Mate Selection in Cyberspace: The Intersection of Race, Gender, and Education

Author(s): Ken-Hou Lin and Jennifer Lundquist

Source: American Journal of Sociology , Vol. 119, No. 1 (July 2013), pp. 183-215
Published by: The University of Chicago Press
Stable URL: https://www.jstor.org/stable/10.1086/673129

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access
to American Journal of Sociology

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
Mate Selection in Cyberspace: The Intersection
of Race, Gender, and Education1

Ken-Hou Lin
University of Texas—Austin

Jennifer Lundquist
University of Massachusetts—Amherst

In this article, the authors examine how race, gender, and education
jointly shape interaction among heterosexual Internet daters. They
find that racial homophily dominates mate-searching behavior for
both men and women. A racial hierarchy emerges in the reciprocating
process. Women respond only to men of similar or more dominant ra-
cial status, while nonblack men respond to all but black women. Sig-
nificantly, the authors find that education does not mediate the ob-
served racial preferences among white men and white women. White
men and white women with a college degree are more likely to contact
and to respond to white daters without a college degree than they are
to black daters with a college degree.

INTRODUCTION
We examine how race, gender, and education jointly shape online interac-
tion among Internet daters. Evidence suggests that the Internet is replac-
ing neighborhood, family, and the workplace as a major venue to meet ro-
mantic partners ðMadden and Lenhart 2006; Rosenfeld and Thomas 2012Þ.

1
This research is supported by a University of Massachusetts Faculty Research Grant.
We thank Kevin Lewis, Joya Misra, Herbert Smith, Donald Tomaskovic-Devey, Robert
Zussman, and the reviewers for their generous comments on the early drafts of this ar-
ticle. Direct correspondence to Ken-Hou Lin, Department of Sociology, University of
Texas, Austin, Texas 78712. E-mail: lin@austin.utexas.edu

© 2013 by The University of Chicago. All rights reserved.

0002-9602/2013/11901-0005$10.00

AJS Volume 119 Number 1 ( July 2013): 183–215 183

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

The immense network and interactional data available on the Internet,

meanwhile, have been proclaimed the new paradigm in data collection, with
the potential to illuminate significant dimensions of social life ðSalganik,
Dodds, and Watts 2006; Watts 2007; Lazer et al. 2009Þ. Utilizing data ex-
tracted from one of the largest dating websites in the United States,
we explore classic sociological questions in the mate selection process: Is
racial homophily or racial hierarchy the chief driving force of assortative
coupling? If racial hierarchies matter, where is the crucial divide among
racial groups in the dating market? How do gender and education intersect
with race in the mate selection process? Empirically, we contrast the simi-
larities and differences in racial preferences between heterosexual men and
women. We also examine the extent to which education as an achieved sta-
tus mediates the ascribed status of race in determining the likelihood of
interaction among Internet daters.
Though many of these questions have been previously examined, ex-
isting literature and theories are largely limited by the empirical evidence.
Studies of interracial marriage and cohabitation observe only the outcome
but not the dynamics in the mate selection process and thus cannot sepa-
rate the effect of racial preferences from that of propinquity in generating
the observed coupling patterns. Studies of self-reported experience and at-
titudes toward interracial relations, on the other hand, are vulnerable ðand
increasingly soÞ to social desirability bias. This article extends and improves
on previous studies. By directly observing the dynamics in the mate selec-
tion process within a bounded social space, we untangle choice from cir-
cumstance and distinguish behaviors from beliefs.
We first review how previous research has approached intermarriage
and interracial dating and their limitations. Building on that literature, we
discuss how theories of homophily, racial hierarchy, and gendered racial for-
mation apply to the interaction of Internet daters. We then introduce our data
set and method, followed by a presentation of the results and a discussion of
the implications of our findings.

INTERRACIAL MARRIAGE: A PARADIGMATIC MEASURE

OF RACIAL DYNAMICS
The vast majority of research on interracial intimate relationships has
focused on marriage. Intermarriage has long been considered the ultimate
indicator of social integration and reduced racial stratification. Scholars con-
clude that marital assimilation played a critical role in the successful Amer-
ican integration of European immigrants, who were once considered “non-
white” at the turn of last century ðFoner 2002; Alba and Nee 2005; Alba
2009Þ. Although still rare, interracial marriage has become more common in
U.S. society and especially among the younger generation. Between 1970

184

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

and 2000, the prevalence of black-white marriage grew more than fivefold
from 65,000 to 363,000 couples, and marital unions between whites and
members of all other races grew almost fivefold from 233,000 to 1.1 million
ðQian 1997; Qian and Lichter 2007Þ.
Marital mate selection generally operates under a strong tendency to-
ward homogamy, where partners are similar to each other on many char-
acteristics, including racial identity. Partly as a result of population size,
intermarriage rates are lowest among whites. Yet when intermarriage oc-
curs, whites are most likely to marry Latinos and Native Americans, fol-
lowed by Asians, and least likely to marry African Americans ðQian and
Lichter 2007Þ. The pattern is consistent with the triracial hierarchy pro-
posed by Bonilla-Silva ð2004Þ, which argues that the new racial stratifi-
cation system has increasingly elevated the status of Asians and Hispanics
above blacks.
However, there is salient gender variation. White men more often inter-
marry with Asian women than they do with black women, while white
women more often marry black men than Asian men ðJacobs and Labov
2002; U.S. Census Bureau 2010Þ. Gendered racial formation theory argues
that such disparities are less explained by a one-dimensional racial hierar-
chy than by deeply instilled societal notions of desirability, which defines
ideal masculinity and femininity according to racial identity. Black men are
stereotypically depicted as hypermasculine while Asian women are often de-
picted as exotic, submissive, and more feminine. Black women and Asian
men, on the other hand, are depicted, respectively, as less feminine and less
masculine ðOmi and Winant 1994; Collins 2004; Nemoto 2006, 2008Þ.
Because intermarriage is numerically rare, most studies focus on white-
minority pairing ðQian 1997; Fu 2001; Rosenfeld 2005; Qian and Lichter
2011Þ. A natural consequence is that existing theories on mate selection are
generalized largely from white-minority coupling patterns. While the tri-
racial hierarchy may explain the likelihood of intermarriage between white
and minority groups, it sheds no light on why black-Asian pairing is equally
rare, though these two groups are claimed to inhabit shorter social dis-
tance. Gendered racial formation theory, similarly, makes claims to societal
notions of masculinity and femininity based solely on white-minority cou-
pling patterns. There is little evidence indicating that similar perceptions of
masculinity and femininity are shared across racial boundaries.
Most important, both triracial hierarchy theory and gendered racial for-
mation theory operate under the implicit assumption that the likelihood
of interracial marriage reflects white racial preference ðand not minority
preferenceÞ. And yet it may be an unrealistic assumption when we know
that continuing societal racial segregation means that marriage markets
are also largely divided. Thus, a major challenge faced by traditional stud-
ies of interracial marriage is that they are unable to detect whether the ob-

185

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

served coupling pattern results from racial preferences or the opportunity

structure prior to marriage, that is, how much exposure individuals had to
those of different backgrounds in their immediate dating market ðBlau
1977; Blau, Blum, and Schwartz 1982; Tomaskovic-Devey and Tomaskovic-
Devey 1988Þ. Indeed, attitudinal survey data show that 86% of Americans
ðand 96% of black AmericansÞ approve of interracial marriage, at least in
theory ðJones 2011Þ. Yet only 4% of U.S. marriages are interracial ðU.S.
Census Bureau 2010Þ.

INTERRACIAL DATING AND COHABITATION:

NEW TRENDS AND IMPLICATIONS
The continued use of interracial marriage as the primary measure of social
integration overlooks the rich and burgeoning life history of intimate part-
nerships. Modern changes in the American family, such as acceptance of
gay unions, marital postponement, cohabitation, and the cultural accep-
tance of nonmarital sexual relations and childbearing, have lessened the
centrality of marriage in the individual’s life course. The 2010 census in-
dicates that married couples represent less than half of all American house-
holds. Even though black-white intermarriage rates are very low, we know
that black-white cohabitation rates are much higher ðQian and Lichter
2007, 2011Þ. While nonmarital relationships may be less stable, an exclusive
focus on marriage ignores significant forms of intimate relations. Prior to
marriage, and occasionally in place of marriage altogether, individuals en-
gage in lengthy periods of dating and serial relationships that, because they
are unofficial unions, are not captured by standard surveys. A growing num-
ber of studies have begun to examine interracial pairing among the hetero-
sexual population, showing that such relationships are indeed more prev-
alent among daters than among married couples ðBlackwell and Lichter
2004; Joyner and Kao 2005Þ. More than 50% of African American, Latino,
and Asian American adults self-report that they have dated someone from
a different racial group ðYancey 2002Þ.
However, similarly to studies of intermarriage, most dating research is
unable to account for differences in group propinquity. For example, in-
terracial dating is elevated among those who have lived in racially mixed
neighborhoods or attended integrated schools ðYancey and Yancey 2002Þ.
Some studies have analyzed racial preferences listed on Internet dating ad-
vertisements as a strategy to avoid unknown levels of interracial marriage
market exposure. They find patterns similar to those found in the inter-
racial marriage literature. Using Yahoo personal ads, one study shows that
black, Hispanic, and Asian straight men are more likely than white men
to have a racial preference for a partner ðPhua and Kaufman 2003Þ. Using
data from the same source, a study comparing white men and women dat-

186

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

ers shows that white women were most likely to exclude Asian men from
their dating preferences while white men were most likely to exclude black
women ðFeliciano, Robnett, and Komaie 2009; Robnett and Feliciano 2011Þ.
Another study shows that Hispanic daters are more open to dating either
whites or blacks than either group is to date one another ðFeliciano, Lee, and
Robnett 2011Þ. Finally, profile data from match.com show that older cohorts
are less willing to date anyone outside their racial group except whites.
Whites and Asians are least willing to date blacks ðTsunokai, Kposowa,
and Adams 2009Þ.
Unfortunately, these studies also suffer the kind of social desirability bias
that typifies traditional survey data on racial attitudes in the “postracial”
United States ðPager and Quillian 2005Þ. A significant portion of Internet
daters might not believe that they themselves have any racial preference;
and, even if they do, they are unlikely to reveal it in their personal ads
because this has potential to reflect negatively on them to other potential
daters. More importantly, very little information is known about actual
dating behaviors, as opposed to stated preference, particularly those from
within the context of a known universe of partner choices.2

