Available online at www.sciencedirect.com

ScienceDirect
Journal of Policy Modeling 37 (2015) 35–46

U.S. casino revenue taxes and short-run labor outcomes
Kahlil S. Philander a,∗ , Bo J. Bernhard a , Bradley S. Wimmer b ,
Ashok K. Singh a , William R. Eadington c
a Department of Hotel Administration, University of Nevada, Las Vegas, 4505 S. Maryland Parkway, Las Vegas,

NV 89154-6037, United States
b Department of Economics, University of Nevada, Las Vegas, 4505 Maryland Parkway, Las Vegas,

NV 89154-6001, United States
c Department of Economics, University of Nevada, Reno, 1664 N. Virginia Street, Reno, NV 89557-0024, United States

Received 15 November 2014; received in revised form 14 December 2014; accepted 16 January 2015
Available online 29 January 2015

Abstract
This study examines the effect of casino tax rate levels on short-run labor decisions by casino firms. Using
a panel data set consisting of all states with legal commercial casino gambling from 1998 to 2009, a 2SLS
fixed-effect model that also uses aggregate and sin-based excise taxes as instruments is estimated. We find
that maximum casino tax rates negatively affect casino employment with an inelastic average effect of −0.6.
We estimate state tax revenue changes per employee from a 1% increase in the gross gaming revenue tax,
finding that states with comparatively low tax rates could increase public revenue with relatively small losses
in employment. Nevada, New Jersey, and South Dakota – each with maximum tax rates currently below
10% – could increase tax revenue by more than $430,000 per estimated employee lost from the tax change.
Results from this study should be included in future analysis of casino tax impacts on economic efficiency.
© 2015 Published by Elsevier Inc. on behalf of Society for Policy Modeling.
JEL classification: H21; H71; L83
Keywords: Gaming tax; Casino taxation; Sin tax; Employment

1. Introduction
Throughout the World and even within a given country, casino gaming markets neither
exhibit uniform tax structures nor do they appear to follow any consistent economic guidelines.
∗ Corresponding author. Tel.: +1 702 895 1265.

E-mail address: kahlil.philander@unlv.edu (K.S. Philander).
http://dx.doi.org/10.1016/j.jpolmod.2015.01.008
0161-8938/© 2015 Published by Elsevier Inc. on behalf of Society for Policy Modeling.

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

Jurisdictional markets vary from monopolies (typically subject to heavy government taxes) to
near perfectly competitive markets (generally taxed at much lower rates). The public revenue
that is generated from gaming taxes is often highlighted by politicians and other stakeholders as
the most important indicator of the economic impact of casino gaming (Chapman et al., 1997;
Eadington, 1996, 1998; National Gambling Impact Study Commission, 1998; Paldam, 2008;
Smith, 1998); while firms often highlight the importance of job creation that may result from a
low tax environment (e.g. Christiansen, 2005).
Two neighboring states, Pennsylvania and New Jersey, are particularly illustrative examples.
In 2010, Pennsylvania (with a 55% gaming tax rate) generated close to four times the tax revenue
of New Jersey’s gaming industry (with a 9.25% gaming tax rate) but employed roughly three
times less people (American Gaming Association, 2011). While there are many different factors
that affect tax revenue and employment in these two states, clearly, there are differing policy
approaches that may not be fully informed about counter factual outcomes. Even policy decisions
that strictly focus on revenue maximization (as opposed to efficiency) will benefit from considering
employment based tax effects from gaming tax policy. Large variation in gaming tax rates is not
limited to the U.S. Gross tax rates on EU lotteries vary from a low of 12% in Estonia to a high
of 50% in Poland (Forrest, 2008), while Albon (1997) has described similar themes of tax rate
variation in the Australian market.1
The welfare implications of developing a better understood gaming tax system are sizeable.
The total worldwide gaming industry is estimated to generate over $110 billion in annual revenue
(PwC, 2011), while the U.S. commercial casino industry alone contributes $7.59 billion in gaming
tax revenue to state and local governments in the U.S. (American Gaming Association, 2011).
Secondary effects that gaming taxation can have are similarly important for maximizing economic
welfare and market performance. The U.S. commercial casino industry employs 340,564 workers
per year totaling $13.1 billion in wages, benefits and tips (American Gaming Association, 2011).
Given the large variation in rates applied across jurisdictions, many regions would likely benefit
if there are obvious changes to taxation rates that would improve efficiency.
Using a partial equilibrium framework, this study estimates an empirical model of the effect
of gross gaming revenue (GGR) excise taxes on short-run labor outcomes in the commercial
casino market. To provide a more complete understanding of this public policy trade-off, and to
provide decision makers with a useful set of estimates for policy decisions, a statistically consistent
empirical model using both panel and instrumental modeling techniques is used. We estimate state
tax revenue changes per employee from a 1% increase in the gross gaming revenue tax, finding
that states with comparatively low tax rates could increase public revenue with relatively small
losses in employment. Nevada, New Jersey, and South Dakota – each with maximum tax rates
currently below 10% – could increase tax revenue by more than $430,000 per estimated employee
lost from the tax change.
2. Review of literature
Gambling markets are typically taxed different than the rest of the economy. Walker and
Jackson (2008) have suggested that governments set gaming tax policy to maximize revenue
1 It should be noted that since gaming taxes are all applied in different manners by each jurisdiction, comparisons can
sometimes be misleading. For example, 12% of the Pennsylvania tax is a transfer to the horse racing industry and their
table games are also taxed at a lower rate; in addition to their ad valorum tax, Nevada also levies fees on gaming devices,
though these are a small portion of the overall revenue generated from the various levies.

