C 2005)
Journal of Abnormal Child Psychology, Vol. 33, No. 3, June 2005, pp. 375–385 (
DOI: 10.1007/s10802-005-3576-2
Peer Effects in Drug Use and Sex Among College Students
Greg J. Duncan,1,6 Johanne Boisjoly,2 Michael Kremer,3 Dan M. Levy,4 and Jacque Eccles5
Received January 13, 2004; revision received July 15, 2004; accepted August 20, 2004
Past research suggests that congregating delinquent youth increases their likelihood of problem
behavior. We test for analogous peer effects in the drug use and sexual behavior of male (n = 279)
and female (n = 435) college students, using data on the characteristics of first-year roommates to
whom they were randomly assigned. We find that males who reported binge drinking in high school
drink much more in college if assigned a roommate who also binge drank in high school than if
assigned a nonbinge-drinking roommate. No such multiplier effect is observed for females, nor are
multiplier effects observed for marijuana use or sexual behavior for either males or females. Students
who did not engage in these behaviors in high school do not appear to be affected by their roommates’
high school behavior.
KEY WORDS: alcohol drinking; substance abuse; peer effects.
A growing literature on juvenile offenders suggests
that congregating deviant youth in classes, juvenile detention facilities or summer camps sparks a “contagion”
process that increases the likelihood of future deviant behavior. Whether peer effects are present among middleclass youth in college is the subject of this paper. Our data
are drawn from surveys of students at a large state university who were randomly assigned first-year roommates.
We focus on several problem behaviors: binge drinking,
marijuana use, and multiple sexual partners.
Peer effects are estimated in a variety of ways.
Among students whose binge drinking, marijuana use, or
sexual behavior had begun during their high school years,
we estimate whether their reports of these behaviors in
their second through fourth college years is greater if they
were assigned first-year roommates who had engaged in
these problem behaviors in high school. And for students
who had not engaged in drug use or sex prior to college
entry, we estimate whether such students might be drawn
into future problem behavior if assigned roommates with
a high school history of problem behavior.
All of the behaviors we investigate have potentially
important consequences. Drug use and unprotected sex
can compromise life chances by threatening health and,
in the case of illegal drug use, risking legal sanctions
(National Institute of Drug Abuse, 2002; National Institute on Alcohol Abuse and Alcoholism, 2000). And yet all
are widespread on US college campuses. Defining binge
drinking as five or more drinks in a row for males and four
or more drinks in a row for females (Wechsler, Dowdall,
Davenport, & Rimm, 1995), Wechsler, Lee, Kuo, and
Lee (2000) estimate that 44% of college students binge
drank in the 2 weeks prior to responding to the 1999
College Alcohol Study. Similar rates were reported in the
1993 wave of the survey. These same surveys showed that
16% of college students reported recent marijuana use in
1999, up from 13% in 1993 (Gledhill-Hoyt, Lee, Strote, &
Wechsler, 2000). Some 14.4% of 18–24-year-old college
students reported four or more sexual partners in their
lifetimes (Centers for Disease Control, 2003).
There are many theories as to why some students
engage in these behaviors while others do not. Some researchers concentrate on personality difficulties such as
anxiety, depression, low self-esteem, and social introversion (Kaplan, 1979; Schall, Kemeny, & Maltzman, 1992;
1 School of Education and Social Policy and Institute for Policy Research,
Northwestern University, Evanston, Illinois.
2 Department of Sociology, University of Quebec at Rimouski, Rimouski
Quebec, Canada.
of Economics, Harvard University, The Brookings Institution, and NBER, Cambridge, Massachusetts.
4 Mathematica Policy Research, Inc., Washington, District of Columbia.
5 Department of Psychology, University of Michigan, Michigan.
6 Address all correspondence to Greg J. Duncan, Institute for Policy
Research, Northwestern University, 2046 Sheridan Road, Evanston,
Illinois 60208; e-mail: greg-duncan@northwestern.edu.
3 Department
375
C 2005 Springer Science+Business Media, Inc.
0091-0627/05/0600-0375/0
376
Valliant, 1995). Others examine college contexts such as
institutional size and competitiveness, and residential factors such as dormitory versus fraternity or sorority residence (e.g., Wechsler et al., 2000).
Still others focus on student beliefs regarding normative behavior among their classmates. If students act,
in part, to conform more closely to their perceptions of
classmates’ behavior and if, as appears to be the case,
many students overestimate the prevalence of problem
behaviors on campus, then it might be possible to affect
behavior with interventions targeted on changing beliefs
(Barnett, Far, Mauss, & Miller, 1996). Many studies of
drug and sex-related problem behaviors concentrate on
the middle and high school periods and examine family influences such as parental monitoring and parental
drinking (Reifman, Barnes, Dintcheff, Farrell, & Uhteg,
1998).
There is ample documentation of continuities in drug
and sexual behaviors across adolescence. Much larger
fractions of binging than nonbinging college students reported binge drinking in high school (Wechsler et al.,
2000, Table III). But while correlations in problem behavior across time are substantial, they are far from perfect: relatively few individuals are chronic binge drinkers
across adolescence and into adulthood, and most eventually stop engaging in these problem behaviors altogether (Schulenberg, O’Malley, Bachman, Wadsworth, &
Johnston, 1996). However, some individuals begin to engage in these behaviors while in college and, more generally, college settings appear to be times of heightened risk
for problem behaviors. Data from the Monitoring the Future study show that while in high school, college-bound
students use less of all classes of substances studied as
compared to classmates not bound for college (Bachman,
Wadsworth, O’Malley, Johnston, & Schulenberg, 1997).
