Page 1
Genetic risk factors correlate with county-level violent crime rates
and collective disadvantage☆
J.C. Barnesa,⁎, Brian B. Boutwellb, Kevin M. Beaverc,d
aSchool of Economic, Political & Policy Sciences, The University of Texas at Dallas, 800 W. Campbell Rd., Richardson, TX 75080, United States
bCollege of Criminal Justice, Sam Houston State University, P.O. Box 2296, Huntsville, TX 77341, United States
cCollege of Criminology and Criminal Justice, Florida State University, 634 W. Call St., Tallahassee, FL 32306, United States
dCenter for Social and Humanities Research, King Abdulaziz University, Jeddah, Saudi Arabia
a b s t r a c ta r t i c l ei n f o
Available online 12 July 2013
Purpose: Social scientists have a rich tradition of uncovering the neighborhood, structural, and ecological cor-
relates of human behavior. Results from this body of evidence have revealed that living in disadvantaged
communities portends myriad negative outcomes, including antisocial behaviors. Though it has long been ar-
gued that associations between neighborhood factors and individual-level outcomes may, at least partially,
reflect genetic selection, a paucity of research has empirically investigated this possibility.
Methods: Thecurrentstudyexaminedwhetherknowngeneticriskfactorsforantisocialbehaviorwerepredictive
ofexposuretodisadvantageandviolentcrimemeasuredatthecountylevel.Drawingongenotypicdatafromthe
National Longitudinal Study of Adolescent Health, a dopamine risk scale was created based on respondents’
genotypes for DAT1, DRD2, and DRD4. County-level disadvantage was measured via Census data and violent
crime rates were measured via the FBI’s Uniform Crime Reports.
Results: Findings revealed that individuals with a greater number of dopamine risk alleles were more likely to
live in a disadvantaged county and were more likely to live in a county with higher violent crime rates.
© 2013 Elsevier Ltd. All rights reserved.
Introduction
Do individuals select the environments to which they are exposed?
Do people migrate toward certain neighborhoods? These questions
have occupied a prominent position in the minds of social scientists for
decades. Consider the following observation from Taft nearly 80 years
ago (1933:700):
Are criminals attracted into delinquency areas more frequently than
into other parts of the city? Are the children of criminals settling in
such areas delinquent entirely because they live in criminogenic
neighborhoods, or largely because they had delinquent parents? Do
people with other abnormal personality traits migrate to such areas
ratherthanelsewhere,andifsoarethesetraitsfactorsintheircriminal
behavior? Inallthese cases thequestionwould be, of course, whether
these personality types were attracted or forced into the areas; or
whether they were such on arrival rather than produced by life in
the region after arrival. In short, are delinquency areas selective, and
ifsoaretheycentersofdelinquencypartlybecausetheyareselective?
The study of the possible selective influence of areas of delinquency
seems to the writer to have been somewhat neglected.
Scholars have examined a litany of factors that might be important
for understanding the clustering of individuals within neighborhoods.
However, these studies have generally focused on social or environ-
mental forces while excluding biological and genetic factors. Early so-
ciological research was motivated to study the factors that underlie
the clustering of individuals within neighborhood settings (Shaw &
McKay, 1942). Preliminary observations led many sociologists to con-
clude that social factors must be predicted by preceding social factors
(Durkheim, 1982). In other words, sociology left no room for biolog-
ical explanations of behavior and decision-making (Udry, 1995),
suggesting instead that the clustering of individuals in certain areas
was the result of social influences.
Contemporary genetics research, however, indicates that this early
sociological explanation may be untenable (Udry, 1995; van den
Berghe, 1990). Itisnow well established thatgenetic and biological fac-
tors play a role in personality development (Bouchard et al., 1990;
Harris,1998) and perhaps evenin individuals’ selection into certain en-
vironmentsandneighborhoodconditions(Rowe&Rodgers,1997;Scarr
Journal of Criminal Justice 41 (2013) 350–356
☆ This research uses data from Add Health, a program project directed by Kathleen
Mullan Harris and designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan
Harris at the University of North Carolina at Chapel Hill, and funded by grant P01-
HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and
Human Development, with cooperative funding from 23 other federal agencies and foun-
dations.SpecialacknowledgmentisdueRonaldR.RindfussandBarbaraEntwisleforassis-
tance in the original design. Information on how to obtain the Add Health data files is
available on the Add Health website (http://www.cpc.unc.edu/addhealth). No direct sup-
port was received from grant P01-HD31921 for this analysis.
⁎ Corresponding author.
E-mail address: jcbarnes@utdallas.edu (J.C. Barnes).
0047-2352/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jcrimjus.2013.06.013
Contents lists available at ScienceDirect
Journal of Criminal Justice
Page 2
& McCartney, 1983). Along these lines, the current study builds on re-
cent genetics research in order to better understand the factors that
underpin peoples’ decisions to live in certain areas and not others. In
order to frame the current focus, we turn first to a discussion of gene-
environment correlation (DiLalla, 2002; Kendler & Baker, 2007; Scarr
& McCartney, 1983).
Gene-environment correlation (rGE)
Thephenomenon ofgene-environmentcorrelation (rGE)hasbeen a
topic of much discussion for at least the past 35 years (Plomin et al.,
1977; Scarr & McCartney, 1983). Researchers have reported that many
environmental variables are partially influenced by genetics (i.e.,
some portion of the variance is attributable to heritable factors;
Kendler & Baker, 2007). In general, two types of rGEs might underlie
the correlation between a person’s genotype and their exposure to cer-
tain neighborhood environments: active rGE and passive rGE (Scarr,
1992; Scarr & McCartney, 1983).
