Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
MARCH MADNESS: NCAA TOURNAMENT PARTICIPATION
AND COLLEGE ALCOHOL USE
DUSTIN R. WHITE , BENJAMIN W. COWAN and JADRIAN J. WOOTEN∗
While athletic success may improve the visibility of a university to prospective
students and thereby benet the school, it may also increase risky behavior in the current
student body. Using the Harvard School of Public Health College Alcohol Study, we nd
that a school’s participation in the NCAA Basketball Tournament is associated with a
47% increase in binge drinking by male students at that school. Additionally, we nd
evidence that drunk driving increases by 5% among all students during the tournament.
(JEL I12, I23, Z28)
I. INTRODUCTION
Alcohol consumption is one of the primary
public health concerns on college campuses in the
United States (NIH-NIAAA 2016; theAmethyst-
Initiative.org 2015). Of particular concern is the
amount of binge drinking—dened as ve or
more drinks at one time—among students. Binge
drinking is associated with increased rates of
drunk driving, sexual assault, and other nega-
tive outcomes among young people (Miller et al.
2007). A survey conducted by Glassman et al.
(2010) revealed that 16% of respondents aged
18–24 consumed more than double the binge
drinking threshold on college football game days
and that 36% of respondents reported binge
drinking during game day festivities.
Studies of alcohol consumption associ-
ated with athletic events have been conducted
previously, but these focus almost entirely on a
single occasion or a single school (Neal et al.
2005; Neal and Fromme 2007). In this paper, we
∗We thank Toben Nelson for providing the College Alco-
hol Study data for this project. We also thank participants
in the Southern Economic Association Conference, Beero-
nomics Conference, and the SES Seminars at Washington
State University for their feedback. All errors are the authors’
alone.
White: Assistant Professor, Economics, University of
Nebraska at Omaha, Omaha, NE 68182. Phone 402-554-
3303, Fax 402-554-2853, E-mail drwhite@unomaha.edu
Cowan: Associate Professor, Economics, Washington State
University, Pullman, WA 99164; NBER, Cambridge, MA
02138. Phone 509-335-2184, Fax 509-335-1173, E-mail
ben.cowan@wsu.edu
Wooten: Assistant Teaching Professor of Economics, Eco-
nomics, The Pennsylvania State University, University
Park, PA 16802. Phone 814-865-7352, Fax 814-863-4775,
E-mail jjw27@psu.edu
explore the impact of the NCAA Men’s Basket-
ball Tournament on student alcohol consumption
during the 1993, 1997, 1999, and 2001 seasons at
more than 40 schools using the Harvard School
of Public Health College Alcohol Study (CAS).
We isolate the effect of tournament participation
on student alcohol consumption by examining
how patterns of consumption change around
the tournament for schools that do and do not
participate in a given year.
Intercollegiate athletics have played a part in
the college experience for more than a century;
currently, over 1,200 colleges and universities
are members of the National Collegiate Athletic
Association (NCAA 2010). The broadcast rights
deal signed by the NCAA for the Division I
Men’s Basketball tournament ($10.8 billion
over 14 years) provides proof of the tremendous
demand both on and off campus for college
basketball (O’Toole 2010). The increasing pop-
ularity of intercollegiate athletics—particularly
football and men’s basketball— drives universi-
ties to invest more in their athletic programs each
year (Berkowitz 2014).
Athletic programs have the potential to
increase the exposure of an institution and
thereby increase the quality of applicants to
their school. Toma and Cross (1998) nd that
winning a national championship in either men’s
basketball or football increases the number of
ABBREVIATIONS
CAS: College Alcohol Study
DD: Difference-in-Differences
GPA: Grade Point Average
NCAA: National Collegiate Athletic Association
1
Contemporary Economic Policy (ISSN 1465-7287) doi:10.1111/coep.12425
© 2019 The Authors. Contemporary Economic Policy published by Wiley Periodicals, Inc. on behalf of the Western Economic Association International.
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution
in any medium, provided the original work is properly cited, the use is non-commercial and no modications or adaptations are made.
2 CONTEMPORARY ECONOMIC POLICY
applicants to a school, and that these increases
were both absolute and relative to peer insti-
tutions. Pope and Pope (2009, 2014) nd that
improved performance in football or basketball
in a given year can increase the number of
Scholastic Assessment Test scores submitted to
a school by up to 10%. A larger pool of appli-
cants allows a school to be more selective when
making admissions decisions.
Lindo, Swensen, and Waddell (2012) exam-
ine variation over time in athletic performance of
football teams at the University of Oregon and
nd that academic performance suffers during
years in which the team performs exceptionally
well. This difference is most pronounced among
males. Responses to a student survey in Lindo,
Swensen, and Waddell (2012) suggest that alco-
hol consumption plays a role in the decline of
grades. This agrees with previous results that nd
alcohol consumption is associated with worse
grades among students and that peer effects are
particularly inuential in determining the extent
of consumption among college students (Eisen-
berg, Golberstein, and Whitlock 2014; Lindo,
Swensen, and Waddell 2013). In a follow-up arti-
cle, Hernández-Julián and Rotthoff (2014) con-
rm the ndings of Lindo, Swensen, and Waddell
(2012) by observing a similar effect at Clemson
University, although they nd that females suf-
fer a larger decrease in academic performance.
While better athletic performance may have a
positive effect for the university in terms of dona-
tions and improving the quality of future classes,
there appears to be an adverse effect on the cur-
rent student body.
More recently, Lindo, Siminski, and Swensen
(2018) investigate the incidence of sexual assault
on college football game days. They nd that
reported sexual assaults increase signicantly
surrounding these events and that alcohol-related
arrests rise as well. The authors suggest that
increased alcohol consumption and a college
party culture are the primary drivers of the spike
in the incidence of sexual assault surrounding
college football games.
