The New York Times has an article on “dating docs”. These are a local phenomenon - I think an ex of mine might have been Patient Zero. I don’t begrudge the Times for writing about them. I’m just surprised they’re considered an interesting phenomenon. What could be more obvious than making sure potential dates know what you’re like?1

Still, reactions have been mixed. From the subreddit:

Jesus. I’m not trying to be insulting at all but even just the existence of a Date Me Doc, or an online dating profile is just so antithetical to how I grew up. (I’m 38) I see people addicted to these apps and now “burning out” from them, it all seems kinda crazy and gamified. The over analysis and “sciencification” of attraction really take the fun out of life, and it happens in every game where people min max to suck the joy out of it. Just talk to people…it is still easy for people from my generation.

I find them weird and off-putting, personally, and I wouldn’t make one because I think most of the people I’d like to date would think the same.

I’ve got to agree. No offence to this woman, but similar things I’ve seen on Twitter give the impression of being more like an ostentatious display of sexual capital rather than a genuine attempt to fill a vacancy. Perhaps that’s not the case in this situation, but I can’t help feeling like this is a poor solution to the problem of being single. While many people have minimum standards for partners like appearance, height, and profession, the criteria that can be captured on paper are blunt tools only effective for a very early filtering process. I don’t see how this stands to be any better than Tinder, unless you consider intellectual compatibility to be of such high importance that you want to find a select group of potential suitors who are perfectly tailored to your way of thinking. That seems like a poor idea for productive dating to me.

But Gwern gives a more scientific counterargument:

My take is that ‘date me’ docs can’t work for their stated purpose of pre-filtering for increased compatibility above & beyond the obvious-to-everyone stuff because compatibility can’t be predicted empirically using exactly the sort of questions/criteria which fill up these date docs/spreadsheets: https://www.lesswrong.com/posts/6yiayg5QWtWme4JN8/anatomy-of-a-dating-document?commentId=ctD7rHdPpwdSW8jMt Dating appears to be a numbers game, where you want as many as possible to find the ‘magic’ match or settle for the best you can get eg https://www.thisamericanlife.org/791/transcript

So, the simplest first-order effect is that they damage your dating prospects by only creating false negatives and shrink your pipeline. If they help, it has to be for a reason other than the ones everyone is so eager to ascribe to it.

Gwern is right that there’s a lot of science purporting to argue that describable preferences can’t help people find matches. I want to start by arguing that this science can’t possibly be right , then look closer into what it is and where it might have gone wrong.

Describable Dating Preferences Can’t Possibly Be Useless

…because there are some basic things people care about matching on a lot

Here are some things that are so obvious they sound like cheating:

  • Unless you’re bisexual, you probably care if someone is a man or a woman

  • Many people care whether their partner prefers polyamorous or monogamous relationships.

  • Many people want a partner of the same religion and level of religiosity.

  • Many people want a partner who matches their preference about whether to have children, and how many to have.

…because in practice, people end up very closely matched on some criteria

  • Only about 4% of marriages are between Democrats and Republicans

  • Only about 3% of high school dropouts marry a spouse with a college degree, compared to more than 80% of PhDs.

  • 90% of whites marry other whites; 80% of blacks marry other blacks.

  • Husbands and wives’ social classes correlate at about 0.8

If 96% of Democrats are marrying non-Republicans, it seems like Democrats must have a strong preference against marrying Republicans, and ought to value having information about someone’s politics before they date them. Realistically, this underestimates the level of political sorting; I don’t think I’d be a good match for an extremely woke person, even if we were both technically “Democrats”.

You could argue that this says nothing about preferences, and that it’s just coincidental sorting; Democrats only meet other Democrats, and so only end up dating them, but they’d be just as happy to date a Republican if only they knew one. I think this fails in several ways: first, many Democrats know plenty of Republicans. Second, many people use dating apps, where it’s easy to date people you don’t know. Third, common-sensically, I still don’t want to date that woke person, or a fundamentalist Christian, or many other types of people with different political views from myself. I won’t deny that there are probably people in those categories I would like if I got to know them. I just think it fails common sense that these have zero predictive power in assessing compatibility.