However, after high school graduation, the increase in
alcohol and marijuana use among college students exceeds that of their former classmates who are not attending
college.
The focus of this paper is on how peers affect drug
use and problematic sexual behavior among college students. Peer influences on problem behaviors in both early
adolescence and college settings have been investigated
extensively, but rarely convincingly. Many empirical studies document the fact that individuals with friends who
abuse drugs are themselves more likely to abuse drugs, but
fail to address problems of self-selection into peer groups
(Manski, 1993; Moffitt, 2001). As long as individuals are
free to choose their friends, it is possible that someone’s
substance abuse behavior or personal characteristics associated with substance abuse are affecting his or her choice
of peer group.
Duncan, Boisjoly, Kremer, Levy, and Eccles
Longitudinal studies have documented that individuals with friends who drink are more likely to begin
drinking subsequently (e.g., Reifman et al., 1998) and
to increase their drinking more rapidly (Curran, Stice, &
Chassin, 1997). While the strength of the evidence for peer
impacts is stronger in longitudinal than cross-sectional
studies, it is still possible that difficult-to-measure characteristics (e.g., thrill seeking) or circumstances (e.g., family
problems) are leading individuals to both choose drinking
peers and display unusually rapid increases in their own
substance use.
A few studies of teen problem behavior have employed two-stage peer effects models in which a firststage peer group equation is estimated and then used to
relate predicted peer group characteristics to teen problem behavior in the second stage equation. Evans, Oates,
and Schwab (1992) estimate models of high-school completion and out-of-wedlock teen child-bearing in which
their contextual variable was the SES of the student body.
They identified student-body SES using characteristics of
the metropolitan area in which the student resided (e.g.,
unemployment rate), and yet one can imagine ways in
which labor market characteristics might influence teen
schooling and fertility choices independently of peer influences. Norton, Lindroth, and Ennett (1998) estimate a
model of peer impacts on teen alcohol and tobacco use
of young adolescents, but they are forced to identify the
model with parental and census-based reports of neighborhood characteristics.
What if selection of peers were beyond the control of the individual? Although no studies of substance
abuse have addressed this question, a growing literature
on delinquent behavior provides disturbing evidence of
unintended (“iatrogenic”) effects of congregating delinquent youth who do not already know one another
(Dishion, McCord, & Poulin, 1999). Bayer, Pintoff, and
Pozen (2003) find that juvenile offenders released from
Florida residential correctional facilities tended to commit the kinds of crimes that had been committed by their
correctional-facility peers. McCord (1995) finds that assigning juvenile offenders to summer camps with the hope
of reducing subsequent criminal behavior in fact had the
opposite effect, when those offenders were compared with
others not afforded the opportunity to attend such camps.
Poulin, Dishion, and Burraston (2001) evaluated the effects of an intervention program that brought deviant teens
together for sessions that emphasized prosocial goals and
self-regulation. Teachers blind to treatment status reported
more problem behavior 3 years later for youth in the treatment group than in a quasi-experimental control group.
If peer effects apply to substance abuse in college,
then we expect that students with high school histories
Peer Effects in Drug Use and Sex Among College Students
of substance abuse will abuse substances more in college
as they have more contact with substance-abusing peers.
National data confirm the stereotype of much more binge
drinking among students living in fraternities and sororities as compared with dormitories (Wechsler et al., 2000),
but here again, since such residential arrangements are
chosen, one cannot conclude that drinking patterns are
caused by fraternity- or sorority-based peer effects. Our
paper uses the “natural experiment” of randomly assigning roommates to get around the confounding effects of
residential choice.
Hypothesis 1: Individuals who abuse drugs or engage in
sexual intercourse prior to college entry will exhibit
greater frequency of these behaviors in college if assigned a roommate who also engaged in these behaviors in high school than if assigned a roommate who did
not.
Note that this peer effects hypothesis applies to youth
who binged, used marijuana, or engaged in sex prior to
entering college. Whether the subsequent problem behavior of students entering college without these experiences
is influenced by the prior experiences of their college
roommates is less clear. Social learning or pressures to
conform might lead innocent youth to adopt the problem
behaviors of their roommates. On the other hand, a bingedrinking roommate may annoy a nondrinking roommate
by making sleep or study more difficult. Illegal drug use
may have a discouraging effect if it increases the risk that
innocent roommates might be arrested.
Hypothesis 2: Whether individuals who do not abuse drugs
or engage in sexual intercourse prior to college entry
are affected by college roommates’ high school behavior is ambiguous and depends on the strength of
encouraging and discouraging factors.
Kremer and Levy (2002), using data from randomly
assigned roommates at a large state university, find that
males assigned to roommates who reported drinking in the
year prior to entering college had one quarter-point lower
GPA than those assigned to nondrinking roommates. Furthermore, they find that the effect of a frequent drinking
roommate on GPA was larger for frequent drinkers than
for nondrinkers and persisted over time, which they take
to support their preference/habit formation model rather
than their disruption model. Because they did not have
data on student drinking during college, they could not
confirm that lower grades were associated with more frequent binge drinking.
Our own paper shifts the focus of the Kremer and
Levy work from college academic outcomes to college
problem behaviors—specifically binge drinking, mari-
377
juana use, and multiple sex partners. Furthermore, the
results on binge drinking will shed light on the mechanism
behind the effects of roommate drinking on academic performance observed by Kremer and Levy.