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

rather than economic welfare. In a similar tone, Smith (1998) has conjectured that the reason
for the observed tax levels in the casino industry is that gaming taxes are set in order to capture
economic rents, rather than to internalize negative externalities or maximize public welfare. The
National Gambling Impact Study Commission (1998) and Meich (2008) both suggest that tax
rates reflect a joint objective, to both raise public funds and to punish an activity characterized as
sinful.
Higher taxation of gaming, as compared to other goods, is often motivated by the negative
externalities associated with problem gambling. This policy design can be optimal if the rates
follow the prescription set by Pigou (1920) and Baumol (1972). Higher (Pigovian) taxes are
efficient if they are applied to offset social costs, or even as a double dividend to reduce other
distorting taxes, such as income taxes (Glomm, Kawaguchi, & Sepulveda, 2008). In practice,
gaming taxes do not appear to follow any sort of Pigovian objective. Chapman et al. (1997) found
that the most socially harmful forms of gaming (e.g. Pokies, slot machines located outside of
casinos) were not taxed at a higher rate in Australia, and similarly distorted tax structures have
been described in the U.S. by Clotfelter (2005). In general, the least harmful form of gambling,
lottery, has been taxed at the highest rate.
An important distinction of the gaming industry compared to most other “sin” goods or industries with negative externalities, is that gaming is often introduced as an export good to other
jurisdictions. Many casinos are built on jurisdictional borders or are operated with restrictions
on the admission of local residents. The domestic state is able to capture the economic rents
from foreign state visitors, while exporting many of the negative externalities when the visitors
return to their foreign homes. This changes the efficient strategy of the home state, and may lead
to economically inefficient proliferation of gaming (Eadington, 1999).2 However, regardless of
any Pigovian objectives, a sound policy decision on appropriate casino tax levels still requires
an understanding of firms’ responses to changes in the tax rate. That is, while jurisdictions have
applied various tax structures (perhaps with Pigovian objectives in mind), appropriate analysis of
the efficient policy decision must also examine how the tax rate policies affect other outcomes
in the gambling industry. To date, no studies have provided an empirically sound analysis of this
relationship. One item of particular interest is how firms will respond in the labor market, given the
employment benefits of casino development are often touted as a key component of their positive
impacts on the economy and local community. In the next section, we provide a description of the
method used to estimate the impact of gross gaming revenue taxes on short-run labor outcomes,
along with the data used as part of the analysis.
3. Data and methodology
Data on the twelve states that offer commercial casino gaming was obtained from the American
Gaming Association (AGA) for the period from 1998 to 2009 to form a panel (American Gaming
Association, 2010). This data consists of an annual census made to all commercial casinos in the
U.S., including their employment, gross gaming revenue, and maximum tax rates. Differences
in data collection after 2009 inhibited use of figures after this year. Since firm level data is not
2 This exportation strategy may also help explain the emergence of small island nations – such as Antigua, the Isle of
Man, and Alderney – as large suppliers of online gaming licenses and regulation. However, this competitive result may
only lead to short-run economic rents and eventually a long-run oversupply of gaming, as neighboring jurisdictions seek
to legalize gaming to capture some of the economic benefits of gaming to offset the negative externalities that are being
exported to their region.