…because empirically, dating sites can sort people very well

The only dating app I ever seriously used was OKCupid, back when it was good. It asked users questions like “Do you like going to big parties?”. They would answer both for themselves, and how an ideal partner would answer (eg if you don’t like parties, but you want to date someone who does). Then it would calculate your match percent with everyone else on the site.

This was a simple, low-tech system. Nobody had done scientific work to establish that the questions it asked were important; many of them obviously weren’t. They were just random questions some people had thought up. Still, it worked uncannily well. For a while, the person in the entire US with the highest match percent with me was my actual girlfriend (who I had met separately, not using the site). She told me I was her second-highest match percent; her ex was #1.

Reading the profiles with high match percents on OKCupid, I usually found them funny, intelligent, interesting, and people I’d be excited to get to know even if I couldn’t date them. Reading the ones with low match percents, I found them alienating, bizarre, and sometimes opening a window into entirely new types of defective people who I didn’t know existed and who I wish I could have stayed in blissful ignorance of.

…because most people have lots of strict preferences that are paradoxically easy to satisfy

Back when I was on the dating market, I was only considering women who met all of the following (estimated percent of people who satisfied each in parentheses):

  • Between ~22 and ~40 (33%)

  • Interested in getting married and having children (50%)

  • Polyamorous (10%)

  • Willing to tolerate a mostly asexual partner (20%)

  • Politically close enough to me that neither of us end up hating each other (60%)

  • Either nonreligious, or at least not considering religion such a big part of their life that it would ruin the relationship for me to be nonreligious (60%)

  • About the same social class and level of education as me (25%)

Multiply all of those out, and it’s about 1/3000 people. This is an overestimate; these aren’t independent criteria, probably people who are more likely to be poly are more likely to be nonreligious, and so on. Let’s say the real number is more like 1/500. I was only even willing to consider one in every 500 people.

Before you start your rant about how this symbolizes the decadence of modern society and I deserve to be forever alone because of my pickiness - it was easy to find these people, I dated several in a row, and I eventually married one. In cities with millions of people, it’s easy to find good matches as long as you don’t dismiss pre-screening as impossible before you even begin.

A final common-sense argument for the value of profiles

Here are fake dating profiles for five women, each a slightly exaggerated version of a real type of person you find on dating apps:

Hiiiiii! I’m Cindy, 29 yo! My favorite things are listening to Taylor Swift ( <3 Taylor 4-ever!) and going out with my friends, maybe I go out a little too much lol. I want a man who treats me like a princess and isn’t afraid of a girl who knows what she wants lol. Good taste in bars and clubs is a must. If you can’t handle me at my worst, you don’t deserve me at my best.

Hi, I’m Larisa. You could say I’m kind of a go-getter. After graduating second in my class at Brown, I was featured in “Twenty Young People To Watch” in TALYNT Magazine. Since then, I’ve started my own eco-conscious footwear line, with branches in five states (soon to be six). I’m looking for someone who moves as fast as I do, a relationship where we motivate and complement each other. Someone who can enjoy a working vacation in Bali, or going skydiving in the Italian Alps (see attached picture). I know there are quality men out there, so book a spot on my calendar if you want to get coffee sometime.

Hi, I’m Sky! I’m an Aries, although my friends say I sometimes act like more of a Virgo. I’m looking for a deep, fulfilling relationship where we inspire each other to become our best selves and face each day’s challenges anew. My interests include Non-Violent Communication, Internal Family Systems therapy, somatic experiencing, and Integral Theory. My ideal partner would be deeply spiritual and interested in co-exploring this beautiful maze we call Life alongside me.

