Matching and Sorting in Online Dating

We take the best academic papers and summarize them in plain English so you can improve your idating site, and help your users make better connections.

Title: Matching and Sorting in Online Dating 
Published In:  American Economic Review 2010
Authors: Günter J. Hitsch, Ali Hortaçsu, and Dan Ariely
Contact: [email protected]
Full Report: http://duke.edu/~dandan/Papers/matchingAndSorting.pdf
Pages: 24

Summary, In Plain English:

Summarized by Kathryn Lord, Romance Coach

This long, dense paper uses data generated by almost 7,000 Boston and San Diego users of an unnamed Internet dating site. All user activity over a three and a half month period in 2003 is observed. The authors used information on the users’ interactions and attributes to determine mate preference.

The study is motivated by two fields of study: 1. market design literature, which looks at the performance of market institutions (economics); 2. matching and sorting patterns in marriage markets (across fields of economics, sociology, social psychology, and anthropology). Online dating sites are "matching markets" used to allocate "indivisible goods" (singles) with no "price" or transfer mechanism. Romantic pairings are not random, but are the result of sorting across many factors.

The first objective was to find out if an economic matching model could predict outcomes on the dating site and how efficient those matchings were (yes).  The second objective was to find out if sorting using mate attributes occurs in online dating (yes), and whether an economic model could predict sorting patterns of couples who met offline (yes).  The authors concluded that data obtained from the online dating site helped understanding of the economic mechanisms underlying match choice and marriage formation. 

The authors appear most concerned about the usefulness of the predictive economic models (whether the models were able to correctly predict the results observed in the date mined). The Gale-Shapely model predicts sorting patterns. The Adachi model "provides a useful stylized description of user behavior on the dating site."

The authors did not seem to be particularly concerned about the individuals per se and how to improve the experience and/or success of the users of the dating site. Pages 137 to 144 were devoted to incomprehensible (to the reviewer – I was lost after the first example) mathematical formulas.  Even the graphs and tables were nearly indecipherable and of marginal help to the reviewer.  Pages 146 through 159 presented data generated using the Gale-Shapely and Adachi models related to racial patterns in sorting, equally incomprehensible to this reader.

Of more use to the Internet dating industry about racial patterns in mating was presented graphically in the January 29, 1011 New York Times article "Who is Marrying Whom". The number crunchers over at OKCupid’s blog also presented some interesting behavior patterns on race and matching in a posting in October 2009.

However, I did find some interesting material in the paper which I will present in the Discussion section below.

Discussion:
The authors mentioned "search friction." They said: "sorting along educational attainment might not reflect a preference for a partner of a certain education level, but rather the fact that many people spend much of their time in the company of others with a similar level of education in school, college, or at work." They describe dating sites as being low in search friction: Since people from all educational backgrounds are presented equally on dating sites, singles have equal access to individuals from many different levels of education.

Here is a definition of "search friction" I was able to find online:

• Search friction: The effects of obstacles to the matching of the supply of a product with the demand for it that arise from the time and cost of the process of finding a match.

This definition (while perhaps not the usage the article authors intended) seems to describe online dating, both from the site providers’ perspective and the singles using the site. How does the dating site attract enough single members to attract even more members, yet keep the searching time-efficient and at a reasonable enough cost not to drive users away? From the users’ perspective, they want the best possible matches for the least time and money invested. Users would be attracted to sites that seem to offer plenty of the highest quality and most attractive singles.

The authors describe Internet dating sites as having minimal search frictions. That may be true, compared to normal real world dating. But site owners and users would likely disagree. Complaints (obstacles) on both sides are rampant. Site owners want paying customers. Users’ most persistent question is “Why doesn’t he/she answer my emails?” when probably the user is persistently contacting the most desirable 1% that everyone else is contacting, too. Can the dating site attract enough 10’s to keep the 1 – 9’s happy? Since the 10’s tend to talk to each other, the 1 – 9’s are bound to be disappointed. The resultant high level of frustration is a disincentive to singles becoming long term paying customers.

The authors mention search costs. Here is a definition I found online:

• Search costs: costs of finding another trader; costs associated with finding a counterparty willing and able to take part in a business transaction, e.g. the costs of advertising.

The authors suggest that in traditional mating, the search costs (the search and the risk of rejection) are considerable, but that online dating is designed to minimize the costs (numbers of singles concentrated in one place, rejection risk minimal because of anonymity and ease of contact, and the only cost being the dating site’s fee and the effort of writing of a witty email). While that is technically true, the sheer numbers involved in online dating (sorting through candidates, writing email after email, and one rejection after another in non-answered emails) can make the perceived costs quite high and exhausting. Even the authors acknowledge the high rates of rejection: in their data, 71% of men’s and 56% of women’s first emails did not receive a reply. And this data was generated in 2003! I suspect the percentages would be even higher now. My clients’ reports seem to indicate that.

Information of interest presented in charts and tables
The charts in Figure 1 present collected data on first emails to the least to the most attractive men and women . Unsurprisingly, the most attractive photos had the greatest chances of being contacted.