The first type of rGE that is applicable to the current focus is active
rGE. Active rGE occurs when a person seeks out environments to suit
their genetic proclivities, a phenomenon more commonly referred to
as niche-picking or self-selection. Research has revealed, for example,
that individuals are likely to select into environments that are compat-
ible with their genetic tendencies (e.g., Beaver et al., 2008). Active rGEs
offer a framework for understanding how genetic factors can influence
the nonrandom selection of people into particular environments, per-
haps even large-scale environments such as neighborhoods, cities, and
counties.
The second type of rGE is referred to as passive rGE. Passive rGE rec-
ognizes that parents pass alongboth an environment and genes to their
offspring. Since the child’s environmentand the child’s genes both orig-
inate from the same source (i.e., their parents) the two are likely to be
correlated. Recall that active rGE allows for the non-random sorting of
people into certain environments based on their genotype. To the ex-
tent that this occurs at the parent-level (i.e., parents choose their own
environments), we would expect the child’s genotype (which is passed
from parent to child) to be correlated with the child’s environment
(which was selected by the parents).
rGE and residential selection
Keeping the discussion of rGE in mind, sociologists have extensively
noted that residential sorting is not the result of a random process nor
does it appear to be strictly the result of environmental forces. Consider
the following comment from Zorbaugh (1929:134-135) more than
eightdecades ago: “Itis often remarked howdifficultit is toget a family
to consent to move out of the slum no matter how advantageous the
move may seem from the material point of view, and how much more
difficult it is to keep them from moving back into the slum.” More re-
cently, Sampson (2008:213) extended this statement by noting:
Humansare agents with thedecision-making power toacceptor re-
ject treatments (Heckman & Smith, 1995). Statistics on the “take-
up” rate show that a majority of MTO [Moving to Opportunity]
families who were offered a voucher did not actually use it. Families
who did use the voucher experienced less neighborhood poverty in
comparison with the noncompliers, but the vast majority remained
within a relatively short distance of their origin neighborhood.
Moreover, many families moved back into poor neighborhoods that
wereverysimilartotheones in whichthey started,surprisingmany
observers. Yet no one should be surprised at these facts. Back in the
1920s, Zorbaugh (1929) noted the “pull” of the slum and how the
strong nature of its social ties kept people returning. It is only from
a middleclass point of view, or what Zorbaugh called the “budget-
minded social agency” (1929, p. 134), that the behavior of those
who have grown up in poverty seems “incalculable.”
Sampson(2008)wascommentingontheresultsfromarecentsetof
experiments,theMTO,whichsoughttodeterminewhethermovingout
of disadvantaged neighborhoods positively impacts the lifestyles, the
health, and the behavior of individuals exposed to those conditions.
The MTO proceeded by offering moving vouchers (Section 8 vouchers)
to selected families, but to the researchers surprise, many families
passed on the opportunity to move out of their neighborhood and
into a less disadvantaged one. As reported by Kling, Liebman, and Katz
(2007), the compliance rate (i.e., the percentage of families offered
vouchers who used the voucher) was between 47 and 60 percent.
Moreover, many of the families that did accept the voucher moved
into a neighborhoodnearby oronethatmirroredtheir original location.
Totheextentthat“optingout” of theMTOisrelated to self-selection
into a neighborhood, these results offer indirect support for the active
rGE hypothesis. Namely, findings from the MTO suggest that neighbor-
hood living conditions are, at least for some, the result of self-selection
and not completely the result of environmental pressures. By extension,
these findings may also support the passive rGE hypothesis. To the ex-
tent that parents select into certain neighborhoods (active rGE), their
children’s genotypes will also tend to be correlated with neighborhood
conditions (passive rGE).
Current study
Juxtaposing the findings from the MTO with the concepts of active
rGE and passive rGE, a conceptual framework begins to emerge. As
the diagrams in Fig. 1 display, it may turn out that individuals self-
select into certain neighborhoods based on their genetic propensities
(i.e., active rGE; Panel A). It may also turn out that parents select into
certainenvironmentsand,therefore,thoseenvironmentsarecorrelated
with their children’s genotypes (i.e., passive rGE; Panel B). This is not to
A: Active rGE
B: Passive rGE
Fig. 1. Conceptual diagram of two rGEs for explaining the correlation between genotype
and environmental conditions.
351
J.C. Barnes et al. / Journal of Criminal Justice 41 (2013) 350–356
Page 3
say that genetic factors are the only influence on neighborhood selec-
tion.Rather, ourhypothesis is that genetic factorsthathave been linked
with antisocial behavior will explain a portion of the variance in differ-
ent measures of structural conditions due to active rGE and/or passive
rGE. No research, of which we are aware, has empirically considered
this hypothesis.
Methods
Data
Data for this analysis were gleaned from the National Longitudinal
Study of Adolescent Health (Add Health; Harris, 2009). The Add Health
data have been described at length elsewhere (Harris etal., 2003, 2006;
Resnick et al., 1997). Briefly, the Add Health is a nationally representa-
tive, longitudinal survey of American youth who were enrolled in mid-
dle and high school during 1995. The study began with a school-level
survey that included all students enrolled in more than 130 schools
across the United States (N ≈ 90,000). From this sample of respon-
dents,asubsampleofroughly20,000wasdrawnandutilizedinthelon-
gitudinal portion of the study. Shortly after the in-school surveys were
completed, the longitudinal subsample was administered a more
lengthy follow-up survey. These interviews were conducted in the re-
spondent’s home (i.e., wave 1). The surveys addressed a range of topics
germane to adolescence such as the respondent’s behavior, their rela-
tionships with peers, and their personality characteristics. Respondents
ranged in age between 11 and 21 years. A second wave of data were
collected a year after wave 1 (i.e., wave 2). Approximately six years
after wave 1 interviews a third wave of data were collected. During
wave 3 interviews, all respondents had reached young adulthood (age
range was between 18 and 26 years).
