Hitting the Jackpot or Hitting the Skids:
Entertainment, Poverty, and the Demand for
State Lotteries
Garrick Blalock, David R. Just, and Daniel H. Simon∗
December 14, 2004

Abstract
State-sponsored lotteries are a lucrative source of revenue. Despite their
low payout rates, lotteries are extremely popular, particularly among lowincome citizens. State officials laud the benefits of lottery proceeds and
promote the fun and excitement of participation. This entertainment value
is one explanation for lottery demand by the poor: individuals with lower
incomes substitute lottery play for other entertainment. Alternatively, lowincome consumers may view lotteries as a convenient and otherwise rare
opportunity for radically improving their standard of living. Bad times
may cause desperation and the desperate may turn to lotteries in an effort
to escape hardship. This study tests these competing explanations. We
examine lottery sales data from 39 states over 10 years and find a strong
and positive relationship between sales and poverty rates. In contrast, we
find no relationship between movie ticket sales, another inexpensive form
of entertainment, and poverty rates.

Keywords: Lottery, Poverty, Entertainment
JEL codes: H30, D12, D81
∗

Cornell University,
Department of Applied Economics and Management,
garrick.blalock@cornell.edu, drj3@cornell.edu, and dhs29@cornell.edu.
Please
direct correspondence to Garrick Blalock, 346 Warren Hall, Ithaca, NY 14853 USA, +1 (607)
255-0307. We thank Kosali Simon and Dongkuk Lim for assistance with data collection and
Christopher Cotton and James E. Blalock for extensive comments.

1

 The Maine State Lottery is proud of its commitment to providing the
citizens of Maine with fun and exciting entertainment.
First sentence of the Maine State lottery homepage.

1. Introduction
Adam Smith wrote in the Wealth of Nations that the British lottery was patronized by players in “the vain hope of gaining some of the great prizes.” Smith was
almost correct. The hope of obtaining great prizes is not completely vain, but the
probability of gaining such prizes is exceedingly small. But for most lottery players, participation is one of the few means by which they can obtain large prizes
from a modest wager with any positive probability. The motivation and demographics of players, and the effects of their participation, are a growing concern
as more states turn to lotteries and other forms of gambling to address budget
shortfalls. The number of states administering lotteries has risen from 14 in 1980
to 39 in 2003. In Massachusetts, the state with the most popular lottery, annual
ticket sales amounted to a staggering $663 per citizen in 2003.1 Even in the state
with the least patronized lottery, Montana, ticket sales per capita were $38. Since
states typically retain about 45% of total sales, the contribution to budgets is
enormous.
While states tout the benefits, typically to education, that are financed by
lottery proceeds, the social costs of lotteries are more difficult to assess. There
1

Sales as reported by the North American Association of State and Provincial Lotteries.
Sales are lottery revenues inclusive of monies returned to players as prizes.

2

 is widespread concern that the poor spend a larger fraction of their income on
tickets than do the rich, and several studies indicate that lotteries are regressive.
The states (and some economists) argue, however, that lotteries are simply a form
of entertainment. For example, the New York State Lottery website advertises
that, “The Lottery is fun. Entertaining. Exciting.” According to this “entertainment hypothesis,” lottery tickets are less of an investment, and more like a ticket
admitting the purchaser to a short, real-life drama. From this perspective, poorer
individuals may purchase more lottery tickets because they cannot afford higherquality entertainment. However, a competing explanation posits that consumers,
especially those in dire economic circumstances, see lotteries as a convenient and
accessible tool for radically altering their standard of living, a government-run,
financial “hail-mary strategy.” In short, bad times may cause desperation and
the desperate may turn to lotteries in an effort to escape hardship. Such behavior
predicted by the “desperation hypothesis” would have the unfortunate effect of
further lowering wealth in households with already declining fortunes.
While anecdotal evidence provides support for both hypotheses, it suggests
that the poor are relatively more likely to see the lottery as a financial investment, and relatively less likely to play for entertainment. Survey respondents in
California were about equally divided between whether they played the lottery
for money or for fun (Los Angeles Times 1986).2 However, among those with less
than $30,000 in income, 25% more respondents cited money rather than fun, while
the reverse was true for those with higher incomes. Moreover, a survey by the
2

As described in Clotfelter and Cook 1990, pg. 109.

3

 Consumer Federation of America and Primerica (1999) indicates that the poor
are more likely to view the lottery as an effective financial investment tool. When
asked what is your “best chance to obtain half a million dollars or more in your
lifetime,” 47% said “save and invest a portion of your income,” while 27% of all
respondents said, “win a lottery or sweepstakes.” However, among respondents
with incomes of $15,000 to $25,000, 45% chose a lottery while only 31% selected
saving as the best opportunity to accumulate half a million dollars.
This study examines how lottery ticket expenditures vary with changes in
economic conditions. Unlike previous studies of lottery participation, we focus
on those around the poverty line in order to test the desperation hypothesis.
Specifically, we regress per-capita ticket sales on the percentage of households
below the poverty line using a panel dataset of lottery ticket sales for 39 states
over 10 years. To test the competing entertainment hypothesis, we also regress
per-capita movie ticket sales on the poverty rate. If poverty causes people to
substitute cheaper entertainment for more expensive entertainment, then we might
expect to see movie ticket sales rise as the poverty rate increases. Our results
provide strong support for the desperation hypothesis and no support for the
entertainment hypothesis. Lottery ticket sales rise with increases in the poverty
rate, while movie ticket sales are uncorrelated with the poverty rate.
The paper proceeds as follows. The next section reviews prior literature on
the demand for lotteries and the models of consumer behavior that motivate our
hypotheses. Section 3 describes the data and Section 4 discusses our econometric
approach and the results. Section 5 concludes with some policy implications of
4

 our findings.