METHODS
Data Sources
Our data are taken from students entering a large,
academically strong state university in the fall of 1998,
1999, and 2000. The university’s housing office provided
information on each student’s housing application and
housing assignment. High school grades, socioeconomic
information, and some behavioral data on students were
gathered from the Cooperative Institutional Research Program’s (CIRP) Entering Student Survey, an annual survey
of the American higher-education system that was started
in 1966 by the American Council on Education and is
now conducted jointly by the Council and the University of California, Los Angeles. In the case of the particular university in our study, entering students filled
in the survey at an orientation session occurring before
classes begin. Although a few students may have met
their roommates first, the large majority of students filled
out this survey over the summer, before meeting their
roommates.
Questions about drug use and sexual behavior in high
school and at the time of the interview were asked in
a survey we administered to students who entered the
university in the fall of 1998, 1999, and 2000 and were
randomly assigned roommates. The timing of our survey
(winter/spring of 2002) provides us with data when students were more than halfway through their second, third,
and fourth years; the average number of years between
college entry and responding to the follow-up survey was
2.4 years. The survey was administered via the Internet
with a telephone follow-up to maximize response rates.
Measures
Dependent Variables
The follow-up survey provides measures of binge
drinking at the time of the survey and whether the student
binge drank at all during high school. In keeping with standard research practice (Wechsler et al., 1995), we defined
binge drinking differently for males and females—five or
more drinks in a row for males and four or more in a row
for females. High school binge drinking was presumed to
take place if the respondent reported that the first time he
378
or she drank the requisite number of drinks was “before
college.” Current binge drinking is measured as times per
month in response to the respective question “Over the
past 2 weeks, on how many occasions have you had [four
if female/five if male] or more drinks in a row?” Responses
are converted to a monthly amount by multiplying by 2.15.
We used information provided by the answers to
these questions to classify respondents and their roommates into the following categories: i) neither binge drank
in high school; ii) the respondent did not binge drink
in high school but the roommate did; iii) the respondent
binge drank in high school but the roommate did not;
iv) both binge drank in high school; and v) roommatedrinking data are not ascertained owing to case or item
nonresponse. The dependent variable is the frequency of
the respondent’s binge drinking at the time of the followup survey.
Similar questions regarding marijuana use provide
measures of any marijuana use in high school and monthly
frequency of marijuana use “during the last 12 months.”
Our measure of current sexual behavior is based on responses to the questions “During the last 12 months, with
how many partners do you estimate you have had sexual
intercourse?” Sex in high school is defined by a “before
college” response to the question “When did you have
sexual intercourse for the first time?”
Duncan, Boisjoly, Kremer, Levy, and Eccles
designation we included respondents who gave Mexican
American/Chicano, Puerto Rican, or Other Latino and no
other response. All respondents marking more than one
category, marking American Indian, or marking “Other”
fall into our “Other” category.
Of all entering students in the 1998 and 1999 cohorts,
about 90% completed the CIRP survey (corresponding
response-rate data for the 2000 cohort are not available).
Of the 10,268 CIRP respondents, 2232 opted to live in
enrichment residence halls, 2029 requested a roommate,
724 requested living alone during their first year, 4134
failed to meet the lottery deadline, and 42 otherwiseeligible students were not assigned a roommate, leaving
1107 students eligible for our lottery sample.
To avoid missing data and other complications of
multiple roommates, we concentrated our analysis on the
990 individuals who were randomly assigned a single
roommate. The follow-up survey response rate among
this sample was 72% and produced an analysis sample of
714. Response rates were considerably higher for females
(76%) than males (67%). Missing data on individual survey items reduced this case count further. The bulk of
nonrespondents could not be located; only 1.8% of all
nonrespondents were successfully contacted but refused
to participate in the study.
Roommate Assignment
Control Variables
CIRP measures used as control variables in our regressions include both self and roommate responses to
questions about: (i) years of father’s education; (ii) years
of mother’s education; (iii) high school grade point average; and (iv) family income.
We also controlled for respondents’ and roommates’
high school test scores. Since some students took only the
SAT, others took only the ACT, and some took both, a
common admissions test score measure was needed as an
academic background variable. We therefore standardized
test scores using the ACT scale based on concordance tables (published by both ACT, Inc. and the College Board),
which are used by many admissions offices around the
country (including the admissions office of the university
used in this study).
Race and ethnicity were asked in the single question:
“Are you (mark all that apply): White/Caucasian, African
American/Black, American Indian, Asian American/
Asian, Mexican American/Chicano, Puerto Rican, Other
Latino, Other.” We coded as “white” respondents who
marked only the first category, “black” respondents who
marked only the second category, and “Asian” respondents
who marked only the fourth category. For our “Hispanic”
Since our identification strategy takes advantage of
the roommate assignment process, it is worth reviewing
this process. Students who met the lottery deadline were
assigned rooms randomly, conditional on gender and four
basic housing preferences: environment (substance-free
housing, nonsmoking roommate, do not mind smoking
roommate, and smoker), room type (single, double, triple
occupancy, and other), geographic area of campus, and
gender composition of hall and corridor. For some of these
preferences, students could indicate a first, second, and
third choice.
For students participating in the lottery, roommate
assignment should be random, conditional on gender, and
basic housing preferences. We call a combination of gender and housing preferences a cell. All of our regression analyses control for the student’s combination of
first choices of housing preferences (using the “absorb”
features of Stata and SAS’s Proc. GLM). In effect, this
amounts to a fixed-effects regression in which peer effect estimates are based only on within-preference-cell
variability.
