Modern Mate Search

Mate search, one of the chief activities of humans, interacts in interesting ways with modern technology. Computers can gather and analyze enormous quantities of data; how best can they be used to improve on realspace search strategies? Sexual norms and behavior have changed a great deal over the past century, in part due to technological progress in contraception; has the result been to make the project of finding a mate easier or harder?

Dating Apps

Subscribers to OKCupid answer questions about themselves, if they wish many questions, and search the answers of other subscribers to find someone who looks like a good fit; other dating apps do much the same thing, with differences in details. That lets you¹ search a much larger number of potential partners than you would encounter in realspace, with part of the work of identifying ones of interest done by the computer. Information provided by potential partners may not be entirely accurate: the pictures displayed are selected to flatter and men sometimes add a few inches to their height. Deception is possible in realspace as well, of course; whether a man is married is not readily discovered by inspection, although a woman can check her date’s ring finger for a pale band. Height and looks are more readily observable.

Most dating apps let you filter by geographical area, age, gender and other characteristics. Additional filtering can be done by the choice of which dating app to use, one for Christians who want to marry Christians (Christian Mingle), Jews who want to marry Jews (Jdate), old people (OurTime or Silver Singles), individuals looking for casual sex (Tinder), interested in marriage (Match, eHarmony), looking for a same-sex partner (Her, Adam4Adam), ...²

Filters exist for realspace mate search too; joining a political movement, becoming involved in a hobby, entering a profession, all put you in contact with potential partners with whom you have something in common. A student at Harvard or Vasser is surrounded by others who have passed the same entrance filter, making a fit much more likely than with random pairing. That makes computerized search less necessary and may be one of the main functions of colleges and universities.

Few of us will ever again walk into a dining hall filled with 100 interesting members of the opposite sex of roughly the same age. (Edward Glaeser)

By the same logic, computer dating is more useful for people for whom suitable partners are scarce, unlikely to be encountered at random, such as someone looking for a same-sex partner. The fraction of same-sex couples that originated with an online meeting is much larger than the fraction of heterosexual couples that did.³ A similar pattern holds for another thin market, couples formed at an age by which most potential partners are married.

In designing a computer program to match people up, the obvious approach is to look for pairs each of whose self-description fits what the other says he is looking for. One problem with that is that self-descriptions are not always accurate; most of us have a biased view of our own faults and virtues and even if my self-view is accurate, I may not think that telling the truth, the whole truth, and all of the truth is the best way of attracting a partner. A second problem is that people may not know what they want — or overestimate how much of what they want they can expect to get.

I want to attract not just partners but the right partners. If what I am looking for is a wife, a woman who is going to break up with me after she discovers things I left out of my profile is a waste of my time as well as hers. That is less of an issue if what you are looking for is casual sex; if you suspect it is not the objective of the women you are interested in, there is an obvious incentive to lie about your intentions. The male strategy of seduce and abandon has been familiar in the realspace context for a very long time.

Instead of believing people about what they are and what they want it might make more sense for a middleman to figure out what characteristics, or, in the online context, what claimed characteristics, work well together. The realspace version is the marriage broker, the shadchen. More common in the past, some still exist in modern day America, mostly in specialized marriage markets — Orthodox Jews, Indian Americans, Chinese, Japanese, well off professionals.⁴

For most of us it is done more informally. A colleague's wife suggests that there are a lot of nice girls at folk dancing. I go, despite not being a dancer, and meet the woman I have now been married to for forty years. I ask the wife of my married son, a perceptive woman who knows lots of people, if she can suggest anyone my single son or single daughter would get along with, and with luck she comes up with someone.

A sufficiently clever algorithm, like a sufficiently competent matchmaker, will match people not by what they say they want but by what predicts a good fit. The dating program, unlike the friend or the shadchen, only knows what the customer tells it, although that might change as artificial intelligence gets better and more and more information becomes available online and searchable.

A program can, however, observe and analyze a much larger population and one alternative to theory is feedback. Imagine a dating app one of whose terms is that, after your first date, you report whether things went well or badly, with further reports after later dates and a final one after you either give up on each other or decide on a long-term relationship. The program uses that information to figure out what characteristics, defined by what each subscriber says he is and wants, predict a successful match. The outcome of my date with someone with a given set of characteristics is evidence of the likely outcome for me of other dates with women with similar characteristics — and for men similar to me.⁵

One commenter on my blog reported what might be a weak version of my approach in a real dating app:

Today, if you disable your OkCupid account, you are prompted to give a reason. One of the options is "I met someone on OkCupid." If you select that option, it will then ask you if it's one of the people you've exchanged messages with lately, along with their pictures.

So perhaps they are using that data as you predict. The App also knows when you sign on again, i.e. when you are single again.

The same commenter pointed out some advantages and disadvantages of online dating that had not occurred to me.

the first main advantage of more dating activity moving online is that it's physically safe to screen out bad/unwelcome messages. You just delete the message and block the person and that's it. If you romantically reject a man in person, there's a small but real chance they're another Elliot Rodgers in the making.