2. Prior Literature on Lottery Demand
Research in economics examining participation in state lotteries has focused on
three questions: (1) Who tends to play the lotteries? (2) How does lottery participation vary with income? And (3) why do individuals choose to buy lottery
tickets? Regarding the first question, a variety of household surveys provide information about the demographic characteristics of lottery participants. Summarizing data from several of these surveys, Clotfelter and Cook (1990) report that
men play more than women, catholics play more than protestants, and blacks and
hispanics play more than non-hispanic whites. In addition, they report that lottery participation and age have a non-monotonic relationship, with middle-aged
adults playing more than young adults (18-25 years old) or the elderly (those over
65), while lottery play decreases with education. One explanation for the negative relationship between education and lottery participation is that less educated
consumers may be less able to evaluate accurately lottery probabilities (Hansen,
Miyazaki, and Sprott 2000).
Of particular interest to policymakers is the relationship between income and
lottery participation: in particular, whether lotteries are a form of regressive taxation. Clotfelter and Cook (1990) summarize survey data which indicate that for
a broad range of incomes, there is no clear relationship between household income
and likelihood of lottery play. However, many studies have found that lotteries are

5

 regressive because lower-income households spend a higher percentage of their income on lottery tickets. For example, Clotfelter (1979) finds that daily and weekly
lotteries are regressive with elasticities of income less than one. Moreover, for daily
lotteries, he finds that the elasticity of income is negative, implying that lottery
sales are an inferior good. Using self-reported expenditure data on lottery winners
in Illinois, Borg and Mason (1988) find that the income elasticity of demand is
between 0 and 0.25, providing further evidence of regressivity. Interestingly, they
find that the income elasticity is lowest for the lowest-income households in their
sample. Hansen, Miyazaki, and Sprott (2000) use several years of county-level
data on lottery sales and demographic characteristics in five different states to assess the relationship between income and lottery participation. In four out of five
states, they find evidence that the lottery is regressive. Moreover, in Minnesota,
they find that the elasticity of income is negative, providing additional evidence
that lottery tickets are an inferior good.
While all of the above studies estimate cross-sectional ordinary least squares
models, two studies employ more rigorous econometric techniques to control for
the decisions of individuals to play the lottery or not. Using microdata on lottery
participation from a survey of 676 Kentucky residents, Scott and Garen (1994)
find no evidence that the lottery is regressive. In fact, they find no relationship
between income and lottery expenditures. They do find, however, that the probability of participation increases with income up to $30,000, and then decreases
with income exceeding $30,000. Additionally, they find that unemployment increases the probability of lottery participation, but reduces lottery expenditures
6

 conditional on participation.
Of particular interest to this study, Scott and Garen (1994) look at the impact
of three variables indicating whether individuals receive public assistance. They
find that individuals receiving AFDC and food stamps buy fewer lottery tickets,
while those receiving Medicaid spend more on lottery tickets. However, the authors note that they have very few observations on individuals playing the lottery
and receiving each of these forms of welfare, therefore they encourage caution in
interpreting the results.
Finally, using microdata on participation in the UK National Lottery, Farrell
and Walker (1999) also consider the nonlinear effect of income on lottery participation and expenditures. Consistent with Scott and Garen (1994), they find that
income initially increases and then reduces lottery participation. However, Farrell
and Walker find this same nonlinear relationship with lottery expenditures (both
conditional on participation and unconditionally). Moreover, their results provide
further evidence that the lottery is regressive, with income elasticities ranging
from 0.01 to 0.45.
The regressivity of lotteries would be less troubling were it not for the additional concern that they exploit consumers’ inability to rationally evaluate the
expected value of a lottery ticket. This concern arises largely because the expected
value of purchasing a lottery ticket is roughly one half the price of the ticket. Such
a poor payout has prompted many to question why so many people buy lottery
tickets, given that they display risk aversion in other settings (Clotfelter and Cook
1990).
7

 There appear to be four primary explanations for why people play the lottery.
The first explanation says that people buy lottery tickets for entertainment as well
as for the opportunity to win prizes (Kearney 2002). As we show above, lotteries
promote their entertainment value, emphasizing that it is fun to play. In addition,
they also highlight the fact that the proceeds are used to support public education
and other valued public goods (Clotfelter and Cook 1990).
A second explanation emphasizes cognitive biases and lack of information.
Langer (1975, 1977) presents evidence for an “illusion of control,” in which individuals believe that they can affect their chances of winning the lottery through
the numbers that they choose. Kahneman and Tversky (1979) argue that individuals place too much weight on very low-probability events. In a typical lotto
game, the odds against hitting the jackpot are several million to one (Clotfelter
and Cook 1990). This means individuals may perceive the probability of winning
the lottery to be several times larger than the true probability. In fact, the extant
literature notes that lottery patrons tend to ignore the probability of winning,
focusing instead on the size of the lottery prize (Forrest, Simmons, and Chesters
2002; Walker and Young 2001), and the frequency of drawings (Cook and Clotfelter 1991). Perhaps these cognitive biases are reinforced by lottery marketing
mechanisms. In a sample of 151 TV and radio lottery ads in the biggest markets,
only 12% mentioned odds of winning, while half mentioned dollar amounts of
prizes (Clotfelter and Cook 1990). Of the 52 ads that portrayed a lottery player,
two thirds showed at least one player winning (Clotfelter and Cook 1990).
Kearney (2002) considers both of these explanations as she examines the im8

