THE SOCIAL DEMOGRAPHY OF INTERNET DATING IN THE UNITED STATES* Jessica M. Sautter, Duke University Rebecca M. Tippett, Duke University S. Philip Morgan, Duke University

This is a preprint of an Article accepted for publication in Social Science Quarterly © 2010 Southwestern Social Science Association.

*All authors contributed equally and share authorship of this article. Direct all correspondence to Rebecca M. Tippett, Department of Sociology, Duke University, PO Box 90088, Durham, NC 27705 (rtippett@soc.duke.edu). Data and coding used in this article are available upon request for those wishing to replicate this study. This research was partially supported by a contract, (N01 HD-3-3354; PI. S. Philip Morgan) "Designing New Models for Explaining Family Change and Variation," with the National Institute of Child Health and Human Development. Data collection was carried out and funded by the Pew Foundation with partial support from Duke University. The authors wish to thank Emilio A. Parrado, Seth Sanders, Lee Rainie, John Horrigan, Lynn Smith-Lovin, Amanda Lenhart, Mary Madden, Douglas Downey, Linda K. George, and Nathan D. Martin for helpful comments and suggestions. Early versions of this article were presented at the 2005 Southern Demographic Association Annual Meeting and the 2006 Southern Sociological Society Annual Meeting.

1

THE SOCIAL DEMOGRAPHY OF INTERNET DATING IN THE UNITED STATES

ABSTRACT Objective: To identify the sociodemographic correlates of internet dating net of selective processes that determine who is “at risk.” We also examine the role of computer literacy, social networks, and attitudes toward internet dating among single internet users. Methods: We use multivariate logistic regression to analyze 3,215 respondents from the first nationally representative US survey of internet dating. Results: Sociodemographic factors have strong effects on internet access and single status but weak effects on use of internet dating services once the sample is conditioned on these factors. For this “at risk” subpopulation, computer literacy and social networks strongly influence the likelihood of internet dating. Conclusions: Internet dating is a common mate selection strategy among the highly selective subpopulation of single internet users and may continue to grow through social networks. Material and virtual elements of the digital divide have direct and indirect effects on internet dating.

2

INTRODUCTION: THE EMERGENCE OF INTERNET DATING In the last decade, searching for romantic partners via internet dating sites has become a visible and common strategy for mate selection; this search strategy emerged amidst important social, demographic, and technological change (Espinoza 2009; Tulsiani, Best, and Card 2008). Specifically, internet dating is growing rapidly in terms of clientele, business returns, and presence in media and pop culture. While it may be too early to assess the impact of this innovation on union formation, we can identify who is adopting this search strategy, with specific focus on the sociodemographic and network correlates of internet dating use. We will show that internet dating is highly selective—only a subset of the total population are potential participants, i.e., “at risk” of using this technology. In our analyses, individuals must have internet access and must be seeking a partner to be considered “at risk” of internet dating. Elements of the “digital divide”— inequality in both physical resources and virtual schemas— and partner/union status condition the likelihood of internet dating. We define internet dating as the use of websites that provide a database of potential partners—typically in close geographical proximity—that one can browse and contact, generally for a fee. These internet dating services facilitate connections that may eventually lead to faceto-face contact and in-person relationships. This is quite different from the earliest forms of online interaction on which much previous research focused. In those studies, singles met in chat rooms and carried out long-distance relationships in cyberspace, often never meeting in person (Ben-Ze’ev 2004; Cooper and Sportolari 1997). Extant work explores a variety of topics, providing important insights about presentation of self and identity (Ellison, Heino, and Gibbs 2006; Whitty 2008), the psychological correlates of online relationships (Anderson 2005; Donn and Sherman 2002), and the utility of mate selection theories of homophily (Fiore and Donath