To verify that the housing assignment process was
indeed random within cells, we first spoke with housing
officers to understand how the assignment process worked
Peer Effects in Drug Use and Sex Among College Students
and the computer software used to make the assignments.
We then reviewed the documentation of the computer software used for the 1997 and 1998 entering cohorts and
checked that it truly randomized within cells. Finally, using techniques discussed more fully in Kremer and Levy
(2002), we verified that, controlling for all housing preference choices, initial roommates’ background characteristics were not significantly correlated.
It is important to note that when we use the term
“roommate” we are referring to the roommate initially
assigned to the student when entering the university. If a
student changed roommates we do not use the information on the new roommates because this would raise the
possibility of self-selection and possibly bias our results.
For example, a student might well eventually switch to
a roommate who is more similar or compatible than the
initial roommate. If this is the case, and we used actual
roommate (instead of initial roommate) information in
our regressions, our peer-effect estimates could reflect
self-selection. University policy does not allow roommate
changes during the first 6 weeks of classes except for extreme cases such as those involving violence, and strongly
discourages any roommate changes during the first year.
Less than 5% of students switch roommates during their
first term, but 82% had switched roommates by the beginning of their second years.
379
widespread at this university. About half of the respondents reported at least some binge drinking in the 2 weeks
prior to completing the follow-up survey. As a point of
comparison, Wechsler et al. (2000) report that 44.1% of
students reported binge drinking in the 2 weeks prior to
their 1999 survey.
Including students with zeroes, current binge drinking averages 3.9 times per month for males and 2.8 times
per month for females. Frequencies are about twice as
high among students who engage in at least some binge
drinking. As shown in the last column of Table I, females
report significantly less binge drinking than do males.
Marijuana use is somewhat less pervasive than binge
drinking, although it is still reported by at least one third of
both male and female students at the time of the follow-up
survey. The average monthly frequency of marijuana use
is considerably lower for females than males, and lower
than binge drinking for both groups.
The distribution of responses to the question on number of current sexual partners is shown at the bottom of
Table I. A little over one third of both males and females
report no sexual intercourse in the 12 months preceding
the follow-up interview, and an additional one third reported sex with only one partner. Roughly one fifth of
both groups report two to three partners. The average
number of partners does not differ significantly between
males and females.
Data on control variables are presented in Table II.
Roughly one third of the roommates of follow-up survey respondents did not respond to the survey. For the
remainder, roughly similar fractions fall into the various
combinations of respondent/roommate binge drinking in
high school. Regarding marijuana use, the modal group
RESULTS
Descriptive Statistics
Table I shows descriptive statistics for our dependent
variables. Binge drinking and marijuana use are fairly
Table I. Summary Statistics of Binge Drinking, Marijuana Use and Sexual Behavior
Male (n = 279)
Current binge drinking (# of times
per month)
Current use of marijuana (# of times
per month in the last 12 months)
Current number of sex partners (# in
the last 12 months)
No sexual intercourse in the last
12 months
1 person
2–3 people
4–5 people
6–12 people
Total
Female (n = 435)
Mean
Std. dev.
% zero
Mean
Std. dev.
% zero
P value of t- or chi-square
test on gender differences
3.86
5.550
48.0
2.81
4.160
53.1
0.005
2.45
6.170
52.0
1.18
3.540
60.1
0.001
1.23
1.430
1.07
1.260
0.120
Distribution
Distribution
36.7
38.4
35.6
20.4
6.6
0.7
38.6
18.6
4.0
0.5
100.0
100.0
0.536
380
Duncan, Boisjoly, Kremer, Levy, and Eccles
Table II. Summary Statistics of Individual and Roommate Characteristics
Respondents to the follow-up surveya
Males
Mean
Respondent and roommate high school behavior
Neither respondent nor roommate binge drank in high school
Respondent but not roommate binge drank in high school
Roommate but not respondent binge drank in high school
Both respondent and roommate binge drank in high school
Neither respondent nor roommate had marijuana in high school
Respondent but not roommate had marijuana in high school
Roommate but not respondent had marijuana in high school
Both respondent and roommate had marijuana in high school
Neither respondent nor roommate had sex in high school
Respondent but not roommate had sex in high school
Roommate but not respondent had sex in high school
Both respondent and roommate had sex in high school
Roommate nonresponse to follow-up survey
Respondent characteristics (all gathered in entering student survey)
Black
Asian
Hispanic
Other
Father’s education
Mother’s education
High school grade point average
Test scores (ACT scale)
Family income (in tens of thousands)
Roommate characteristics (all gathered in entering student survey)
Non-white roommate
Roommate’s father’s education
Roommate’s mother’s education
Roommate’s high school grade point average
Roommate’s test scores (ACT scale)
Roommate’s average family income (in tens of thousands)
a All
Females
Std. dev.
Mean
Std. dev.