In this paper, we examine self-reported alco-
hol use during a major college sporting event
at schools that do and do not participate in the
tournament in a given year using a difference-in-
differences (DD) framework. We are thus able
to study a direct link between college sports and
drinking. Furthermore, while previous papers
have largely focused on college football, we
examine the effects of a post-season basket-
ball tournament that involves many schools
simultaneously and takes place in the spring
rather than the fall.1
Identication in our paper is established by
comparing the change in drinking that occurs
among students at participating (treatment)
schools during the time of the tournament (mid-
March to early April in our sample years) to any
change that occurs at nonparticipating (control)
schools around that time. Thus, time-invariant
differences in alcohol use across schools by
participation status as well as seasonal changes
in drinking that are common to all college stu-
dents are “differenced out” of our estimated
treatment effect.
In the next section of this paper, we present
our empirical model to estimate the impact of
NCAA tournament participation and games on
alcohol consumption. We then provide a sum-
mary of the data used in our analysis. In the
results section, we show that tournament par-
ticipation increases the binge drinking rate of
male students by approximately 47% (relative to
the average binge rate among males at tourna-
ment schools). Furthermore, we nd that students
are more likely to self-report drunk driving dur-
ing the tournament. We conclude by consider-
ing the meaning of these results and their pol-
icy implications.
II. EMPIRICAL MODEL
The NCAA Men’s Basketball Tournament
extends the basketball season for participating
teams by up to six games during the years studied
in this paper. The winner of the tournament is
deemed the national champion for that season.2
These games are watched or attended by tens of
millions of fans.3For college students, if alcohol
consumption is in fact complementary to viewing
or attending (important) games by the college’s
team (as is suggested by Lindo, Swensen, and
Waddell 2012), then drinking among students
will rise when the school’s team participates in
tournament games.
Using reports of alcohol consumption tak-
ing place between January and May in each of
1. This is a necessity of working with the CAS data, as
surveys are collected each spring.
2. Information on the selection of participating teams can
be found at http://www.ncaa.com/content/di-principles- and-
procedures-selection
3. In our sample years, television viewership for the
national championship game alone was between 23 and 33
million people: http://www.sportsmediawatch.com/ncaa-
nal-four-ratings-history-most-watched-games-cbs- tbs-
nbc/
WHITE, COWAN & WOOTEN: MARCH MADNESS 3
our sample years, we can observe alcohol con-
sumption of students during the NCAA tourna-
ment. We exploit both time differences (whether
a respondent’s survey covers the time period for
the tournament) as well as school differences
(whether the respondent was attending a tourna-
ment school in that year) in a DD framework to
identify the effect of NCAA tournament partici-
pation by an individual’s institution on their alco-
hol consumption.
Our regression model is expressed as follows:
Dismy =β
0+β
1•treatedismy +γ
sy
+μ
m+λ•Xismy +ϵ
ismy
Drepresents the drinking behavior of interest,
and i,s,m, and yare indices for individual,
school, month, and year, respectively. We focus
primarily on the number of occasions that an
individual binged (measured as at least ve drinks
at one time) over a 2-week period, but consider
other measures of alcohol use as well.
Our treatment variable, treatedismy, is a binary
variable that is equal to one if the individual’s
retrospective survey covers the date(s) of one or
more games played by their school team in the
tournament that year and is zero otherwise. Thus,
it is zero for individuals attending schools that did
not participate in the tournament in a given year
as well as for individuals attending tournament
schools but whose survey did not cover any of
that school’s tournament games. μmrepresents
month xed effects, and γsy represents school by
year xed effects that account for school-specic
trends across years. Xismy is the vector of all other
individual and school explanatory variables, and
ϵismy is a disturbance term.
Our main parameter of interest, β1, is identi-
ed through any change in drinking that occurs in
tournament schools during the time of the tourna-
ment compared to other times within the year, all
relative to the same difference at nontournament
schools.4
Time-invariant differences in alcohol con-
sumption by institution are not a threat to
identication: if students at tournament schools
tend to drink more than students at nontourna-
ment schools, but this difference does not change
during the tournament, we will fail to reject
the hypothesis that β1=0, and the difference
4. We include a full set of month effects in our model
rather than an indicator for “during tournament.” We also
performed specications with unique dummies for all month-
year pairs as a robustness check (the results are very similar
to those of our baseline model; see Section IV.C).
will be reected in the school-year xed effects.
Similarly, seasonal changes in drinking behavior
across all schools cannot account for a positive
β1, since those changes will be absorbed by the
month xed effects. We note that student drink-
ing could rise during spring break (when school
is out), but we do not observe that tournament
schools are more likely to hold their spring break
during the time of the tournament.5
Xismy includes factors that are commonly
associated with alcohol consumption by college
students, according to Glassman et al. (2010):
race, gender, age, membership in a fraternity or
sorority, and year in school. Xismy also includes
measures of grade point average (GPA), mari-
tal status, men’s basketball regular season win
percentage, and athletic conference (because
conference afliation varies for some teams
during our study period, it is not collinear with
school xed effects). As we show in the results
section, the inclusion of these variables makes
little difference in our estimates of interest.
III. DATA
We use the Harvard School of Public Health
CAS to examine the relationship between col-
lege basketball postseason play and the amount
of alcohol consumed by an institution’s student
body. CAS constitutes a nationally representa-
tive sample of full-time college students attend-
ing 4-year institutions in 1993, 1997, 1999, and
2001.6Forty-one of the institutions in the data
participated in NCAA Division I athletics dur-
ing the sample frame; students from these insti-
tutions make up the sample used for this paper.
The names of the schools included in our sample
can be found in Table A1 of Appendix S1, Sup-
porting Information. In each year, CAS students
were sent a mail survey, which included a series
of questions regarding alcohol and other drug use,
5. In 2015, 79% of spring breaks across 529 institutions
occurred entirely within the month of March and 97% over-
lapped March (STSTravel 2014). Though we do not know the
spring break dates of all the school-year pairs in our data,
we think it is unlikely that schools schedule spring break to
overlap the NCAA tournament from year to year. Thus, our
analysis in Section IV.C is unlikely to be biased by any dif-
ferences in spring break dates by tournament status, and we
provide alternative specications to support this opinion.