3

2005) and complementary needs (Hitsch, Hortaçsu, and Ariely 2006). This research, however, focuses solely on individuals who use the internet or are already active internet daters. Sociological research has focused more on macro-level social factors (e.g., increasing singles population, reduced dating networks, and growing mobility) and their impact on internet dating (Brym and Lenton 2001; Whitty 2006); internet daters identify these social factors and the desire to find love as stronger motivations to turn to internet dating than personality factors (Barraket and Henry-Waring 2008). Drawing on recent data from the Pew Internet & American Life Project’s first nationally representative survey about use of internet dating and attitudes towards this technology (Madden and Lenhart 2006), this paper expands upon sociodemographic research by modeling the selective processes leading to use of internet dating. Specifically, we identify sociodemographic correlates of being “at risk” of using internet dating, i.e., having access to the internet and being single in the past five years during the technology’s growth. We then model the likelihood of internet dating net of these selective processes, examining the impact of sociodemographic factors, computer literacy, and social network effects.
THE STRUCTURED GROWTH OF THE INTERNET AND INTERNET DATING

Internet dating has grown rapidly. As of 1999 only two percent of American singles had used some form of online personals services. By 2002, one in four singles had used internet dating services in their search for a potential partner and internet dating had grown into a $304 million business (Orr 2004). A $1 billion industry in 2008, internet dating is projected to grow an average of 10% annually through 2013 (Tulsiani et al. 2008); in spite of global economic recession, the industry continues to grow (Carpenter 2008). This growth in internet dating services and patterns of utilization reflect three factors: i) technological change making internet dating available and efficient, coupled with growing

4

computer literacy that makes it increasingly accessible; ii) demographic change producing a greater number and variety of persons who might be searching for romantic partners (people “at risk” of using this technology); and iii) social change making internet dating more acceptable. Technological Change and the Digital Divide A “perfect storm” of events produced the rapid growth of internet dating: widespread internet access, growing computer literacy, improved real-time chat programs, and digital cameras. Internet access and use have increased dramatically since 1997 (Fox, Anderson, and Rainie 2005; Morales 2009). According to Current Population Survey data, 37% of US households had at least one computer in 1997; by 2003, this number was 62%. Over the same time period, households with internet access increased from 18% to 55% (Day, Janus, and Davis 2005). As the internet becomes an increasingly important part of many Americans’ lives, used as a tool for everyday activities such as shopping, finding information about new products and places, and a myriad of other tasks (Horrigan and Rainie 2006; Victory and Cooper 2002), it is logical that individuals may turn to the internet to search for a partner. Although proliferation of computers and related technology has been dramatic, it has also been unequal. This stratified system of computer and internet access, marked by racial and class-based divisions, is known as the “digital divide” (Day et al. 2005; Morgan and Welsh 2006). The result of the digital divide is unequal access to information and technology that have the potential to improve individuals’ lives. The unequal distribution and growth of technology use across sociodemographic groups is well documented in social science research (e.g., Fountain 2005) and evidence shows that gaps in internet usage remain structured along sociodemographic lines (Morales 2009). Those who are most likely to be online have higher

5

incomes and education, are more likely to live in a suburb or a city, and are more likely to be white (Morgan and Welsh 2006; Orr 2004; Rainie and Bell 2004). While disparities in material resources are fundamental drivers of the “digital divide,” recent critiques of this concept point out that differences in internet use are not merely a function of differential access to technological hardware such as computers and high-speed modems (DiMaggio, Hargittai, Neuman, and Robinson 2001; Peter and Valkenburg 2006; van Dijk 2006; Warschauer, 2003). Internet use is also stratified by computer skills, which shape whether individuals are able to effectively use internet technology. New material resources, such as computers and the internet, must be linked to schemas, ways of knowing and understanding how to use material resources that individuals often implement without noticing (Johnson-Hanks, Bachrach, Kohler, and Morgan 2006; Katz and Rice 2002; Sewell 1992). Computer literacy, or virtual schemas for computer/internet use, must be acquired: a set of concepts must be learned (e.g., logging-on, uploading files) and strategies formed (e.g., appropriate search queries, saving and archiving located information) (Mossberger, Tolbert, and Stansbury 2003; Warschauer 2003). Similar to material resources, virtual schemas that facilitate internet use are also structured by income, education, and race (Kvasny 2006; Peter and Valkenburg 2006). Both material and virtual components of the digital divide—access to computers, propensity toward use, and strength of computing skills—indicate that breadth of internet use will be strongly conditioned by social and economic variables (Horrigan 2008; van Dijk 2006; Willis and Tranter 2006; Zillien and Hargittai 2009). As a technology becomes more common and diffuses throughout the population, disparities become less prevalent (Fountain 2005). We argue, however, that use of internet dating websites is still in the early stages of diffusion, and thus will be stratified by socioeconomic factors.