.143
.118
.118
.190
.224
.141
.141
.090
.269
.100
.100
.115
.394
.351
.324
.324
.393
.418
.348
.348
.287
.444
.301
.301
.319
.490
.230
.133
.133
.159
.348
.109
.109
.095
.301
.124
.124
.092
.317
.421
.340
.340
.366
.477
.312
.312
.294
.459
.330
.330
.289
.466
.014
.050
.025
.057
16.552
16.055
3.781
28.952
12.128
.119
.219
.157
.233
1.819
1.907
.246
2.499
5.930
.021
.097
.039
.034
16.391
15.728
3.782
27.744
11.924
.143
.296
.194
.183
2.007
2.164
.259
2.647
6.286
.151
16.581
16.016
3.765
28.729
13.176
n = 279
.358
1.751
1.826
.262
2.608
6.335
.161
16.445
15.848
3.774
27.578
12.333
n = 435
.368
1.975
2.059
.261
2.726
6.499
randomly assigned one roommate.
consisted of respondent/roommate pairs in which both
had never had marijuana in high school. In the case of
sexual behavior, the modal group consisted of respondent/roommate pairs in which both were virgins in high
school.
The remaining rows of Table II show the affluent
nature of the sample, with high paternal and maternal education and family incomes averaging more than $100,000.
Test scores and high school grade-point averages are
high. Relatively few of the students were from minority
groups.
Peer Effects in Binge Drinking
Regression models of current binge drinking are reported in Table III. The “respondent but not roommate
binge drank in high school” category is used as the reference group, so coefficients on the included variables show
regression-adjusted differences in college binge drinking
relative to this group. In order to test whether peer effects
exist for those who binge drank in high school (hypothesis 1), the key explanatory variable of interest is “both
binge drank in high school.”
In all cases, we present separate models for males
and females and run what amount to fixed-effects regressions with controls for all combinations of first housing preferences. Since we were unable to secure software
that simultaneously: (i) provided a Tobit regression procedure to handle the substantial number of “zero” responses
in our dependent variables (Table II); (ii) enabled us to
use multiple imputation procedures for missing data on
our independent variables; and (iii) enabled us to adjust
Peer Effects in Drug Use and Sex Among College Students
381
Table III. Individual and Roommate Characteristics as Predictors of Current Binge Drinking (Number of Times Per Month)
OLS regressiona
Male
Respondent and roommate high school behavior
Neither respondent nor roommate binge
−2.227∗
1.224
drank in high school
1.067
Roommate but not respondent binge drank
−1.893∗
in high school
Respondent but not roommate binge drank
in high school (reference)
1.042
Both respondent and roommate binge
3.830∗∗
drank in high school
Roommate nonresponse to follow-up
survey
Respondent characteristics (all gathered in entering student survey)
Black
−.985
2.481
Asian
−.382
1.926
Hispanic
−1.392
2.075
Other
−1.791
1.606
Father’s education
−.213
.221
Mother’s education
.196
.219
High school grade point average
−1.243
1.526
Test scores (ACT scale)
.075
.141
Family income (in thousands)
−.200∗∗
.063
Roommate characteristics (all gathered in entering student survey)
Non-white roommate
−.154
.945
Roommate’s father’s education
−.167
.208
Roommate’s mother’s education
−.093
.203
Roommate’s high school grade point
−.567
1.328
average
Roommate’s test scores (ACT scale)
−.081
.131
Roommate’s family income (in thousands)
−.097∗
.055
N = 279
Tobit regressionb
Female
Male
Female
−3.447∗∗
.604
−2.637∗∗
.641
−2.463∗∗
.335
−3.365∗∗
.720
−3.142∗∗
.417
−2.147∗∗
.280
−.424
.886
2.582∗∗
1.121
−.126
.528
.066
.778
−1.271∗∗
.410
.847
.661
−.521
1.548
.298∗∗
−.380∗∗
−1.978∗∗
.155∗
.058
1.482
.685
1.057
1.180
.121
.105
.820
.082
.037
−.744
−1.368
1.497
−.195
−.143
−.003
−1.379
−.062
−.034
2.072
.954
2.145
1.110
.176
.164
1.136
.116
.049
.214
.223
−.681
1.825
.149
−.243∗∗
−1.183
.034
.078∗∗∗
1.445
.669
.736
1.248
.105
.090
.706
.071
.030
.220
−.077
−.048
−.760
.557
.117
.106
.790
−.780
−.166
−.057
.591
.699
.171
.160
1.058
.454
.024
.012
.216
.524
.102
.090
.669
−.005
.007
N = 435
.079
.033
−.059
−.054
N = 271
.109
.044
−.059
−.054
N = 426
.068
.027
Note. All regressions include controls for housing preferences, cohort and test taken; values not shown.
a With control for all combinations of 1st preferences and missing data imputation, but no adjustment for roommate clustering.
b Coefficients shown are marginal effects. Tobit regressions include controls for a restricted number of combinations of 1st preferences. Missing
values assigned to the mean and controlled for by missing value indicators; values not shown.
∗ p ≤ .10. ∗∗ p ≤ .05.
with Huber-White methods for the lack of independence
caused by roommate clustering, we present both OLS as
well as Tobit models. Our SAS-based OLS models incorporate multiple imputation for missing data but no HuberWhite clustering adjustment, while our Stata-based Tobit
coefficients (expressed as marginal effects) incorporate
Huber-White clustering adjustments but use missing data
dummies to handle missing data.
High school binge drinking is a powerful predictor
of college binge drinking. Both male and female respondents entering college with a history of binge drinking
report much more frequent binge drinking at the time of
the follow-up interview than respondents entering college
without a history of binge drinking. In the case of respondents assigned nondrinking roommates, respondents who
binge drank in high school averaged 2.2–3.4 more binge
drinking episodes per month at the time of the follow-up
survey than respondents who did not drink in high school,
as reflected in the coefficient of the variable “neither respondent binge drank in high school” across the various
regression specifications.