Two unique features of the Add Health design are capitalized upon
by the current research. First, a host of county-level measures that can
be linked with the individual respondents are available to Add Health
researchers. These measures are available for wave 1, wave 2, and
wave 3. Second, genotypic information was gathered for a subsample
of the wave 3 respondents. Twins and full siblings (where both sib-
lings were participating in the study) were asked to provide buccal
cell samples so that they could be genotyped. Genotypic information
was available for 2,574 respondents (Cohen, Feng, Florey, et al., n.d.).
After eliminating one twin from each monozygotic twin pair (to avoid
artificially decreasing standard errors; Haberstick et al., 2005) and
after eliminating cases with missing data, final analytic sample sizes
ranged between 2,212 and 2,268.
Measures
Genetic risk variable
Dopamine risk. Scholars have shown that certain dopaminergic genes
may be related to criminal and antisocial behavior (Beaver et al., 2007;
Craig & Halton, 2009). These genetic risk factors, via rGE processes, may
be related to the individual’s social environment. Included in the Add
Health was genotypic information for three dopamine polymorphisms:
DAT1, DRD2, and DRD4. Based on information gleaned from these three
polymorphisms, a dopamine risk scale was generated. DAT1 is a dopa-
mine transporter gene that has two common alleles; the 9-repeat allele
and the 10-repeat allele. The 10-repeat allele has been identified as the
risk allele (Guo et al., 2010; Rowe et al., 2001) and, as such, DAT1 alleles
were coded so that 0 = 9-repeat allele and 1 = 10-repeat allele. Respon-
dents with any other allele were assigned a missing value and were
omitted from the analyses (Hopfer et al., 2005).
DRD2 is a dopamine receptor polymorphism that has one allele
known as the A1 allele and another known as the A2 allele. The A1 al-
lele has been identified as the risk allele (Guo et al., 2007). Thus, re-
spondents with two A2 alleles were coded as 0, those with one A1
allele were coded as 1, and respondents with two A1 alleles were
coded as 2.
The DRD4 polymorphism is a dopamine receptor gene that has
been tied to antisocial and criminal behavior. The 7-repeat allele is
regarded as the risk allele (Faraone et al., 2001; Rowe et al., 2001).
Building on previous research, DRD4 alleles were coded so that the
7-repeat allele (along with the 8-, 9-, and 10-repeat alleles) = 1 and
the4-repeatallele (alongwiththe2-, 3-,5-, and6-repeatalleles) = 0.
In order to generate the dopamine risk scale, respondents’ scores on
the three dopamine polymorphisms (DAT1, DRD2, and DRD4) were
summed together. The polymorphisms were coded co-dominantly,
meaning that the value for each polymorphism reflected the number
ofriskallelescarriedbytherespondent.Thedopamineriskscaleranged
from a minimum of 0 (i.e., no dopamine risk alleles) to a maximum of
6 (i.e., six dopamine risk alleles).
County-level variables
Violent crime rate. Each year, the Federal Bureau of Investigation col-
lects crime rate data from across the U.S. These data are published
annually in the Uniform Crime Reports (UCR) and the county-level
measures were included as part of the Add Health data collection at
wave 1, wave 2, and wave 3. The violent crime rate variable was a
composite index reflecting the number of robberies, aggravated as-
saults, rapes, and homicides per 100,000 residents in the respondent’s
county. Violent crime rate data from wave 1 and wave 3 were ana-
lyzed in the present study.
Collective disadvantage. Two county-level indicators of collective dis-
advantage were calculated based on prior research from Sampson
et al., (1997). The first was generated using data from wave 1 by draw-
ing on the 1990 Census data (the decennial census closest to wave 1
data collection). The second county-level measure of collective disad-
vantage was drawn from data available at wave 3, which reflected
data drawn from the Census of Population and Housing in 2000. To cre-
ate the collective disadvantage scale, the following measures were fac-
tor analyzed: the percentage of Black residents, the percentage of
female headed households, the percentage of residents with an income
under$15,000, thepercentage ofresidentsonpublic assistance,andthe
unemployment rate. Factor analysis revealed that the correlation struc-
ture of the five items was best explained with a single latent construct
for both the wave 1 measure and the wave 3 measure. All factor load-
ings(exceptpercentonpublic assistanceatwave3 whichhad a loading
of .67) were greater than or equal to .75 and the reliability coefficient
was .92 for the wave 1 measure and .89 for the wave 3 measure. Both
scales were created using regression scoring based on the factor analy-
sis results. Higher values reflected more collective disadvantage for
both measures.
Controls
The respondent’s sex was included as a dichotomous variable
(0 = female, 1 = male) as was the respondent’s race (0 = non-Black,
1 = Black). Information on respondent race was gleaned from wave 3
interviewer reports of the respondent’s race.