6

Demographic Change and the Single Population Since 1970, the general U.S. singles population, comprised of individuals who have never been married and those who are currently divorced or widowed, has grown for both men and women. As of 2003, 48% of women and 45% of men were single compared to 40% and 35%, respectively, in 1970. This growth is due in part to steady increases in the median age at first marriage, rising from 20.8 to 25.3 years for women and 23.2 to 27.1 years for men between 1970 and 2003 (U.S. Census Bureau 2004). Concomitant with the delay of first marriage has been increasing geographic mobility as individuals travel to college and abroad and frequently move to meet the job demands of a global economy (Brym and Lynton 2001; Furstenberg 2003; Rosenfeld 2007). Relocation moves individuals away from established social networks and traditional avenues for mate selection. A majority of individuals currently in a committed relationship report meeting their partner or spouse through in-person social networks such as family, friends, work, or school (Madden and Lenhart 2006). As individuals face new life demands that may remove them from these more traditional ways of meeting potential partners, they may become more reliant on intermediaries to facilitate union formation. Social Networks, Stigma, and Diffusion Although the use of intermediaries in the marriage market is not new (Ahuvia and Adelman 1992; Sindberg, Roberts, and McClain 1972), many people hold negative attitudes toward formal marriage market intermediaries that may be anchored in American cultural understandings of love and partnering. Dominant ideologies purport that love “happens,” it cannot be planned, bought, or found through a search engine (Hardey 2002) and it is supposedly distinct from the commercial activities that permeate capitalist societies (Zelizer 2005). Internet

7

dating, like other approaches that create a rationalized short-cut to finding a compatible mate, is at odds with this traditional notion of romantic, spontaneous love (Hollander 2004). Negative attitudes specific to internet dating are related to safety and deception on the internet and the type of people who use internet dating. Many Americans express these concerns. Madden and Lenhart’s (2006) analysis of Pew Internet & American Life data finds that 55.7% of respondents believe that a lot of people who use online dating lie about whether they are married and 67.1% of respondents think that internet dating is dangerous because it puts your personal information on the internet. Though fewer respondents indicated concerns about the quality of internet dating users, 29% agreed that people who use online dating are desperate. Results from a recent cross-national survey show similar levels of stigma. Of the respondents who claim they will never use an online personal advertisement or online dating service, 30% report they refuse to do so because “it could be dangerous,” while 22% claim that the Internet should not “be used to find someone a date” (Barnhoorn 2007). Stigma against internet dating may be a deterrent for some, particularly if it originates from close friends and family members (e.g., Wildermuth 2004). Just as social networks can spread stigma and discourage individuals from internet dating, networks with enthusiastic members may do the opposite, facilitating adoption of internet dating. Orr (2004) and Fallows (2004) both note that the early users of new services and technologies serve as evangelists, spreading the good word to the rest of society. This is consistent with Rindfuss, Choe, Bumpass, and Tsuya’s (2004) analysis of family change in Japan. Drawing from literature on network diffusion and social movements, the authors explain how innovative family behaviors characteristic of the second demographic transition spread throughout the population. Following Rindfuss and colleagues (2004), we conceptualize internet dating as an initially rare and

8

marginalized family behavior that is in transition to becoming more prevalent and socially acceptable. New behavioral norms spread as more people know someone who has engaged in new behaviors, changing attitudes towards those behaviors. In this way, micro-level interpersonal interactions link demographic processes and social change. This theory is applicable to the spread of internet dating: as a changing population of singles turns to the internet as a mate selection strategy and tells others about their success, internet dating may become an increasingly normative path to union formation DATA AND MEASURES Data We use data from a December 2005 national telephone survey sponsored by the Pew Internet & American Life Project. The Pew Internet & American Life Project collects and analyzes data on the impact of the internet on family, community, and public life. Princeton Survey Research Associates International, an independent research company, conducted 3,215 telephone interviews using standard list-assisted random digit dialing methodology. Detailed information about sampling and methodology, including sample design, contact procedures, weighting, design effects, and response rates, can be found in Madden and Lenhart (2006). Unlike prior work on internet dating, our description and analysis are based on a nationally representative sample, not a sample drawn solely from an internet dating site or internet users (e.g., Fiore and Donath 2005). We argue that sociodemographic factors select who will be at risk for internet dating. By analyzing both respondents who have used internet dating and those who have not, our sample allows us to examine the sociodemographic factors associated with internet dating net of the selection processes leading one to be at risk of its use.

