Current work
We are currently pursuing two research tracks related to online dating. The first is post hoc quantitative analysis of messaging behavior and profile characteristics, an expansion in scope and method of the work begun in Andrew Fiore's master's thesis. The second is a series of surveys administered to online dating users about their perceptions and expectations of potential partners whom they meet through online personal ads. In the following sections, we describe both tracks of research.
Track I: Quantitative behavioral analysis
Analysis of the characteristics and behavioral patterns of online dating users has the potential to yield rich insight into what users seek in a date and how they evaluate potential partners. Studying users of large real-world systems in particular would give us reliable evidence about online dating behavior because we can study a large group from a diverse population, ensuring that we can generalize confidently from our findings.
Quantitative analysis allows us to characterize the users of an online dating Web site and compare them to other populations, offline and online. Analysis of communicating pairs (cf. Klohnen & Mendelsohn 1998) lets us infer what characteristics people are actually looking for in a potential partner and how these demonstrated preferences compare with those they claim to seek. It's possible, too, that different subpopulations of users within the site are seeking entirely different things and using different evaluative techniques. Breaking down our analysis into demographic subgroups would allow us to compare these approaches.
Specifically, we plan to answer the following questions for as many data sets as we can obtain:
What predicts how many messages a given user will receive?
What fraction of dyadic exchanges are "successful"? What attributes differentiate "successful" from "unsuccessful" exchanges? (We can define success as the length of the exchange, or rate of transition to another medium, like telephone or off-site email.)
What is the effect of having a photo on communication success? Of having a video clip?
How do men and women differ in message-sending behavior? Do they value different attributes?
What about different ethnic or cultural groups, or different regions of the country?
Analyses like these will help inform better designs of online dating systems. For example, knowing what characteristics users actually use to make decisions about whom to contact would allow us to create profiles that more accurately present and even highlight these characteristics, making the decision process quicker and easier. Information about which characteristics successful couples have in common can be used to improve suggested matches. Additionally, knowledge about the differing behavior of various subpopulations of users could facilitate adaptive interfaces that adjust to the needs of individual users.
Method
We will employ a variety of analytic methods to make sense of a large volume of social-transactional data from online dating Web sites. We begin with descriptive techniques, examining central tendencies and distributions, and data visualization to spot trends. Relational (correlation, regression) and comparative (t-tests, non-parametric significance test) tools permit the exploration specific hypotheses about the factors that influence messaging behavior. Techniques like clustering and factor analysis will identify emergent patterns of communication and classes of user behavior.
New techniques for dyadic analysis (e.g., Klohnen and Mendelsohn 1998, Fiore and Donath 2005) allow us to compare the characteristics of selected dyads, like a romantic couple or a pair of users who are communicating online, to the characteristics of all random pairings of people in the sample. These techniques compensate for the baseline similarity that most groups of people exhibit in personality and attitudinal characteristics, which varies substantially by characteristic but is typically much greater than zero.
Furthermore, as in Fiore's master's thesis, we will continue to develop novel software tools for visualization and interactive data analysis. For that project, Fiore built a database-backed visualization application that combined user-configurable scatterplot arrays with transparent, social-network-style connection overlays. This tool presented visually how different subsets of the population were communicating with each other and among themselves, inspiring the dyadic analysis described above.
Track II: Examining perception and expectation
Interpersonal perception relies on a variety of sophisticated mechanisms that humans have evolved over time. Computer-mediated communication, however, offers only a limited set of communicative channels compared to face-to-face interaction, forcing users to employ other (historically unusual) means to evaluate potential partners. What are the implications of this restriction for mate perception and selection? Are users of online dating choosing qualitatively different mates than they would if they were meeting them offline? This question has not only personal but also societal implications for our notions of love, sex, marriage, and family. More concretely, users of online dating systems often express disappointment in their experiences, particularly after meeting in person a potential partner with whom they have corresponded only online. We hypothesize that this disappointment stems from users' systematic misperception of one another through mediated channels in general and online dating profiles in particular. Walther et al. (2001) show that communication over sparse channels can lead to idealization; if this is the case, disappointment in online dating could stem from overly optimistic expectations formed prior to meeting in person. This research seeks to identify on what dimensions, to what degree, and for what reasons online dating users might misperceive each other, in particular when the misperception leads to disappointment. Knowledge of the mechanism behind this process will enable designers to create online dating systems that present information and communications in such a way as to prevent the distorted perception of the misperceived cues, or at least mitigate their consequences. Some of the dimensions on which users might misperceive one another that this work will investigate include demographics such as age, ethnicity, religion, height, weight; physical attractiveness; hobbies and interests; political affiliations; similarity to one's "ideal partner"; style of relationship desired, e.g., marriage vs. casual dating; and personality traits as indexed by the canonical Big Five -- Extraversion, Agreeableness, Openness to experience, Neuroticism, and Conscientiousness.