6. For details on the survey design, see Wechsler et al.
(2002). Cowan and White (2015) provide a comparison of
CAS with Monitoring the Future and the National Longitudi-
nal Survey of Youth (1997 cohort), and nd similar drinking
patterns and other characteristics of college students across
the data sources.
4 CONTEMPORARY ECONOMIC POLICY
TABLE 1
Mean Characteristics for Students and Schools by Tournament Status
Characteristics
Nontournament
Colleges
Tournament
Colleges
Nontournament
Respondents
Tournament
Respondents
Number of binge occasions in past 2 weeks 1.370 1.526 1.477 1.812
Drank last month 0.651 0.697 0.709 0.632
Number of drinks in past month 24.142 27.234 26.078 33.980
Survey overlaps any of college team’s tournament games 0 0.146 0 1
Number of tournament games overlapped by survey 0 0.235 0 1.606
Men’s basketball season win percentage 40.617 67.586 67.832 66.156
Age 21.043 20.676 20.685 20.627
Belongs to fraternity/sorority 0.153 0.207 0.210 0.190
Married 0.098 0.068 0.069 0.060
Female 0.577 0.576 0.575 0.584
1993 respondent 0.268 0.591 0.671 0.125
1997 respondent 0.315 0.115 0.054 0.471
1999 respondent 0.297 0.230 0.216 0.315
2001 respondent 0.121 0.064 0.060 0.089
Freshman 0.204 0.219 0.216 0.236
Sophomore 0.202 0.217 0.213 0.238
Junior 0.248 0.237 0.238 0.229
Senior 0.236 0.228 0.231 0.209
White 0.767 0.857 0.862 0.829
Black 0.064 0.036 0.035 0.043
Asian 0.075 0.048 0.048 0.051
Other race 0.093 0.059 0.055 0.077
GPA 3.159 3.151 3.147 3.177
N21,987 3,990 3,406 584
Notes: Bold values indicate that the difference in means is signicant at the 5% level. Tournament respondents are those
individuals whose survey overlaps at least one of their own institution’s team’s games in the NCAA tournament. Nontournament
Respondents are individuals at Tournament Colleges whose survey covers dates either before or after the tournament. Students at
Tournament Colleges attended institutions whose NCAA Division I Men’s Basketball Team participated in that season’s NCAA
tournament. Students at Nontournament Colleges also attended institutions with an NCAA Division I Men’s Basketball Team,
but their team did not participate in the NCAA tournament in that year.
experiences in college, and limited demographic
and family background information.
Of the 31,184 student observations attending
Division I institutions, 25,977 have nonmissing
values for all variables used in this paper.7This
is our regression sample. Student surveys were
distributed early in the year and returned over the
next several months: 4,396 surveys were returned
in February, 7,804 in March, 11,054 in April, and
2,527 in May, accounting for about 99% of our
sample over the 4 years.
According to Table 1, the majority of respon-
dents reporting during a year in which their
school’s team participated in the NCAA
Tournament are surveyed in 1993 (approxi-
mately 60% of such responses). The bulk of
7. In order to determine that missing values are not more
prevalent among tournament or nontournament schools, we
calculate the number of observations with missing values for
each group, and present those results in Table A2 of Appendix
S1. Nontournament schools have a higher frequency of obser-
vations with missing values, with March responses (prior to
the tournament) being the primary driver of this difference.
nontournament responses occurred in the 1997
and 1999 surveys, simply due to the fact that
fewer schools that participated in the CAS were
invited to participate in the NCAA tournament
during those survey years, especially in compari-
son to 1993. Because the primary response years
for each type of respondent differ, it may also
be that the response dates within years for each
type of respondent also vary. Table 2 provides
the average response date for tournament and
nontournament school-years. There is certainly
variation between the response dates, but the
average response date for tournament respon-
dents is within one standard deviation of the
response date for nontournament respondents in
each year (and the overall sample), mitigating
concerns that there are systematic differences
in response patterns between tournament and
nontournament schools.
A limitation of our analysis is that we only
know the date a survey was processed by CAS
administrators; we do not know the exact dates
on which individuals completed their survey. As
WHITE, COWAN & WOOTEN: MARCH MADNESS 5
TABLE 2
Average Response Date by Year and Tournament Status
1993 1997 1999 2001 Overall
Average response date — tournament school-years March 5 April 30 April 2 April 12 March 17
(17 days) (10 days) (18 days) (26 days) (25 days)
Average response date — nontournament school-years March 12 April 27 April 8 April 4 April 2
(21 days) (19 days) (23 days) (27 days) (26 days)
Note: Standard deviations are provided in parentheses.
a result, when questions refer to retrospective
alcohol consumption—such as binge drinking
over the previous 2 weeks— we do not know
the exact period of time to which a student
is referring.
In lieu of information on the dates over which
each student is measuring their drinking, we
assume that 4 weeks pass between the completion
of the survey by the individual and its process-
ing date. This accounts for any lag between the
date a student lled out the survey and the date
they mailed it, time spent in the mail, and any
lag between receipt of the survey and its process-
ing by CAS administrators. The question “How
many times did you binge drink in the past 2
weeks?” would therefore be interpreted as “How
many times did you binge drink between 28 and
42 days prior to the processing date of the sur-
vey?,” and questions regarding drinking in the
previous month would be interpreted with respect
to the time period 28–58 days prior to the pro-
cessing date.