 pact of lotteries on consumer behavior. She finds that lottery sales increase with
the expected value of the lottery game, suggesting that consumers are able to
properly evaluate the nature of the gamble. Additionally, she finds that consumers respond to non-wealth creating characteristics of the lottery games; consumers prefer picking more numbers to fewer numbers, and prefer older to newer
games. As she notes, this latter result is consistent with the notion that consumers
value the entertainment aspect of lotteries. Alternatively, this may indicate that
consumers irrationally believe that these characteristics affect the likelihood of
winning.
The third explanation comes from Friedman and Savage (1948), who use
expected utility theory to address why people both play the lottery and buy
insurance—seemingly inconsistent behaviors. The fourth explanation, which comes
from prospect theory, argues that people who have suffered a financial shock may
be more likely to play the lottery. The following section develops these last two
explanations more fully, and examines how they apply to our particular focus on
lottery participation by the poor.

2.1. Modeling Risky Choices
Several formal theories exist on why individuals play lotteries. Most prominently,
Friedman and Savage (1948), attempting to explain how an individual may be
risk averse enough to buy insurance, yet risk loving enough to purchase lottery
tickets, proposed that utility of wealth functions must take on an inverted s-shape.
This expected utility approach supposes that over lower levels of wealth a utility
9

 function will be concave, reflecting an aversion to risk, yet convex over higher
levels of wealth, reflecting risk loving behavior. In the context of our paper, it
is important to determine the behavioral implications over average wealth levels,
versus those below the poverty level. According to this theory, an individual solves

maxU X L , P L =
L

u xLi pLi ,

(1)

where L is the index of possible risky choices, X L is a vector of money outcomes
for lottery L, and P L is a vector of probabilities associated with those outcomes.
In the case of a lottery, there is typically a very small probability of a very large
reward. A small price is charged for the chance to engage in the gamble, but
enough that the expected value of the lottery is negative. This has been the
dominant theory of risky choice for several decades, and Friedman and Savage’s
explanation of lottery behavior has received wide ranging support in the literature.
A condition necessary for playing the lottery is

∆U = pu (w + x − k) + (1 − p) u (w − k) − u (w) > 0,

(2)

where w is the current level of wealth, x is the jackpot, p the probability of
winning, and k is the price of a lottery ticket. If the price of a lottery is small, say
one unit of wealth, and the probability of winning is small, we can approximate
this inequality using
∆U ≈ pu (w + x) > u (w) .

10

(3)

 Let wl < wh represent lower and middle wealth levels respectively. If poor individuals will play the lottery, but middle-income individuals will not, then

u (wh ) > pu (wh + x) > pu (wl + x) > u (wl ) .

(4)

This implies that middle-income individuals have a higher marginal utility of
wealth than poor individuals. As a result, demand for lottery tickets increases as
wealth declines.
Prospect theory provides a substantially different explanation for the behavior
described by Friedman and Savage. Kahneman and Tversky (1979) observed that
individuals appear to exaggerate the pain from financial losses, leading them to
take great risks to avoid or rebound from losses. On the other hand, the pleasure
from financial gains are under emphasized, leading to more conservative behavior
following an increase in financial well-being or when attempting to increase wealth.
This type of behavior has become known as loss aversion. Loss averse behavior is
often modeled using prospect theory, which supposes that individuals compare all
outcomes to some reference level of wealth. Thus, no utility of wealth function is
defined. Rather, the individual makes decisions according to a utility of gains or
losses function as measured against the reference wealth. Kahneman and Tversky
(1979) propose that individuals solve the following optimization

maxV X L , P L  r =
L

11

v xLi  r π pLi ,

(5)

 where L is the index of possible risky choices, X L is a vector of money outcomes
for lottery L, P L is a vector of probabilities associated with those outcomes, r
is some reference level of wealth, often interpreted as the current wealth. The
probability weighting function π is a function mapping the unit interval into the
unit interval. This function represents a distortion of the probability distribution.
The function is such that π (p) > p is p < p, and π (p) < p is p > p. Further,
π (0) = 0, π (1) = 1, and π (0) < π (1) . The function v is given by

v (x r) =



 ug (x − r) if x ≥ r

(6)


 ul (x − r) if x < r,
Here, ug is a concave function with ug (0) = 0, and ul is a convex function with
limx↓0 ul (x) = 0. Further, it is assumed that limul (x) > limug (x) , and thus
x↑0

x↓0

individuals discount small losses heavily. The probability weighting function π is
assumed to be increasing, exaggerating small probabilities and diminishing larger
probabilities. In the case of a standard state lottery, a ticket will be purchased if