9

Measures Outcome Variables “Single” Status. Single respondents are: 1) currently divorced, separated, widowed, or never married, or 2) married or in a committed relationship for fewer than five years. This captures a range of people who could have used internet dating since its rapid expansion in 2000, including people who are currently married. Internet Users are respondents who have ever used either internet or email at least occasionally. Single Internet Users are internet users who have been single in the past five years. An Internet Dater is a respondent who has ever gone to/used an online dating website where you can meet someone online. A “yes” response indicates that an individual has used an internet dating website at some point; our data do not allow us to distinguish between current and previous users. Respondents who have never used an online dating website, and those who do not use the internet, are coded as zero. Internet dating site is not defined in the survey questionnaire; it was up to the respondent to decide whether sites she or he used qualified as dating sites. Respondents reported using a wide range of sites, including traditional pay sites (match.com), specialized sites (blackplanet.com, gay.com), social networking sites (myspace.com), and general chat rooms. Sociodemographic Variables Gender is measured dichotomously. Race categories include white, black, and other (including Asian or Pacific Islanders, mixed race, Native American, and other). Age is measured continuously; we also include a squared term to capture non-linear effects of age based on likelihood ratio tests of model fit. Education is measured with four categories: less than high school, high school, some college (including technical, trade, or vocational school), and college

10

degree or higher. Income refers to total family income from all sources before taxes in 2004. As is typical in survey research, we had a great deal of missing responses on the income measure (21.5%). To prevent loss of these cases, we included missing as an analytical category instead of imputing. Categories include missing, less than $30,000, $30,000-$75,000, and $75,000 and higher. Community type is measured categorically and includes urban, suburban, and rural. Religion is measured categorically and includes Protestant, Catholic, and other religion. We also include indicators of being currently divorced and being a part-time or full-time student. Computer Literacy First we create a measure of how often respondents use the internet from home or work. Frequent users are respondents who use the internet once a day or more from home or work. Medium users are respondents who use the internet between two and five days a week from home or work. Respondents who use the internet frequently from one location and less frequently at the other are classified as frequent users. Respondents who are neither frequent nor medium users are categorized as infrequent users. Next, we create an activity index that represents the number of distinct activities for which the respondents used the internet. Respondents were asked if they used the internet for the following seven activities: sending or reading email; participating in chat rooms; creating or reading a blog; rating a product, person, or service; using online classified sites; or using online social or professional networking sites. The index ranges from 1-7, with higher scores indicating that the respondent uses the internet for a wider range of activities. Lastly, duration is the number of years the respondent has been using the internet. Less than one year of use is coded as zero.

11

Social Networks To address social network effects on the likelihood of internet dating, we create dichotomous measures of whether respondents know someone who is an internet dater. All respondents were asked whether they know anyone (aside from themselves) who has used an online dating website (Know). Respondents who know someone who has used internet dating were further asked if they know someone who has 1) gone on a date with someone they met through a dating website (Know Dater), or 2) been in a long-term relationship with or married someone they met through a dating website (Know Long-Term). Internet Dating Attitudes All internet users were asked whether they agree or disagree with six statements about internet dating: 1) “Online dating is a good way to meet people,” 2) “Online dating allows people to find a better match for themselves because they can get to know a lot more people,” 3) “Online dating is easier and more efficient than other ways of meeting people,” 4) “People who use online dating are desperate,” 5) “A lot of people who use online dating lie about whether they are married,” and 6) “Online dating is dangerous because it puts your personal information on the internet.” We create a scale of overall internet dating attitudes to summarize these responses; this scale ranges from zero (all negative attitudes) to six (all positive attitudes). Methods First, we report descriptive statistics for the full sample and subsamples of internet users and internet using singles, providing chi square tests for differences between groups. We then report sequential multivariate logistic regressions to examine three dependent variables that characterize the selective processes leading to internet dating. We test the associations of sociodemographic factors with the likelihood of internet dating in the general population and

