalachian
Regional Commission: An empirical assessment of 26 years of regional development plan-
ning. Journal of the American Planning Association 61 (3): 345-64.
Mofidi, Alaeddin, and Joe A. Stone. 1990. Do state and local taxes affect economic growth?
Review of Economics and Statistics 72 (4): 686-91.
Mullen, John K., and Martin Williams. 1994. Marginal tax rates and state economic growth.
Regional Science and Urban Economics 24 (6): 687-705.
Niskanen, William A. 1992. The case for a new fiscal constitution. Journal of Economic Perspec-
tives 6 (2): 13-24.
Plaut, Thomas R., and Joseph E. Pluta. 1983. Business climate, taxes and expenditures, and state
industrial growth in the United States. Southern Economic Journal 50 (1): 99-119.
Rasmussen, D. W., and T. W. Zuehlke. 1990. Sclerosis, convergence, and taxes: Determinants of
growth among U.S. states. Environment and Planning 8 (1): 1-11.
Romans, Thomas, and Ganti Subrahmanyam. 1979. State and local taxes, transfers and regional
economic growth. Southern Economic Journal 46 (2): 435-44.
Sobel, Russell S. 1997. Optimal taxation in a federal system of governments. Southern Economic
Journal 64 (2): 468-85.
312 PUBLIC FINANCE REVIEW

Tax Foundation, Inc. 1960-1961. Facts & figures on government finance (11th ed.). New York:
Tax Foundation, Inc.
. 1991. Facts & figures on government finance. Baltimore: Johns Hopkins University
Press.
. 1993. Facts & figures on government finance. Washington, DC: Tax Foundation, Inc.
U.S. Bureau of the Census. 1961. United States statistical abstract. Washington, DC: Govern-
ment Printing Office.
. 1962. United States statistical abstract. Washington, DC: Government Printing Office.
. 1965. United States statistical abstract. Washington, DC: Government Printing Office.
. 1992. United States statistical abstract. Washington, DC: Government Printing Office.
. 1995. United States statistical abstract. Washington, DC: Government Printing Office.
U.S. Department of Labor. 1990. Geographic profile of employment and unemployment. Wash-
ington, DC: U.S. Department of Labor.
Vedder, Richard K. 1990. Tiebout, taxes, and economic growth. Cato Journal 10 (1): 99-108.
———. 1995. State and local taxation and economic growth: Lessons for federal tax reform.
Joint Economic Committee, 104th Congress, First Session.

Randall G. Holcombe is a DeVoe Moore professor of economics at Florida State Univer-

sity. He is the author of ten books and more than 100 articles in academic and profes-
sional journals. His primary areas of research interest are public finance and public
choice. His most recent book, From Liberty to Democracy: The Transformation of Amer-
ican Government, was published in 2002.

Donald J. Lacombe is an assistant professor of economics at Ohio University. His re-

search has focused on public choice and state and local public finance, and his work has
been appeared in Public Choice, the Journal of Economic History, and other journals.