RACE IN CYBERSPACE: A NEW FRONTIER

Using interactional data from online sources is one way to examine what
people actually do at the beginning of the mate selection process. Since on-
line dating websites contain the full potential market from which users se-
lect and the service providers maintain complete records of the online ac-
tivities of their users, they provide a unique window to observe both the

2
Preference is an elusive concept. One could certainly make the argument that stated
preference is closer to one’s true preference because it is less contingent on the perceived
opportunity and cost structure. On the other hand, if one’s preferences are informed by
stereotypes, how one reacts to that person’s profile and online dating persona may reflect
a “truer” preference because it is based on more accurate information. The contrast be-
tween perception and reality is most salient in the case of Middle Eastern Americans.
While previous studies on dating profiles ðe.g., Robnett and Feliciano 2011Þ indicate that
white daters tend to exclude Middle Eastern American daters in their stated racial pref-
erences, our preliminary analysis shows that white daters do not avoid interacting with
Middle Eastern American daters at all. A potential explanation for this result is that Mid-
dle Easterners are “whiter” and more acculturated than they are portrayed in the media.
In any case, if we define preference as a principle of action, it is reasonable to believe that
the private interaction choice in a large dating website better captures one’s racial prefer-
ence than publicly stated preferences shown on the dating profile. We would even spec-
ulate that dating websites provide one of the best social settings to observe underlying
racial preferences, not only because no other social settings provide a comparable pool of
potential mates in both quantity and diversity but also because the cost of potential re-
jection is fairly low, compared with rejections that take place in the workplace, church,
and neighborhood.

187

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

opportunity structure and racial preference among daters at the interaction

level. Evidence shows that Internet dating is becoming an important and
popular technology for mate searching. The Pew Research Center’s Inter-
net and American Life Project shows that 74% of American Internet users
who are currently single and looking for romantic partners have used the
Internet to find dates, about half of whom have actively dated people they
met online, with 17% entering into long-term or marital relationships as a
result ðMadden and Lenhart 2006Þ. Another study finds that, other than
having access to the Internet and a certain level of computer literacy, In-
ternet daters differ little from daters who meet face to face ðSautter, Tippett,
and Morgan 2010Þ. A nationally representative survey also finds that the
Internet has begun to rival neighborhood, family, and the workplace as one
of the most common venues for couples to meet ðRosenfeld and Thomas
2012Þ. As Internet usage has become a regular feature in most Americans’
lives, researchers have increasingly come to view cyberspace activity less as
a novel activity and more as an extension of normal social behavior.
A few scholarly studies have utilized interaction data from online sources
ðe.g., Wimmer and Lewis 2010Þ. Using data from a German online dating
website, Skopek, Schulz, and Blossfeld ð2011Þ find that educational homo-
phily is the dominant pattern among daters’ initial contacts and replies. In
the minority of cases in which educational dissymmetry does occur, it is
gendered, with men contacting women of lower education. Most relevant
to our study is Hitsch, Hortaçsu, and Ariely’s analysis ð2010a, 2010bÞ on
mate preferences in an online dating setting, in which they observe strong
preferences for racial similarity among Internet daters. We improve on their
study in three important ways. First, Hitsch et al. limit their analysis to pat-
terns of initiation and thus yield incomplete knowledge of the mate selec-
tion process. This is especially problematic when the majority of all initial
contacts are made by men. Our study analyzes both the initiating and
reciprocating behaviors and shows how racial preferences are contingent
on gender and the stage of action. Second, similarly to Skopek et al. ð2011Þ,
Hitsch et al. limit the risk set of interaction to only those profiles that an
Internet dater has chosen to view more closely after first considering the pro-
file picture and summary information. This methodological approach ob-
scures the fact that the browsing behavior is inherently part of the selection
process. We instead construct an opportunity space for each individual
that includes all the probable dyads in the same metropolitan area. Finally,
because of sample size limitation, Hitsch et al. are unable to estimate the
racial preferences of the minority groups with precision. Our analysis, by
contrast, utilizes a significantly larger and geographically more diverse sam-
ple and thus provides a better understanding of the racial preferences of the
minority groups.

188

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

RESEARCH QUESTIONS
We answer three empirical questions in this article. First, do people prefer
those of similar racial identity or those with a more dominant racial status?
Second, if racial hierarchies are in operation, where is the crucial divide
among racial groups in the dating market? Finally, we explore how gender
and education intersect with race. In the next section, we discuss these
questions and the related hypotheses.

Homogamy: Homophily, Hierarchy, or Propinquity?

It is well established that homogamy, the marriage or union between in-
dividuals who are similar to each other, is the dominant pattern of mate
selection. Couples often share similar characteristics such as educational
attainment and racial identity. However, because only the end result is ob-
served, there is no direct evidence showing that homophily, the intrinsic
tendency of individuals to associate with similar others, is the driving force
behind such patterns. At least two other theories are proposed to explain the
prevalence of homogamy ðKalmijn 1994, 1998Þ. Applying a utility maximi-
zation framework, some researchers argue that individuals do not seek sim-
ilarities but, rather, higher status in the mate selection process ðe.g., Elder
1969; Becker 1991; Oppenheimer 1997Þ. When this theory is applied within
the rubric of racial hierarchies, people of marginalized racial status may
search for partners of the dominant racial group, who, in turn, may avoid
or reject those of marginalized racial status. Thus, homogamy is prevalent
only because market competition prevents those of marginalized racial
status from partnering with those of dominant racial status. Other scholars
emphasize the importance of propinquity in the partner-searching process.
That is, individuals select mates from those who are proximate in social
space. Since people of similar backgrounds have greater, if not exclusive,
contact with each other in various social settings, no intrinsic preference is re-
quired to produce a homogamous outcome ðBlau et al. 1982; Tomaskovic-
Devey and Tomaskovic-Devey 1988Þ. Analyzing pictures on Facebook, a
recent study also shows that friendship formation is less linked to race than
to dorm coresidence and past elite high school attendance ðWimmer and
Lewis 2010Þ.
These explanations yield distinct predictions on the behavior of Internet
daters with regard to racial preference. If homophily theory is the principle
of mate selection, we should observe that Internet daters, in general, are
more likely to interact with members of their own group than daters of all
other groups. If racial hierarchy is the main driving force of mate selection,
we should see that Internet daters of marginal racial status are more likely
to approach and respond to those of dominant status, while the latter avoid

189

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

and reject the former.3 Finally, if propinquity is the sole social force that
generates observed racial homogamy, we should see no significant role of
racial identity in online interaction.4

Racial Hierarchy in Dating Markets: Where Is the Critical Line Drawn?

Second, we explore whether and how racial hierarchies in the United States
shape mate selection processes. Because of sample size limitations, most em-
pirical studies focus on only black-white intermarriage, where blacks are
considered to be the marginal group and whites the dominant group. This
dichotomous model has been challenged by contemporary racial dynamics,
particularly the influx of Hispanic and Asian immigrants since the 1960s. One
might ask where the critical line is now being drawn among racial groups as
the United States gradually transforms from a biracial to a multiracial so-
ciety. Some argue that the divide in the U.S. racial hierarchy is between
whites and racialized others ðBlauner 2001Þ; some argue that the divide is
between blacks and nonblacks ðFeagin 2000Þ; and others advocate a tri-
racial stratification model, where Asian and Hispanic Americans are in
the middle of the hierarchy ðBonilla-Silva 2004Þ.
Existing evidence supports the latter two theories more than the first.
Hispanic and Asian Americans have higher intermarriage and cohabita-
tion rates with whites than blacks do ðQian and Lichter 2007Þ. Reported
racial preference studies also show that whites and Asians are less willing
to date blacks ðTsunokai et al. 2009Þ, while white and Hispanic daters are
preferred by each other ðFeliciano et al. 2009, 2011Þ. In employment out-
comes, Asians tend to be occupationally close to whites, with blacks the
most distant ðStainback and Tomaskovic-Devey 2012Þ. The literature gen-
erates hypotheses largely about the behavior of white Internet daters. If
the racial divide is mainly between whites and nonwhites ðBlauner 2001Þ,
we should see whites preferring to date only other whites, with equal de-
grees of rejection for all minority racial groups. Alternatively, if the black

3
It should be noted that white daters are expected to behave similarly in both scenarios
since homophily and hierarchy are indistinguishable among members of the dominant
racial group. Therefore, the critical test of these two hypotheses is whether minority
daters prefer daters from their own group over those of more dominant racial status.
4
Because the propinquity hypothesis focuses on opportunity structure instead of in-
trinsic preference, observing tendencies toward homophily or racial preference does not
suggest that this hypothesis should be rejected. In this article, we do not directly test the
effect of propinquity. However, the absence of homophily or racial preference would in-
dicate a high likelihood that propinquity is a major explanation for observed racial ho-
mogamy. Note that while Internet dating may reduce local barriers such as work and
neighborhood segregation, daters are still likely to limit their mate searches to those they
will be able to meet in person. Thus, in our analyses we limit the online dating market
opportunity structure to the city level.

190

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

and nonblack framework is most applicable ðFeagin 2000Þ, white daters

should be open to dating all racial groups except blacks. On the other hand,
the triracial hierarchy theory ðBonilla-Silva 2004Þ suggests that whites
should prefer to date other whites first and foremost, followed by Asians and
Hispanics, and least likely to date blacks. How racial hierarchies might
characterize the preferences of minority groups is an open question. While
much of the literature suggests that minority groups should prefer to date
whites, it is unclear how ethnic minorities would differentially rank other
minority groups in their dating preferences. We therefore also explore hier-
archy patterns among minority Internet daters.