12

then model the structural factors that condition being at risk of internet dating: factors associated with likelihood of being an internet user and likelihood of being single among internet users. Finally, we model sociodemographic, computer literacy, social network, and attitudinal factors associated with likelihood of internet dating among the at-risk population—single internet users. RESULTS The Pew Internet & American Life Project provides the first ever nationally representative data on the prevalence of internet dating in the U.S. population. Figure 1 illustrates our approach and estimates the prevalence of internet dating. Only 5.6% of the full sample has ever used internet dating. We divide the full sample into those who use a computer (72%) and those who do not (28%). Among computer users, almost all use the internet or email (97%). Only 70% of the original sample uses the internet and is therefore eligible to be an internet dater. Among internet users, just under half were single or married less than five years. This highly selective process leaves 32% of the original sample—single internet users—at risk for internet dating. Once we restrict analysis to the population at risk for internet dating, over 17% of single internet users have used internet dating websites. Figure 1: Conceptual Approach Descriptive Statistics. Table 1 presents sociodemographic characteristics of the full sample and subsamples. We illustrate how the sociodemographic composition significantly changes with increasingly selective samples. Compared to the full sample, internet users have a higher proportion of males, whites, suburban and urban residents, respondents with a college degree or higher, and respondents who make $75,000 or more per year. This indicates the presence of a material digital divide structured by socioeconomic status. Internet users are also, on average, 5 years younger than the full sample (47.51 vs. 52.34).

13

Examining the subsample at risk for internet dating—single internet users—we see that they are significantly different from both the full sample and internet users. For example, even though internet users have a higher proportion of whites (84.8%) compared to the full sample (83.7%), there is a lower percentage of white respondents among single internet users (79%), reflecting racial differences in likelihood of being single. Single internet users are also younger (42.44 vs. 47.51), reflecting age patterns of marriage, and have nearly twice the proportion of respondents currently enrolled in school. Few sociodemographic characteristics distinguish internet daters from single internet users. Internet daters are, on average, 2.5 years younger than single internet users and have a higher proportion of both respondents who make $75,000 or more per year and respondents who are currently divorced. Thus, among a relatively younger at risk population pursuing higher education but with moderate income, those who are younger with higher earnings are more prevalent in the population choosing internet dating to find a mate. Table 1: Sociodemographic Descriptives Table 2 presents percentages and means of computer literacy, social networks, and attitudinal measures among internet users, single internet users, and internet daters. Compared to all internet users, single internet users report a higher average activity index but shorter duration of use. These patterns most likely reflect cohort differences in internet usage, with younger cohorts having less time (in years) exposure to the internet but greater integration into their daily lives. A greater proportion of single internet users know someone who has used internet dating (46.6% vs. 30% of all internet users) and they report, on average, more positive attitudes about internet dating. These differences may reflect the influence of both age and social networks.

14

The distributions of computer literacy, social networks, and attitudes are significantly different between single internet users and internet daters. Compared to single internet users, more internet daters use the internet once or more a day (80.9% vs. 63%), use the internet for a greater average number of activities (3.15 vs. 2.2), and have a longer duration of internet use (8.29 year vs. 7.18 years). More internet daters know someone who has used internet dating and they hold more positive attitudes about internet dating. Internet daters are distinct from internet users and single internet users in that they have high computer literacy, have social networks that include internet daters, and hold more positive attitudes toward internet dating. Table 2: Computer Literacy, Social Network, and Attitudes Multivariate Analyses. We use weighted multivariate logistic regression to examine the selective processes that make internet daters a demographically distinct subsample of the general population. Table 3 presents odds ratios for the effects of sociodemographic variables on the likelihood of: 1) ever being an internet dater (in the full sample); 2) being an internet user; 3) being single in the past five years, conditional on being an internet user; and 4) ever being an internet dater, conditional on being an internet user and being single. Note that the results for each dependent variable are based on separate logistic equations. The first column shows that a number of variables are significantly associated with the likelihood of being an internet dater. As previously discussed, however, use of internet dating is contingent on internet access and being single. We turn now to an examination of the sociodemographic correlates of these selective processes. Analyzing internet use among the full sample (column 2) provides strong evidence for the digital divide. Both education and income are positively associated with likelihood of internet use. Even after controlling for socioeconomic status, black respondents are less likely