Analysis plan
In order to examine the relationship between the dopamine risk
scale and the county-level measures of violent crime and collective dis-
advantage, the analysis proceeded in four steps. First, observed mean
levels of violent crime and observed mean levels of collective disadvan-
tage were calculated for respondents who had different levels of dopa-
mine risk. Second, zero-order correlation coefficients were calculated
between the dopamine risk scale, the violent crime rate, and the collec-
tive disadvantage scale. Third, OLS regression models were estimated.
In these models, the violent crime rate variables and the collective
352
J.C. Barnes et al. / Journal of Criminal Justice 41 (2013) 350–356
Page 4
disadvantage scales were each (separately) regressed on the dopamine
risk scale. The estimates gleaned from these regression models were
used to plot predicted violent crime rates and predicted collective dis-
advantage scores based on each respondent’s level of dopamine risk.
Fourth, the relationship between dopamine risk and the county mea-
sures was assessed after statistically controlling for the respondent’s
sex and race.
One final point is worth noting before proceeding to the results.
Recall that wave 1 data collection occurred when all of the Add Health
respondents were in middle or high school and the age range was
11-21 years old. Due to this feature of the data collection, any corre-
lation between the respondent’s genotype and the county-level mea-
sures (violent crime rate and collective disadvantage) is likely to be
the result of passive rGE processes. Wave 3 interviews, however,
were conducted when the respondents had reached young adulthood
and the age range was 18-26. Although it is possible that some re-
spondents remained living with their parents, it is likely that many
had moved into homes of their own. Thus, any correlation between
respondent genotype and the wave 3 county-level variables is likely
to be the result of active rGE processes. Preliminary evidence of this
point is found in the number of unique counties represented in the
analyses. At wave 1, respondents hailed from 117 different counties.
At wave 3, however, respondents lived in 387 different counties.
Findings
Before proceeding to the analysis, it was important to determine
whether the dopamine risk scale varied significantly across counties. If
not, the remainder of the analysis would be futile in that we would be
trying to explain a variable with a constant. In order to determine
whether dopamine risk and total genetic risk varied at the county
level, a two-pronged approach was followed. First, a one-way ANOVA
wasestimatedwherethedopamineriskscalewastheresponsevariable
and the respondent’s county identifier was the factor variable. The re-
sults revealed that the dopamine risk scale (F = 2.03, df = 116,
p b .0001) varied significantly across the different counties. The second
approach was to aggregate the dopamine risk scale to the county level
(by calculating the mean dopamine risk score for respondents in each
county) and observe the distributional properties of the aggregated
scale. This approach revealed that when aggregated to the county-
level thedopamineriskscale had a distribution that approximated nor-
mality (see Appendix A). Thus, a lack of variation on the dopamine risk
scale should not hinder the analyses presented below.
Mean scores on the violent crime rate and on the collective disad-
vantage scale are presented in Table 1. Notice that the mean scores
are presented separately for respondents with different levels of dopa-
mine risk. The top portion of the table presents the mean scores of vio-
lent crime rate and neighborhood disadvantage at wave 1. The bottom
portion presents the mean scores on the violent crime rate and neigh-
borhood collective disadvantage at wave 3. Presented in the first two
columns of the top portion of Table 1 are the sample size (N) and the
mean violent crime rate at wave 1. As can be seen, there were 330 re-
spondentswhohadeithera0ora1onthedopamineriskscale(because
only 30 respondents scored a 0, the 0 and 1 categories were collapsed
together. The substantive findings for all of the analyses were identical
when the categories were not collapsed). For these respondents, the
mean violent crime rate in their county of residence during wave 1
was 656.17 per 100,000 residents. As we move down the column into
higher scores on the dopamine risk scale, a pattern begins to emerge.
Specifically, mean violent crime rates at wave 1 appear to increase as
scoresonthedopamineriskscaleincrease.Thissuggeststhatdopamine
risk may be tied to county-level violent crime rates.
Presented in the last two columns of the top portion of Table 1
are sample sizes and the mean collective disadvantage scores for
respondents with different levels of dopamine risk. In broad strokes,
respondents who carried more dopamine risk tended to live in
counties with a higher level of collective disadvantage.
Moving to the bottom portion of Table 1, we see that respondents
who scored lower on the dopamine risk scale tended to live in
counties with lower (relatively) rates of violent crime and in counties
with lower (relatively) levels of collective disadvantage. This pattern
was not, however, consistent across all levels of dopamine risk. For in-
stance, respondents with 2 dopamine risk alleles, on average, lived in
counties with less violent crime than respondents who had 0 or 1 do-
pamine risk alleles. A similar finding emerged for the collective disad-
vantage measure.
Presented in Table 2 are the zero-order correlation coefficients be-
tween dopaminerisk, thecountyviolentcrimerate,and thecountycol-
lective disadvantage scale. As can be seen, the dopamine risk scale was
positively and significantly (p b .05, two-tailed) correlated with the
wave 1 violent crime rate (r = .09), with the wave 3 violent crime
rate (r = .09), with the wave 1 collective disadvantage scale (r = .10),
and with the wave 3 collective disadvantage scale (r = .09).