I’m Hana. I’m a grad student in economics, studying how poor countries develop infrastructure. I’ve been kind of obsessed with it lately. In order to motivate me to do my chores, I name the rooms of my house after underdeveloped countries and tell myself things like “Kitchenya has a food import-export imbalance, you need to buy more rice right now”. I promise I can think about other things. Sometimes I play around with AI art or try degenerate crypto betting schemes that I always lose money on. Looking for someone who will help mesolve the gender imbalance and fertility crisis in Bedroomrundi go on a few dates with me and see what develops (no pun intended).

I’m Jane. My favorite animes are Full Metal Alchemist, Attack On Titan, My Hero Academia, Code Geass, Neon Genesis Evangelion, Gurren Lagann, and Fate: Stay Night. My favorite video games are Super Smash Bros, Final Fantasy, Stardew Valley, Minecraft, and Fortnite. I’m really shy and don’t leave the house a lot but my family says I should get more into dating. Let me know if you want to hang out and play something and get to know each other better.

I predict most people will have strong preferences for one of these people over another. I think the preferences people get from ads like these are valid and reflect real long-term relationship compatibility.

(I also predict some people will galaxy-brain themselves into saying things like “Well, Larisa sounds like someone I can’t stand, but how can I be sure that, after spending months with her, I wouldn’t find she’s actually very nice and I’m deeply in love with her?” But just because something isn’t impossible doesn’t mean you should bet on it.)

And all of this is separate from the types of preferences mentioned above - ie it’s not just the easy things like race, religion, income, number of children desired, politics, sexual compatibility, etc. Everything here is after the 1/500 even-getting-started number listed above! So learning about people from profiles must allow an even stronger filter than that!

So What Are The Supposed Studies Saying You Can’t Predict Romance With Definable Criteria?

Gwern lists some of them here. I won’t go too much into any individual study, except to note that Sparks (2020) is a great name for someone researching the causes of romantic attraction, and Wood & Furr sounds like a children’s cartoon about adorable animals. I’ll separate them, plus some related work, into a few designs:

A first group asks people their preferences on some large battery of “objective” questions, then has them do speed dating, then demonstrates that their supposed preferences have no relationship to who they select in the speed dating session.

Sometimes this reaches an almost nonsensical level. For example, Kurzban & Weeden** purport to find that people’s supposed preferences for age, race, religion, education, and whether a potential partner already has children are all meaningless. Not only that, but there is no correlation at all between Partner #1’s age, race, education, etc, and Partner #2’s! A 20 year old white Ivy League man is exactly as likely to desire a 60 year old black high school dropout with two kids, as to desire anyone else.

Here a lot of the problem is that most of the selection is being done by the speed dating event itself. The paper admits that most of the events it looks at are already age-segregated, some are explicitly race- or religion- segregated, and all are in specific neighborhoods that probably have some level of class and income segregation. Consider for example educational preferences. The study tells us “event average” was correlated at 0.73, subject’s own education level was correlated with their date’s at 0.03, and subject’s preferences were correlated with their date’s features at 0.03. I find it hard to read this as anything other than “events were already extremely education-segregated, such that obvious well-known features like educational assortment failed to materialize, so we got nonsense data about the value of educational preferences.”

But this can’t fully explain the poor predictive value of things like whether the person involved already had children. I think to some degree here we have to bring in the speed dating format itself, which allowed only three minute conversations between participants and a binary yes-no decision. The researchers noted that mostly people just said yes to the attractive people, regardless of anything else. I don’t know if this is a great model for long-term relationship formation.

Joel 2017 is a better speed-dating study. It asks participants (all undergraduates) about:

…a wide range of psychological constructs, including personality measures (e.g., the Big Five personality dimensions, attachment style, perceptions of one’s own mate value), well-being assessments (e.g., positive affectivity, negative affectivity, satisfaction with life), mating strategies (e.g., sociosexuality, interest in long-term relationships), values (e.g., traditionalism, conservatism), and self-reported traits (e.g., warmth, physical attractiveness) along with ideal-partner-preference items for those same traits.

Some of these should matter a lot. For example, “interest in long-term relationships” sounds like whether someone is looking for a casual fling vs. marriage, a frequent dealbreaker on dating sites. And “values eg traditionalism and conservatism” sounds like politics - and again, we know only about 4% of Democrats marry Republicans and vice versa2.