∆V = π (p) ug (w − r + x − k) + π (1 − p) v (w − k r) − v (w r) > 0,

(7)

where w is the current level of wealth, and k is the price of a lottery ticket.
Prospect theory makes no real distinction between the behavior of rich and poor.
Rather, it suggests that differences in behavior will derive from whether individuals are above or below their reference level of wealth. The reference level of wealth
is more stable than wealth itself, and hence when individuals face negative income
12

 shocks, we would expect them to be more likely to play the lottery. To see this,
if an individual is below his reference level of wealth, then we can replace v with
ul in the left hand side of 7. We can find the effect of a negative shock by taking
the derivative of the left hand side of 7 to obtain
∂∆V
= π (p) ug (w − r + x − k) + π (1 − p) ul (w − k − r) − ul (w − r) .
∂w

(8)

Because the price of lottery tickets should be small relative to changes in wealth,
we can approximate this as
∂∆V
≈ π (p) ug (w − r + x − k) − (1 − π (1 − p))ul (w − r) .
∂w

(9)

This must be less than zero if the shock in wealth is relatively small compared
to the possible lottery winnings, and if p is small. To see this, note that ug is
concave and thus has a declining slope. For the size of prize involved in lotteries, the marginal utility of wealth should be close to zero. Alternatively, the
slope of ul is increasing as w approaches r. Thus, if w is close enough to r,
ul (w − r) > ug (w − r + x − k) . Because π (0) < π (1) , π (p) < 1 − π (1 − p)
if p is small enough. Thus, increasing the distance between current wealth and
reference wealth can lead an individual to play a lottery they otherwise would
not. If an individual falls below the poverty line (decrease in w), he will be more
likely to play the lottery until he either returns to his reference wealth, or adjusts
his reference wealth downward.

13

 If individuals behave consistently with prospect theory, we expect to find increases in lottery sales as individuals face negative income or wealth shocks.3 This
means that lotteries should behave as an inferior good, decreasing in ticket sales as
income increases. In the following section, we examine the empirical relationship
between poverty and lottery participation, and we compare this to the relationship
between poverty and entertainment consumption.

3. Data
We employ three data sources. First, we use an annual panel of 39 state lottery ticket sales totals from 1990 to 2002.4 The figures represent dollars crossing
the counter from consumers to lottery sales agents. Some of those dollars are
then returned to players as winnings. The difference between sales revenue and
winnings is the “handle” in industry terms. Because the payout percentage is
typically set by state law and static, total sales figures and the “handle” should
be extremely collinear and this is borne out in our data. We obtained the data
from Christiansen Capital Advisors (CCA) LLC, a consulting firm specializing in
3

Ideally, we would like to distinguish between income and wealth. In practice, however,
our data include only income measures. Because individuals with very low incomes typically
have little savings, we believe income changes to be closely correlated with changes in perceived
wealth. Furthermore, we have done our analysis using levels of household receiving welfare, a
more direct measure of wealth. We find very similar results. However, because welfare qualification rules changed during the time period of the panel, we believe the poverty rate is a more
consistent measure of households facing economic hardship.
4
Including the District of Columbia, four of the 39 states introduced a lottery during the time
period of our panel. No state ended its lottery. We use an unbalanced panel in our analysis.
However, our results change very little if we limit our sample to a balanced panel of only states
with a lottery throughout the study period.

14

 the gaming industry. CCA gathered the data from primary state lottery board
sources.
Second, we use several government datasets to control for demographic and
economic changes at the state level. We used state-year household poverty rates
and other demographic characteristics (race, marital status, age, education, etc.)
from the United States Bureau of Labor Statistic’s (BLS) March supplement of the
Current Population Survey. Further, we use the BLS annual seasonally adjusted
measure of state unemployment. Finally, we employ data on state tax revenues
published by the U.S. Census Bureau.5
Third, to test the entertainment hypothesis, we obtained annual state movie
box office sales from Nielsen EDI, a division of ACNielsen that specializes in
motion picture industry data. Nielsen EDI collects box office sales figures through
electronic reporting from cinemas nationwide. Data are available from 1991 to
2002.
Table 1 shows our three key variables: lottery sales, the poverty rate, and
box office sales, by state for 2002. Table 2 displays descriptive statistics for these
variables averaged across all states by year. Table 3 shows the correlation of key
demographic variables.
It is important to note that our data are aggregate and do not reveal which
individuals purchased lottery tickets. Therefore, we cannot link individuals’ eco5

Because of small sampling issues in its annual survey, BLS recommends that researchers use
3-year moving averages when comparing CPS poverty rates across states. We have performed
all of our estimations with both annual and 3-year averages and find very little difference in the
outcome. The results reported here are based on annual data.

15

 nomic conditions with their lottery consumption. Instead, we relate state-level
economic conditions to state-level lottery sales. We discuss the limitations of this
approach in the next section.