The next step of the analysis was to generate predicted rates of vi-
olent crime as a function of dopamine risk. In order to do so, an OLS
regression model utilizing the wave 1 violent crime rate as the depen-
dent variable and the dopamine risk scale as the independent variable
was estimated and the coefficient estimates were used to generate
predicted values (i.e., predicted violent crime rates for individuals
with different levels of dopamine risk). The predicted wave 1 violent
crime rates are presented graphically in Fig. 2, along with the coeffi-
cient estimates gleaned from the regression models in the figure cap-
tion. Consistent with the above findings, the dopamine risk scale was
positively associated with wave 1 violent crime rates. At wave 1, re-
spondents with the lowest level of dopamine risk (those with a 0 or
a 1 on the dopamine risk scale) were predicted to live in a county
with a violent crime rate of 640.37 per 100,000 residents. Respondents
Table 1
Mean county violent crime rate and mean county collective disadvantage by dopamine
risk
MeanMean
Violent Collective
Dopamine Risk
N
Crime Rate
N
Disadvantage
Wave 1
0 – 1
2
3
4
5 – 6
330
787
729
309
68
656.17
685.92
755.75
807.73
882.81
345
804
737
314
68
-.09
-.11
.03
.16
.28
Wave 3
0 – 1
2
3
4
5 – 6
346
786
712
307
63
508.13
498.36
572.54
581.14
665.84
346
786
712
307
63
-.07
-.10
.02
.15
.25
Table 2
Correlations between individual-level dopamine risk and county-level violent crime
rate and collective disadvantage
Dopamine
Risk
W1
Violent
Crime Rate
W3
Violent
Crime Rate
W1
Collective
Disadvantage
Dopamine Risk
W1 Violent Crime Rate
W3 Violent Crime Rate
W1 Collective Disadvantage
W3 Collective Disadvantage
-
.09⁎
.09⁎
.10⁎
.09⁎
-
.72⁎
.53⁎
.58⁎
-
.59⁎
.73⁎
-
.78⁎
⁎ p b .05 (two-tailed).
353
J.C. Barnes et al. / Journal of Criminal Justice 41 (2013) 350–356
Page 5
withthehighestlevel ofdopaminerisk(thosewitha5ora6)werepre-
dictedtolivein acountywith aviolentcrimerateof864.97per100,000
residents; an increase of more than 35% over those with the lowest do-
pamine risk.
AsecondOLSregression model predictedwave 3 violentcrimerates
with the dopamine risk scale. The results from this analysis are also
presented in Fig. 2. As with the wave 1 violent crime rate, dopamine
risk was positively and significantly associated with wave 3 violent
crimerates.Respondentswiththelowestdopamineriskwerepredicted
to live in a county with 483.09 violent crimes per 100,000 residents at
wave 3. Respondent with the highest level of dopamine risk were pre-
dicted to live in counties with 632.04 violent crimes per 100,000 resi-
dents at wave 3.
The resultsfrom the nexttwoOLS regressionanalyses are presented
in Fig. 3. In the first analysis, the wave 1 collective disadvantage scale is
utilized as the dependent variable and the dopamine risk scale is uti-
lized as the predictor variable. Predicted scores on the collective disad-
vantage scale are plotted as a function of dopamine risk. A positive
relationship was identified such that those with the lowest dopamine
risk were predicted to live in counties with a disadvantage score of
-.17 and those with the greatest dopamine risk were predicted to live
in counties with a disadvantage score of .23. Recall that the collective
disadvantage scale was generated as a factor score. Thus, the values
roughly correspond to z-scores (i.e., standard deviation units).
The fourth OLS regression model utilized the wave 3 collective dis-
advantage measure as the outcome variable. The same familiar pattern
emerged. In short, respondents with fewer risk alleles on the three do-
pamine variables tended to live in relatively less disadvantaged
counties at wave 3. Respondents carrying zero or one risk allele were
predicted to live in counties with a disadvantage score of -.14 while
those with five or six risk alleles lived in counties with a predicted dis-
advantage score of .20.
The final analysis re-estimated each of the regression models
presented in the figures, but this time the respondent’s sex and the re-
spondent’s race wereentered intotheregression models ascontrol var-
iables. The results of these models are presented in Table 3, which is
split into two panels. Panel A presents the results from the regression
models where the violent crime rate variables served as the dependent
variables.Model1reportsthebaselineestimatesfortheeffectofthedo-
pamine risk scale on the wave 1 violent crime rate (which are identical
to those presented in Fig. 2). Model 2 entered the sex variable into the
regression equation. Respondent sex does not significantly predict
wave 1 violent crime rate and the effect of dopamine risk was un-
changed when this variable was included. Model 3 enters the respon-
dent’s race. As can be seen, when the respondent’s race is accounted
for,thedopamineriskscalenolongerexertsastatisticallysignificantin-
fluence on wave 1 violent crime rates. Models 4 through 6 perform the
same series of analyses with the wave 3 violent crime rate variable
serving as the dependent variable. A substantively identical pattern
emerged where the effect of the dopamine risk scale was rendered sta-
tistically insignificant when the respondent’s race was controlled.
Finally, PanelB presents theresults fromsix OLS regressionmodels
where the collective disadvantage variables were utilized as the de-
pendent variables. As with the violent crime rate measures, models 7
and 10 indicated that dopamine risk significantly predicted wave 1
collective disadvantage (model 7) and wave 3 collective disadvantage
(model 10). Models 8 and 11 revealed that controlling for the respon-
dent’s sex did not affect the dopamine risk scale estimate. Models 9
and 12, however, indicated that the dopamine risk scale no longer sig-
nificantly predicted either outcome once the respondent’s race was
entered into the regression model.1
Discussion
Social scientists have long noted the impact of structural/
contextual factors (e.g., county-level variables) on individual-level
behavior (Durkheim, 1982). Despite mounds of research into these
influences, statistical significance can often be fleeting (Harden et
al., 2009) and the theoretical links between macro-level factors and
individual-level behavior often leaves much to be desired. This is
not to say that neighborhood/contextual factors have no impact on
individuals’ behavior; a series of ingenious experiments by Keizer et
al., (2008) provide strong evidence that context influences norm-
violating behavior. Instead, available research indicates that the associ-
ation between contextual factors and individual-level behavior is more
complex than has traditionally been recognized.