Peer effects in binge drinking is present if college
drinking for students who entered college with a history
of heavy drinking is magnified when those students are
assigned roommates who had similar high-school histories (Hypothesis 1). Since “respondent but not roommate binge drank in high school” is the reference category, the coefficient on “both respondent and roommate
binge drank in high school” reflects our estimate of peer
effects. The first column, on the basis of OLS estimation
382
Duncan, Boisjoly, Kremer, Levy, and Eccles
for males, shows a large and statistically significant deviant peer effect—almost four times more binge drinking episodes per month. The magnitude of the analogous
coefficient based on the Tobit specification for males is
smaller (2.58) but still statistically significant. But for
females, the coefficient is insignificant and even has an
unexpected negative sign (on both the OLS and Tobit
specifications).
Are nondrinking students susceptible to peer influence if matched with drinking roommates? Here the relevant coefficients are on the “roommate but not respondent
binge drank in high school” and “neither respondent nor
roommate binge drank in high school” categories, which
contrast nondrinkers who were and were not paired up
with drinking roommates. Since supplemental calcula-
tions showed that these two coefficients are insignificantly
different from one another in all cases, Table III provides
no evidence of peer effects for nondrinkers.
Peer Effects in Marijuana Use and Sexual Behavior
In some respects, the patterns of marijuana use parallel those of binge drinking (Table IV). Marijuana use
in high school and college are highly correlated, male
respondents who had not used marijuana in high school
were, if anything, turned off rather than turned on by
marijuana-using roommates, and there is no evidence of
peer effects for females. But a key difference is that there
is no evidence of multiplier effects for marijuana use
Table IV. Individual and Roommate Predictors of Current Use of Marijuana (Number of Times Per Month in the Last 12 Months)
OLS regressiona
Male
Respondent and roommate high school behavior
Neither respondent nor roommate used
−3.672∗∗
.978
marijuana in high school
1.085
Roommate but not respondent used
−3.899∗∗
marijuana in high school
Respondent but not roommate used
marijuana in high school (reference)
Both respondent and roommate used
.092
1.560
marijuana in high school
Roommate nonresponse to follow-up
survey
Respondent characteristics (all gathered in entering student survey)
Black
−1.436
2.609
Asian
.090
1.615
Hispanic
−.419
2.120
Other
−.005
1.688
Father’s education
−.055
.242
Mother’s education
−.008
.221
High school grade point average
−2.697∗
1.575
Test scores (ACT scale)
.117
.152
Family income (in thousands)
.074
.066
Roommate characteristics (all gathered in entering student survey)
Non-white roommate
.614
.978
Roommate’s father’s education
−.193
.215
Roommate’s mother’s education
−.133
.211
Roommate’s high school grade point
−.520
1.372
average
Roommate’s test scores (ACT scale)
−.081
.143
Roommate’s family income (in thousands)
.101∗
.057
N = 279
Tobit regressionb
Female
Male
Female
−2.028∗∗
.759
−2.366∗∗
.458
−1.443∗∗
.253
−1.934∗∗
.742
−2.886∗∗
.333
−.811∗∗
.202
−.190
1.029
−.105
.847
.095
.393
−1.572∗∗
.565
−.597∗∗
.257
−.598
.279
−1.039
−.894
.222∗
−.130
−.696
.028
−.037
1.555
.686
1.073
1.186
.125
.103
.834
.095
.037
−.807
−.515
−.798
2.571
−.138
−.129
−1.265
.263∗∗
.080∗
1.525
.975
1.053
1.573
.149
.145
1.008
.104
.042
−.061
−.150
−.906∗∗
−.605∗
.110∗
−.063
−.485
.025
−.027
.789
.354
.203
.322
.066
.057
.427
.046
.019
−.398
.027
.075
−.796
.553
.119
.107
.807
−.216
−.078
−.009
−.504
.666
.146
.133
.867
−.203
.009
.094∗
−.004
.283
.064
.057
.412
−.072
.003
N = 435
.079
.035
−.072
−.001
N = 277
.093
.037
−.011
−.003
N = 431
.042
.017
Note. All regressions include controls for housing preferences, cohort and test taken; values not shown.
a With control for all combinations of 1st preferences and missing data imputation, but no adjustment for roommate clustering.
b Coefficients shown are marginal effects. Tobit regressions include controls for a restricted number of combinations of 1st preferences. Missing
values assigned to the mean and controlled for by missing value indicators; values not shown.
∗ p ≤ .10. ∗∗ p ≤ .05.