4. Methods and Results
We first test the desperation hypothesis by regressing state lottery sales on state
poverty rates:

lottery sales per capitait = β1 poverty rateit + Xi βit + αi + λt +

it

(10)

where Xi represents a vector of state demographic factors, αi is a fixed effect
capturing the static unobservables of state i, λt is an indicator for year t, and

it

is an error term.
Column 1 of Table 4 displays the results of this estimation. The coefficient
on the poverty rate, 230.088, suggests that per-capita lottery sales increase about
$2.30 for each one percentage point increase in the poverty rate.
Although the fixed effect αi controls for time-invariant, unobserved state characteristics, it is likely that time-varying demographic characteristics such as age
and education are correlated with both poverty and lottery participation. Similarly, other measures of household economic conditions including income and
unemployment are certainly correlated with both poverty and lottery participation. In order to more precisely test the desperation hypothesis, Column 2 of

16

 Table 4 controls for state per-capita disposable income and the state unemployment rate, as well as other demographic characteristics including age, eduction,
percent white and black, percent married, percent with children, and percent female headed households. By including income as well as unemployment, we are
able to separate the impact of poverty from the overall impact of income on lottery
consumption.
The inclusion of these variables has little impact on the effect of the poverty
rate. At the same time, lottery sales increase with disposable income and decrease
with the unemployment rate. This suggests that while lottery tickets are an inferior good over the whole range of income, lottery tickets generally are a normal
good—consumers tend to buy more tickets as their income increases. This is consistent with the desperation hypothesis, which would not be expected to apply to
individuals at middle and upper income levels. The negative relationship between
unemployment and lottery sales is interesting, as it directly contradicts the prediction generated by prospect theory, that individuals become more risk loving when
they suffer a financial shock. Combined with the positive effect of the poverty
rate and the positive effect of income, these results suggest that it is a financial
shock that pushes an individual into poverty that triggers the desperation, which
leads to increased lottery consumption.
Although the results of Table 4 are consistent with the desperation hypothesis, they are also consistent with the entertainment hypothesis. Substitution of
lottery tickets for more expensive entertainment suggests a much more benign
explanation for our findings: consumers economizing on their entertainment ex17

 penditures, rather than desperate individuals pursuing a hail-mary strategy. To
test this possibility, we examine the relationship between movie ticket sales and
the poverty rate. To the extent that movies also provide affordable entertainment,
we should expect box office sales to also increase with poverty if consumers are
simply seeking inexpensive entertainment.6,7 Columns 1 and 2 of Table 5 consider
this possibility. Column 1 shows the effect of poverty on box office sales, which
is near zero. In other words, we find no evidence that poor consumers increase
their consumption of movie tickets when their incomes decline. At the same time,
the strong positive effect of disposable income remains, confirming movies to be a
normal good. Column 2 conditions the relationship between poverty and lottery
sales on movie box office sales. Specifically, we include box offices sales as a regressor capturing variation in entertainment consumption. The inclusion of a movie
box office control has little effect on our results, suggesting that the desperation
hypothesis is invariant to the entertainment value of lottery play.
Another alternative explanation for our findings could be unobservable efforts
by states to promote lotteries during budget shortfalls. If budget shortfalls were
driven by negative economic shocks that also increased the poverty rate, our es6

Indeed, there is evidence that low-income households attend movie theaters. For example,
the Census Bureau reports that in 1997, when the poverty line for a family of four was about
$16,000, 46% of households with incomes between $10,000 and $20,000 attended a movie theater.
7
The identifying assumption here is that poor individuals substitute both lottery tickets
and movies for more expensive forms of entertainment. It does not require that the same
individuals purchase both lottery tickets and movie tickets. Lancaster (1966) in his theory of
demand proposes that consumers derive utility from the basic characteristics of goods, such as
their entertainment value. Goods that provide similar basic characteristics will thus operate
as substitutes in consumption. Exogenous factors increasing the consumption of entertainment
should affect both lottery and movie ticket sales if both are forms of entertainment.

18

 timation of the effect of poverty would be inconsistent. To check this possibility,
Column 3 of Table 5 includes the log of state tax revenues from all sources. Because we expect tax revenues to respond to changes in population, we also include
the log of state population.8 The effect of poverty remains strongly positive. Interestingly, lottery sales decrease with increases in tax revenue. One explanation
for this finding could be that states choose to promote lotteries more when tax
revenues decline and additional government income is most needed. Although we
cannot conclude this from our limited analysis, the possibility warrants further
research.
Our results so far show that lottery tickets are generally a normal good, but
are an inferior good around the wealth level of the poverty line. Our measure
of the poverty rate includes households on the edge of poverty as well as those
below the line. Table 6 splits the poverty rate into the percentage below and
above 50% of the poverty income level. Column 1 shows the estimation using the
standard BLS poverty rate definition for reference, Column 2 displays the effect of
just the percentage of the population below 50% of the poverty line, and Column
3 includes both those above and below the 50% mark. The magnitude of the
effect of those in “extreme” poverty in isolation is much lower than the overall
poverty effect and insignificant. One explanation could be that the number of
individuals in extreme poverty is insufficient to generate a measurable effect on
8

Another approach would be to regress on per-capita tax revenue. Because some state expenses, such as highway maintenance, may respond slowly to population changes, we believe
total tax revenue to be the more relevant measure of state financial health. However we have
estimated regressions using both measures. The coefficient of per-capita tax revenue is near
zero.

