Skip to main content

Modern Romance: Online Dating as a two-sided market

The dynamics of dating have been discussed and dissected extensively, but in this post we can hopefully gain some new insights into it by viewing it as a networked market. Online dating specifically, renders itself quite naturally as a market (albeit a non-traditional and subtle one) and as of 2015, up to 38% of single Americans report having participated in it[1]. At its most abstract, online dating is a network connecting people participating in a two sided market where potential partners can “match” with each other. What makes this market quite interesting is that there is no explicit currency, price, or transfer mechanism. The transaction in question is a social transaction, not an economic one, and can be viewed as an exchange of social/ psychological rewards or an allocation of relationships. If we think of currency as social currency based on attractiveness or desirability, we get an interesting market in which even though currency is required for transactions and more currency can lead to more desirable outcomes, the currency is never actually “spent” as the exchanged “good” itself is indivisible.


This idea of currency itself is dynamic as it is a function of both the individual’s attributes and the potential partners’ preferences. Hence, an agent may have high currency or desirability for one person and low desirability for another, and the preferences may not necessarily be monotonically related to their attributes. This makes matching in the market quite interesting as individual preferences are likely to be heterogenous. Efficient matching in this market thus relies on the existence of pairs of mutually desirable agents in a setting where all preferences are heterogenously distributed. An interesting question one may ask is how, then, is the efficiency of dating markets so high in real life? One possible explanation, as Hitsch, Hortacsu and Ariely suggest[2], might be that there is natural sorting in dating markets based on attributes, and also possibly because people tend to prefer partners who are “similar” to them, and thus the market naturally resolves into pairs of mutual desirability. An interesting sidenote is that if our previous assumption of heterogenous preferences breaks down, it would indicate that preferences are in fact monotonically related to attributes, and certain attributes are more “universally preferred” than others.


In creating this market, the network effect is fundamental. The main appeal of online dating is the availability of many potential mates, and achieving a critical mass is key to the success of an online dating platform. Agents’ utility increases if there are more users on the “other side” of the market, as there is a higher number of potential transactional partners, and this is a positive cross-side effect[3]. However, as more users join “their side”, users’ utility decreases due to higher competition, and this is a same-side negative effect. In general, additional users add to congestion and increase search costs (along with competition if on the same side). Thus, the efficiency of the network depends not just on number of users, but also on their potential to be a match for others, and a good strategy for the platform might be to limit users to those likely to find matches.


Because of the importance of network effects to the efficiency of the market, it is common to see design choices made to exploit this factor. By their very nature, creating online platforms involves making embedded design choices which influence not just potential matchings, but also structure information in a way that deliberately creates asymmetry. With regards to network effects, a common design choice on online dating platforms is to conceal knowledge about participants, so the total number of potential partners is hidden and only a limited number of recommended partners are displayed everyday. This artificially boosts the cross side positive effect, as it creates the illusion of unlimited potential partners somewhere in the network (as a sidenote, this technique is also utilized quite effectively by Netflix, and you’ll never see the total count of movies on their website). Similarly, the users are also not told how many other users are on “their side”, and withholding this information hides the true severity of competition in the market.


The platform operator may also choose to subsidize participants on one side of the market, as was the case with Ashley Madison, which offered free subscriptions for women[]. This helps balance the split of the market, keeping both sides interested. Another debatable design choice is the use of recommendation algorithms vs letting users search for partners. While searching encourages users to filter through all users and attempt to form matchings according to their own beliefs over their preferences, recommendation algorithms take the opposite approach and try to present a limited set of potential matchings based on inferred user preferences. The search vs recommend design decision also determines knowledge other users have of preferences – while the search design allows users to directly observe each others’ preferences, the recommendation design forces users to only infer preferences. We can argue that recommendation algorithms are more efficient, as they would only show users to each other if they believed that both users could find each attractive, and thus they would reduce search costs. Moreover, limiting the view of the market to just a few options at a time may also make those options seem more attractive. However, this leads to certain agents never even knowing about the existence of others, and heavily impacts the final matchings agents end up with.


There is also a considerable amount of information asymmetry on both sides of the market, as users have an incentive to present a biased view of themselves on their online profiles. Furthermore, design decisions may actually encourage information asymmetry, such as in the case of Tinder, on which matches are judged based on a few pictures and minimal profile information. Since the app is also designed to be “fast-paced” and displays many potential matches in quick succession, it encourages users to find a breadth of potential matchings rather than depth.


Finally, we must ask ourselves the question, is the matching produced by online dating services actually “good”? If we define “good” to be each agent getting their top choice, then it is unclear, as we can only observe a binary view of preferences rather than rank-order preferences. However, if we define it to be a “stable” matching, then evidence suggests that these are actually pretty good matchings. In their 2010 paper[4], Hitsch, Hortaçsu, and Ariely use the Gale-Shapley algorithm based on estimated male preferences to predict the sorting patterns found in online dating. They find a significantly strong correlation, and conclude that the stable matching predicted by the Gale-Shapley algorithm can be seen as the limit outcome of this two-sided search and matching model. Due to the properties of the Gale-Shapley algorithm, this has some powerful implications, such as that the matching produced by online dating is a stable matching, and is also Pareto-optimal match, within the set of stable matches, for the side of the market (men or women) that makes the offers in the deferred acceptance procedure.


The correlation of online dating outcomes with the predictions from stable matching algorithms definitely indicate their efficiency and usefulness, but we have also seen several sources of inefficiency and information asymmetry in the market. Perhaps a reasonable improvement could be a decrease in information gaps, maybe by making preferences more transparent and profiles more informative. Another improvement could be finding a better balance between searching and recommendations, so that users have more control over matchings but are still shown only what is most relevant to them, in order to reduce search costs. Although flawed and overly reliant on design and implementation choices, online dating is still a complex and multi-faceted network that warrants further discussion.


– Bhai Jaiveer Singh



1) Ansari, Aziz, & Klinenberg, Eric. “Modern Romance.” New York: Penguin Press, 2015.

2) Burtch,Gordon & Ramaprasad, Jui. “Assessing and Quantifying Local Network Effects in an Online Dating Market.” Working Papers, NET Institute, 2016.


4) Hitsch, Günter J., Hortaçsu, Ali, & Ariely, Dan.”Matching and Sorting in Online Dating.” American Economic Review, 2010.


[1] Ansari, Aziz, & Klinenberg, Eric. “Modern Romance.” New York: Penguin Press, 2015.

[2] Hitsch, Günter J., Hortaçsu, Ali, & Ariely, Dan.”Matching and Sorting in Online Dating.” American Economic Review, 2010.

[3] Burtch,Gordon & Ramaprasad, Jui. “Assessing and Quantifying Local Network Effects in an Online Dating Market.” Working Papers, NET Institute, 2016.


[4] Hitsch, Günter J., Hortaçsu, Ali, & Ariely, Dan.”Matching and Sorting in Online Dating.” American Economic Review, 2010.


Leave a Reply

You must be logged in to post a comment.