Since the choice of a 4-week lag to estimate
the dates of the retrospective drinking period is
somewhat arbitrary, we also estimated our mod-
els using 2-, 3-, 5-, and 6-week lags. Figure 1
shows how our point estimate of interest (the
effect of the tournament on binge drinking occa-
sions) and corresponding condence intervals
change with the choice of lag. As seen in the
gure, point estimates diminish as the lag is
moved away from 4 weeks in either direction,
with 2- and 6-week lags yielding a near-zero
coefcient. Figures 2 and 3 show that this pattern
holds among male respondents, but that the binge
drinking behavior of female students appears to
have no relationship to the tournament window.8
8. The following is a quote from Wechsler et al. (2002)
regarding the mailing and collection of CAS surveys in all
four survey years: “Following the same practice as that used
in the 3 prior surveys, we mailed questionnaires directly to
students beginning in February and sent 3 separate mailings
within a minimum span of 3 weeks. The initial mailing con-
sisted of a letter of invitation to participate in the study and
It is unfortunate that we cannot corroborate
our assumption about the lag between survey
completion and survey processing, especially
because our results are sensitive to this assump-
tion. In all years, there are many more responses
received prior to the NCAA tournament than
after it. It is comforting, however, to nd that
the results (for men) diminish symmetrically
with a shorter or longer lag, since this means
that shifting individuals who respond late to
the survey into the “treatment” group does not
produce a different result than shifting treat-
ment toward individuals responding somewhat
earlier. Thus, concerns about self-selection of
response times and its own correlation with
drinking behavior, even if that relationship
is correlated with tournament school status,
are diminished.
A. Drinking Measures
We focus on three drinking measures to eval-
uate the impact of the NCAA Men’s Basketball
Tournament on alcohol consumption patterns:
(1) the number of binge drinking incidents
reported in the past 2 weeks, (2) whether or
not an individual reported drinking at all in the
past month, and (3) the number of alcoholic
beverages an individual reported consuming
in the past month. Two of our dependent vari-
ables are quasi-continuous measures based
on binned responses to questions regarding
alcohol consumption, and the third depen-
dent variable is a binary value based on these
original measures.
Our measure of binge drinking is based on
respondents’ answer to the question “Think back
a questionnaire. We followed this mailing with a reminder
postcard and a separately mailed second questionnaire. Mail-
ings were different for each school, and we scheduled them
to avoid the period immediately preceding and following
spring break to capture behavior that occurred on campus
and to avoid responses that reected behavior during spring
vacation.”
6 CONTEMPORARY ECONOMIC POLICY
FIGURE 1
Varying the Response Lag (w/95% CI)
FIGURE 2
Varying the Response Lag among Male Respondents (w/95% CI)
WHITE, COWAN & WOOTEN: MARCH MADNESS 7
FIGURE 3
Varying the Response Lag among Female Respondents (w/95% CI)
over the last two weeks. How many times have
you had ve or more drinks in a row?” Potential
answers are 0, 1, 2, 3 to 5, 6 to 9, and 10+
times. For answers with single values, that value
is coded as the number of binge occasions in the
past 2 weeks. For bins with multiple values, the
midpoint is taken, and for individuals choosing
the 10+bin, a value of 10 binge occasions in the
past 2 weeks is assigned.
The number of drinks consumed in the past
month (30 days) is generated by combining the
values of two questions. First, “On how many
occasions have you had a drink of alcohol in the
past 30 days?” Responses to this question were
categorized as 0, 1 to 2, 3 to 5, 6 to 9, 10 to 19,
20 to 39, and 40+occasions. Where bins were
bounded, we coded the number as the midpoint
value. When using the top code, a response was
coded as 40 drinks. The next question, “In the
past 30 days on the occasions when you drank
alcohol, how many drinks did you usually have?”
had possible responses of 1, 2, 3, 4, 5, 6, 7,
8, 9+drinks per occasion. Only the top code
needed clarication, and any response using the
top code was assigned a value of 9 drinks per
occasion. The number of drinks consumed in the
past month was then calculated as the product of
the number of days an individual drank in the past
30 days and the number of drinks an individual
typically consumed when they did drink.
The binary variable indicating any alcohol
consumption in the past month (30 days) is gener-
ated from the question about the number of occa-
sions on which the respondent consumed alcohol.
Where this value is 0, the binary variable is also
0; where the response was a value greater than 0,
the binary variable was coded as 1.
B. Preliminary Data Evaluation
Summary statistics for variables included
in our analysis by tournament school status
(whether an individual’s team participated in
the tournament in that year, or not) are pro-
vided in Table 1. Students at tournament schools
are more likely to have consumed alcohol
and, on average, engaged in more drinking
and binge drinking than students at nontour-
nament schools. There are also signicant
differences in the demographics of these two
groups, with students from tournament schools
being whiter, younger, less frequently married,
and more likely to be members of fraternities
and sororities.
Table 1 also shows how respondents whose
survey covers games played by their institution’s
8 CONTEMPORARY ECONOMIC POLICY
TABLE 3
Effects of Own-Institution NCAA Tournament Participation on Student Drinking Behaviors
All Observations Males Females
Number of
Binges Drink?
Number of
Drinks
Number of
Binges Drink?
Number of
Drinks
Number of
Binges Drink?
Number of
Drinks
Primary specication
(n=25,438)
0.353** −0.001 5.041** 0.724** 0.037 10.947** 0.062 −0.031 0.645
(0.122) (0.020) (1.331) (0.206) (0.031) (4.046) (0.146) (0.027) (1.923)
Including school*month FE’s
(n=25,438)
0.334** 0.03 0.912 0.766** 0.069** 4.867 0.022 0.002 −2.288
(0.121) (0.017) (1.061) (0.205) (0.026) (3.177) (0.144) (0.023) (1.576)
Including month*year FE’s
(n=25,438)
0.297** −0.04** 3.669** 0.596** −0.007 7.179** 0.062 −0.064** 0.818
(0.106) (0.014) (0.919) (0.179) (0.022) (2.696) (0.127) (0.02) (1.397)
Primary specication, tournament
school subsample (n=9,774)
0.209** −0.001 2.919** 0.435** 0.023 6.985** 0.011 −0.024 −0.402
(0.077) (0.013) (0.872) (0.124) (0.019) (2.391) (0.093) (0.018) (1.256)
Primary specication, including
spring break (n=22,970)
0.321*−0.009 5.682** 0.774** 0.046 12.713** −0.04 −0.053 0.286
(0.138) (0.022) (1.417) (0.23) (0.034) (4.187) (0.165) (0.03) (2.148)
Notes: Standard errors are robust to heteroskedasticity and clustered at the institution level. Other controls included in the regressions but not
shown include all those described in Section IV.A.