…

The flip side of that benefit is reducing false-positive date requests (asking someone out when they don't want you to). If you match with someone online, you know that they are:

A) single

B) potentially interested in you

C) Open to being asked out on a date.

…

The main thing lost from online dating, IMO, is you don't have whisper networks/gossip. As they say, on the internet, nobody knows you're a dog, so if a man's college classmates all think he's a creep, if he can write a nice message, you don't know that.

Hacking Data Apps

My mate search ended when dating apps were still in their infancy so I cannot offer any first hand advice to those who are single and would rather not be but I can point them at advice by others. One source is a Ted Talk by Amy Webb. The early part, describing how she estimated the number of potential husbands for her in the city of Philadelpha (35), reminded me of a similar calculation I did more than forty years ago. My much smaller pool was Blacksburg, the location of Virginia Polytechnic Institute, where I was an assistant professor recently divorced. I concluded that the number of women in Blacksburg that I would have to go out with at least once to eliminate as possible mates was larger than the number I was likely to go out with in any reasonable length of time — I was less picky in my calculations than Ms Webb. It followed that the important constraint was not the size of the pool to be searched but my approach to searching it; I needed to do a better job of locating good candidates. The best approach I could think of at the time was to expand my non-romantic social network and use it to troll the pool. That led me to the suggestion by a colleague’s wife which in turn led to a successful outcome to my search. Thinking over the odds later, after estimating that the woman I married was about a one in ten thousand catch, I concluded that I had been very lucky.

Amy Webb goes on to describe creating a formula to rank potential dates, discovering that none of the men who fit it wanted to date her, constructing ten fake male profiles on OKCupid designed to be the sort of men she was looking for and seeing what women they attracted. While that gave her information about her competition it told her nothing at all about what sort of women her ideal man would be attracted to. That made me wonder why she did no instead construct fake female profiles, see which ones worked best, and tweak hers accordingly.

She did, however, end up with a husband.

More interesting is a piece titled "How a Math Genius Hacked OK Cupid to Find True Love." Chris McKinlay was a doctoral candidate in math at UCLA, probably not a genius but clearly an enthusiast, with access to a lot of computing power. He data mined OK Cupid, used statistical techniques to analyze data on 20,000 women, clustered them on the basis of the questions they answered and how they answered them, and identified the cluster that looked like the sort of women he was interested in. He then data mined that cluster to figure out what they were like, rewrote his bio to emphasize features that would appeal to them and tweaked his profile to answer questions they appeared interested in. After doing all of that he ran a search of women in Los Angeles sorted by how well their profiles matched his. OKCupid users are notified when someone visits their page, giving them a chance to respond, so he wrote a computer program to visit the pages of the high scorers — a thousand a day. It took more than fifty dates to find one that worked.

They got married.

While he was not lying to the women he wanted to date he was reporting that subset of true statements about himself that he thought would appeal to them, answering only questions where his true answer fit theirs. Whether that was the right amount of honesty for his purposes depends on the tradeoff between quantity and quality of dates. A woman who would date him on the basis of his profile but dump him after a few dates was a waste of his time, which is an argument for showing the bad as well as the good in advance. On the other hand, the information from OKCupid, good and bad, would not be enough for a final judgement, and the more women willing to date him the more chances he would have to improve on it. He ended up with more women willing to date him than he had time to date, which suggests that he would have been better off being a little more honest, would still have had plenty of dates and might have found what he was looking for after thirty instead of fifty.

A third online account of gaming OK Cupid, by someone posting as Jacob Putanamit, is titled “Dating: a Research Journal.⁶” The author started by finding, on the OKCupid site, the formula the program used to calculate the match percentage between two people. It was based on how important you rated each question you answered and how you answered it; whether a woman agreed with you on your “very important” questions and you with her on hers almost entirely determined the match percentage. As he wrote, if you want to raise your match percentage with everyone you might be interested in:

you can probably find a few unequivocal questions that almost everyone in your relevant dating pool will answer the same way. For me, it’s things like How many children do you have? Is homosexuality a sin? Are you racist? Do you put more weight in science or faith? and the horrifying Would you sabotage contraceptives to have kids even though your mate doesn’t want kids?

Marking a bunch of those as “very important” and giving the obvious answers raises your match percentage with almost everyone. Marking questions that you care a lot about as “very important” lowers your match percentage with women you probably don’t want to date. Marking all the rest of the questions you answered as “somewhat important” means that the more of those a woman agrees with you on, the higher the match percentage will be.

Someone who does not disagree with you on any of your very important questions will have a high match percentage from your half of the calculation — but it could be low from her side if you disagreed with her on one or more of the questions she had marked as very important.⁷ The more of your questions she agreed with, the higher the match percentage. So women you might be interested in would see a high match percentage, making them more willing to date you, while just how high would tell you which ones you wanted to date.