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46
80%
Colorado

70%

Illinois
60%

Indiana
Iowa

50%

Louisiana
40%

Michigan
Mississippi

30%

Missouri
20%

Nevada
New Jersey

10%

Pennsylvania

0%

South Dakota

Fig. 1. State maximum gaming tax rates from 1998 to 2009.

provided by all gaming commissions, maximum tax rates are the most representative figure for
marginal rates, as they indicate the marginal tax paid by the largest operators. The maximum tax
rates are illustrated in Fig. 1. Six states showed variation in tax rates during the period of study,
while the remaining six states whose rates did not change during the period of study, served as
controls.
Employment is used as a proxy variable to identify short-run labor decisions. Employment is
commonly used as a variable to measure economic impacts (e.g. Daley, Ehrlich, Landefeld, &
Barker, 1997; Lehmus, 2014), and may even be a better economic impact response variable to
changes in tax rates than, say, capital expenditures since they can be more readily adjusted by
operators. Although capital investment is an interesting dependent variable to examine because
it reflects operators investments decisions, this variable tends to be relatively static since casino
investments can be quite lumpy – most modern integrated-resorts require upfront capital investments ranging from hundreds of millions to billions of dollars. Short-run changes in employment
should be adequately captured within a given year of any change in the tax rate, and employment
is not restricted by casino license quotas. Reliable estimates of employment are also reliably
available from all U.S. commercial casino gaming markets due to the AGA’s annual census.
Eq. (1) defines the underlying relationship between economic output and gaming taxes:
yit = α0 + β1 · Taxit + xit · λ + d t · δ + vi + uit

(1)

where yit is the indicator of casino gaming development, employment; α0 is the model constant,
and Taxit is the primary variable of interest, including the maximum gaming tax. xit is a vector
of factors that affect gaming development that vary by state and year; dt is a vector of nonstate specific factors that affect gaming development and vary by year (proxied by year dummy
variables); δ are the coefficients for non-state specific factors that affect gaming development and
vary by year; vi are state specific constant factors that affect gaming development; and uit is the
model error term.
Given that we have a panel of data to analyze, a first-differencing of Eq. (1) along with the
use of time dummy variables would produce consistent estimates, if there were no time varying
unobserved variables that are correlated with Tax. However, this seems unlikely to be the case.
Casino gaming is a highly politicized and regulated industry. Changes in policy design that are

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

difficult to identify with proxy variables, such as licensing or responsible gambling requirements,
seem likely to affect both taxation rates and casino employment. Coefficient estimates from a
fixed-effect model without other corrections would therefore be subject to omitted variable bias.
To measure Eq. (1), we therefore adopt an instrumental variable approach in addition to the
fixed-effects, two-stage least squares (2SLS).
The 2SLS estimator consistently estimates the endogenous tax rate coefficient by using an
instrument to purge out the correlation between the explanatory variable and the model error term.
Two necessary conditions need to be satisfied by valid instruments. The first condition is a correlation of zero between the instrument and the structural model error term, i.e. Cov(StateTaxRateit ,
uit ) = 0 and Cov(AlcoholTaxRateit , uit ) = 0. Second, the instrument and the endogenous variable
must be correlated, i.e. Cov(StateTaxRateit , GamingTaxRateit ) =
/ 0 and Cov(AlcoholTaxRateit ,
/ 0. Two instruments are proposed to correct the endogenous gaming tax
GamingTaxRateit ) =
variable. The first instrument is the total state wide tax revenue as a percentage of state GDP. The
second instrument is the total state wide alcohol tax revenue, as a percentage of state GDP. This
second variable is similar to the first, but focuses on revenue from another specific “sin tax” and
thus may explain a difference aspect of the variance in gaming taxes.
It seems likely that the overall statewide tax rate will not provide any additional explanatory
power to gaming employment beyond what is already addressed by the model through the gross
gaming revenue tax, economic control variables, and the fixed-effect variables. For example, if the
presence of a particularly pronounced recession in a single state causes both the gaming tax and
the statewide tax to change simultaneously, the economic control variables such as real GDP or
the unemployment rate will likely capture this variation in the structural regression model. With
regards to the alcohol tax rate variable, this seems even more likely to satisfy the first necessary
condition of valid instruments. The variable focuses on a tax attributable to another industry
and should not have a direct effect on casino employment. An empirical overidentification test
(Hansen, 1982) is also conducted on the data to support the validity of the instruments.
The second condition of a non-zero correlation between the gaming tax and statewide tax
has not been established empirically in the literature; however, Furlong (1998) did find that tax
collections contributed to casino adoption at the state level. Calcagno, Walker, and Jackson (2010)
also found that other fiscal variables affected casino policy adoption. With regards to the alcohol
tax rate, it seems plausible that states that choose to heavily tax one sin good will also heavily
tax other legal sin goods, so a positive relationship is expected between these variables. As a
robustness check, empirical tests for this assumption will also be conducted, including an F-test
of weak identification (Staiger & Stock, 1997).
Model control variables include gross state product and real personal income per capita (US
Bureau of Economic Analysis, 2011), the state unemployment rate (US Bureau of Labor Statistics,
2011), the state population (US Census Bureau, 2011), gaming availability proxied by number
of instate commercial casinos, and productivity proxied by gross gaming revenue per employee
(American Gaming Association, 2010).
4. Results
4.1. Fully specified empirical model
The first stage results of the fully specified empirical model are provided in Table 1. The model
appears to fit the data well. Both instruments, the natural logarithm of alcohol tax revenue per unit
of state domestic product (Log Alcohol Tax) and the natural logarithm of all state tax revenue per