∗Signicant at the 5% level.
∗∗Signicant at the 1% level.
team in the tournament compare to other respon-
dents at tournament schools (who returned their
survey at different times of the year).9We actu-
ally see a lower probability of drinking among
during-tournament respondents, but that is
accompanied by more binge drinking and drinks
over the past month, indicating that during-
tournament respondents consume more alcohol
conditional on drinking at all. Other than per-
centages of respondents by year (which indicates
that responses covering the tournament were
more common in some years than others), there
are few signicant differences across other vari-
ables between during-tournament respondents
and other respondents at tournament schools.
This assuages concerns that those who choose
to return their survey at a date that indicates
tournament overlap are selected on unobserved
factors that are correlated with drinking behavior.
IV. RESULTS
A. Primary Specication Regression Results
The results from our primary regression anal-
ysis are presented in the rst row of Table 3. All
9. Table 1 indicates that the bulk of Tournament-
observing respondents at tournament schools reported during
the 1997 and 1999 surveys. According to the information pro-
vided by the surveyors in Wechsler et al. (2002), this is due
primarily to randomness in the response rates of the surveyed
individuals. This is unlikely to affect our results, since each
specication will include year xed effects to account for dif-
ferences across time.
regressions are performed using ordinary least
squares with robust standard errors clustered
by school. Each column represents a differ-
ent dependent variable (drinking measure).10
Control variables for this specication include
team win percentage, membership in a frater-
nity or sorority, marital status, gender, year in
school, race, GPA, age, and xed effects for
athletic conference, month, and school-year
interaction terms. Detailed regression results
for our primary specication are presented in
Table A3 of Appendix S1. “Treated” respon-
dents attended a school that participated in the
NCAA tournament during their survey year,
and had surveys that covered the date(s) of
at least one of their school team’s tournament
games.
We nd that treatment raises the number
of binge occasions in the past 2 weeks by
roughly 0.35, or 23% at the mean for tourna-
ment schools.11 The number of drinks in the
past month rises by 5 (a 19% increase). Treat-
ment appears to have no signicant effect on
the likelihood that students drank in the past
month.
10. While not reported in the tables, our results are robust
to using a binary measure of “binge drank in the past 2 weeks”
as a dependent variable. These results are available upon
request.
11. While the results presented in Table 3 employ linear
models, the results are robust to using other estimation tech-
niques. The results of negative binomial and logistic regres-
sion models are available in Table A6 of Appendix S1.
WHITE, COWAN & WOOTEN: MARCH MADNESS 9
As presented in Table A3 of Appendix S1,
other coefcients in our models are gener-
ally as expected: students who are members
of fraternities and sororities drink more alco-
hol; married and female students drink less
(DeSimone 2007). Freshmen are less likely to
have drunk at all in the previous month but are
more likely to binge. White students consume
substantially more alcohol than “other race”
students, who in turn consume more than black
and Asian students. A higher GPA is associated
with less drinking (according to any measure).
Lastly, the fact that the men’s basketball team
regular season win percentage has little effect
on drinking behaviors suggests that participation
in the tournament, rather than the quality of
the team itself, drives the observed increase in
alcohol consumption.
B. Effects by Gender
When results are estimated on samples of
men and women separately, stark differences
emerge. As seen in Table 3, the measured
increase in binge drinking and number of drinks
consumed is concentrated almost exclusively
among male students.
The observed increase in the number of binge
drinking occasions in the past 2 weeks for men
represents an approximately 47% increase at
the mean among students at tournament schools
(which is just under two binge occasions). Males
report consuming almost 11 additional drinks in
the past month when their college team partici-
pated in the tournament (a 40% increase at the
mean). Because past and emerging research sug-
gests that drinking surrounding sporting events is
associated with highly negative outcomes, these
results are important from a public health per-
spective (Lindo, Siminski, and Swensen 2018).
C. Robustness
In this subsection, we examine the robustness
of our main results in several different ways.12
We begin by examining how sensitive our treat-
ment effect estimate is to the set of controls in
the model.
The rst three rows of Table 3 show, respec-
tively, results from models including (1) the
12. In addition to the robustness tests described in this
section, we also include results from a falsication analysis
and analysis of various subgroups of schools in Table A4 of
Appendix S1. Each of these analyses supports the generality
of our ndings above.
primary specication, which includes separate
dummies for each school-year pair (school and
year xed effects are interacted), (2) interacted
xed effects for each school-month pair (interact-
ing school and month rather than school and year
xed effects), and (3) interacted xed effects for
each year-month pair (interacting year and month
rather than school and year xed effects). The
primary specication controls for the possibility
that some schools experience a different trend
in alcohol use over our sample years than oth-
ers, which could be correlated with tournament
status. The second accounts for the possibility
of different seasonal variation between schools.
The third is the most exible way to account
for differences in drinking over time (within
and across years). Each specication afrms the
results from our primary specication.