Findings from the current analysis offer a new perspective on the
influence of structural and contextual influences. Specifically, the
question of whether individuals self-select into certain environments
was directly addressed by analyzing the association between molecu-
lar genetic data and exposure to county-level crime rates and
county-level rates of collective disadvantage. Two important conclu-
sions were gleaned from the analysis. First, the results of a series of
statistical tests indicated that individual-level genetic risk predicted
0
100
200
300
400
500
600
700
800
900
1000
0-12345-6
Predicted Violent Crime Rate
Dopamine Risk
Wave 1
Wave 3
Fig. 2. Predicted violent crime rate as a function of dopamine risk. Note: Wave 1:
bdopamine risk= 56.15, SE = 22.08, p b .05 (two-tailed); Wave 3 bdopamine risk= 37.24,
SE = 14.59, p b .05 (two-tailed); Standard errors were corrected for the clustering of
cases within counties.
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0-12345-6
Predicted Collective Disadvantage
Dopamine Risk
Wave 1
Wave 3
Fig. 3. Predicted collective disadvantage as a function of dopamine risk. Note: Wave 1
bdopamine risk= .10, SE = .04, p b .05 (two-tailed); Wave 3 bdopamine risk= 09, SE = .03,
p b .05 (two-tailed); Standard errors were corrected for the clustering of cases within
counties.
354
J.C. Barnes et al. / Journal of Criminal Justice 41 (2013) 350–356
Page 6
county-level measures of violent crime rates and collective disadvan-
tage. These findings are important in that they suggest the sorting of
individuals into certain areas (i.e., counties) is not random and may
not be completely due to social/environmental forces. Instead, indi-
viduals who carry more “risk” alleles for antisocial behavior are
more likely to live in areas where violent crime is more prevalent
and where collective disadvantage is greater. Taken together, these
findings provide support for gene-environment correlation (rGE) hy-
potheses. The findings from the wave 1 analysis lend support for the
passive rGE hypothesis because all respondents were still in middle or
high school at the time and, therefore, were most likely to be living
with their parents. The findings from the wave 3 analysis lend sup-
port for the active rGE hypothesis because all respondents had
reached young adulthood by this time and, therefore, were more like-
ly to be living in areas of their own choosing.
The second important finding to emerge from the analysis was
that the correlation between genetic risk and the county-level vari-
ables was completely explained away by the respondent’s race.
There are two possible explanations for this finding. The first explana-
tion highlights the possibility that dopamine risk has an indirect in-
fluence on neighborhood selection, but dopamine risk is correlated
with race, thereby masking the genetic effect. In short, the first expla-
nation states that the genetic effect is not a statistical artifact. The sec-
ond explanation states that dopamine risk is a proxy for race and, due
to population stratification, blacks are more likely to live in high
crime/disadvantaged areas meaning that the genetic effect is a statis-
tical artifact. Unfortunately, identifying which of these two processes
accurately explains the current findings is empirically impossible
with the Add Health data. Thus, we are left only with theoretical con-
siderations to ponder. There is certainly evidence to support both ex-
planations and, realistically, it is likely that both processes can be
credited. To be sure, genomic data, while clearly evincing the point
that genetic variation is greatest within race rather than across race,
has also shown that racial characteristics correlate with genotype at
an impressive rate. Tang, Quertermous, Rodriguez, et al. (2005) re-
vealed a nearly perfect overlap in self-reported race and groups iden-
tified with a genetic cluster analysis. At the same time, U.S. history is
fraught with examples of racial discrimination (both historical dis-
crimination and contemporary discrimination (Wilson, 1987)) and
segregation and we would be extremely negligent if we did not con-
sider the lingering effects that these past policies, attitudes, and per-
ceptions might have on today’s society (Massey & Denton, 1993;
Wilson, 1987). In that respect, it is important to note a recent analysis
by Warren et al. (2012) which demonstrated that Whites were less
willing to reside in communities with Black residents if they also
viewed Blacks as being criminal and representing an economic liabil-
ity. Wilson (1987) appropriately summarized this position when he
noted that “…long periods of racial oppression can result in a system
of inequality that may persist for indefinite periods of time even after
racial barriers are removed” (p. 147).
We encourage future scholars to revisit these issues and we hope,
at a minimum, that these findings will spark a discussion about the
interplay between genetic factors, race, and exposure to certain
structural/contextual factors such as county-level crime rates and col-
lective disadvantage. We are sensitive to the fact that these issues are
controversial and, by some, are considered taboo. Indeed, Kemper
(1994) noted nearly 20 years ago that, “…if sociology and biology
have not been on speaking terms in general, sociological disdain for
the biological reaches its apogee when it comes to social stratifica-
tion” (p. 48). We hope that analyses such as the present one will dis-
solve some of this disdain and will allow criminology and her scholars
to contribute progressive, ethical, and moral research to 21st century
scientific discourse.