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

Table 1
First-stage model.
Log of Maximum GGR Tax Rate

Log of alcohol tax revenue per GDP
Log of all tax revenue per GDP
Log of number of commercial casinos
Log of population
Log of average GGR per casino
Log of real income per capita
Log of unemployment rate
Years 1999–2009 (dummy variables)

b

se

z

p

0.27***

0.08
0.24
0.07
0.17
0.07
0.35
0.08
–

3.40
1.70
0.88
−4.30
0.48
−2.20
1.69
–

0.00
0.09
0.38
0.00
0.63
0.03
0.09
0.00

0.41*
0.06
−0.74***
0.03
−0.77**
0.14*
Yes***

Heteroskedasticity robust standard errors provided.
* p < 0.10.
** p < 0.05.
*** p < 0.01.
Table 2
Tests of IV assumptions.
Instruments: Log Alcohol Tax/GDP & Log All Tax/GDP
Observations
F-test of weak identification
Hansen-J statistic
Hansen-J p-stat

15.04
0.01
0.93

unit of state domestic product (Log All Tax), have positive and statistically significant coefficients;
z(1 1 4) = 3.40, p = .001 and z(1 1 5) = 1.70, p = .09, respectively. This finding supports the second
necessary condition for the instruments. To further establish the joint validity of the instruments, as
the alcohol tax variable initially shows more power than the all tax variable, additional statistical
tests were conducted and are provided in Table 2. The F-test of weak identification = 15.04,
exceeded the benchmark value of 10, suggesting that these variables jointly satisfied the second
necessary condition of strong instruments (Sovey & Green, 2011; Staiger & Stock, 1997). The
Hansen J statistic, HJ = 0.01, p = 0.93 (also known as the overidentification test), failed to reject
the assumption that Log Alcohol Tax and Log All Tax do not belong in the structural model. This
finding supports the first necessary condition of valid instruments. The presence of the dummy
year variables is indicated instead of fully described, to improve readability (Table 3).
The structural model produced results that were consistent with theory. The elasticity of casino
employment with respect to the maximum gaming tax rate is estimated to be z = −0.60, and is
statistically significant, z(1 1 5) = −2.99, p = .003. A 1.0% increase in the maximum tax rate is
predicted to reduce casino employment by 0.6%. Findings from other gaming variables appeared
to be reasonable. The gaming availability variable (log of number of commercial casinos) produced
a reasonably inelastic coefficient (0.67), and the productivity variable (log of average GGR per
casino) also produced an inelastic estimate (0.36). The findings support the assertion that U.S.
casinos may be subject to economies of scale in labor. A doubling of the gaming revenue per
casino would only increase employment by an estimated 36%. Indeed, Gu (2001) has previously
provided evidence that casino operations are subject to economies of scale in payroll and other
expenses.

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

Table 3
Full model specification.
Log of employees

Log of Maximum GGR Tax Rate
Log of number of commercial casinos
Log of average GGR per casino
Log of population
Log of real income per capita
Log of unemployment rate
Years 1999–2009 (dummy variables)

b

se

z

p

−0.60***

0.20
0.09
0.08
0.21
0.46
0.09
–

−2.99
7.17
4.24
2.10
0.58
1.92
–

0.00
0.00
0.00
0.04
0.56
0.05
0.19

0.67***
0.36***
0.45**
0.27
0.16*
Yes

Heteroskedasticity robust standard errors are provided.
* p < 0.10.
** p < 0.05.
*** p < 0.01.