Gender and Education

The third topic we examine in this article is the role of gender and edu-
cation in the interracial mate selection process. Previous studies have ex-
amined gender differences in the likelihood of forming interracial relations
but found mixed results. Attitudinal surveys show that women report being
more open to interracial relationships than men and the gender difference
is more salient among whites than among blacks ðJohnson and Marini 1998;
Meier, Hull, and Ortyl 2009Þ. Yet the census shows that white men are more
likely to out-marry than white women ðU.S. Census Bureau 2010Þ. Research
on dating profiles also finds that white women are more likely than white
men to exclude nonwhites from their stated racial preference ðFeliciano et al.
2009; Robnett and Feliciano 2011Þ. Studies on online ðHitsch et al. 2010bÞ
and speed dating ðFisman et al. 2006, 2008Þ also show that women have
stronger in-group preferences than men. This motivates us to further ex-
amine the difference in homophily tendencies between men and women.
Furthermore, race often interacts with gender in the marriage market.
Asian women are more likely than Asian men to marry whites, even when
the cases of “war brides” are controlled for; the reverse gender pattern is
true for blacks, though the gap has been decreasing over time ðJacobs and
Labov 2002Þ. Gendered racial formation theory attributes such results to
societal notions of masculinity and femininity that vary by race ðCollins
2004; Nemoto 2006, 2008Þ. Black men are stereotypically depicted as hy-
permasculine and Asian women are commonly described as feminine and
submissive, while Asian men and black women are, respectively, charac-
terized as emasculated and unfeminine. If this claim is valid, we should ob-
serve that Asian women and black men receive more positive responses
not only from whites but also from other minority groups. On the other
hand, the theory predicts that Asian men and black women should receive
fewer responses from whites, as well as other minority groups, in the dat-
ing market. Thus, gendered racial formation theory introduces added nu-
ance around the previous triracial hierarchy hypothesis, predicting that

191

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

men are likely to place black women at the bottom of the preference hi-
erarchy, while women are likely to do so with Asian men.
In addition to testing how gender interacts with racial-ethnic identity,
we examine whether educational status mediates the effect of race. Recent
demographic studies ðSchwartz and Mare 2005; Fu and Heaton 2008; Hou
and Myles 2008Þ show that, while there has been a slow decrease in racial
homogamy over time, educational homogamy has risen since the 1960s.
This development is consistent with the thesis that achieved status is be-
coming increasingly more important than ascribed status in the mate se-
lection process ðKalmijn 1991Þ. Other studies analyze interactions between
men and women on a German online dating website and find that educa-
tional homophily is the most dominant mechanism in online mate choice
ðSkopek et al. 2011Þ. We thus contrast the effects of racial homophily and
those of educational homophily in our analysis. If educational homophily
is prioritized over racial homophily, we should observe that white daters
prefer minority daters with similar education status over white daters of
lower educational status. If racial homophily remains the most dominant
sorting mechanism, we should observe that white daters prefer white dat-
ers over minority daters, regardless of their educational status.5

DATA, VARIABLES, AND METHOD

Data
We obtained the data from one of the largest U.S. dating and social net-
working websites, which facilitates both heterosexual and same-sex dating
for millions of active users. Similarly to most dating websites, registered
users can create a personal profile, search and view other users’ profiles, and
contact fellow users through a website-based messaging system. A typical
user profile contains basic information such as sex, sexual orientation, geo-
graphical location, age, race, height, body type, religion, language, lifestyle,
and socioeconomic status, as well as photographs and short essays. Unlike
most large dating websites that charge a membership fee to contact other
users, this website places no restriction on searching, viewing, sending, and
responding to messages, which, we believe, makes this website one of the
best data sources for studying online dating behaviors in the United States.
It should also be noted that this website does not recommend potential
matches by ethnic-racial status. The only criteria used to select which pro-

5
We are aware of the recent debate on status exchange theory among intermarriage
scholars ðRosenfeld 2005; Gullickson and Fu 2010; Kalmijn 2010Þ. However, since the
theory provides only post hoc prediction on the socioeconomic asymmetry of interracial
couples, it is not clear how to test its validity at the beginning of the mate selection pro-
cess. We thus plan to explore this topic with a different set of analyses in future.

192

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

files to display are age, sexual orientation, and the matching score that is
derived from personality questions.
Though we do not claim that our data set is representative of the general
population, it is more powerful than conventional survey data in a number
of ways.6 First, this data set contains actual interactions among Internet
daters, which allows us to observe what people do instead of what people
say. Second, since our data set is generated from interactions within a defi-
nite population, it allows us to examine how race determines the likelihood
of interaction in a bounded probability space. Third, because all the vari-
ables are extracted from digital records, our data set is largely immune to
measurement problems such as social desirability bias and recall errors
that are common in conventional survey data. Fourth, the size of our data
set gives us the opportunity to explore not only the interaction between
whites and minorities but also the interaction among minority groups. Fi-
nally, because we have access to almost as much information as the users
on the website, we are confident that our estimates are less biased by un-
observed variables.
The original data set consists of approximately 9 million registered users
worldwide and 200 million messages, from November 2003 to October 2010.
In essence, the data set consists of numerous social networks in which the
users are nodes with various attributes and the messages are directional
ties that connect nodes. However, in contrast to typical social network data,
both our nodes and ties have a temporal property: each user has a definite
lifetime and each tie is formed at a specific time point.
To facilitate the analysis, we filter the users in four steps. First, we limit
our scope to users who reside in the 20 largest metropolitan areas in the
United States.7 This facilitates the reconstruction of opportunity structure

6
We consider two types of sample selection biases when analyzing this data set. First,
among those searching for potential partners, online daters are likely to be younger and
of higher socioeconomic status than those who do not have access to the Internet. Since
age and education status are both associated with more positive attitudes toward in-
terracial relations, we would expect Internet daters to have more inclusive racial pref-
erences than the general population. Second, Internet daters who use general dating
websites are likely to have more liberal attitudes toward dating across racial lines than
those who use ethnic dating websites exclusively. Thus, taking these together, we expect
to underestimate the significance of racial preferences in the U.S. dating market.
7
The metropolitan areas alphabetically include 12060 Atlanta–Sandy Springs–Marietta,
Ga.; 12420 Austin–Round Rock, Tex.; 12580 Baltimore-Towson, Md.; 14460 Boston-
Cambridge-Quincy, Mass.–N.H.; 16980 Chicago-Naperville-Joliet, Ill.-Ind.-Wis.; 19100
Dallas–Fort Worth–Arlington, Tex.; 19820 Detroit-Warren-Livonia, Mich.; 26420
Houston–Sugar Land–Baytown, Tex.; 31100 Los Angeles–Long Beach–Santa Ana,
Calif.; 33100 Miami–Fort Lauderdale–Pompano Beach, Fla.; 33460 Minneapolis–St.
Paul–Bloomington,Minn.-Wis.;35620 NewYork–NorthernNewJersey–LongIsland,N.Y.-
N.J.-Pa.; 37980 Philadelphia-Camden-Wilmington, Pa.-N.J.-Del.-Md.; 38060 Phoenix-
Mesa-Scottsdale, Ariz.; 38900 Portland-Vancouver-Beaverton, Ore.-Wash.; 41740 San

193

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

ðdiscussed belowÞ and brings down the sample size to about 3 million dat-
ers. Second, we exclude users who did not send or receive at least one mes-
sage, who did not upload at least one photograph, who listed their birth year
later than 1992 or earlier than 1911, or who fit the profile of spammer users.8
The reason is that, similarly to most free membership websites, some of the
users did not actively engage with or even return to the website after initial
registration and a few users are likely to be fake identities created by
spammers. We thus retain only genuine dating website members, that is,
users who had the opportunity to legitimately interact with other users in
the data set. Third, we exclude daters who were looking only for casual sex
or platonic relationships to ensure that the patterns observed among the
daters reflect the mate selection process. Finally, we exclude from the anal-
ysis in this article users who identified as gay or bisexual, a population we
explore in a separate paper. Our final sample consists of 528,800 straight
men and 405,021 straight women.
We identify users’ racial identity using the information on their personal
profiles. There are 10 ethnicity boxes the users can check when they fill out
their personal profiles. The options are Asian, Middle Eastern, black, Na-
tive American, Indian, Pacific Islander, Hispanic/Latino, white, other, and
undeclared.9 Users can check as many boxes as they prefer. We categorize
those who did not check any box as undeclared and those who checked more
than one box as multiracial. Our initial sample thus consists of 11 ethnic-
racial groups, with the 10 default categories and the multiracial group.
Table 1 presents the ethnic and racial composition of our analytical sam-
ple. About half of the users self-identify as white.10 About 24% of the sam-
ple did not specify any ethnic and racial identity.11 Blacks and Hispanics

Diego–Carlsbad–San Marcos, Calif.; 41860 San Francisco–Oakland–Fremont, Calif.;

42660 Seattle-Tacoma-Bellevue, Wash.; 45300 Tampa–St. Petersburg–Clearwater, Fla.;
47900 Washington-Arlington-Alexandria, D.C.-Va.-Md.-W.V.
8
After consulting with the managers of the website, we identify spammers as users ð1Þ who
did not answer any personality question, ð2Þ whose profiles were deleted or blacklisted, and
ð3Þ who had an account lifetime that was shorter than an hour.
9
It should be noted here that, while Indians are conventionally categorized as Asian,
they are treated as an independent ethnic group in the data set and are not included in
the Asian category throughout our analysis.
10
From now on we use white and non-Hispanic white interchangeably.
11
We conducted three sensitivity tests around these respondents and determined that
their missing ethnic identity more reflects tendency in reporting behavior than conscious
omission. First, we had a group of undergraduates examine photographs for respondents
who left their race missing. In a sample of 150 profiles, no systematic pattern emerged in
which people of one particular ethnicity were any more or less represented. Second, we
conducted factor and logistic analyses predicting the correlates of missing on race and
found that being missing on height, education, smoking behavior, and drinking behav-
iors correctly predicted being missing on race 93% of the time. Third, as table 1 indicates,
racially undeclared daters receive among the fewest e-mails, indicating that other daters