19

 overall sales, even if a true relationship between extreme poverty and lottery
sales exists. Alternatively, the small measured effect could be due to the strong
correlation between the two poverty measures. When both measures are included,
the effect of extreme poverty falls by about half, whereas the effect of those above
the 50% mark remains consistent with the overall poverty measure. In general, the
evidence suggests that it is those in the upper end of poverty who are affecting
lottery sales. This explanation is consistent with the constraints of consumers’
budgets—if income falls into the realm of extreme poverty, it may be insufficient
to permit the purchase of large numbers of lottery tickets. In addition, individuals
in extreme poverty are likely to come from the ranks of those already in poverty
and already desperate. As a result, moving from the upper half to the lower half of
the poverty distribution is unlikely to result in a substantial increase in risk-taking
behavior.
As noted above, we use state-level data that does not identify individuals buying lottery tickets. As a result, we cannot directly link household economic conditions with lottery consumption. This limitation allows for an alternative explanation for the positive relationship between poverty and lottery sales: changes in
poverty rates could be correlated with changing demographic characteristics that
increase in lottery consumption. Although we include most of the demographic
characteristics that previous research has shown to predict lottery consumption,
it is still possible that unobserved changes in the make-up of poor could affect
lottery sales. However, it seems unlikely that the composition of those in poverty
would systematically change to yield the results that we observe. Nonetheless,
20

 one would need panel microdata on a large random sample with sufficient observations of individuals moving into and out of poverty to conclusively rule out this
alternative explanation. Unfortunately, we are not aware of the existence of such
a dataset.

5. What have we learned?
Two explanations may account for the widely observed phenomenon that the
poor spend a disproportionate amount of their income on lottery tickets. First,
lower-income individuals may substitute lottery play for more expensive forms of
entertainment. Second, low-income consumers may view lotteries as a convenient
and otherwise rare opportunity for radically improving their standard of living.
Bad times may cause desperation and the desperate may turn to lotteries in an
effort to escape hardship. Our study tests these two competing hypotheses. We
find no evidence to support the former entertainment hypothesis. In contrast, we
find strong and robust evidence in support of the desperation hypothesis.
In relation to established theories of consumer behavior under uncertainty,
our results are consistent with Friedman and Savage and partly consistent with
prospect theory. On one hand, the findings support prospect theory because
negative income shocks that result in poverty increase lottery ticket sales. On
the other hand, the evidence counters prospect theory because negative shocks
to employment and average income do not increase lottery sales. In sum, the
reference level of wealth at which a shock occurs matters.

21

 Finding that desperation motivates lottery consumption by the poor has some
troubling policy implications. Although it is understandable that state officials
emphasize benevolent explanations for why people play the lottery, our results
suggest that these explanations do not apply to the poor. Rather than seeking
fun and exciting entertainment, the poor appear to play because of an ill-conceived
belief that participation will improve their financial well being. This cost to the
poor must be carefully balanced against the benefits of lottery revenue.
Finally, the results indicate that it is individuals falling just below the poverty
line that contribute the largest increase to lottery sales. This observation suggests
that lottery participation is strongest among those in poverty who seemingly have
the greatest chance of escaping it. With its low expected return, lottery participation may be a factor tending to frustrate policy efforts to lift the poor out of
poverty.

References
Borg, Mary O., and Paul M. Mason (1988): “The Budgetary Indidence of
a Lottery to Support Education,” National Tax Journal, 41(1), 75–82.
Clotfelter, Charles T. (1979): “On the Regressivity of State-Operated
”Numbers” Games,” National Tax Journal, 32(4), 543–548.
Clotfelter, Charles T., and Philip J. Cook (1990): “On the Economics
of State Lotteries,” The Journal of Economic Perspectives, 4(4), 105–119.
Consumer Federation of America and Primerica (1999): “New Study:
Typical American Household Has Net Financial Assets of $1000,” Press release.
Cook, Philip J., and Charles T. Clotfelter (1991): “The Peculiar Scale
22

 Economies of Lotto,” Working Paper 3766, National Bureau of Economic Research, Cambridge, MA.
Farrell, Lisa, and Ian Walker (1999): “The Welfare Effects of Lotto: Evidence from the UK,” Journal of Public Economics, 72(1), 99–120.
Forrest, David, Robert Simmons, and Neil Chesters (2002): “Buying
a Dream: Alternative Models of Demand for Lotto,” Economic Inquiry, 40,
485–496.
Friedman, Milton, and Leonard J. Savage (1948): “The Utility Analysis
of Choices Involving Risk,” Journal of Political Economy, 56, 279–304.
Hansen, Ann, Anthony D. Miyazaki, and David E. Sprott (2000): “The
Tax Indidence of Lotteries: Evidence from Five States,” The Journal of Consumer Affairs, 34(2), 182–203.
Kahneman, Daniel, and Amos Tversky (1979): “Prospect Theory: An Analysis of Decision under Risk,” Econometrica, 47(2), 263–292.
Kearney, Melissa Schettini (2002): “State Lotteries and Consumer Behavior,” Working Paper 9330, National Bureau of Economic Research, Cambridge,
MA.
Lancaster, Kelvin J. (1966): “A New Approach to Consumer Theory,” Journal of Political Economy, 2(74), 132–157.
Langer, Ellen J. (1975): “The Illusion of Control,” Journal of Personality and
Social Psychology, 32, 311–328.
(1977): “Psychology of Chance,” Journal for the Theory of Social Behavior, 7(2), 185–207.
Los Angeles Times (1986): Poll 104.
Scott, Frank A., and John E. Garen (1994): “Probability of Purchase,
Amount of Purchase, and the Demographic Indidence of the Lottery Tax,”
Journal of Public Economics, 54(1), 121–143.
Walker, Ian, and Juliet Young (2001): “An Economist’s Guide to Lottery
Design,” Economic Journal, 111(475), 700–722.