The population variable produced a significant and positive elasticity coefficient (0.45), which
suggests a positive relationship between total population growth and growth in short-run casino
development. The economic variable, log of real income per capita, failed to produce a significant
coefficient. Log of the unemployment rate was significant, z(1 1 5) = 1.92, p = .05, with an elasticity
of 0.16. Part of the impact of these variables may be explained by an indirect effect on casino
employment through the average GGR variable. That is, higher income and employment levels
will lead to increased spending on casino gaming, thereby increasing employment through the
average GGR variable, in lieu of a directly measured effect.3 Other statistical tests produced
results consistent with a well-defined model.4
4.2. Regression-adjusted models
The robustness of the coefficient estimates were further validated by estimating various
regression-adjusted models. Table 4 includes several of these regression-adjusted (second-stage)
models, along with a non-2SLS model (model 1). A few noteworthy results arise. First, the inclusion of simple year dummy variables ensures that the coefficient on the variable of interest (Log of
Maximum GGR Tax Rate) is statistically significant, negative in direction, and inelastic. Second,
the addition of Average GGR per Casino leads to a reduction of significance on the Unemployment
Rate variable. This may indicate that the effect of the economy on casino employment is indirectly
occurring through the change in gross gaming revenue. The change in the Bayesian information
criterion (BIC) further supports this assertion, as this value increases (decreases in absolute value)
from model (7) to (8), despite the addition of a statistically significant population variable. The
3 Additional variables were tested for significance as part of the empirical analysis, but were omitted from the results due

to high collinearity with the fixed-effects and the other explanatory variables. These excluded variables included the AGA
measures of national gaming acceptability, the admission tax dummy variable, real gross state product, and alternative
transformations of all variables.
4 The Skewness/Kurtosis, χ2 (2, n = 133) = 2.21, p = .331, and Shapiro–Francia W, z(1 3 3) = 0.854, p = 0.196, tests failed
to reject the assumption of normality. To examine serial correlation, several regression-adjusted models were estimated
with Parzen kernel estimators that are robust to the presence of arbitrary heteroskedasticity and one period serial correlation
(Baum, Schaffer, & Stillman, 2007). Only minor differences appeared in the magnitude of the t-stats, and the differences
did not cause a change in the interpretation of hypotheses tests for any variable of interest.

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

Table 4
Regression-adjusted structural models.
Log of employees
(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

−0.13

−0.85

−0.72**

−1.07***

−0.86***

−1.13***

−0.56***

−0.62***

−0.60***

0.62***

–

–

0.69***

0.67***

0.64***

0.66***

0.69***

0.67***

0.36***

–

–

–

–

–

0.35***

0.37***

0.36***

0.70
0.73

–
–

–
–

–
–

−0.06
–

–
−0.11

–
–

0.46**
–

0.45**
0.27

0.12*

–

–

–

–

0.32***

–

–

0.16*

Yes*

No

Yes**

Yes***

Yes***

Yes**

Yes

Yes

Yes

No

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

N/A
N/A
N/A
−255.5

1.24
3.92
0.05
−64.4

10.35
2.59
0.11
−56.7

9.28
4.23
0.04
−84.63

15.19
5.63
0.02
−103.4

10.00
1.33
0.25
−79.0

9.22
1.59
0.21
−196.9

14.56
0.20
0.65
−187.1

15.04
0.01
0.93
−184.7

(1)
Log of Maximum
GGR Tax Rate
Log of number of
commercial
casinos
Log of average
GGR per casino
Log of population
Log of real
income per
capita
Log of
unemployment
rate
Years 1999–2009
(dummy
variables)
Instrumental
variables
Observations
Weak ID (F-Test)
Hansen-J statistic
Hansen-J p-stat
BIC

Heteroskedasticity robust standard errors provided in parentheses.
* p < 0.10.
** p < 0.05.
*** p < 0.01.

non-2SLS model (1) does not show a statistically significant estimate of the maximum GGR tax
rate, likely the result of a biased coefficient estimate.
There is a noteworthy change in the 2SLS maximum GGR tax rate coefficient when average
GGR per casino is added to the equation. Average GGR per casino is an imperfect proxy variable
for the average casino size (and therefore the number of employees needed to operate the facilities).
Average GGR is affected by both supply and demand factors, and therefore there is a possibility
that the variable inclusion may be over-controlling the model, reducing the (absolute value) impact
of the tax rate. It is unclear if the change in the maximum GGR tax rate coefficient is due to a
non-orthogonal relationship with supply side aspects of the casino size proxy variable, or due to
a better controlled model, but controlling for casino size is a crucial element of the model and
therefore we proceed under an assumption of theoretical validity.
4.3. Log-linear model
As an alternate means to estimate the robustness of both the estimates and the tax rate elasticities, a log-linear model was fitted. The dependent variable (employment) is transformed by the
natural logarithm, but the right hand side variables all appear in level form. As shown in Table 5,