Table 1 on page 131-2 presents characteristics of the users of the dating site (age, race, marital status, education, and income), compared to the general population.

Table 2 on page 136 presents weight, height, and BMI (all by age) for the men and women on the dating site (self reported, of course), compared with those of the general population.  The figures suggest that the self reporting is not strictly accurate, though to my reading, not terribly distorted. (The self reported weight of women between 30 and 60 seemed to have the greatest variation from the general population –up to 24 pounds less – but perhaps they all dieted rigorously before signing up.)

Table 4 and Figure 2 (page 149) summarized user behavior:

• 3,004 men browsed 385,470 profiles, sent 49,223 first emails (16.4 per person, average, 12.7% of those profiles browsed). Of those first emails, 2,130 indicated interest in meeting (the authors’ definition of a “match”), 4.3% of the first contacts made.
• 2,783 women browsed 172,946 profiles, sent out 14,178 first emails (5.09 per person average, 8.2% of profiles browsed). Of those first emails, 914 resulted in interest in meeting, 6.4% of first contacts made.
• The median number of emails sent before indications of interest in meeting was 6, with a mean of about 12.

Discussion of Table 4 and Figure 2: Men work harder than women to get matches, but women’s efforts are more successful. In an almost equal number of men and women, women looked at less than half the number of profiles than men did (44%). Women sent out first emails at a rate of 31% compared to men.  Yet their success rate on those first emails was 148% better than those of the men.

Kathryn Lord: Based on this information, you should:

1. Recognize that a realistic customer will more likely be a satisfied customer. While photos of "10’s" might be more likely to catch initial interest, non-response to first emails works against having faithful paying customers.
2. Persistently educate your customers about searching, "favoriting," and making first contacts. While it is normal and understandable that asingle would "favorite" and contact the highest quality candidates first, the likelihood of a positive reply is small and discouraging, a good way to lose a paying customer.
3. Make sure that your customers understand that a high non-response rate to first email contacts is normal and to be expected.
4. Suggest that customers systematically widen their search criteria until they have consistent positive returns to their first emails. These responders are the individual’s best market.
5. In particular, encourage women to make first contacts. Men appreciate the effort, and women have better returns on those first email efforts.

Comments

[Fernando Ardenghi]

There is a collection of papers trying to explain stable marriage as an economic function (as a complex mental process).
They propose: a stable marriage can be predicted using mainly Gale_Shapley algorithm, General Equilibrium Theory and Non_cooperative Game Theory.
The Gale-Shapley algorithm is not good enough to solve the stable marriage problem because is based on initial preferences (attraction, fantasies, infatuation) and does not take into account temporal patterns of relationship variables.

papers:
* "What Makes You Click? – Mate Preferences in Online Dating" (2010)
Appendix
* "Matching and Sorting in Online Dating" (2010)
* "Marriage and Online Mate-Search Services: Evidence From South Korea" (2009)
* "What Makes You Click? — Mate Preferences and Matching Outcomes in Online Dating" (2006)
* "Matching and Sorting in Online Dating" (2008)
* "Who Marries Whom and Why"
* "Lifecycle marriage matching: Theory and evidence"
* "What matchings can be stable? The refutability of matching theory"
* "Equilibrium matching with ageing and uncertain careers: who marries whom and when?"

Gale_Shapley algorithm, General Equilibrium Theory and Non_cooperative Game Theory do not take into account temporal patterns of romantic relationship development nor personality traits. Nor the discovery uncovered by Eastwick and Finkel 2008; also Kurzban and Weeden, 2007; Todd, Penke, Fasolo, and Lenton, 2007 who found that people often report partner preferences that are not compatible with their choices in real life. Marital stability does not include marital satisfaction. A marriage can be stable but miserable. Latest Research in Theories of Romantic Relationships Development shows: compatibility is all about a high level on personality similarity between prospective mates for long term mating with commitment.

From the paper "Matching and Sorting in Online Dating" (2010)

"In Hitsch et al. (2010), we discuss and present evidence that shows sorting patterns in short- and long-term relationships are similar, which suggests mate preferences do not substantially differ across short- and long-term relationships."

(FLAWED) Researchers should read
1) "Human oestrus" Gangestad & Thornhill (2008)
"Only short-term but not long-term partner preferences tend to vary with the menstrual cycle"

2) "Does the contraceptive pill alter mate choice in humans?" Alvergne & Lummaa (2009)
".. whereas normally cycling women express a preference for MHC (Major Histocompatibility Complex) dissimilarity in mates, pill users prefer odours of MHC-SIMILAR men, indicating that pill use might eliminate adaptive preferences for genetic dissimilarity." ................ "Recently, Roberts et al. attempted to eliminate these potential confounds by adopting a within-subject design in which women's mate preferences were assessed before and after they began taking the pill. Women starting the pill showed a significant preference shift towards MHC SIMILARITY compared with three months before the pill was taken, a shift that was not observed in the control group of normally cycling women."

Please read
TOP papers for the Online Dating Industry:
http://onlinedatingsoundbarrier.blogspot.com/2011/05/top-papers-for-online-dating-industry.html