194

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

TABLE 1
Ethnic-Racial Composition of the Sample

Race Men ð%Þ Women ð%Þ

Asian . . . . . . . . . . . . 3.28 3.69
Middle Eastern . . . . . .36 .26
Black . . . . . . . . . . . . 3.35 3.86
Native American . . . . .23 .26
Indian . . . . . . . . . . . . .81 .47
Pacific Islander . . . . . .25 .24
Hispanic . . . . . . . . . . 4.97 4.66
White . . . . . . . . . . . . 52.05 53.67
Other . . . . . . . . . . . . 1.47 1.48
Undeclared . . . . . . . . 25.24 24.36
Multiracial . . . . . . . . 7.99 7.05
N ........... . . . . 528,800 405,021

are significantly underrepresented in our sample. About 4% of the users

identified themselves as black ðthis is only slightly higher when excluding
“undeclared”Þ in contrast to 12% in the U.S. population. About 5% of the
users identified as Hispanic/Latino or white Hispanic in contrast to 16% of
the U.S. population.12 One group that is overrepresented is multiracial users
ðat between 7% and 8%Þ, who are often counted as 2% in the national
population.13 For the following analyses, we focus only on Asian, black, His-
panic, and white users.14

Dependent Variables
We focus our analysis on initial messages exchanged between any two users
who both reside in the same metropolitan area. These filters yield 2,750,893

dismiss them as illegitimate or less engaged prospects because their profile is incomplete
on other important factors as well.
12
Since Hispanic is often conceptualized as an ethnic rather than a racial category, those
who are conventionally counted as Hispanic might identify with both Hispanic and white
on the website. Our analysis shows that such inclusion would add two additional per-
centage points for the Hispanic population.
13
Our ongoing analysis of biracial groups shows that users who identify as white-Asian,
white-Hispanic, and white–Native American behave similarly to those who identify
only as white and do not show a particular preference for users of their minority identity.
However, those who identify as white-black show a preference for both users who iden-
tify as white only and those who identify as black only.
14
Though Hispanic is officially defined as an ethnic rather than a racial category, we
believe that it is analytically meaningful to juxtapose it with Asian, black, and white as a
distinct racial group. The reason is not only that Hispanics, along with Asians, are often
considered as occupying the racial middle ðBonilla-Silva 2004; O’Brien 2008Þ but also
that scholars argue that the Hispanic population has been increasingly racialized in the
past decades ðMassey 2007Þ.

195

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

initial messages from straight men to straight women and 1,180,260 from
straight women to straight men. Although we do not know the content of
these messages, we believe that they are an indication of romantic attraction
between users, not just because the website is explicitly geared toward dat-
ing singles, but also because of the related statistics on these initial messages
ðsee table 2Þ. Initiation and response rates are, on the whole, quite low across
all the groups, which indicates that when solicitations and responses do oc-
cur, they are meaningful behaviors in this setting. Furthermore, we find
asymmetrical interaction between men and women. Men in our sample sent
80% more messages and received two-thirds fewer messages than women.
Looking at total numbers of initial messages sent and response rates among
men and women, we see that only 2.6% of the messages sent by men were
responded to, in contrast with 5.6% of messages sent by women. Some of this
difference is a result of the gender imbalance of supply and demand on the
website. Not only are there 30% more male members, but male members, on
average, have a 35% longer account lifetime than female members ðsee ta-
bles 1 and 3Þ.
The second half of table 2 presents the means and the standard deviations
of the number of initial messages received by gender and racial groups. We
find that race operates differently between men and women. Among men,
white men receive significantly more initial messages than any other group.
Asian men, on the other hand, receive the fewest unsolicited messages, fol-
lowed closely by Hispanic and black men. For women, Asian women, on av-
erage, received the most messages, followed by white and Hispanic women.
Black women received the fewest messages.
A comparison of this sort reveals an unequal opportunity structure
among daters of different racial groups in a mainstream dating market and
is in alignment with the basic hierarchical ordering predicted by gendered
racial formation theory. However, because it focuses solely on the receiving

TABLE 2
Descriptive Statistics on Initial Messages Exchanged among Users

STRAIGHT MEN STRAIGHT WOMEN

Mean SD Mean SD
Average no. messages received . . . . . . . . . . . . . 2.232 4.694 6.792 10.244
Average no messages sent . . . . . . . . . . . . . . . . 5.202 30.142 2.914 12.356
Total no. messages sent . . . . . . . . . . . . . . . . . . 2,750,893 1,180,260
Others’ response rate ð%Þ . . . . . . . . . . . . . . . . 2.55 5.62
Average no. messages received by racial identity:
Asian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.309 2.931 8.825 11.54
Black . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.489 3.159 3.246 5.016
Hispanic . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.554 3.225 7.052 9.888
White . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.959 5.392 7.959 10.74

196

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

side of the interaction, it does not shed light on how unequal group com-
position and the likely variance in racial preference across racial groups
may jointly generate the observed differences. This shortcoming motivates
our multivariate analysis. Following previous studies on interracial friend-
ship ðHallinan and Teixeira 1987; Hallinan and Williams 1989Þ, we shift
the analytical focus to dyadic interaction. We first examine, among all prob-
able dyads, how the racial identity of the potential sender and that of the
receiver jointly predict whether an initial message is sent. To do so, we ran-
domly sample a subset of users and reconstruct their opportunity structure
on the website, which generates all probable dyads on the website for this
subset of users.15 We then merge these dyads with the initial messages that
were actually sent, yielding a binary outcome in which 1 indicates that the
probable dyad was realized and 0 otherwise.
We also examine the likelihood of responding to an initial message.
Particularly, we estimate how the likelihood of a response is conditional on
both the racial identity of the sender and that of the receiver. The sample of
this analysis is all initial messages sent among the daters, each with a bi-
nary outcome in which 1 indicates that the initial message was responded
to by the receiver and 0 otherwise.

Control Variables
Since racial identity tends to correlate with other critical mate selection
variables such as parental status, socioeconomic status, and body type, the
observed racial discrepancies might be biased upward. That is, some ob-
served racial disparities might reflect preference on other important as-
pects of social life that are correlated with race rather than racial prefer-
ence. Four sets of variables are utilized to control for confounding factors:

15
We randomly sample 300 users by racial identity, by gender, and from each metro-
politan area who joined the website in 2009. We construct the opportunity space as such:
Say there are n men of racial group i and m women of racial group j in a given metro-
politan area. Presumably, each man can send m initial messages, and so the total number
of potential combinations of initial messages from a man of group i to a woman of group
j is n  m. Among these dyads, we exclude any cases in which the account lifetime of the
potential sender did not overlap with that of the potential receiver. We also exclude the
cases in which the potential receiver is younger or older than the potential sender’s de-
fault age range. By the website’s default, the minimum age a man sees on the website is
his age/2 1 7, and the maximum is ðhis age 1 2Þ  10/9; for women, the minimum is her
age  9/10 2 2 and the maximum is ðher age 2 7Þ  2. We then calculate the overlap
between two users for each dyad as a measure of exposure. Finally, for ease of compu-
tation, we randomly draw 1 million dyads from each sender group as the sample of
analysis. It should be noted that, by doing so, we make the assumption that the daters do
not discriminate spatially within their own metropolitan area. This is an assumption we
need to make because of the lack of more detailed information on daters’ location.

197

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

basic demographic information, lifestyle, socioeconomic status, and the

degree of online engagement. Basic demographic information includes age,
height, body type, and geographic location.16 Lifestyle variables include
smoking, drinking, drug use habits, and parental status/preferences. So-
cioeconomic status consists of education level, income level, and the num-
ber of languages one speaks. In addition, we control for the daters’ degree of
engagement, which includes total time spent on the website, total account
lifetime ðfrom registration to the last log-inÞ, the number of photographs
uploaded to the website, and the number of personality questions answered
on the website.
Table 3 presents the descriptive statistics for the control variables. Be-
cause we are mainly interested in these variables only for how they might
confound our variable of interest, race, we review just a few highlights from
the table. As one might expect, the average age of the sample is younger than
that of the general population. The users also have higher educational at-
tainment: 35% of men and 40% of women report a college degree or more,
in contrast with 32.8% of Americans ages 25–34.17 The heights of male and
female users are also about 10 centimeters/4 inches higher than the aver-
age American ðhttp://www.cdc.gov/nchs/fastats/bodymeas.htmÞ. We suspect
that this is due to the tendency for users to overreport their height rather
than the selectivity of Internet daters. In any cvse, inconsistencies of this
sort should not bias the results since other daters are also using this same
information, however exaggerated, to determine whether or not to interact
with any given online dater.
Table 3 also shows that there are several differences between men and
women, all of which are reflective of gender-specific social expectations.
For example, a possible indication of the greater scrutiny to which female
bodies are subject, women users are more likely than men to consider them-
selves overweight, while men are more likely to describe themselves as fit.
Gender differences also extend to reporting behaviors. Female daters more
often report their parental status or attitude toward children than their male
counterparts do. Male daters, on the other hand, are more prone to report
their income level than female daters. There are also differences in online
engagement. Although women spend similar amounts of time on the web-

16
Owing to anonymity concerns, we do not have access to birth dates beyond year for
each user. The age here is calculated as the difference between one’s birth year and the
year he or she last logged in to the website. If someone did not report his or her height,
we impute the average height minus a standard deviation by gender. For very few cases,
we also top- or bottom-code one’s height by gender if the reported number is dramat-
ically above or below the mean. Three dichotomous variables are added in our model to
indicate imputation, top-coding, and bottom-coding in the regression analysis.
17
U.S. Census Bureau, 2010 Current Population Survey, table 3 ðhttp://www.census.gov
/hhes/socdemo/education/data/cps/2010/tables.htmlÞ.