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

Table 5
Log-Lin model specification.
Log of employees

Maximum GGR tax rate
Number of commercial casinos
Average GGR per casino (millions)
Population (millions)
Real income per capita (thousands)
Unemployment rate
Years 1999–2009 (dummy variables)
Observations
F-test of weak identification
Hansen-J statistic
Hansen-J p-stat

b

se

z

p

−2.365***

0.752
0.001
0.033
0.127
0.017
1.673
–

−3.145
2.446
4.809
0.615
−0.116
0.742
–

0.002
0.014
0.000
0.539
0.908
0.458
0.187

0.003**
0.159***
0.078
−0.002
1.241
0.081
12.18
0.82
0.37

** p < 0.05.
*** p < 0.01.

the results appear to be consistent with the log–log model estimations. The F-test of weak identification = 12.18, exceeded the benchmark value of 10, and the Hansen J statistic, HJ = 0.82,
p = 0.37, again failed to reject the assumption of the instruments as structural variables. All of the
explanatory variables that were significant at the .05α level in the log–log model are significant in
the log-linear model, with the exception of population. It may be the case that a linear relationship
between the log of employment and population does not adequately capture the true relationship
of these variables.
The advantage of the log-linear model specification is that it allows for the computation of
non-constant elasticity values. In particular, elasticities of employment can be computed as a
function of the state tax rate, allowing for non-average values to be estimated, which may be
more useful for policy making at the state level. The elasticity of employment with respect to
the maximum tax rate is quite inelastic at the lowest levels of tax rates that are observed in the
sample (e.g. Nevada at a 6.75% tax rate is estimated to have an elasticity of −0.16). Estimated
elasticities reach the unitary level at roughly a 42% rate. Pennsylvania’s 55% rate is estimated to
have an elasticity of −1.3.
5. Discussion
To date, no study has empirically estimated the effect that casino excise taxes have on the
decisions made by firms. Using an estimation method that controlled for potential endogeneity in
the selection of a tax rate, this study found strong evidence that maximum gross gaming revenue
taxes have an adverse, but relatively inelastic, effect on casino employment. The study revealed
a constant elasticity model estimate of roughly −0.6, with the entire 95% confidence interval
narrowly lying in the inelastic range [−0.999, −0.208]. The log-linear model revealed relevant
elasticity estimates for all presently observed state tax rates. Findings from other gaming related
variables were also consistent with expectations. The gaming availability variable (log of number
of commercial casinos) produced an inelastic coefficient (0.67), and the productivity variable (log
of average GGR per casino) also produced an inelastic estimate (0.36). A doubling of the gaming
revenue per casino would only increase employment by an estimated 36%. This helps validate

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

Table 6
Summary of change in tax income and employment from a 1% change in state GGR tax rate. Land-based and riverboat
casinos.

Colorado
Illinois
Indiana
Iowa
Louisiana
Michigan
Mississippi
Missouri
Nevada
New Jersey
Pennsylvania
South Dakota

Maximum
GGR tax rate
(%)

Estimated
elasticity

Baseline
casino
employment

Estimated loss
of employees

Estimated tax
Inc. per lost
employee

21.5
6.75
9.25

−0.47
−1.18
−0.95
−0.52
−0.51
−0.45
−0.28
−0.50
−0.16
−0.22
−1.30
−0.21

15,857
10,961
177,397
36,377

8.3
41.8
60.3
11.5
8.8
11.5
19.2
7.4
65.3
0.3
8.3
41.8

$123,014
$118,690
$145,710
$265,331
$309,694
$364,807
$342,264
$407,505
$434,263
$470,098
$142,374
$629,882

Note: Elasticities based on results shown in Table 5. Estimated changes in tax revenue assume 1% increase in the total
tax revenue.