It was clever, probably correct, and now almost entirely useless because since the article was written OKCupid has changed its algorithm and other features of the program.⁸ It does, however, demonstrate an approach that might be applied to other dating programs, perhaps including the current version of OKCupid, if you could find the relevant information.

Jacob’s article does not describe the outcome of his search but elsewhere on his blog he makes it clear that he is married, polyamorous, and still looking for both lovers and friends. He offers interested blog readers forms to fill out and send him, one for would-be lovers, one for would-be friends.

As that example shows, dating apps are not the only way to use the internet to find a mate. For another example, consider a web forum devoted to some shared interest — science fiction, fruit trees, right or left-wing politics. Participants, like the members of a realspace interest group, have been prefiltered for a common interest. The forum lets you observe a potential mate interacting with others while you remain an invisible spectator; you may even be able to look backwards in time by reading every post he made since he joined the forum. You cannot observe how good looking someone is or how good a dancer but how well someone thinks, argues, behaves, are things that some of us might consider more relevant. I date my interest in the woman I am now married to to hearing her explain a point in calculus to another student, revealing a beautifully clear, logical mind. I fell in love on the spot.

That, at least, is how I like to remember it.

If I were Twenty-five and Single

It has been more than forty years since I had any practical use for a dating app but I find thinking about them intriguing. The accounts I found dealt with particular apps but the underlying principles apply to all of them.

You have two objectives. One is to figure out which women you want to date, the other is to persuade them to date you. You have two mechanisms for achieving those objectives. One is the information you provide to the app, some or all of which will be available to potential dates. The other is the decisions you make of which women to like, message, propose dates to, based on the information they provide you.

You cannot adequately satisfy the first objective by which women you select to message⁹ because there is not enough time; even if the site lets you browse as many profiles as you want, there are far too many. You need to filter the data, reduce the number of candidates for your attention to a number you can actually look at. Even if you have enough time you do not have all the information needed to do the job perfectly since not all of it is in their profiles.

Explicit filters chosen by you, such as age and geographic location, reduce the pool to be searched.  The app may further assist your search by calculating a match percentage or something similar and showing you those it thinks best match you. You look over the profiles that the computer shows you, decide which women to message. Each woman you message looks at your profile and message and decides how to respond.¹⁰

This is the point at which your two objectives may come into conflict. For example …

I believe that current concern with climate issues is vastly exaggerated, that global warming is unlikely to be a catastrophe and might even turn out to be a net positive.

¹¹ That is an opinion unlikely to be shared by most of the women I would want to date, so expressing it on my profile would reduce the number willing to respond positively to my message. But while it reduces the number willing to date me it raises the average quality of those who remain, since a woman who is either skeptical of the current orthodoxy or interested in why someone else is would be a better gamble for me than one who is neither.

If leaving out that information gets me thirty positive responses and including it gets me ten I should probably leave it out — some of the twenty might find my view less off-putting once they knew more about it and me. If including it reduces responses from thirty to twenty-five, it is probably a good way of saving myself some wasted time. If leaving it out gives me three positives and including it gives me none, I am certainly better off leaving it out.

At the start of the search for a mate online I do not know how many positives I will get, which is an argument for weak filtering, at least until I have more information, so I should start by giving only information that provides a hard filter, repels women I probably shouldn’t be dating, omitting information that provides only a weak filter, probably including my views on climate. If I turn out to be more popular than I expected I can add it to my profile later and see what happens.

1. Or your agent. We are used to doing mate search for ourselves but in a society where contact between single men and women is strictly limited the search for a bride may be done by the a man's mother and sister. In Saudi Arabia the process is simplified by the fact that endogamy within the clan (‘Akila) sharply reduces the number of candidates.

2. Aaps listed are examples, not recommendations. I have not used any of them and by the time this is published some may have changed their specialty or vanished entirely.

3. “Searching for a Mate: The Rise of the Internet as a Social Intermediary,” Michael J. Rosenfeld and Reuben J. Thomas, American Sociological Review 77(4): 523-547, 2012

4. One of the latter that advertises its services in airline magazines is “It’s Just Lunch.”

5. I first came up with this idea after hearing a talk on multi-dimensional voting theory, matching the positions of voters as shown by who they vote for with the positions of politicians as shown by who votes for them. My idea at the time was to use the approach to match readers with books they would like, but matching people with partners they might like would be an even more valuable application.

6. Part 1,  Part 1.5. Part 3. The interesting stuff is in the second part.

7. The formula applied the same approach to each side and defined the match percentage as the square root of the product of the two results.

8. As my, possibly biased, user sources put it, the program stopped being run by geeks and the suits wrecked it.

9. I am assuming that you are, like me, a male heterosexual, but the logic should generalize to other cases.

10. Different aaps differ in the details of how one party expresses interest that the other can respond to. On OKCupid when Chris McKinley was using it, he signaled his interest in a woman by looking at her profile and then letting her respond or not — although if she didn’t, he could presumably have given a stronger signal by messaging her. On some other programs a woman only knows you are interested when you message her.

11. For some of my reasons for that belief see my first post and several  later ones.