As discussed in Section II, a potential threat
to the interpretation of our results is spring break,
a time (usually a week) off from school given
to students at most U.S. institutions, typically
during the month of March. Spring break at
many colleges likely coincides with part of the
NCAA tournament, and alcohol use increases
during spring break, particularly among males
(Lee, Lewis, and Neighbors 2009).13
Because the vast majority of schools hold
spring break during the month of March, we
expect that any “spring break” effect on drinking
is likely to be absorbed by the month effects in our
model. This may confound our results if spring
break is more likely to be scheduled over NCAA
tournament dates at tournament schools than at
nontournament schools. Since some schools con-
sistently play in the tournament each year while
others rarely do, it is possible that these teams
schedule spring break during the tournament
while others do not, which would challenge our
ability to accurately estimate a treatment effect.
To counter this possibility, we estimate our
model with only students from schools whose
team plays in the tournament in at least 1 year
of our data using only those students whose sur-
vey overlaps the tournament dates in a given
year (thus, month effects are removed from the
13. Another potential confounding factor in our analysis
is the possibility that students do not consume more alcohol
overall when they observe the NCAA tournament. It is possi-
ble that they instead change the time period during which they
consume alcohol. In Table A7 of Appendix S1, we attempt to
detect intertemporal substitution of alcohol consumption, and
nd little evidence that students consume less alcohol before
or after the NCAA tournament in order to compensate for con-
sumption during the tournament window.
10 CONTEMPORARY ECONOMIC POLICY
FIGURE 4
Spring Break Dates by School-Year
model). Identication in this model comes from
year-to-year variation in drinking at schools who
reach the tournament in some years but not oth-
ers. Because academic calendars are set well in
advance of any invitation to participate in the
NCAA tournament, we think it is highly unlikely
that a college’s spring break is more likely to
occur during the tournament in years in which
their team participates than in years in which they
do not.
The results of this exercise are contained in the
fourth row of Table 3. Treatment is again dened
based on overlap with any tournament games (as
in all other rows of the table). The point estimates
are smaller in magnitude than in the primary
specication, and some are less precisely esti-
mated (owing to the much reduced sample size).
The broad picture is remarkably similar, how-
ever: binge drinking and total drinking rise during
the tournament when a student’s own school team
participates, and this effect is again focused on
male students.
Finally, we successfully obtained detailed
spring break dates during our study years from
34 schools,14 in order to include a control for
spring breaks in our regression analysis. A his-
togram of spring break dates by the tournament
status of schools is presented in Figure 4, and
shows that the spring break dates of schools
participating in the NCAA tournament closely
14. Some schools indicated that they do not maintain
archives of academic calendars that reach back 25 years,
while others simply did not respond to our request for
information when past academic calendars were unavailable
online.
resemble the distribution of spring break dates
for nonparticipating school-year observations.
In order to control for spring break in our
regression analysis, we created a binary vari-
able for whether or not the survey response
window overlapped the school’s reported spring
break in respondents’ survey year. The results of
the specication including this measure can be
found in the nal row of Table 3. These estimates
are nearly identical to our primary specication.
The evidence of each of our specications tar-
geting potential interference resulting from the
timing of spring break suggests that the effect
of spring break on our measurement of alco-
hol consumption during the NCAA tournament
is negligible.
D. Effects of Individual Rounds
In the rst row of Table 4, the denition of
treatment is changed from a binary variable indi-
cating whether an individual’s survey overlapped
with any games played by their college team in
the tournament to the number of the team’s games
(rounds) overlapped by the student’s survey. The
results are qualitatively and quantitatively con-
sistent with those from our baseline treatment
denition, since the average number of rounds a
team plays in the tournament is a little less than
two games.
The rest of the rows in Table 4 display
the effects of overlapping exactly one round,
two rounds, and three or more rounds sepa-
rately (within the same regression). Though the
increases are not perfectly monotonic in all cases,
and some individual coefcients are no longer
WHITE, COWAN & WOOTEN: MARCH MADNESS 11
TABLE 4
Effects of Own-Institution NCAA Tournament Rounds on Student Drinking Behaviors, by Gender
All Observations Males Females
Number of
Binges Drink?
Number of
Drinks
Number of
Binges Drink?
Number of
Drinks
Number of
Binges Drink?
Number of
Drinks
Treated (number of rounds) 0.196*−0.003 2.585** 0.414** 0.025 6.677** −0.001 −0.031 −0.918
(0.077) (0.013) (0.847) (0.123) (0.019) (2.356) (0.091) (0.018) (1.231)
One round observed 0.290*−0.011 3.734*0.520*0.027 8.846 0.110 −0.039 −0.199
(0.147) (0.023) (1.463) (0.25) (0.035) (4.702) (0.178) (0.03) (2.044)
Two rounds observed 0.303 0.039 8.311*0.821*0.059 13.921 −0.202 −0.002 2.437
(0.231) (0.049) (3.537) (0.389) (0.071) (8.526) (0.267) (0.07) (6.815)
Three plus rounds observed 0.586 −0.01 4.301 1.139*0.066 16.548 0.088 −0.092 −6.698
(0.331) (0.051) (3.403) (0.501) (0.072) (9.023) (0.431) (0.074) (4.742)
N25,438 25,438 25,438 10,845 10,845 10,845 14,593 14,593 14,593
R2.179 .124 .148 .165 .112 .129 .149 .132 .126
Notes: Standard errors are robust to heteroskedasticity and clustered at the institution level. Other controls included in the
regressions but not shown include all those described in Section IV.A.
∗Signicant at the 5% level.
∗∗Signicant at the 1% level.
statistically signicant at conventional levels, the
broad pattern is that “dosage” matters: when an
individual’s survey overlaps more tournament
games, they engage in more binge drinking and
total drinking. As before, these effects are highly
concentrated among male respondents.
E. Drunk Driving
Our last analysis deals with the question of
whether the increase in drinking as a result of the
tournament leads to behaviors that could cause
harm to others. We are limited in the behaviors
we can analyze given that many questions in CAS
ask respondents about behaviors or experiences
over the course of the past year (which is not
well suited to our analysis of changes around
tournament time). However, we are able to exam-
ine self-reported drunk driving and self-reports of
riding in a car with a drunk driver over the past
month. Table 5 shows the results of this analysis.