Method
It is likely that users of online personals systematically misperceive the people with whom they communicate through these systems, yet we do not know on what dimensions or to what degree this misperception happens. To find out, we plan to survey users about their perceptions and expectations of the people they meet through online dating services both before and after they meet face-to-face. The surveys will include questions about personality, attractiveness, and perceived commonalities with the communication partners.
To gain the fullest picture of the changes in expectations that users might experience, we will survey both people in a communicating pair of users both before and after they meet in person. The full sequence is: Person A and Person B begin communicating through the online dating site. The system detects their communication and sends them each a pre-date survey about the other -- that is, Person A reports on his/her perceptions and expectations of Person B, and vice versa. Person A and Person B meet face-to-face and learn more about each other. Subsequently, the system sends Person A and Person B each a post-date survey about the other to gauge their perceptions and expectations following their face-to-face meeting. To analyze the survey responses, we will compare the participants' pre-date and post-date responses to determine how their initial perceptions changed with in-person interaction, and on what dimensions they were pleasantly surprised or disappointed.
To accomplish this, automated survey software will obtain informed consent from an initial random sample of online dating users. Because we wish to gather data from dyads, not just individuals, however, it is insufficient to take a random sample, because both members of any given communicating pair are unlikely to belong to it. So it is necessary to employ a technique similar to "snowball" sampling -- the survey system will automatically detect when a previously consenting user contacts or is contacted by another user. If that user is not part of the survey pool, the system will automatically send him or her an invitation to participate. Thus, the sample will expand in a way that maximizes the number of complete dyads it includes.
Theoretical background
Many theories of computer-mediated communication describe the medium as impoverished in one way or another -- it might be less than real time, or the information conveyed might be of limited richness (e.g., text only), compared to face-to-face interaction. How does this affect the user's experience of communicating through such a channel? Joseph Walther (1996) proposed a theory of "hyperpersonal" communication that might explain what happens: when the channel is impoverished, users can't get as much information as quickly as they would face-to-face, so they fill in the blanks optimistically about their conversational partner. That is, in some sense they idealize them in light of incomplete information. Walther's later work established that this effect is most powerful -- generating the greatest interpersonal affinity -- in long-term online interactions in which participants never see photos of each other (Walther et al. 2001). By contrast, Walther et al. found the least social affinity occurred in short-term online interactions without photographs -- a finding particularly germane to online dating, where many users remain unwilling or unable to post photos of themselves with their profiles.
Research in social psychology has shown that similarity in attitudes, political beliefs, and factors like religiosity predict attraction. (For good summaries, see Fisher 1992; Brehm et al. 2002; Buss 2003; and Buss & Barnes 1986). Proximity, and with it frequent exposure to a person, also plays a major part. Personality may be more complicated, but some recent work suggests that similarity in some personality traits is desirable, but in other traits, complementarity is better. Klohnen and Mendelsohn (1998) found that people prefer similarity in traits that they like in themselves, and complementarity in traits that they dislike in themselves -- more precisely, those traits that do not match their conception of ideal self. Work in this area is ongoing, and some studies suggest that those characteristics that predict initial attraction might not predict long-term relationship satisfaction, and vice versa (Klohnen & Luo 2003; Watson et al. 2004).
For more information about the theory behind this work, please see our annotated bibliography, Attraction, Selection, and Satisfaction -- Understanding what makes happy couples, which provides summaries of 20 important articles about relationship research.
Conclusion: The matching problem
The problem of matching two entities for mutual benefit goes beyond the realm of romance. Employers and employees undertake a similar process in trying to pair a good worker with a good job. Real estate buyers and sellers, too, go through similar steps: search, presentation and perception, and communication -- in this case, negotiation1. What this research reveals about the efficacy of computer systems to mediate the matching problem in a romantic context may well bring insight into how to solve similar problems in a variety of important domains. Thus, although the present work focuses on interpersonal relationships, the analytic techniques and design principles that result will likely apply more broadly.