198

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

site, they have shorter account lifetimes. This might reflect the gender dy-
namic of the website that female users are in greater demand and thus can
more quickly find potential dates. Men, on the other hand, tend to have
longer account lifetimes and answer more personality questions on their
profiles.

Analytical Strategy
We examine how racial identity on both ends of the dyad jointly predicts
the likelihood of sending and responding to an initial message. We estimate
a separate model for each race-gender group ðsuch as Asian women, black
men, etc.Þ and thus allow racial preference to vary by racial identity and
gender. We specify the model predicting the sending behavior for users of
group k as
 
Pðyi; j 5 1Þ q
ln 5 a1; k 1 b1;k Gj 1 o bp;2;k Xp; j
1 2 Pðyi; j 5 1Þ p51

1 b3;k Wi; j 1 b4;k Mi 1 εi; j ;

where yi; j denotes whether an initial message was sent from i to j, G

denotes the group membership of user j, X denotes all other attributes of
user j in table 3, and q denotes the number of attributes. The variable W
denotes the log-transformed overlapping website membership periods
between i and j, which is a measure of exposure time in which the potential
sender and potential receiver in the same metropolitan area could have
encountered one another on the website. We expect W to be positively
associated with the likelihood of i sending an initial message to j but
with diminishing returns. The variable M is a popularity control for the
sender i, denoting the number of messages user i received from other daters
per 100 days. A given dater’s popularity is likely to influence how inclined
he or she is to initiate contact with others. Since we fit a separate model for
each race-gender group, we allow all the coefficients to vary by race-gender
group. Similarly, we specify the model predicting the responding behavior
for users of group k as
 
Pðyj;i 5 1Þ q
ln 5 a2;k 1 b5;k Gi 1 o bp;6;k Xp;i
1 2 Pðyj;i 5 1Þ p51

1 b7;k Sj;i 1 b8;k Mj 1 εj;i ;

where yj;i denotes whether user j responded to the initial message sent by
user i, G denotes the group membership of user i, X denotes all other at-
tributes of user i in table 3, q denotes the number of attributes, S denotes
the matching score between j and i, and M denotes the number of messages

199

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 TABLE 3
Descriptive Statistics on Control Variables

Straight Men Straight Women

Age . . . . . . . . . . . . . . . . . . . . . 30.1 29.54
ð9.25Þ ð9.47Þ
Height ðcmÞ . . . . . . . . . . . . . . . 177.36 165.29
ð7.85Þ ð6.59Þ
Height not specified . . . . . . . ... .208 .191
Height top coded . . . . . . . . . ... .0009 .0007
Height bottom coded . . . . . . ... .0012 .0006
Body type:
Thin . . . . . . . . . . . . . . . . . . . .061 .084
Overweight . . . . . . . . . . . . . . .056 .173
Average . . . . . . . . . . . . . . . . . .165 .159
Fit . . . . . . . . . . . . . . . . . . . . .243 .108
Unspecified . . . . . . . . . . . . . . .475 .476
Region:
Northeast . . . . . . . . . . . . . . . .328 .375
Southeast . . . . . . . . . . . . . . . . .107 .093
Midwest . . . . . . . . . . . . . . . . .136 .132
West . . . . . . . . . . . . . . . . . . . .329 .308
Southwest . . . . . . . . . . . . . . . .100 .092
Smoking:
Yes . . . . . . . . . . . . . . . . . ... .212 .193
No . . . . . . . . . . . . . . . . . ... .548 .587
Unspecified . . . . . . . . . . . ... .24 .22
Drinking:
Often . . . . . . . . . . . . . . . . . . .106 .106
Socially . . . . . . . . . . . . . . . . . .482 .516
Rarely . . . . . . . . . . . . . . . . . . .12 .122
Not at all . . . . . . . . . . . . . . . .069 .054
Unspecified . . . . . . . . . . . . . . .223 .202
Drug use:
Never . . . . . . . . . . . . . . . ... .445 .413
Sometimes . . . . . . . . . . . . ... .281 .24
Unspecified . . . . . . . . . . . ... .274 .347
Parental status/preference:
Has children . . . . . . . . . . . . . .092 .131
Likes children . . . . . . . . . . . . .373 .399
Doesn’t like/want children . . . .067 .072
Unspecified . . . . . . . . . . . . . . .468 .398
Education:
High school or less . . . . . . . . . .301 .287
Some college . . . . . . . . . . . . . .04 .034
College . . . . . . . . . . . . . . . . . .277 .305
Professional . . . . . . . . . . . . . . .075 .096
Unspecified . . . . . . . . . . . . . . .307 .278
Income:
<$20,000 . . . . . . . . . . . . . . . . .057 .054
$20,000 –$50,000 . . . . . . . . . . .124 .082
$50,000 –$80,000 . . . . . . . . . . .061 .026
$80,000 –$150,000 . . . . . . . . . . .035 .008
>$150,000 . . . . . . . . . . . . . . . .019 .006
Unspecified . . . . . . . . . . . . . . .704 .824

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

TABLE 3 (Continued )
Straight Men Straight Women
Engagement:
Online time ðin 15 min.Þ . . . . . 145.14 149.67
ð535.87Þ ð531.82Þ
Account lifetime ðin daysÞ . . . 342.21 253.73
ð458.05Þ ð386.05Þ
Photos uploaded . . . . . . . . . . 3.453 3.595
ð2.526Þ ð2.589Þ
Questions answered . . . . . . . . 183.44 148.98
ð360.15Þ ð305.5Þ
NOTE.—Numbers in parentheses are SDs.

user j received per 100 days.18 We expect S to be positively associated and

M to be negatively associated with the likelihood of j responding to the
initial message from user i.
Since interaction decisions are nested within individuals ði in the sending
model and j in the responding modelÞ, a dependence structure is expected.
We thus model both the sending and the responding behaviors by fitting
a series of generalized estimating equations ðGEEs; Liang and Zeger 1986;
Hanley et al. 2003; Zuur et al. 2009Þ with the logit link function and an
exchangeable correlation structure.19

RESULTS
Figure 1 presents the results in two 44 matrices ðestimates are shown in
table A1 in the appendixÞ, where the rows represent the race of the poten-
tial sender and the columns represent the race of the potential receiver. With
the diagonal used as the reference, cell ½i, j represents the predicted odds
ratios of a user of group i sending an initial message to a user of group j.
The left matrix of figure 1 presents the sending pattern of female users.
Within each matrix, the darker the shading in the cell, the more likely the
sender ðleftÞ is to send a message to the receiver ðtopÞ. Looking first at Asian
women, we see that they are most likely to send initial messages to Asian
men followed by white men and least likely to message Hispanic and black

18
The matching score is determined by the answers to the personality questions on the
website; it is shown to both users as an indicator of compatibility.
19
There are advantages to analyzing our data with the GEE approach. First, the GEE
approach addresses dependency among observations and optimizes the statistical power
of the correlated data by estimating clustered correlations. In contrast to mixed effects or
hierarchical models, the GEE approach makes little demand of within-cluster variance
and thus is more suitable in our situation in which the participation of the users follows a
power-law distribution and a significant number of our observations are singletons. We
believe that exclusion of the singletons would create serious selection bias and therefore
do not think that the random intercept approach is suitable for our analysis.

201

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

F IG . 1.—Predicted odds ratios of sending an initial message ðdarker cells represent

higher probabilitiesÞ.

men. Black women show the highest levels of homophily. They rarely mes-
sage white, Asian, and Hispanic men. Hispanic women are also most likely
to message their coethnics, though the tendency is not as strong as it is for
black women. Hispanic women’s second preference is white men, and they
rarely initiate contact with Asian or black men. Finally, white women most
prefer white men, their second preference is Hispanic men, and they rarely
send initial messages to other minority men. Stated from the men’s per-
spective, white men have the best odds of being contacted by women even
if all racial groups are equally represented on the dating website, largely
because they are among the top choice groups for Asian, Hispanic, and
white women. Asian and black men, on the other hand, receive messages
only from their coethnics.
Among men, we observe a similar but weaker tendency toward homo-
phily. Asian men contact primarily Asian women and, secondarily, white
and Hispanic women; they are unlikely to contact black women. Similarly,
black men contact other black women first and foremost but also send mes-
sages to Hispanics and Asians and, slightly less frequently, to white women.
Hispanic men likewise prefer their own group first but also send messages
to whites and Asians while least frequently contacting black women. Finally,
white men contact other white women most but secondarily Hispanic and
Asian women. Like Asians, they almost completely avoid contacting black
women. Told another way, the main story here is the experience of black
women, who receive the lion’s share of their messages from black men, a tiny
amount from Latino men, and practically no messages from either Asian or
white men. Asian and white women, on the other hand, consistently receive
messages from all men, both inside and outside their ethnic group.

