Does eHarmony really use science to match people, or is that just marketing speak? This is not the first time we have heard this question, so we asked our Chief Scientist Dr.Steve Carter to educate, and set the record straight. His detailed response is below. But for those interested in the short answer about whether eHarmony truly uses science and research to help people find love, the answer is a resounding yes. The proof is in the Ph.D’s, who have been working with the company since its inception in 2000.
- The Ph.D’s on our staff have published numerous articles and, yes, actually are scientists! I’ve seen them here in the building myself. And, the founder of our company was a licensed psychologist for 35 years.
- Why don’t we share our matching algorithm for peer reviews? From a business perspective, sharing our algorithm with the world, and our competitors, seems like a pretty silly thing to do. Hence, the “secrecy.”
Dr. Steve Carter, who has been with eHarmony from the beginning, will take it from here…
The primary question we asked when developing eHarmony was whether the scientific study of married couples could lead to models which “predict” marital success. Before eHarmony launched, Dr. Neil Clark Warren (a clinical psychologist with over 35 years of experience in counseling couples) collaborated with a group of researchers for almost three years collecting data on nearly five thousand married couples, looking at their core personality traits and key values. This data, along with a measure of marital satisfaction, was analyzed to see if the levels and similarities within and between married individuals could predict the level of marital quality. This empirical research resulted in statistical models which were then associated with cut-off thresholds for scores that indicated a high probability of being in a successful relationship if married. These models and rules became the compatibility algorithms which eHarmony used to predict which singles should be presented to each other as matches.
The core assumption that eHarmony makes when using models based on data from married couples is that you can generalize their “fit” to data collected from singles. The validity of this assumption has been successfully tested multiple times since our launch. Each time, we have observed that couples who met through eHarmony are significantly happier, and notably less likely to have divorced, when compared to couples who did NOT meet on eHarmony (Carter & Buckwalter, 2009; eHarmony, 2012).
Since our launch in 2000, we have surveyed more than 50,000 married couples in more than 23 different countries, and systematically improved our compatibility matching system year after year. This is really what makes eHarmony unique: We strive to only match individuals who have the best chance at a successful relationship based on scientific research into what combinations of personality, values, and beliefs lead to the happiest couples.
However, as with most things, eHarmony is not without its critics.
Criticism #1: Similarity is rarely observed to account for relationship satisfaction in the literature.
It would be a mistake to think that similarity is the only factor that can be or is used in the eHarmony compatibility matching system. The importance of “main effects” (i.e., the potential direct impact of your traits and your partner’s traits on your and their feelings) are strongly leveraged when the eHarmony system computes the potential quality of a relationship between you and a potential match. However, similarity is a key, and often mentioned component of our system, and rightly so.
In previous research looking at married couples, the results regarding the impact of similarity on the Big Five personality traits (Openness, Conscientiousness, Extroversion, Agreeableness, and Neuroticism) have been mixed, and at times negligible. In most of these studies, the sample sizes were less than 100 couples, so the power to show significance for weaker effects may have been lacking. Our research, using sample sizes of thousands of couples, has generally shown that small but significant similarity effects exist for 3 of the 5 personality traits from the Big Five, including Agreeableness, Extroversion, and Conscientiousness. We have not found similarity effects for Openness or Neuroticism in our research.
In our research and models, we have also included more characteristic adaptations (Costa & McCrae, 1994), which include beliefs, attitudes, values, and social and relationship roles. Emerging research has shown stronger similarity components for these characteristics, including Becker (2013). Similarity in these characteristic adaptations account for much more variance in relationship satisfaction than similarity in the Big Five, accounting for about 5% of the variance. While these effects still aren’t as strong as actor and partner effects (i.e., the direct impact of your traits on your own feelings, and the direct effect of your partner’s traits on your feelings, respectively), the larger effect sizes for the impact from these traits shows that similarity does in fact make a worthy contribution to predicting relationship quality.
Examples of some of these characteristic adaptions:
- Importance of monogamy
- Sexual affect
- Importance of romantic gestures
- Emotional intimacy
Criticism #2: Longitudinal random-assignment validation (i.e., drug-trial style) studies are lacking.
Some critics have also taken issue with eHarmony’s reliance on cross-sectional data, and observational design. In cross-sectional research, cause-and-effect must at some level be inferred from covariance or correlation. For example, if I ask a group of ten-year-old children what they ate yesterday, and how much they weigh, I may observe in the data that there is a correlation between calories consumed and weight, and infer that eating more calories causes greater weight. However, in a longitudinal, random-assignment study, I might randomly assign different groups of children 2 different diets: one high in calories and one low in calories, at time A I would then observe at time B whether my experimental manipulation actually caused a difference in average weight.
The benefits of a longitudinal design where you experimentally control the conditions that you theorize are causing an effect are inarguable when it comes to gaining knowledge. Unfortunately, longitudinal designs can also be quite difficult and expensive to construct, and can raise ethical concerns about the value of the data versus the potential negative impact on participants. This is particularly true in the case of eHarmony. One can imagine a simple experimental design where some eHarmony users are randomly selected to be given “compatible matches” that are actually NOT optimized in any way, and then their relationship outcomes are compared to eHarmony users who got the “real deal.” If the pattern of benefits that have been cross-sectionally observed and validated were once again observed longitudinally, it would be at the cost of seeing these users in significantly worse relationships than their peers. This would obviously not be an ethical business practice, or an ethically defensible research design.
Criticism 3: eHarmony won’t divulge it’s actual matching algorithms.
Publishing our core intellectual property has always been a daunting prospect for eHarmony. While we love to participate in an open, scientific dialogue with our peers (and often do in direct communication and collaboration with researchers around the world), making our “secret sauce” public would, by necessity, make the duplication of our product by others unavoidable. However, making at least some of our models available for academic review is under consideration, and we are always working to find ways to better share our methodologies and analyses with our users.
Carter S. R., Buckwalter J. G. (2009). Enhancing mate selection through the Internet: A comparison of relationship quality between marriages arising from an online matchmaking system and marriages arising from unfettered selection. Interpersona: An International Journal on Personal Relationships, 3, 105–125
Costa, P.T., Jr., & McCrae, R. R. (1994). Set like plaster? Evidence for the stability of adult personality. In T. F. Heatherton & J. L. Weinberger (Eds.), Can personality change? (pp. 21-40). Washington, DC: American Psychological Association.
Becker, O. A. (2013). Effects of similarity of life goals, values, and personality on relationship satisfaction and stability: Findings from a two-wave panel study. Personal Relationships, 20, 443-461.