the wave of firm consolidation in the U.S. casino industry during the late 20th century and early
21st century.
States with relatively low maximum tax rates should re-examine their current tax structure, if
there is a need for public revenue. For example, Nevada is currently considering implementation
of a revenue tax on all businesses (Deloitte, 2013), in order to fund education initiatives in the state.
While a comprehensive review of state tax structure is beyond the scope of this study, given that
taxes on intermediary businesses have been shown to be non-optimal (e.g. Diamond & Mirrlees,
1971a, 1971b), and the relatively small predicted impact of a change in gaming taxes in this
study, it seems likely that the tax currently under consideration is sub-optimal. To provide further
guidance for policy makers, we summarize estimated state elasticities for each state included
in this study, based on the results shown in Table 5. We also estimate the tax revenue change
per employee, from a 1% increase in the state GGR tax revenue.5 As shown in Table 6, states
with comparatively low tax rates could increase public revenue with relatively small losses in
employment. Nevada, New Jersey, and South Dakota – each with maximum tax rates currently
below 10% – could increase tax revenue by more than $430,000 per estimated employee lost from
the tax change.
Using these estimates, policy makers should make more informed and explicit decisions about
their casino tax policy. Even a policy framework that ignores wider welfare effects and strictly
focuses on public revenue maximization could find value in these results, as they are applicable to
a study of net tax changes, comparing direct tax revenue from gaming excise taxes against wider
sources of tax revenue (e.g. comparing losses in employment related payroll/income taxes against
gains in GGR taxes). Of course, tax based policies are just one part of an efficient casino public
policy framework.

5 As a point of clarification, the 1% increase is in GGR, not a full percentage point increase in the applied rate. For
example, this should be interpreted as a change from 10% to 10.1%, not 10% to 11%.

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

While we note that this study focused on the short-run responses in employment to changes
in casino excise taxes, long-run capital investment changes in response to tax policy may ultimately lead to different employment equilibriums than are observed in the short-run. For example,
increases in tax costs may incentivize casinos to pursue a more capital intensive production function over the long-run. Of course, given the lumpy financial investments that are made in modern
resort-style casinos, and the quotas imposed by casino licenses, the long-run response to changes
in tax rates may take many years or decades to occur. Therefore, the current results may hold the
most meaningful policy implications.
Future study using firm level data would be useful in attempting to estimate tax effects on
long-run investment decisions, such as capital allocation. Such analysis could benefit from an
ability to closely examine accounting data figures. Using firm level data, a full panel of the U.S.
may not be entirely necessary, as relatively straightforward difference-in-difference studies could
yield accurate estimates. Admission taxes also deserve study, as their effect on both operator
and consumer incentives are interesting. In a jurisdiction with a 20% GGR tax and an average
10% theoretical win, $500 in coin-in would be required to generate the same per player tax
revenue as a $10 admission tax. Given the limited use of admission taxes in the U.S. casino
industry, an empirical test of how behavior changes in response to these different excise taxes
could yield evidence of many potential pareto improvements. While the most widely used taxation
tool in the gaming industry is an ad valorem tax on GGR, specific excise taxes may yield welfare
improvements.
Finally, expanding this study beyond commercial casinos would be useful in order to increase
reliability and generalizability of the findings. Tribal gaming, lotteries, video lottery terminals,
online casinos, and other forms of gaming all may react in different ways to changes in their cost
structure, and this is worth exploring.
This study was also limited to the use of a comparative static approach to policy modeling. A
general equilibrium approach would yield better insight into the secondary effects of casino tax
policy. For example, the global competitiveness of online gaming may exacerbate the sensitivity
of those firms behavior to changes in the tax rate. Understanding these subtleties will be important
for the development of optimal gaming policy mechanisms.
Acknowledgement
Funding assistance for the first and second author was provided by the Caesars Foundation.
References
Albon, R. (1997). The efficiency of state taxes. Australian Economic Review, 30(3), 273–287.
American Gaming Association. (2010). State of the states: The AGA survey of casino entertainment. Washington, DC:
American Gaming Association.
American Gaming Association. (2011). State of the states: The AGA survey of casino entertainment. Washington, DC:
American Gaming Association.
Baum, C., Schaffer, M., & Stillman, S. (2007). Enhanced routines for instrumental variables/gmm estimation and testing.
Boston, MA: Boston College Department of Economics Working Paper No. 667.
Baumol, W. (1972). On taxation and the control of externalities. American Economic Review, 62(3), 307–
322.
Calcagno, P., Walker, D., & Jackson, J. (2010). Determinants of the probability and timing of commercial casino
legalization in the United States. Public Choice, 142(1), 69–90.
Chapman, R., Beard, J., Jones, C., Cuthbertson, S., Brett, D., & Aitchison, G. (1997). A framework for national competition
policy reviews of gaming legislation. Sydney, Australia: Centre for International Economics.