15

than white respondents to use the internet or email at least occasionally (OR=0.59), reflecting racial disparities in access to technology. Younger individuals are more likely to use the internet than older respondents (OR=0.94 per year of age), a finding consistent with ideas that younger people may be more willing to adopt newer technologies. This finding is also in sync with theories of the importance of computer literacy or virtual schemas for effective technology use. Not only have younger individuals been exposed to computers and related technologies for a larger proportion of their lives, many have also received formal computer education, unlike their older counterparts, thus enhancing their abilities to utilize these technologies. The third column of Table 3 highlights the second selective process leading individuals to be at risk of internet dating—odds of being single (i.e., not currently in a committed relationship or entered one within the last five years) among internet users. The odds of being single vary dramatically with age—they are very high at young ages, decline until later middle-age, and then increase again as individuals become more likely to experience divorce or the deaths of their partners. Although education and income are highly correlated, they exert distinct effects on the probability of being single. Respondents with higher education have a higher likelihood of being single than the reference group of high school graduate, reflecting delays in marriage due to education (students are much more likely to be single than non-students, OR=1.48). In contrast, higher income is associated with decreased likelihood of being single. The fourth column of Table 3 estimates likelihood of internet dating net of these selective processes. When we restrict analyses to the population at risk—single internet users— sociodemographic factors are no longer associated with likelihood of internet dating. The only significant predictor of internet dating is being divorced (OR=2.10), perhaps because divorced respondents are still embedded in social networks created during their marriage and are seeking

16

new avenues to find a partner. Ultimately, patterns of internet dating are influenced by extant patterns of stratification: age, education, income, and race are associated with both internet use and likelihood of being single. Once we control for internet access and single status, however, these stratifying forces no longer predict who utilizes internet dating to search for a partner. Table 3: Multivariate Logistic Regressions Next we turn to an examination of computer literacy, social networks, and attitudes toward internet dating to better understand how other social factors affect differential propensity toward internet dating among our subsample at risk. As previously discussed, computer literacy, a virtual component of the digital divide, should be positively related to likelihood of internet dating, as should social networks that include internet daters and positive attitudes toward internet dating. Table 4 presents results of weighted multivariate logistic regressions that incorporate these proximate determinants in the estimation of the likelihood of internet dating among single internet users. The first three columns show these determinants modeled individually; in Model 4 their effects are estimated simultaneously. Model 1 indicates a significant effect of computer literacy on likelihood of internet dating. For each additional activity the respondent does on the internet, the odds of internet dating increase by 1.81, indicating that the more individuals use the internet for a wide variety of day-to-day activities, the more likely they are to also turn to the internet as a tool for finding potential partners. Duration of internet use shows no significant relationship with likelihood of internet dating, suggesting that factors other than mere exposure to technology may influence who utilizes internet dating. Model 2 demonstrates the influence of social networks on internet dating. Respondents who know someone who has used internet dating websites and those who know individuals who

17

have been in a long-term relationship with someone they have met through an internet dating website have significantly greater odds of ever internet dating themselves (OR of 3.48 and 2.04, respectively). Individuals who are embedded in social networks in which internet dating is an acceptable and successful behavior are much more likely to turn to internet dating. Last, Model 3 shows that, not surprisingly, stigma toward internet dating is associated with decreased likelihood of internet dating; the more positive attitudes toward internet dating one holds, the more likely one is to internet date (OR=1.47). The effect of divorce status remains significant across these three models, but no other sociodemographic covariates show consistently significant effects. In the final model, we include the full set of these proximate determinants. In general, the effects described above persist. Specifically, no sociodemographic variables significantly influence the likelihood of internet dating. These results confirm that while sociodemographic variables strongly condition those who are at risk of internet dating, ultimately, they do not predict whether a respondent adopts this technology or not. In contrast, the more proximate variables, using the internet for a variety of activities (OR=1.55), knowing someone who has used internet dating websites (OR=2.19), and holding positive attitudes toward internet dating (OR=1.32) show independent positive effects on the likelihood of internet dating. Table 4. Multivariate Logistic Regressions with Proximate Determinants DISCUSSION Our analyses show that the prevalence and correlates of internet dating change dramatically when considering the various factors leading one to be at risk of internet dating. Initial analysis of all respondents, regardless of internet use and marital status, indicates that 5.6% of all respondents have used internet dating, and that some sociodemographic variables—