Peer Effects in Drug Use and Sex Among College Students
383
Table V. Individual and Roommate Predictors of Current Number of Sexual Partners (Nos. in the Last 12 Months)
OLS regressiona
Male
Neither respondent nor roommate had sex in
high school
Roommate but not respondent had sex in high
school
Respondent but not roommate had sex in high
school (reference)
Both respondent and roommate had sex in high
school
Roommate nonresponse to follow-up survey
Tobit regressionb
Female
Male
Female
−.756∗∗
.282
−1.168∗∗
.191
−.893∗∗
.204
−.903∗∗
.137
−.433
.346
−.969∗∗
.215
−.742∗∗
.204
−.710∗∗
.124
.399
.355
.251
.235
.337
.278
.240
−.473∗
.244
−.542∗∗
.143
.842∗∗
Respondent characteristics (all gathered in entering student survey)
Black
−.049
Asian
−.383
Hispanic
−.720
Other
−.317
Father’s education
−.012
Mother’s education
−.109∗
High school grade point average
.074
Test scores (ACT scale)
−.061∗
Family income (in thousands)
.005
Roommate characteristics (all gathered in entering student survey)
Non-white roommate
−.075
Roommate’s father’s education
.009
Roommate’s mother’s education
−.032
Roommate’s high school grade point average
−.394
Roommate’s test scores (ACT scale)
.003
Roommate’s family income (in thousands)
−.018
N = 279
.680
.423
.555
.452
.060
.057
.401
.037
.016
−.073
−.325
.334
.031
−.021
.010
−.363
.043∗
−.011
.479
.209
.325
.370
.039
.032
.261
.025
.012
−.347
−.817∗∗
.155
.428
−.025
−.077∗
−.084
−.071∗∗
.016
.612
.218
.539
.375
.049
.047
.339
.034
.014
−.303
−.452∗∗
.357
.233
−.068∗
.026
−.511∗∗
.041∗
−.001
.369
.158
.323
.320
.033
.031
.232
.023
.010
.260
.055
.051
.358
.035
.014
−.145
−.012
.037
.020
−.013
.000
N = 435
.168
.038
.033
.255
.024
.011
−.412∗∗
−.025
−.015
−.578∗
.015
−.007
N = 275
.190
.048
.044
.308
.031
.013
−.239∗
−.018
.013
−.052
.001
−.001
N = 425
.141
.033
.030
.222
.022
.010
Note. All regressions include control for housing preferences, cohort, test taken; values not shown. Missing values assigned to the mean and controlled
for by missing value indicators; values not shown. Standard errors adjusted for room clustering using Huber-White robust estimations except for Tobit
model.
a With control for all combinations of 1st preferences and missing data imputation, but no adjustment for roommate clustering.
b Coefficients shown are marginal effects. Tobit regressions include controls for a restricted number of combinations of 1st preferences. Missing values
assigned to the mean and controlled for by missing value indicators; values not shown.
∗ p ≤ .10. ∗∗ p ≤ .05.
among males who entered college having used marijuana
in high school. Individuals in this group who were paired
with marijuana-using roommates reported no significantly
greater use of marijuana at the time of the follow-up survey
than those paired with nonmarijuana-using roommates.
There is no conclusive evidence of peer effects in the
sexual behavior outcomes either (Table V). Both males
and females who lost their virginity in high school report
more sexual partners at the time of the follow-up survey than high-school virgins. And while the nonvirgins
who are paired with nonvirgins report a somewhat larger
number of sexual partners than nonvirgins paired with
virgin roommates; the difference is only statistically significant for males under the OLS specification. In the case
of sex coupled with binge drinking, both binge drinking
and sex in high school boost the reported frequency of
this combination.
DISCUSSION
Intervention research on juvenile offending is uncovering disturbing evidence that congregating offending
youth into treatment groups may spur a kind of “deviancy
training” that increases rather than reduces future problem
behavior (Dishion et al., 1999). Whether a similar process
might be taking place in college dormitories is the subject
of this paper.
We find important but rather selective evidence of
the dangers of grouping college students who exhibited
384
problem behavior in high school. Pairing up young men
who binge drank in high school appears to promote binge
drinking in college. No such multiplier effect is observed
for females, nor are multiplier effects observed for marijuana use or sexual behavior for either males or females.
Theory is more ambivalent about the consequences
of roommate pairings for first-year college students with
no prior history of problem behavior. We uncovered
no evidence that being paired with a roommate with
problematic high school behavior had any effect on a
student’s own problematic behavior in college.
Our evidence on peer effects on drinking behavior
raises important questions about why males who binge
drink in high school are so vulnerable to roommate influences. Our interviews provide some data on the nature of
social interactions and compatibility with first-year roommates, as well as beliefs regarding normative behavior. We
intend to explore possible mediational processes in our future work. We also plan to use additional data to explore
whether the observed peer effects on drinking behavior
are also present in the first year of college and whether
these effects grow or diminish over time.
Also of interest is why peer effects are observed for
binge drinking but neither marijuana nor sexual behavior.
A possible methodological reason is that our threshold for
high school drinking (i.e., binge amounts) is considerably
higher than for marijuana use (i.e., any such use before
college) or sex (i.e., any sexual intercourse before college).
We investigated this by grouping roommates according to
whether each reported any drinking during high school.
When we ran regressions that were similar to those presented in Table III, the multiplier effects disappeared. Men
who reported any drinking in high school were no more
likely to binge drink in college if assigned a roommate
who drank at all in high-school than not. Thus, the deviant peer effect appears to apply only to students who
had drunk heavily while in high school. We lacked information on the extent of marijuana use or sexual behavior
in high school.
A final methodological issue is whether the sex difference in the definition of binge drinking (five drinks
in a row for males and four for females) might account
for the differential peer impacts estimated for males and
females. Since our survey followed the prevailing practice
for defining thresholds for binge drinking, we are unable
to assess the sensitivity of the results to this difference.
Our results suggest that, for the most part, students,
parents, and college administrators need not fear that
roommate assignments will promote problem behavior
during college. Indeed, our other work with these data
suggests that shuffling the roommate deck and pairing
students who did not know one another prior to college en-
Duncan, Boisjoly, Kremer, Levy, and Eccles
try may promote social understanding (Boisjoly, Duncan,
Kremer, Levy, & Eccles, 2004). An important exception
is that pairing young men with drinking problems may
aggravate those problems. Despite the logistical difficulties in identifying entering students with binge drinking
histories, our results suggest substantial benefits to ensuring that binge-drinking young men do not room together
in their first college year.