202

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

Thus far, our finding is consistent with previous findings and shows that
racial homophily dominates daters’ sending behavior. Women in general
send messages only to their coethnics or to white men, and men, while ap-
pearing to cross some ethnic boundaries with relative fluidity, draw the line
at black women. These results shed light on the gendered racial disparities
observed in the descriptive statistics ðtable 2Þ. The disadvantage suffered
by Asian men, black men, and black women is not entirely driven by white
preference as many previous studies have assumed. Instead, the experience
of Asian and black daters on this online dating site is driven by the sending
behaviors of nonwhite groups as well. Furthermore, these results show that
the reason black men receive more messages than Asian men in table 2 is
not that black men are more popular in general but that black women have
greater homophily tendency than Asian women. Overall, our results con-
tradict the popular belief that black men prefer white women over black
women and white men prefer Asian women over white women. Black men
in fact demonstrate the strongest homophily tendency among male daters.
Our next set of logistic models focuses on the response behaviors of dat-
ers when the sender has demonstrated interest. That is, we ask, among all
initial messages received, how does the racial identity of the sender and
receiver affect which messages are more likely to receive a response? We
refer to this as our reciprocal models.
Figure 2 shows the results of the response models with two 44 matrices
ðestimates in table A2 in the appendixÞ. Figure 2 is set up similarly to figure
1 with the diagonal as the reference, but here cell ½ j, i represents the odds
ratios of group j responding to the initial message sent by group i. We first
focus on the likelihood of women responding to men who have initiated
contact. Looking first at the responses of Asian women, it becomes clear
that, when given a choice, Asian women are most likely to respond to white
men, followed by Asian men. They are less likely to respond to Hispanic
men or black men. Black women, by contrast, respond to daters who con-
tact them fairly equally, with a preference for white men. The responding
behavior of Hispanic women is comparable to that of Asian women. They
are most responsive to white men, followed by their coethnics, and least re-
sponsive to black men. White women’s reciprocal behaviors look little dif-
ferent from their sending behaviors. They respond predominantly to white
men. In brief, black men are least likely to receive responses from anyone
except black women, Hispanic and Asian men are somewhere in the mid-
dle, and white men enjoy the highest likelihood of response.
Next we examine the reciprocal behavior among men. Asian men’s re-
sponse patterns indicate a willingness to respond equally to white and co-
ethnic women. But they tend to ignore messages from black and Hispanic
women. Black men, like their female coethnics, operate with the most equal
opportunity across the board ðthe differences are not statistically signifi-

203

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

F IG . 2.—Predicted odds ratios of responding to an initial message ðdarker cells rep-

resent higher probabilitiesÞ.

cant; see table A2Þ. Hispanic men, when approached, appear to be most
interested in Asian women, but the likelihood is not statistically greater
from that of Hispanic women. Similarly to their sending behavior, Hispanic
men particularly avoid responding to black women. Finally, white men are
equally likely to respond to Asian, Hispanic, and white women but not to
black women.
Overall, the reciprocal models present a picture very different from that
of the sending models. While homophily dominates the sending behavior,
racial hierarchy overshadows the responding process. Daters tend to re-
spond to those of the dominant or same racial status while rejecting racially
marginalized groups. Messages from white men and women are likely to be
reciprocated by daters of other groups, but white women reciprocate only
to white men. Black daters, particularly black women, tend to be ignored
when they contact nonblack groups, even though they do not discriminate
against any out-groups.
Finally, we examine how education might mediate the observed racial
preferences by adding an interaction term between race and educational
level. To simplify the analysis, we focus on white men and white women,
the largest group of daters. We analyze these interaction effects first in
sending behaviors in figure 3 and then in responding behaviors in figure 4
ðestimates are shown in table A3 in the appendixÞ.
Figure 3 shows the predicted likelihoods that white daters with and
without a college degree contact each of the racial and educational sub-
groups. The results show that, regardless of their own educational level,
white women are still more likely to contact white men than any other

204

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

F IG . 3.—Predicted probability of sending an initial message, white daters

group. College-educated white women even prefer non-college-educated

white men over college-educated Asian men. White men show similar
preferences as we saw in the previous sending models. Black women, with
or without a college degree, are marginalized as the least contacted group.
When it comes to response patterns, as shown in figure 4, we again see
persistent racial preference. White women are more likely to respond, over-
all, to men with a college degree than to men without; however, this behav-
ior does not break the constraints of race. White women respond most often
to white men above all other ethnicities. College-educated white women
treat college-educated minority men similarly to those without a college de-
gree. This tendency to privilege a man’s whiteness over his achieved status
is even more pronounced among non-college-educated women, who are
even more likely to respond to white men’s messages regardless of their
level of education.
Educational attainment of potential dating partners also means little to
white men in mitigating the ethnic response patterns we saw earlier in

205

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

F IG . 4.—Predicted probability of responding to an initial message, white daters

figure 3. White, Hispanic, and Asian women are equally likely to receive
responses from white men whether they have a college degree or not, while
black women are least likely to receive any response. One notable excep-
tion is that college-educated white men appear to be more responsive to
college-educated black women than to black women who lack a college
degree. Overall, figures 3 and 4 indicate that educational homophily in gen-
eral matters far less than racial homogamy in predicting white women’s
and men’s sending and responding behavior.

DISCUSSION
Utilizing data from one of the largest dating websites in the United States,
we examine how race, gender, and education jointly shape the interaction
among Internet daters. Existing studies of interracial marriage and cohab-
itation observe only the outcome of mate selection and thus provide limited
knowledge on how racial preferences shape the dynamics in the searching

206

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

process. Attitudinal studies, on the other hand, are vulnerable to social de-
sirability bias. This article improves on previous studies by directly observ-
ing the dynamics in the mate selection process within a bounded social space.
We extend previous studies using online dating data by examining both the
initiating and reciprocating behaviors. Our analysis shows that racial pref-
erences are not only gendered but contingent on the stage of action.
We find that homophily dominates the searching behavior for both men
and women. When Internet daters search for potential mates, they are most
likely to approach those who have the same racial identity. Yet a salient
racial hierarchy dominates in the reciprocating process. White men’s and
women’s messages are likely to be reciprocated by daters of other groups,
but white women reciprocate mostly only to white men. Black daters, par-
ticularly black women, tend to be ignored when they contact nonblack groups,
even though they respond to out-groups no less frequently. Asian and His-
panic daters seem to be at the middle of the racial hierarchy. They are re-
sponsive to whites, their coethnics, and to some extent each other but not to
black daters. Importantly, we find that education does not mediate the ob-
served racial preferences among white men and women. White men and
women with a college degree prefer to contact and reciprocate to white dat-
ers without a college degree over black daters with a college degree.
While previous studies show that the preference for similarity is the
dominant pattern of interaction ðHitsch et al. 2010b; Skopek et al. 2011Þ,
our results suggest that both homophily and racial hierarchies are at play
in generating the observed pairing pattern. Daters tend to search for po-
tential mates within racial boundaries, even when the social and physical
barriers to contacting other groups are relatively weak in cyberspace. Ra-
cial hierarchy, on the other hand, manifests itself most strongly in the re-
ciprocating process. Conditional on confounding factors, daters tend to re-
spond to those of equal or more dominant racial status and ignore those
of more marginalized groups. These findings suggest that racial bound-
aries operate differently at various levels of social interaction. At the stage
of initiation, the distinction is placed between in-groups and out-groups,
while at the stage of reciprocation, the actors are likely to consider those of
dominant racial status as equals or more desirable while rejecting mar-
ginalized groups.
Where is the crucial divide among racial groups in the dating market?
Our results suggest that the racial divide is gendered. For American men,
the racial hierarchy of preference and privilege appears to fall into Bonilla-
Silva’s ð2004Þ triracial stratification model, with white men at the top, re-
ceiving the most contacts and fewest rejections, followed by Asians and
Hispanics in the middle and black men trailing at the bottom of the hier-
archy. But for American women on the dating market the racial divide in
preference and privilege appears to be in line with Feagin’s ð2000Þ black

207

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

versus nonblack model. Most importantly, our results show that although
these patterns differ by gender, a common thread is that black women
and black men are clearly disadvantaged in the dating market relative to
other ethnic groups. Being black on the dating market—particularly being
a black female—means that one’s invitations are most likely to be ignored.
Black men and women are the only ones who regularly respond to one an-
other.
While some attitudinal surveys suggest that women have more liberal
attitudes toward interracial relationships ð Johnson and Marini 1998; Meier
et al. 2009Þ, our results are consistent with studies of stated preferences
ðFeliciano et al. 2009; Robnett and Feliciano 2011Þ and studies of online
interaction ðHitsch et al. 2010b; Skopek et al. 2011Þ, indicating that men,
in fact, are more willing than women to date out-groups. We are hesitant,
however, to conclude that men are less race conscious than women, given
that men and women confront a differing terrain of demand and supply in
the dating market. On the basis of the fact that women receive many more
messages than men and that there are more men than women populating
dating websites, men may simply be less able to be as selective as women
can. Furthermore, since women’s own social status has been historically
more dependent on that of their partners because of the norms of patriar-
chal tradition ðSpickard 1991; Root 2001Þ, it may still be more socially ac-
ceptable for men to date out-groups than for women.
Our findings show some support for gendered racial formation theory
ðOmi and Winant 1994; Collins 2004; Nemoto 2006, 2008Þ, but only in-
sofar as racial preferences are conditioned by gendered conceptions of race.
While Asian women receive responses from men of all groups, Asian men
do not receive responses from white women. Yet when we see how white
women respond to other minorities, we find that Asian men are no more
marginalized than black and Hispanic men. Our results also contradict
some of the other predictions of gendered racial formation theory, such as
the notion that black men should prefer black women less than they do
Asian and white women or that white and Hispanic women should be less
responsive to Asian men than they are to black men. This is not to deny the
prominence of controlling images or racialized and gendered stereotypes
that are associated with certain groups. Rather, when it comes to this par-
ticular economy of romantic interaction, the pattern is more monolithic
than that stipulated by gendered racial formation theory. Simply stated,
white women prefer white men over nonwhite men while white men pre-
fer nonblack women over black women. Furthermore, because gendered
racial formation theory juxtaposes the racialized experience of minority
women with that of their minority male counterparts, it has the potential
to obscure the magnitude of distinctions. For example, an implicit sugges-
tion of gendered racial formation theory is that Asian men’s marginaliza-

208

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

tion is equivalent to that of black women. Our results, however, make a

strong case that the discrimination against black women is the single lar-
gest marginalization of note.
Does education, an increasingly important achieved status, trump racial
preferences in the dating market? The answer is a resounding no, at least in
our sample. Though college-educated daters in general receive more un-
solicited messages than their non-college-educated counterparts, white men
without a college degree still receive more messages than college-educated
black and Asian men. College-educated black women receive fewer mes-
sages than other women of any education level. Furthermore, we find that,
for white male and female daters, race of potential daters has a far greater
effect than education does in predicting an online interaction. White men
and women with a college degree are more likely to contact and recipro-
cate to white daters without a college degree over black daters with a col-
lege degree.
Overall, our study has three theoretical implications for future studies
of intermarriage and racial relations. First, we show that mate selection
should be conceptualized as a multistep process. An emphasis on outcomes
such as marriage or cohabitation is useful in delineating national trends of
racial relations but provides little insight into how race operates in the
mate selection process. Our analysis suggests that racial boundaries are
more salient when daters are searching for potential mates and more po-
rous when responding to others’ initiation. That is, though daters in general
do not search outside their coethnics, women are open to interacting with
those of a similar or more dominant racial status and men are open to in-
teracting with all but black women.
Second, previous theories often presume that coupling patterns reflect
white preferences. Our study suggests that the racial preferences of minor-
ities are likely to be as consequential in generating the observed patterns.
For example, gendered racial formation theory attributes the prevalence
of Asian women–white men pairing to white men’s preference toward ste-
reotypically submissive women. Yet we do not find that white men show
particular preference for Asian women. Instead it is Asian women who are
more responsive to white men.
Finally, our study finds that the racial hierarchy pattern does not man-
ifest itself as a difference in desirability but as a difference in exclusivity in
the dating market. Daters of all groups are first and foremost attracted to
daters of their own group. Yet when attracted to other groups, only some
daters are welcomed to cross the borders. In this sense, racial boundaries
function similarly to one-way turnstile gates. While nonblack daters, par-
ticularly white men, are well received when they contact daters of other
groups, black daters, particularly black women, are largely confined to a
segregated dating market.