23

 A. Tables

24

Table 1: Descriptive statistics for 2002
--------------------------------------------------------------------state  
lottery sales
poverty rate
box office sales
---------------------+----------------------------------------------Arizona  
54.03
0.14
31.31
California  
86.28
0.13
35.44
Colorado  
88.14
0.10
31.31
Connecticut  
255.20
0.08
28.12
Delaware  
145.22
0.09
27.50
District of Columbia  
369.81
0.17
29.45
Florida  
162.48
0.13
28.18
Georgia  
286.13
0.11
19.83
Idaho  
71.12
0.11
21.36
Illinois  
124.30
0.13
23.31
Indiana  
101.69
0.09
16.75
Iowa  
59.12
0.09
13.50
Kansas  
74.31
0.10
20.47
Kentucky  
160.10
0.14
16.16
Louisiana  
68.47
0.17
16.81
Maine  
122.03
0.13
15.63
Maryland  
244.49
0.07
25.32
Massachusetts  
655.47
0.10
32.35
Michigan  
170.31
0.12
25.44
Minnesota  
75.13
0.06
25.61
Missouri  
103.16
0.10
22.36
Montana  
33.39
0.14
17.91
Nebraska  
146.47
0.11
19.48
New Hampshire  
168.51
0.06
17.50
New Jersey  
249.72
0.08
27.03
New Mexico  
71.96
0.18
23.31
New York  
248.13
0.14
29.21
Ohio  
174.18
0.10
19.92
Oregon  
90.58
0.11
24.61
Pennsylvania  
156.80
0.09
20.08
Rhode Island  
193.51
0.11
19.55
South Carolina  
176.99
0.14
17.61
South Dakota  
34.53
0.12
18.11
Texas  
138.25
0.16
24.54
Vermont  
132.96
0.10
10.47
Virginia  
151.92
0.10
26.84
Washington  
72.27
0.11
29.66
West Virginia  
114.83
0.17
10.74
Wisconsin  
78.58
0.09
20.70
---------------------------------------------------------------------

25

 Table 2: Descriptive statistics by year, 1990-2002. Box office sales figures are only
available from 1991.
Lottery sales
-------------------------------------------------------------------------------year  
freq.
mean
min
max
s.d.
----------+--------------------------------------------------------------------1990  
34
95.30
19.09
259.59
59.88
1991  
35
95.01
29.13
267.33
58.24
1992  
36
104.27
30.36
309.58
60.10
1993  
37
118.92
38.05
375.91
70.41
1994  
37
132.07
36.93
434.28
81.54
1995  
37
138.33
37.27
463.65
85.99
1996  
38
136.81
35.00
497.64
91.05
1997  
38
137.35
31.42
507.29
93.45
1998  
38
139.65
33.01
514.25
94.84
1999  
38
140.36
33.43
542.52
99.67
2000  
38
142.19
31.29
610.98
107.49
2001  
38
143.72
30.51
614.87
108.00
2002  
39
151.55
33.39
655.47
111.33
-------------------------------------------------------------------------------Poverty rate
-------------------------------------------------------------------------------year  
freq.
mean
min
max
s.d.
----------+--------------------------------------------------------------------1990  
34
0.12
0.06
0.21
0.03
1991  
35
0.13
0.07
0.19
0.03
1992  
36
0.14
0.08
0.25
0.04
1993  
37
0.14
0.09
0.26
0.04
1994  
37
0.13
0.08
0.26
0.04
1995  
37
0.12
0.05
0.22
0.04
1996  
38
0.13
0.06
0.25
0.04
1997  
38
0.13
0.08
0.22
0.04
1998  
38
0.12
0.07
0.22
0.04
1999  
38
0.11
0.07
0.21
0.03
2000  
38
0.11
0.05
0.17
0.03
2001  
38
0.11
0.06
0.18
0.03
2002  
39
0.11
0.06
0.18
0.03
-------------------------------------------------------------------------------Box office sales
-------------------------------------------------------------------------------year  
freq.
mean
min
max
s.d.
----------+--------------------------------------------------------------------1991  
35
10.84
3.88
26.42
4.55
1992  
36
11.37
4.69
24.93
4.10
1993  
37
12.44
4.59
27.02
4.49
1994  
37
13.13
5.92
25.74
4.19
1995  
37
13.81
6.31
27.43
4.44
1996  
38
14.41
6.45
27.24
4.48
1997  
38
15.65
6.52
29.97
5.18
1998  
38
17.94
7.98
27.93
4.93
1999  
38
19.47
9.54
30.06
5.21
2000  
38
19.25
9.12
30.05
5.27
2001  
38
20.23
9.12
32.66
5.79
2002  
39
22.65
10.47
35.44
6.03
-------------------------------------------------------------------------------26

 Table 3: Correlation table.
  lottery sales
box office sales
-------------------+-------------------------------lottery sales
 