Limitations of the current study must be noted. First, data were
gleaned from the Add Health study, meaning that we were limited
tothegeneticinformation includedbytheAddHealthresearchers. Ex-
tant evidence clearly indicates that dopamine genes are not the only
genetic factors that correlate with antisocial behavior and, therefore,
it is unlikely that the three dopamine genes examined here are the
only factors that correlate with county crime rates and county collec-
tive disadvantage. Second, and related to the first, is that this analysis
was restricted to two county-level measures (violent crime rates and
collective disadvantage). The choice of these two county variables
wasobvious;a largeliteraturehasdiscussedthepotentialrolethatex-
posure to crime and disadvantage has on an individual’s behavior
(e.g., Sampson et al., 1997). However, this does not mean that other
county-level variables are unimportant and that rGEs do not exist be-
tween other genes and other environmental variables. Future work
Table 3
OLS regression estimates before and after including control variables
Panel A: Violent Crime Rate
Wave 1Wave 3
Model 1Model 2Model 3Model 4Model 5Model 6
b(SE)b(SE)b(SE)b(SE)b(SE)b(SE)
Dopamine Risk56.15⁎
(22.08)
56.15⁎
(22.13)
-.32
(22.98)
29.90
(24.27)
.37.24⁎
(14.59)
37.18⁎
(14.59)
-6.09
(14.28)
16.08
(10.57)
Male (=1)
Black (=1)458.51⁎
(145.86)
418.60⁎
(160.09)
Panel B: Collective Disadvantage
Wave 1Wave 3
Model 7Model 8 Model 9Model 10Model 11Model 12
b(SE)b(SE) b(SE)b(SE) b(SE)b(SE)
Dopamine Risk.10⁎
(.04)
.10⁎
(.04)
.02
(.04)
.03
(.03)
.09⁎
(.03)
.09⁎
(.03)
.04
(.04)
.03
(.02)
Male (=1)
Black (=1)1.16⁎
(.31)
1.02⁎
(.26)
Note: All standard errors were corrected for the clustering of cases within counties.
⁎ p b .05 (two-tailed).
355
J.C. Barnes et al. / Journal of Criminal Justice 41 (2013) 350–356
Page 7
should expand on the current study by including different genetic
markers and different environmental measures (preferably at differ-
ent levels of aggregation such as the block-group level).
Scholars have long questioned whether residential selection is the
result of social forces, individual self-selection, or some combination
of both. The logical answer is that both social and individual-level fac-
tors are at play and researchers have wrestled with trying to account
for both types of influences when analyzing the impact of structural
factors on individual-level behavior. The current study suggests that
self-selection, perhaps even at the molecular genetic level, is impor-
tant to consider when traversing these issues.
Appendix A. Histogram of county-level dopamine risk
Note
1. A Bonferroni correction was carried out to adjust for multiple testing bias. We
analyzed the impact of the dopamine risk scale on two different outcomes collected
at two time periods. Correcting the critical value for the four tests suggests a p-value
of .05/4 = .013 be used for hypothesis testing. Nearly all of the significant findings
highlighted in the paper held up when this more conservative critical value was uti-
lized. Specifically, the exact p-value for model 1 and model 2 was .014, the p-value
for model 4 and model 5 was .011, the p-value for model 7 and model 8 was .011,
and the p-value for model 10 and model 11 was .002.
References
Beaver,K.M.,Wright,J.P.,&DeLisi,M.(2008).Delinquentpeergroupformation:Evidence
of a gene X environment correlation. The Journal of Genetic Psychology, 169, 227–244.
Beaver, K. M., Wright, J.P., DeLisi, M., Walsh, A., Vaughn, M. G., Boisvert, D., et al. (2007). A
gene X gene interactionbetween DRD2 and DRD4 isassociated with conduct disorder
and antisocial behavior in males. Behavioral and Brain Functions, 3, 30.
Bouchard, T. J., Lykken, D. T., McGue, M., Segal, N. L., & Tellegen, A. (1990). Sources of
human psychologicaldifferences:The Minnesota Studyof Twins Reared Apart. Science,
250, 223–228.
Cohen, M. S., Feng, Q., Florey, F. A., Ford, C. A., Harris, K. M., Hewitt, J. K., et al. (n.d.). Bio-
markers in Wave III of the Add Health Study. The Add Health Biomarker Team.
Craig, I. W., & Halton, K. E. (2009). Genetics of human aggressive behavior. Human
Genetics, 126, 101–113.
DiLalla, L. F. (2002). Behavior genetics of aggression in children: Review and future
directions. Developmental Review, 22, 593–622.
Durkheim, E. (1982). The rules of sociological method (8th ed.). Chicago: University of
Chicago Press.
Faraone, S. V., Doyle, A. E., Mick, E., & Biederman, J. (2001). Meta-analysis of the associa-
tion between the 7-repeat allele of the dopamine D4 receptor gene and attention
deficit hyperactivity disorder. The American Journal of Psychiatry, 158, 1052–1057.
Guo, G., Cai, T., Guo, R., Wang, H., & Harris, K. M. (2010). The dopamine transporter
gene, a spectrum of most common risky behaviors, and the legal status of the be-
haviors. PLoS One, 5, e9352.
Guo, G., Roettger, M. E., & Shih, J. C. (2007). Contributions of the DAT1 and DRD2 Genes
to Serious and Violent Delinquency among Adolescents and Young Adults. Human
Genetics, 121, 125–136.
Haberstick, B. C., Lessem, J. M., Hopfer, C. J., Smolen, A., Ehringer, M. A., Timberlake, D.,
et al. (2005). Monoamine Oxidase A (MAOA) and antisocial behaviors in the
presence of childhood and adolescent maltreatment. American Journal of Medical Ge-
netics. Part B, Neuropsychiatric Genetics, 135, 59–64.