In the rst row, drunk driving and riding with a
drunk driver are pooled together— the dependent
variable is equal to one if the student reported
either behavior in the previous month. The sec-
ond row shows the results for drunk driving only,
and Row 3 shows the results for riding with a
drunk driver only.
Both males and females experience a roughly
4 percentage point increase in the probability of
experiencing any kind of drunk driving episode
(a 9.1% difference at the mean for males and a
TABLE 5
Effects of Own-Institution NCAA Tournament
Participation on Drunk Driving Involvement
All
Observations
Males
Only
Females
Only
Self-reported drunk
driving incident
(n=25,438)
0.038 0.036 0.038
(0.022) (0.034) (0.029)
Self-reported driver
(n=25,438)
0.052*0.047 0.053*
(0.021) (0.034) (0.027)
Self-reported
passenger
(n=25,438)
0.024 0.049 0.005
(0.02) (0.033) (0.026)
Notes: Standard errors are robust to heteroskedasticity
and clustered at the institution level. Other controls included
in the regressions but not shown include all those described
in Section IV.A. Self-reported drunk driving incidents are
calculated based on responses to a question asking “In the past
30 days, how many times did you...,” where the items related
to drunk driving are “drive after drinking alcohol,” “drive
after having 5 or more drinks,” and “ride with a driver who
was drunk or high.” For each item, respondents selected from
options of “Not At All,” “Once,” and “Twice or More.” The
rst two behaviors were combined to generate our measure
of drunk driving, while the third behavior was used as our
measure of riding with a drunk driver. Since we utilized a
binary dependent variable in each case, the responses “Once”
and “Twice or More” (to at least one question) triggered a
value of 1, while “Not At All” triggered a value of 0. The
value of the rst row, measuring any drunk driving incidents
was assigned a value of 1 if the response to any of the
three questions is “Once” and “Twice or More,” and was 0
otherwise.
∗Signicant at the 5% level.
∗∗Signicant at the 1% level.
12 CONTEMPORARY ECONOMIC POLICY
13.9% difference for females). However, statisti-
cal signicance in these models is weak; we only
observe signicant effects on the rate of drunk
driving for the group as a whole, and for the
female only group.
Overall, these results suggest that the increase
in (binge) drinking occurring during the NCAA
tournament among students at participating
schools may in fact lead to behavior that is harm-
ful to others, although there is too much noise
to make this assertion with condence. Future
work with more focused information on student
behavior during tournament time could measure
how the tournament affects other problematic
behaviors that are associated with alcohol use
and abuse.
V. CONCLUSION
Athletic success presents many exciting
opportunities for schools, but this paper adds
to a growing body of research suggesting that
participation and success in major sports is
associated with increased binge drinking (and
its associated negative consequences) among
the current student body. Our ndings suggest
a large increase in binge drinking episodes
reported by male students during the NCAA
men’s basketball tournament.
Determining how to combat the increase in
drinking arising during the NCAA tournament
among students at participating schools is beyond
the scope of this paper. However, we note a few
possibilities here. Given the enormous popular-
ity of the NCAA tournament on and off cam-
pus, we believe that a dramatic change to the
event itself is highly unlikely.15 As such, col-
lege administrators and other interested parties
might focus on interventions targeting the sup-
ply or demand of alcohol around major sporting
events. For example, changing perceptions about
what constitutes “normal” drinking (Haines and
Spear 1996Perkins 2002) and providing moti-
vational interventions (Borsari and Carey 2000)
have been shown to reduce drinking (at least in
the short run) in other contexts. Increase in law
enforcement (surrounding underage drinking or
drunk driving) during the tournament is another
15. It should be noted that a modest expansion in the
number of tournament teams, from 64 to 68, has taken place
in recent years. A large-scale expansion in the number of
tournament teams is sometimes proposed (AP 2010). Our
results suggest such a move would increase drinking around
the tournament, though we cannot know for sure, since our
data do not include such an expansion.
possibility. The difference in behavior between
men and women suggests that the lowest cost
intervention would be to target interventions at
male students, since the increase in risky alco-
hol consumption observed during the tournament
occurs among males.
Intervention to reduce risky alcohol consump-
tion becomes more urgent when considering the
increased incidence of trafc fatalities surround-
ing major sporting events observed by Wood,
McInnes, and Norton (2011). Interventions,
such as the beer sales ban at the University of
Colorado’s Folsom Field in 1996, have also
been demonstrated to reduce negative outcomes
(Bormann and Stone 2001). Additional methods
of intervention might include harm reduc-
tion through education (e.g., regarding sexual
assault), ride services, or other means, adminis-
tered just prior to or during the tournament itself.
Future work might examine how these different
policy options substitute for or complement each
other and determine cost-effective ways to limit
high levels of alcohol use and its consequences
during the NCAA Men’s Basketball Tournament.
REFERENCES
AP. “NCAA Continues Talks of Expanding Tournament.”
2010. Accessed January 6, 2017. http://www.espn.com/
mens-college-basketball/news/story?id=4887774.
Berkowitz, S. “University of Texas Athletics Posts Record
Revenue, Expenses.” USA Today. 2014. http://
www.usatoday.com/story/sports/college/2014/01/
30/university-of-texas- longhorns-athletics- revenue-
expenses/5062161/.
Bormann, C. A., and M. H. Stone. “The Effects of Eliminating
Alcohol in a College Stadium: The Folsom Field Beer
Ban.” Journal of American College Health, 50(2), 2001,
81– 88.
Borsari, B., and K. B. Carey. “Effects of a Brief Motivational
Intervention with College Student Drinkers.” Journal
of Consulting and Clinical Psychology, 68(4), 2000,
728– 33.
Cowan, B. W., and D. R. White. “The Effects of Merit-
Based Financial Aid on Drinking in College.” Journal
of Health Economics, 44, 2015, 137–49.