209

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
APPENDIX
TABLE A1
Partial Coefficients and SEs of Logistic Regressions Predicting Sending
an Initial Message with Full Controls ðCorresponds to Fig. 1Þ

Sending Pattern Asian Black Hispanic White

Female:
To Asian men . . . . . . . … 21.904*** 21.186*** 21.166***
ð.198Þ ð.209Þ ð.213Þ
To black men . . . . . . . 21.229*** … 2.996*** 2.960***
ð.250Þ ð.176Þ ð.172Þ
To Hispanic men . . . . . 2.843*** 21.564*** … 2.396**
ð.207Þ ð.158Þ ð.135Þ
To white men . . . . . . . 2.544** 21.613*** 2.375** …
ð.167Þ ð.125Þ ð.114Þ
Unique senders . . . . . . 2,766 3,036 3,571 8,961
Male:
To Asian women . . . . . … 2.862*** 2.490*** 2.182*
ð.0847Þ ð.0810Þ ð.0927Þ
To black women . . . . . 21.933*** … 21.292*** 21.134***
ð.155Þ ð.106Þ ð.127Þ
To Hispanic women . . . 21.166*** 2.639*** … 2.381***
ð.104Þ ð.0690Þ ð.0872Þ
To white women . . . . . 2.842*** 21.180*** 2.483*** …
ð.0777Þ ð.0660Þ ð.0587Þ
Unique senders . . . . . . 2,257 3,027 3,582 7,977
NOTE.—We estimate a separate logistic regression for each race-gender group. Error terms
are assumed to be correlated within each sender with an exchangeable structure. The in-group
is the reference group in all eight models. N 5 1,000,000. Numbers in parentheses are SEs.
* P < .05.
** P < .01.
*** P < .001.

TABLE A2
Partial Coefficients and SEs of Logistic Regressions Predicting Responding
to an Initial Message with Full Controls ðCorresponds to Fig. 2Þ

Responding Pattern Asian Black Hispanic White

Female:
To Asian men . . . . . . … .0475 .0829 2.636***
ð.196Þ ð.120Þ ð.0386Þ
To black men . . . . . . 2.599*** … 2.421*** 2.615***
ð.166Þ ð.113Þ ð.0392Þ
To Hispanic men . . . . 2.197 .0851 … 2.504***
ð.109Þ ð.124Þ ð.0282Þ
To white men . . . . . . .198** .234** .213*** …
ð.0651Þ ð.0731Þ ð.0573Þ
N ............... 101,727 40,202 102,409 1,325,340
Unique responders . . . 12,717 10,043 15,108 178,405
Male:
To Asian women . . . . … 2.0893 .161 .0368
ð.181Þ ð.102Þ ð.0272Þ

210

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 TABLE A2 (Continued )
Responding Pattern Asian Black Hispanic White
To black women . . . . 2.482** … 2.585*** 2.282***
ð.165Þ ð.123Þ ð.0408Þ
To Hispanic women . . . 2.224 .138 … .0111
ð.153Þ ð.127Þ ð.0289Þ
To white women . . . . . .00724 .0912 2.113 …
ð.0767Þ ð.0722Þ ð.0596Þ
N ............... 15,828 19,170 28,243 567,733
Unique responders . . . 6,611 7,331 10,929 157,556
NOTE.—We estimate a separate logistic regression for each race-gender group. Error terms
are assumed to be correlated within each responder with an exchangeable structure. The in-
group is the reference group in all eight models. Numbers in parentheses are SEs.
* P < .05.
** P < .01.
*** P < .001.

TABLE A3
Partial Coefficients and SEs of Logistic Regressions with
Full Controls, White Daters ðCorresponds to Figs. 3 and 4Þ

WHITE MALE WHITE FEMALE

Noncollege College Noncollege College
Sending pattern:
To Asian noncollege . . . . 2.272* 2.512*** 2.776** 21.375***
ð.117Þ ð.147Þ ð.249Þ ð.394Þ
To Asian college . . . . . . 2.772*** 2.312*** 21.491*** 21.138***
ð.120Þ ð.0818Þ ð.253Þ ð.184Þ
To black noncollege . . . . 21.283*** 21.359*** 21.245*** 21.489***
ð.141Þ ð.178Þ ð.184Þ ð.259Þ
To black college . . . . . . . 21.367*** 2.983*** 21.252*** 2.543***
ð.160Þ ð.123Þ ð.255Þ ð.165Þ
To Hispanic 2.365*** 2.478*** 2.671*** 21.179***
noncollege . . . . . . . . . ð.0868Þ ð.111Þ ð.143Þ ð.222Þ
To Hispanic college . . . . 2.527*** 2.468*** 2.476* 21.205***
ð.121Þ ð.107Þ ð.192Þ ð.241Þ
To white noncollege . . . . … 2.208** … 2.382**
2.079 ð.127Þ
To white college . . . . . . 2.254*** … .0851 …
ð.0715Þ ð.102Þ
Responding pattern:
To Asian noncollege . . . . 2.0244 .0224 2.413*** 2.776***
ð.0941Þ ð.0749Þ ð.118Þ ð.122Þ
To Asian college . . . . . . .0702 .00178 2.409*** 2.799***
ð.0818Þ ð.0396Þ ð.0893Þ ð.0646Þ
To black noncollege . . . . 2.344*** 2.304** 2.668*** 2.884***
ð.0973Þ ð.100Þ ð.0920Þ ð.105Þ
To black college . . . . . . . 2.206 2.374*** 2.518*** 2.596***
ð.116Þ ð.0741Þ ð.118Þ ð.0744Þ
To Hispanic .0272 2.130* 2.420*** 2.866***
noncollege . . . . . . . . . ð.0610Þ ð.0656Þ ð.0598Þ ð.0708Þ

211

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

TABLE A3 (Continued )
WHITE MALE WHITE FEMALE
Noncollege College Noncollege College
To Hispanic college . . . . .0135 .0179 2.422*** 2.497***
ð.0984Þ ð.0586Þ ð.0915Þ ð.0613Þ
To white noncollege . . . . … 2.078*** … 2.289***
ð.0213Þ ð.0215Þ
To white college . . . . . . .0411 … .101*** …
ð.0267Þ ð.0233Þ
NOTE.—We estimate a separate logistic regression for each gender-education group. Error
terms are assumed to be correlated within each responder with an exchangeable structure.
The in-group is the reference group in all eight models. Numbers in parentheses are SEs.
* P < .05.
** P < .01.
*** P < .001.

REFERENCES
Alba, Richard. 2009. Blurring the Color Line: The New Chance for a More Integrated
America. Cambridge, Mass.: Harvard University Press.
Alba, Richard, and Victor Nee. 2005. Remaking the American Mainstream: Assimila-
tion and Contemporary Immigration. Cambridge, Mass.: Harvard University Press.
Becker, Gary Stanley. 1991. A Treatise on the Family. Cambridge, Mass.: Harvard Uni-
versity Press.
Blackwell, Debra L., and Daniel T. Lichter. 2004. “Homogamy among Dating, Cohab-
iting, and Married Couples.” Sociological Quarterly 45 ð4Þ: 719–37.
Blau, Peter M. 1977. Inequality and Heterogeneity: A Primitive Theory of Social Struc-
ture. New York: Free Press.
Blau, Peter M., Terry C. Blum, and Joseph E. Schwartz. 1982. “Heterogeneity and
Intermarriage.” American Sociological Review 47:45–62.
Blauner, Bob. 2001. Still the Big News: Racial Oppression in America. Philadelphia:
Temple University Press.
Bonilla-Silva, Eduardo. 2004. “From Bi-racial to Tri-racial: Towards a New System of
Racial Stratification in the USA.” Ethnic and Racial Studies 27 ð6Þ: 931–50.
Collins, Patricia Hill. 2004. Black Sexual Politics: African Americans, Gender, and the
New Racism. New York: Routledge.
Elder, Glen H. 1969. “Appearance and Education in Marriage Mobility.” American
Sociological Review 34 ð4Þ: 519–33.
Feagin, Joe R. 2000. Racist America: Roots, Current Realities, and Future Reparations.
New York: Routledge.
Feliciano, Cynthia, Rennie Lee, and Belinda Robnett. 2011. “Racial Boundaries among
Latinos: Evidence from Internet Daters’ Racial Preferences.” Social Problems 58 ð2Þ:
189–212.
Feliciano, Cynthia, Belinda Robnett, and Golnaz Komaie. 2009. “Gendered Racial
Exclusion among White Internet Daters.” Social Science Research 38 ð1Þ: 39–54.
Fisman, Raymond, Sheena S. Iyengar, Emir Kamenica, and Itamar Simonson. 2006.
“Gender Differences in Mate Selection: Evidence from a Speed Dating Experiment.”
Quarterly Journal of Economics 121 ð2Þ: 673–97.
———. 2008. “Racial Preferences in Dating.” Review of Economic Studies 75 ð1Þ: 117–32.
Foner, Nancy. 2002. From Ellis Island to JFK: New York’s Two Great Waves of Im-
migration. New Haven, Conn.: Yale University Press.