K.S. Philander et al. / Journal of Policy Modeling 37 (2015) 35–46

Christiansen, E. M. (2005). The impacts of gaming taxation in the United States. Washington, DC: American Gaming
Association.
Clotfelter, C. T. (2005). Theory and practice of excise taxation: Smoking, drinking, gambling, and driving. New York:
Oxford University Press.
Daley, W. M., Ehrlich, E. M., Landefeld, J. S., & Barker, B. L. (1997). Regional multipliers: A user handbook for the
regional input–output modeling system (RIMS II). http://www.bea.gov/scb/pdf/regional/perinc/meth/rims2.pdf
Deloitte. (2013). Nevada legislature takes no action on margin tax initiative, thus placing the measure
on the 2014 ballot; revisions to taxation of mining also proposed. http://www.deloitte.com/assets/DcomUnitedStates/Local%20Assets/Documents/Tax/us tax multistate Nevada Alert 3-19-2013.pdf
Diamond, P. A., & Mirrlees, J. A. (1971a). Optimal taxation and public production: I. American Economic Review, 61(1),
8–27.
Diamond, P. A., & Mirrlees, J. A. (1971b). Optimal taxation and public production: II. American Economic Review, 61(3),
261–278.
Eadington, W. R. (1996). The legalization of casinos: Policy objectives, regulatory alternatives, and cost/benefit considerations. Journal of Travel Research, 34, 3–8.
Eadington, W. R. (1998). Contributions of casino-style gambling to local economies. Annals of the American Academy
of Political and Social Science, 556, 53–65.
Eadington, W. R. (1999). The economics of casino gambling. Journal of Economic Perspectives, 13(3), 173–192.
Forrest, D. (2008). Gambling policy in the European Union: Too many losers? Public Finance Analysis, 64, 540–
569.
Furlong, E. (1998). A logistic regression model explaining recent state casino gaming adoptions. Policy Studies Journal,
26, 371–383.
Glomm, G., Kawaguchi, D., & Sepulveda, F. (2008). Green taxes and double dividends in a dynamic economy. Journal
of Policy Modeling, 30(1), 19–32.
Gu, Z. (2001). Economies of scale in the gaming industry: An analysis of casino operations on the Las Vegas strip and in
Atlantic City. Journal of Hospitality Financial Management, 9(1), 1–15.
Hansen, L. (1982). Large sample properties of generalized method of moments estimators. Econometrica, 50(3),
1029–1054.
Lehmus, M. (2014). Distributional and employment effects of labour tax changes in Finland. Journal of Policy Modeling,
36(1), 107–120.
Meich, B. F. (2008). The power to destroy: The psychology of gaming taxation. Gaming Law Review and Economics,
12(5), 458–465.
National Gambling Impact Study Commission. (1998). The regional economic impacts of casino gambling: assessment
of the literature and establishment of a research agenda. http://govinfo.library.unt.edu/ngisc/reports/ecoimprpt.pdf
Paldam, M. (2008). The political economy of regulating gambling. New York: Palgrave Macmillan.
Pigou, A. (1920). The economics of welfare. London: MacMillan and Co.
PwC. (2011). Global gaming outlook. Retrieved from http://www.pwc.com/gr/en/publications/global-gaming-outlook.
jhtml
Smith, J. (1998). Gambling taxation in Australia. Discussion paper number 16. Division of Economics and Politics,
Research School of Social Sciences, Australian National University.
Sovey, A., & Green, D. (2011). Instrumental variables estimation in political science: A readers’ guide. American Journal
of Political Science, 55(1), 188–200.
Staiger, D., & Stock, J. (1997). Instrumental variables regression with weak instruments. Econometrica, 65(3), 557–586.
U.S. Bureau of Economic Analysis. (2011). Regional economic accounts. Washington, DC: U.S. Bureau of Economic
Analysis. http://www.bea.gov/regional/
U.S. Bureau of Labor Statistics. (2011). Databases, tables & calculators. Washington, DC: U.S. Bureau of Labor Statistics.
http://www.bls.gov/data/
U.S. Census Bureau. (2011). State & county quickfacts. Washington, DC: U.S. Census Bureau. http://quickfacts.census.
gov/qfd/index.html
Walker, D. M., & Jackson, J. D. (2008). Do U.S. gambling industries cannibalize each other? Public Finance Review, 36,
308–333.