DeSimone, J. “Fraternity Membership and Binge Drinking.”
Journal of Health Economics, 26(5), 2007, 950– 67.
Eisenberg, D., E. Golberstein, and J. L. Whitlock. “Peer
Effects on Risky Behaviors: New Evidence from Col-
lege Roommate Assignments.” Journal of Health Eco-
nomics, 33, 2014, 126– 38.
Glassman, T. J., V. J. Dodd, J.-J. Sheu, B. A. Rienzo, and A. C.
Wagenaar. “Extreme Ritualistic Alcohol Consumption
among College Students on Game Day.” Journal of
American College Health, 58(5), 2010, 413– 23.
Haines, M., and S. F. Spear. “Changing the Perception of the
Norm: A Strategy to Decrease Binge Drinking among
College Students.” Journal of American CollegeHealth,
45(3), 1996, 134– 40.
Hernández-Julián, R., and K. W. Rotthoff. “The Impact of
College Football on Academic Achievement.” 2014.
https://pirate.shu.edu/rotthoku/papers/Big
WHITE, COWAN & WOOTEN: MARCH MADNESS 13
Lee, C. M., M. A. Lewis, and C. Neighbors. “Preliminary
Examination of Spring Break Alcohol Use and Related
Consequences.” Psychology of Addictive Behaviors,
23(4), 2009, 689– 94.
Lindo, J. M., I. D. Swensen, and G. R. Waddell. “Are Big-
Time Sports a Threat to Student Achievement?” Ameri-
can Economic Journal: Applied Economics, 4(4), 2012,
254– 74.
. “Alcohol and Student Performance: Estimating the
Effect of Legal Access.” Journal of Health Economics,
32(1), 2013, 22– 32.
Lindo, J. M., P. M. Siminski, and I. D. Swensen. “College
Party Culture and Sexual Assault.” American Economic
Journal: Applied Economics, 10(1), 2018, 236– 65.
Miller, J. W., T. S. Naimi, R. D. Brewer, and S. E. Jones.
“Binge Drinking and Associated Health Risk Behaviors
among High School Students.” Pediatrics,119(1), 2007,
76– 85.
NCAA. “Membership.” 2010. http://www.ncaa.org/about/
who-we- are/membership.
Neal, D. J., and K. Fromme. “Hook’em Horns and Heavy
Drinking: Alcohol Use and Collegiate Sports.” Addic-
tive Behaviors, 32(11), 2007, 2681– 93.
Neal, D. J., D. E. Sugarman, J. T. Hustad, C. M. Caska,
and K. B. Carey. “It’s All Fun and Games... Or Is It?
Collegiate Sporting Events and Celebratory Drinking.”
Journal of Studies on Alcohol, 66(2), 2005, 291– 94.
NIH-NIAAA. “College Drinking— National Institute on
Alcohol Abuse and Alcoholism (NIAAA).” 2016.
Accessed August 30, 2016. https://www.niaaa.nih
.gov/alcohol-health/special- populations-co- occurring-
disorders/college-drinking.
O’Toole, T. “NCAA Reaches 14-Year Deal with CBS/Turner
for Men’s Basketball Tournament, which Expands to
68 Teams for Now.” USA Today. 2010. http://content
.usatoday.com/communities/campusrivalry/post/2010/
04/ncaa-reaches- 14-year- deal-with- cbsturner/1.
Perkins, H. W. “Social Norms and the Prevention of Alcohol
Misuse in Collegiate Contexts.” Journal of Studies on
Alcohol, (14), 2002, 164– 72.
Pope, D., and J. Pope. “The Impact of College Sports Success
on the Quantity and Quality of Student Applications.”
Southern Economic Journal, 75(3), 2009, 750– 80.
. “Understanding College Application Decisions:
Why College Sports Success Matters.” Journal of
Sports Economics, 15(2), 2014, 107– 31.
STSTravel. “Spring Break 2015 Dates.” 2014. Accessed
March 24, 2015. http://www.ststravel.com/college-
spring-break- dates.
theAmethystInitiative.org. “The Amethyst Initiative.”
2015. Accessed March 24, 2015. http://www
.theamethystinitiative.org/statement/.
Toma, J. D., and M. E. Cross. “Intercollegiate Athletics
and Student College Choice: Exploring the Impact
of Championship Seasons on Undergraduate Applica-
tions.” Research in Higher Education, 39(6), 1998,
633– 61.
Wechsler, H., J. E. Lee, M. Kuo, M. Seibring, T. F. Nelson,
and H. Lee. “Trends in College Binge Drinking during a
Period of Increased Prevention Efforts: Findings from 4
Harvard School of Public Health College Alcohol Study
Surveys: 1993– 2001.” Journal of American College
Health, 50(5), 2002, 203–17.
Wood, S., M. M. McInnes, and D. A. Norton. “The Bad
Thing about Good Games: The Relationship between
Close Sporting Events and Game-Day Trafc Fatal-
ities.” Journal of Consumer Research, 38(4), 2011,
611– 21.
SUPPORTING INFORMATION
Additional supporting information may be found online
in the Supporting Information section at the end of the article.
Appendix A1: NCAA Division I Schools Surveyed in the
Harvard School of Public Health College Alcohol Study.
Appendix A2: Missing Values by Month and School
Tournament Status.
Appendix A3: Effects of Own-institution NCAA Tourna-
ment Participation on Student Drinking Behaviors.
Appendix A4: Falsication Test of NCAA Tournament
Participation on Student Drinking Behaviors.
Appendix A5: Effects of Own-institution NCAA Tour-
nament Participation on Student Drinking Behaviors: Alter-
native Specications.
Appendix A6: Effects of Own-institution NCAA Tour-
nament Participation on Student Drinking Behaviors Using
Count and Binary Models.
Appendix A7: Intertemporal Substitution of Alcohol
Consumption Surrounding NCAA Tournament.