Harden, K. P., D’Onofrio, B., Van Hulle, C., Turkheimer, E., Rodgers, J. L., Waldman, I. D.,
et al. (2009). Population density and youth antisocial behavior. Journal of Child Psy-
chology and Psychiatry, 50, 999–1008.
Harris, J. R. (1998). The nurture assumption: Why children turn out the way they do. New
York: Free Press.
Harris, K. M. (2009). The National Longitudinal Study of Adolescent Health (Add Health),
Waves I & II, 1994–1996; Wave III, 2001–2002; Wave IV, 2007-2009. [machine-
readable data file and documentation]. Chapel Hill, NC: Carolina Population Center,
University of North Carolina at Chapel Hill.
Harris, K. M., Florey, F., Tabor, J., Bearman, P. S., Jones, J., & Udry, J. R. (2003). The National
Longitudinal Study of Adolescent Health: Research Design. [www document]. URL:
http://www.cpc.unc.edu/projects/addhealth/design
Harris, K. M., Halpern, C. T., Smolen, A., & Haberstick, B. C. (2006). The National Longi-
tudinal Study of Adolescent Health (Add Health) Twin Data. Twin Research and
Human Genetics, 9, 988–997.
Heckman, J. J., & Smith, J. A. (1995). Assessing the case for social experiments. Journal of
Economic Perspectives, 9, 85–110.
Hopfer, C. J., Timberlake, D., Haberstick, B., Lessem, J. M., Ehringer, M. A., Smolen, A.,
et al. (2005). Genetic influences on quantity of alcohol consumed by adolescents
and young adults. Drug and Alcohol Dependence, 78, 187–193.
Keizer, K., Lindenberg, S., & Steg, L. (2008). The spreading of disorder. Science, 322,
1681–1685.
Kemper, T. D. (1994). Social stratification, testosterone, and male sexuality. In L. Ellis
(Ed.), Social Stratification and Socioeconomic Inequality, volume 2, Reproductive
and Interpersonal Aspects of Dominance and Status (pp. 47–61). Westport, CT:
Praeger.
Kendler, K. S., & Baker, J. H. (2007). Genetic influences on measures of the environ-
ment: A systematic review. Psychological Medicine, 37, 615–626.
Kling, J. R., Liebman, J. B., & Katz, L. F. (2007). Experimental analysis of neighborhood
effects. Econometrica, 75, 83–119.
Massey, D. S., & Denton, N. A. (1993). American apartheid: Segregation and the making of
the underclass. Cambridge, MA: Harvard University Press.
Plomin, R., DeFries, J. C., & Loehlin, J. C. (1977). Genotype-environment interaction and
correlation in the analysis of human behavior. Psychological Bulletin, 84, 309–322.
Resnick, M. D., Bearman, P. S., Blum, R. W., Bauman, K. E., Harris, K. M., Jones, J., et al.
(1997). Protecting adolescents from harm: Findings from the National Longitudinal
Study of AdolescentHealth.Journal of the AmericanMedical Association, 278, 823–832.
Rowe, D. C., & Rodgers, J. L. (1997). Poverty and behavior: Are environmental measures
nature and nurture? Developmental Review, 17, 358–375.
Rowe, D. C., Stever, C., Chase, D., Sherman, S., Abramowitz, A., & Waldman, I. D. (2001).
Two dopamine genes related to reports of childhood retrospective inattention and
conduct disorder symptoms. Molecular Psychiatry, 6, 429–433.
Sampson, R. J. (2008). Moving to Inequality: Neighborhood effects and experiments
meet social structure. The American Journal of Sociology, 114, 189–231.
Sampson, R. J., Raudenbush, S., & Earls, F. (1997). Neighborhoods and violent crime: A
multilevel study of collective efficacy. Science, 277, 918–924.
Scarr, S. (1992). Developmental theories for the 1990s: Developmental and individual
differences. Child Development, 63, 1–19.
Scarr, S., & McCartney, K. (1983). How people make their own environments: A theory
of genotype➔environment effects. Child Development, 54, 424–435.
Shaw,C. R.,& McKay, H. D.(1942).Juveniledelinquencyandurban areas.Chicago:University
of Chicago Press.
Taft, D. R. (1933). Testing the selective influence of areas of delinquency. The American
Journal of Sociology, 38, 699–712.
Tang, H., Quertermous, T., Rodriguez, B., Kardia, S. L. R., Zhu, X., Brown, A., et al. (2005).
Genetic structure, self-identified race/ethnicity, and confounding in case-control as-
sociation studies. American Journal of Human Genetics, 76, 268–275.
Udry, J. R. (1995). Sociology and biology: What biology do sociologists need to know?
Social Forces, 73, 1267–1278.
van denBerghe, P. L. (1990). Whymost sociologists don’t (and won’t) think evolutionarily.
Sociological Forum, 5, 173–185.
Warren,P.Y., Stewart, E.A., Tomaskovic-Devey,D., & Gertz,M. (2012).White’s residential
preferences: Reassessing the relevance of criminal and economic stereotypes. Race
and Justice, 2, 231–249.
Wilson, W. J. (1987). The truly disadvantaged: The inner city, the underclass, and public
policy. Chicago: University of Chicago Press.
Zorbaugh, H. (1929). The gold coast and the slum: A sociological study of Chicago’s near
north side. Chicago: University of Chicago Press.
356
J.C. Barnes et al. / Journal of Criminal Justice 41 (2013) 350–356
Download full-text 