ACKNOWLEDGMENTS
Financial support from the W.T. Grant Foundation,
the John D. and Catherine T. Arthur Foundation, and the
NICHD Child and Family Well-Being Research Network
(2 U01 HD30947-07) is gratefully acknowledged. We
thank Sean McCabe, Carol Boyd, and William Zeller for
their contributions in the early stages of this research;
Brian Madden, Deanna Maida, and Bessie Wilkerson for
research assistance; and John Schulenberg and members
of the Executive Session on Deviant Peer Influences for
their comments.
REFERENCES
Bachman, J., Wadsworth, K., O’Malley, P., Johnston, L., & Schulenberg,
J. (1997). Smoking, drinking and drug use in young adulthood: The
impacts of new freedoms and new responsibilities. Mahwah, NJ:
Erlbaum.
Bayer, P., Pintoff, R., & Pozen, D. (2003). Building criminal capital behind bars: Social learning in juvenile corrections (Working Paper).
Yale, CT: Yale University, Economics Department.
Barnett, L., Far, J., Mauss, A., & Miller, J. (1996). Changing perceptions of peer norms as a drinking reduction program for college students. Journal of Alcohol and Drug Education, 41, 39–
62.
Boisjoly, J., Duncan, G., Kremer, M., Levy, D., & Eccles, J. (2004).
Empathy or antipathy? The impact of diversity (Working paper).
Institute for Policy Research, Northwestern University.
Centers for Disease Control. (2003). Youth online comprehensive
results. Youth Risk Behavior Surveillance System. Retrieved from
http://apps.nccd.cdc.gov/yrbss/QuestYearTable.asp?ByVar=CI&
cat=4&quest=Q60&loc=XX&year=2003.
Curran, P., Stice, E., & Chassin, L. (1997). The relation between adolescent alcohol use and peer alcohol use: A longitudinal random
coefficients model. Journal of Consulting and Clinical Psychology,
65, 130–140.
Dishion, T., McCord, J., & Poulin, F. (1999). When interventions harm:
Peer groups and problem behavior. American Psychologist, 54,
755–764.
Evans, W. N., Oates, W. E., & Schwab, R. M. (1992). Measuring peer
group effects: A study of teenage behavior. Journal of Political
Economy, 100, 966–991.
Gledhill-Hoyt, J., Lee, H., Strote, J., & Wechsler, H. (2000). Increased
use of marijuana and other illicit drugs at US colleges in the 1990s:
Results of three national surveys. Addiction, 95, 1655–1667.
Kaplan, M. (1979). Patterns of alcoholic beverage use among college
students. Journal of Alcohol and Drug Education, 24, 26–40.
Kremer, M., & Levy, D. (2002). Peer effect from alcohol use among
college students. Mimeo, Harvard University.
Peer Effects in Drug Use and Sex Among College Students
Manski, C. (1993). Identification of endogenous social effects: The reflection problem. Review of Economic Studies, 60, 531–542.
McCord, J. (1995). Some unintended consequences of summer camps.
Paper presented at the Society for Research on Child Development,
Indianapolis, March 30th to April 2nd, 1995.
Moffitt, R. A. (2001). Policy interventions, low-level equilibria and social interactions. In S. Durlauf & P. Young (Eds.), Social dynamics.
Cambridge, MA: MIT Press.
National Institute of Drug Abuse. (2002). NIDA Research Report—
Marijuana Abuse (NIH Publication No. 02-3859). National Institute of Health.
National Institute on Alcohol Abuse and Alcoholism. (2000). Highlights
from the Tenth Special Report to Congress. Alcohol Research and
Health, 24.
Norton, E., Lindroth, R., & Ennett, S. (1998). Controlling for the endogeneity of peer substance use on adolescent alcohol and tobacco
use. Health Economics, 7, 439–453.
Poulin, F., Dishion, T., & Barraston, B. (2001). 3-Year iatrogenic effects
associated with aggregating high-risk adolescents in cognitivebehavioral preventive interventions. Applied Developmental Science, 5, 214–224.
385
Reifman, A., Barnes, G., Dintcheff, B., Farrell, M., & Uhteg, L.
(1998). Parental and peer influences on the onset of heavier drinking among adolescents. Journal of Studies on Alcohol, 59, 311–
318.
Schall, M., Kemeny, A., & Maltzman, I. (1992). Factors associated with
alcohol use in university students. Journal of Studies on Alcohol,
53, 122–136.
Schulenberg, J., O’Malley, P., Bachman, J., Wadsworth, K., & Johnston,
L. (1996). Getting drunk and growing up: Trajectories of frequent
binge drinking during the transition to young adulthood. Journal of
Studies on Alcohol, 57, 289–304.
Valliant, P. (1995). Personality, peer influence and use of alcohol and
drugs by first-year university students. Psychological Reports, 77,
401–402.
Wechsler, H., Lee, J., Kuo, M., & Lee, H. (2000). College binge drinking
in the 1990s: A continuing problem. Journal of American College
Health, 48, 199–210.
Wechsler, H., Dowdall, G., Davenport, A., & Rimm, E. (1995).
A gender-specific definition of binge drinking among college students. American Journal of Public Health, 85, 982–
985.
Purchase answer to see full
attachment