1.0000
box office sales
 
0.4263
1.0000

  poverty rate
above 50% poverty
below 50% poverty
unemployment rate disposable income
-------------------+------------------------------------------------------------------------------------------poverty rate
 
1.0000
above 50% poverty  
0.9616
1.0000
below 50% poverty  
0.8700
0.7014
1.0000
unemployment rate  
0.5792
0.5646
0.4902
1.0000
disposable income   -0.3432
-0.4643
-0.0572
-0.2951
1.0000

27

 Table 4: Effect of poverty on lottery sales per capita. Year and state indicators
are included but not reported.
Per. below poverty

lott. sales per capita
(1)
230.088
(89.037)∗∗

Per. white
Per. black
Per. male
Per. married
Per. w/ children
Per. female head of household
Average age
Average age squared
Per. w/ less than high school education
Per. w/ high school education but no college
Disposable income per capita
Unemployment rate
Obs.
R2
F statistic
No. states

483
.319
15.501
39

Absolute value of t statistics in parentheses
* significant at 5%; ** significant at 1%

28

lott. sales per capita
(2)
244.959
(92.956)∗∗
-184.385
(107.152)
187.156
(105.93)
166.698
(216.834)
-219.19
(151.505)
7.695
(28.105)
-53.039
(218.792)
-4.212
(69.808)
.084
(.776)
-53.988
(112.496)
65.558
(98.726)
17.389
(2.475)∗∗
-726.591
(196.339)∗∗
483
.446
13.472
39

 Table 5: Estimations controlling for entertainment value of lottery purchases and
state tax revenues. Year and state indicators are included but not reported.
Per. below poverty

movie sales per capita
(1)
-3.042
(4.81)

-27.96
(5.498)∗∗
-4.227
(5.364)
-25.044
(11.176)∗
4.364
(7.922)
1.929
(1.444)
-8.427
(11.29)
5.613
(3.589)
-.066
(.04)
29.714
(5.929)∗∗
3.22
(5.056)
.353
(.13)∗∗
8.977
(10.207)

lott.˜sales per capita
(2)
192.31
(90.688)∗
-1.809
(.959)
-244.56
(107.014)∗
167.235
(101.16)
171.859
(211.967)
-238.515
(149.345)
28.508
(27.279)
-69.45
(212.892)
8.191
(67.843)
-.069
(.754)
13.836
(115.294)
49.731
(95.319)
15.699
(2.478)∗∗
-822.639
(192.535)∗∗

449
.908
157.996
39

449
.434
11.819
39

Movie sales per capita
Per. white
Per. black
Per. male
Per. married
Per. w/ children
Per. female head of household
Average age
Average age squared
Per. w/ less than high school education
Per. w/ high school education but no college
Disposable income per capita
Unemployment rate
Log total taxes
Log population
Obs.
R2
F statistic
No. states
Absolute value of t statistics in parentheses
* significant at 5%; ** significant at 1%

29

lott.˜sales per capita
(3)
363.146
(90.82)∗∗

-156.442
(117.488)
92.541
(113.264)
267.951
(221.395)
-47.06
(152.146)
2.131
(23.186)
180.654
(222.567)
15.821
(71.8)
-.183
(.798)
87.88
(107.273)
-14.05
(98.918)
6.743
(1.182)∗∗
-819.327
(148.349)∗∗
-7.883
(4.021)∗
-169.148
(45.572)∗∗
470
.388
17.595
38

 Table 6: Effect of extreme poverty on lottery sales. Year and state indicators are
included but not reported.
Per. below poverty

lott. sales per capita
(1)
244.959
(92.956)∗∗

Per. under 50% of poverty line

lott. sales per capita
(2)

lott. sales per capita
(3)

110.364
(192.314)

55.458
(191.751)
318.03
(113.217)∗∗
-184.444
(107.117)
192.471
(105.999)
179.565
(217.061)
-224.52
(151.529)
8.293
(28.1)
-49.049
(218.748)
-4.616
(69.785)
.09
(.776)
-43.704
(112.827)
65.33
(98.694)
17.549
(2.478)∗∗
-712.737
(196.657)∗∗
483
.447
13.011
39

Per. in upper 50% of poverty range
Per. white
Per. black
Per. male
Per. married
Per. w/ children
Per. female head of household
Average age
Average age squared
Per. w/ less than high school education
Per. w/ high school education but no college
Disposable income per capita
Unemployment rate
Obs.
R2
F statistic
No. states

-184.385
(107.152)
187.156
(105.93)
166.698
(216.834)
-219.19
(151.505)
7.695
(28.105)
-53.039
(218.792)
-4.212
(69.808)
.084
(.776)
-53.988
(112.496)
65.558
(98.726)
17.389
(2.475)∗∗
-726.591
(196.339)∗∗
483
.446
13.472
39

Absolute value of t statistics in parentheses
* significant at 5%; ** significant at 1%

30

-221.419
(107.175)∗
180.218
(106.777)
188.791
(218.814)
-208.769
(152.665)
16.949
(28.16)
10.864
(219.489)
-10.274
(70.328)
.149
(.782)
16.851
(111.655)
86.916
(99.2)
17.515
(2.499)∗∗
-649.214
(196.952)∗∗
483
.437
13.002
39