212

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

Fu, Vincent Kang. 2001. “Racial Intermarriage Pairings.” Demography 38 ð2Þ: 147–59.
Fu, Xuanning, and Tim B. Heaton. 2008. “Racial and Educational Homogamy: 1980 to
2000.” Sociological Perspectives 51 ð4Þ: 735–58.
Gullickson, Aaron, and Vincent Kang Fu. 2010. “Comment: An Endorsement of Ex-
change Theory in Mate Selection.” American Journal of Sociology 115 ð4Þ: 1243–51.
Hallinan, Maureen T., and Ruy A. Teixeira. 1987. “Opportunities and Constraints:
Black-White Differences in the Formation of Interracial Friendships.” Child Devel-
opment 58 ð5Þ: 1358–71.
Hallinan, Maureen T., and Richard A. Williams. 1989. “Interracial Friendship Choices
in Secondary Schools.” American Sociological Review 54 ð1Þ: 67–78.
Hanley, James A., Abdissa Negassa, Michael Edwardes, and Janet E. Forrester. 2003.
“Statistical Analysis of Correlated Data Using Generalized Estimating Equations: An
Orientation.” American Journal of Epidemiology 157 ð4Þ: 364–75.
Hitsch, Günter, Ali Hortaçsu, and Dan Ariely. 2010a. “Matching and Sorting in Online
Dating.” American Economic Review 100 ð1Þ: 130–63.
———. 2010b. “What Makes You Click?—Mate Preferences in Online Dating.”
Quantitative Marketing and Economics 8 ð4Þ: 393– 427.
Hou, Feng, and John Myles. 2008. “The Changing Role of Education in the Marriage
Market: Assortative Marriage in Canada and the United States since the 1970s.”
Canadian Journal of Sociology 33 ð2Þ: 337–66.
Jacobs, Jerry A., and Teresa G. Labov. 2002. “Gender Differentials in Intermarriage
among Sixteen Race and Ethnic Groups.” Sociological Forum 17:621–46.
Johnson, Monica Kirkpatrick, and Margaret Mooney Marini. 1998. “Bridging the Ra-
cial Divide in the United States: The Effect of Gender.” Social Psychology Quarterly
61 ð3Þ: 247–58.
Jones, Jeffrey M. 2011. “Record-High 86% Approve of Black-White Marriages.” Gallup.
http://www.gallup.com/poll/149390/record-high-approve-black-white-marriages.aspx.
Joyner, Kara, and Grace Kao. 2005. “Interracial Relationships and the Transition to
Adulthood.” American Sociological Review 70 ð4Þ: 563–81.
Kalmijn, Matthijs. 1991. “Status Homogamy in the United States.” American Journal of
Sociology 97 ð2Þ: 496–523.
———. 1994. “Assortative Mating by Cultural and Economic Occupational Status.”
American Journal of Sociology 100 ð2Þ: 422–52.
———. 1998. “Intermarriage and Homogamy: Causes, Patterns, Trends.” Annual Re-
view of Sociology 24:395–421.
———. 2010. “Comment: Educational Inequality, Homogamy, and Status Exchange in
Black-White Intermarriage.” American Journal of Sociology 115 ð4Þ: 1252–63.
Lazer, David, et al. 2009. “Computational Social Science.” Science 323 ð5915Þ: 721–23.
Liang, Kung-Yee, and Scott L. Zeger. 1986. “Longitudinal Data Analysis Using Gen-
eralized Linear Models.” Biometrika 73 ð1Þ: 13–22.
Madden, Mary, and Amanda Lenhart. 2006. “Online Dating.” Pew Internet and Amer-
ican Life Project. http://www.pewinternet.org/Reports/2006/Online-Dating.aspx.
Massey, Douglas S. 2007. Categorically Unequal: The American Stratification System.
New York: Russell Sage Foundation.
Meier, Ann, Kathleen E. Hull, and Timothy A. Ortyl. 2009. “Young Adult Relationship
Values at the Intersection of Gender and Sexuality.” Journal of Marriage and Family
71 ð3Þ: 510–25.
Nemoto, Kumiko. 2006. “Intimacy, Desire, and the Construction of Self in Relationships
between Asian American Women and White American Men.” Journal of Asian Amer-
ican Studies 9 ð1Þ: 27–54.
———. 2008. “Climbing the Hierarchy of Masculinity: Asian American Men’s Cross-
Racial Competition for Intimacy with White Women.” Gender Issues 25 ð2Þ: 80–100.
O’Brien, Eileen. 2008. The Racial Middle: Latinos and Asian Americans Living beyond
the Racial Divide. New York: New York University Press.

213

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
American Journal of Sociology

Omi, Michael, and Howard Winant. 1994. Racial Formation in the United States: From
the 1960s to the 1990s. New York: Routledge.
Oppenheimer, Valerie Kincade. 1997. “Women’s Employment and the Gain to Marriage:
The Specialization and Trading Model.” Annual Review of Sociology 23:431–53.
Pager, Devah, and Lincoln Quillian. 2005. “Walking the Talk? What Employers Say
versus What They Do.” American Sociological Review 70 ð3Þ: 355–80.
Phua, Voon Chin, and Gayle Kaufman. 2003. “The Crossroads of Race and Sexuality:
Date Selection among Men in Internet ‘Personal’ Ads.” Journal of Family Issues 24
ð8Þ: 981–94.
Qian, Zhenchao. 1997. “Breaking the Racial Barriers: Variations in Interracial Marriage
between 1980 and 1990.” Demography 34 ð2Þ: 263–76.
Qian, Zhenchao, and Daniel T. Lichter. 2007. “Social Boundaries and Marital Assim-
ilation: Interpreting Trends in Racial and Ethnic Intermarriage.” American Socio-
logical Review 72 ð1Þ: 68–94.
———. 2011. “Changing Patterns of Interracial Marriage in a Multiracial Society.”
Journal of Marriage and Family 73 ð5Þ: 1065–84.
Robnett, Belinda, and Cynthia Feliciano. 2011. “Patterns of Racial-Ethnic Exclusion by
Internet Daters.” Social Forces 89 ð3Þ: 807–28.
Root, Maria. 2001. Love’s Revolution: Interracial Marriage. Philadelphia: Temple Uni-
versity Press.
Rosenfeld, Michael J. 2005. “A Critique of Exchange Theory in Mate Selection.” Amer-
ican Journal of Sociology 110 ð5Þ: 1284–1325.
Rosenfeld, Michael J., and Reuben J. Thomas. 2012. “Searching for a Mate: The Rise
of the Internet as a Social Intermediary.” American Sociological Review 77 ð4Þ:
523–47.
Salganik, Matthew J., Peter Sheridan Dodds, and Duncan J. Watts. 2006. “Experi-
mental Study of Inequality and Unpredictability in an Artificial Cultural Market.”
Science 311 ð5762Þ: 854–56.
Sautter, Jessica M., Rebecca M. Tippett, and S. Philip Morgan. 2010. “The Social
Demography of Internet Dating in the United States.” Social Science Quarterly 91
ð2Þ: 554–75.
Schwartz, Christine, and Robert D. Mare. 2005. “Trends in Educational Assortative
Marriage from 1940 to 2003.” Demography 42:621– 46.
Skopek, Jan, Florian Schulz, and Hans-Peter Blossfeld. 2011. “Who Contacts Whom?
Educational Homophily in Online Mate Selection.” European Sociological Review 27
ð2Þ: 180–95.
Spickard, Paul R. 1991. Mixed Blood: Intermarriage and Ethnic Identity in Twentieth-
Century America. Madison: University of Wisconsin Press.
Stainback, Kevin, and Donald Tomaskovic-Devey. 2012. Documenting Desegregation:
Racial and Gender Segregation in Private Sector Employment since the Civil Rights
Act. New York: Russell Sage Foundation.
Tomaskovic-Devey, Barbara, and Donald Tomaskovic-Devey. 1988. “The Social Struc-
tural Determinants of Ethnic Group Behavior: Single Ancestry Rates among Four
White American Ethnic Groups.” American Sociological Review 53 ð4Þ: 650–59.
Tsunokai, Glenn T., Augustine J. Kposowa, and Michele A. Adams. 2009. “Racial Pref-
erences in Internet Dating: A Comparison of Four Birth Cohorts.” Western Journal of
Black Studies 33 ð1Þ: 1–15.
U.S. Census Bureau. 2010. America’s Families and Living Arrangements. Table FG-4.
http://www.census.gov/population/www/socdemo/hh-fam/cps2010.html.
Watts, Duncan J. 2007. “A Twenty-First Century Science.” Nature 445 ð7127Þ: 489.
Wimmer, Andreas, and Kevin Lewis. 2010. “Beyond and Below Racial Homophily:
ERG Models of a Friendship Network Documented on Facebook.” American Journal
of Sociology 116 ð2Þ: 583–642.

214

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms
 Mate Selection in Cyberspace

Yancey, George. 2002. “Who Interracially Dates: An Examination of the Characteristics

of Those Who Have Interracially Dated.” Journal of Comparative Family Studies 33
ð2Þ: 179–90.
Yancey, George, and Sherelyn Whittum Yancey. 2002. Just Don’t Marry One: Inter-
racial Dating, Marriage, and Parenting. Valley Forge, Pa.: Judson.
Zuur, Alain F., Elena N. Ieno, Neil Walker, Anatoly A. Saveliev, and Graham M.
Smith. 2009. Mixed Effects Models and Extensions in Ecology with R. New York:
Springer.

215

This content downloaded from

103.197.103.156 on Mon, 09 Jun 2025 06:21:14 UTC
All use subject to https://about.jstor.org/terms