18

being male, having education beyond the high school level, and living in an urban or suburban community—are associated with higher likelihood of ever using internet dating websites. When we focus analysis on internet use, strong evidence of the digital divide emerges: race, education, and income are all significantly associated with the likelihood of internet use. Further, among the sample of internet users, race, education, and income are associated with likelihood of being single. Among the sub-sample at risk of internet dating, 17.5% of internet using singles have used internet dating. What is most compelling, however, is that once the digital divide and relationship status are taken into account, variables previously associated with likelihood of internet dating are no longer significant predictors of its use. Sociodemographic characteristics ultimately have indirect effects on likelihood of internet dating through their direct effects on internet use and single status. They define the populations at risk for internet dating, but fail to determine who dates online and who does not among the at risk population. By analyzing a sample of all Americans, not just internet users or internet daters, this study highlights the distinct and sometimes counterbalancing effects of sociodemographic factors on likelihood of internet use, likelihood of being single, and likelihood of internet dating. We find support for both material and virtual aspects of the digital divide. First, we find that internet use among the full sample is stratified by socioeconomic status and race. However, we find no significant socioeconomic predictors of internet dating within the sub-population of internet using singles. Beyond the digital divide, socioeconomic status does not determine use of internet dating among the at-risk population. Second, our analyses show that computer literacy, a non-material component of the digital divide, is positively associated with use of internet dating websites among single internet users. More experienced internet users, i.e., those who use the internet for a variety of tasks, may be more comfortable using internet dating websites, which

19

require skills such as creating an account and logging on, uploading pictures, and searching a database. Additionally, experienced users may be more comfortable with using the internet in the task of finding a partner because it is a tool they utilize in their daily activities. We also find indirect support for the network diffusion of novel family practices through social networks and changing attitudes (Rindfuss et al. 2004). Knowing someone who has used internet dating and holding positive attitudes about internet dating were both strongly and positively associated with use of internet dating websites. The existence of social network effects on likelihood of internet dating extends prior research on predictors of internet dating that focused more exclusively on personality characteristics (e.g., Whitty 2006). While we cannot address the causal order of these relationships, our results are consistent with the idea that positive attitudes and stories of success are spread through social networks, increasing participation in internet dating among people who already know an internet dater. CONCLUSION We show that social, economic, and demographic variables are associated with whether a person has ever used internet dating websites. These effects operate primarily on the conditioning variables, whether a person is an internet user and whether they are single. Net of these variables that define the “at risk” population, the distal social, economic and demographic variables have modest effects. Instead, proximate variables, such as computer literacy, social networks, and attitudes towards internet dating technology, strongly influence an individual’s propensity to use internet dating. Our results concerning social networks suggest that internet dating will continue to grow as internet daters tell friends and family about their experiences, thus encouraging other internet using singles to use internet dating. Stratification and inequality

20

in internet dating due to the material and virtual aspects of the digital divide will continue but may attenuate as internet use becomes more common. Although we use data from the first study of internet dating from a nationally representative sample of Americans, there are still some limitations to our analyses, including cross-sectional data and lack of information about relationships that are initiated through internet dating. Future longitudinal data that sample both internet daters and non-internet daters will be valuable because they will allow one to assess the temporal order of key variables like knowing an online dater and respondent’s reported use. Also, longitudinal data would allow description of how internet-facilitated relationships differ from other relationships in outcomes such as satisfaction, homogamous matching, and longevity. Given that participation in internet dating shows no signs of slowing in the upcoming years, it is important that researchers devote more attention to understanding whether the digital divide will have increasing influence on rates of union formation and whether internet-facilitated mate selection processes differ substantively from more traditional ones.

21

Figure 1: Conceptual Approach for Estimating Use of Internet Dating Sites Full Sample n=3,215 Computer Use Internet or Email Use Single or Committed