Then they made everyone go on four-minute speed dates with twelve other people and rank them on a 1-9 scale. They found that they could explain 10 - 20% of “value”, in the sense of which people were consistently more desired than others, but about 0% of “relationship desire”, ie the degree to which specific people preferred specific partners beyond their generic value.

I agree this is a strong study. But again, its results are bizarre. Not only do things which we know matter (preference for long-term vs. casual relationships, liberal vs. conservative values) not matter, but they were only about to determine 10 - 20% of people’s sexual market values, even from a panel of questions including things like “attractiveness”. If you accept this as proof that explicit questions can’t predict compatibility, you should also accept it as proof that explicit questions can barely predict sexual market value - which I think most people would have a hard time swallowing.

So what could have gone wrong? For one thing, this study was done on undergraduates, dating other undergraduates at their same institution - so we’ve already lopped off most variation in age, education, class, previous sexual/marital history, and maybe even politics. For another, once again, “liked this person after a four minute speed date” was considered the gold standard of true romantic compatibility, even though realistically probably people just chose whoever was hottest and maybe most personable, and didn’t even have time to ask about the deep values questions they were assessing.

A second group asks people for traits their idea partner must have, then shows that they don’t really use those traits when selecting partners.

So for example, we have Sparks 2020, where 138 undergrads were asked to name three qualities that their ideal partner would have. Then they sent them on blind dates, and asked them to rank their partner on various traits, as well as rank how interested they were in their partner. The researchers found that score on the subject’s supposedly-important preferences did no better at predicting the subject’s romantic interest than score on someone else’s supposedly-important preferences which the subject didn’t share.

They kind of admit this design has too low power to demonstrate much, so they try a different design, where they ask the subject to rate their friends on various traits. Then they ask which friends the subject is romantically interested in, and find their supposedly-important preferences don’t predict this any better than anything else. There was potentially a small effect for actual romantic partners to fit the subject’s ideal traits unusually well, although this only appeared on some models and not others.

What should we make of this? Plausibly 19 year olds describing their ideal partners to researchers are not especially self-aware or honest. The most popular ideal traits in the study were “good sense of humor”, “intelligent”, “honest”, “attractive” and “kind”. It doesn’t seem too surprising to me if 19 year olds saying they wanted an “honest” partner don’t really show a strong preference for honest people over kind people compared to those saying they want a “kind” partner. Add in that our only knowledge of the partner’s true qualities are 1-2 undergrads rating them on a 1-11 scale after a first date, and I don’t know if we should expect a stronger correlation than this.

A third group study twins. Identical twins raised together are similar in appearance, social class, various psychological traits, and whatever preferences are genetically or familially determined. Should we expect them to be attracted to similar people? Should we expect similar people to be attracted to them?

Lykken and Tellegen do this study. In preliminary research, they find that of all variables, couples are most likely to sort along IQ/educational attainment, attractiveness, conservative/religious values, and a factor representing interest in outdoor sports.

They investigate whether spouses of identical twins are correlated; that is, if Alice and Beth are identical twins, and Alice marries Charles and Beth marries Daniel, will Charles and Daniel be similar to each other?

Their summary is “no”, but this table doesn’t look completely negative to me. Given that the Charles→Daniel correlation is the product of three primary correlations - Charles→Alice, Alice→Beth, and Beth→Daniel - this looks pretty respectable to me, though admittedly the identical twins don’t look more strongly correlated than the fraternal ones. Broader questions about interests and talents seem much weaker than the larger factors.

They move on to a more interesting question: are identical twins’ spouses attracted to the other twin? That is, is Charles (Alice’s husband) attracted to Beth? They find that twins’ husbands are attracted to their sister-in-laws slightly more than chance, but twins’ wives are not attracted to their brother-in-laws more, which they explain by men being more driven by physical attraction. But their data are confusing, and as far as I can tell they misprint the table where they present them, so I can’t draw many conclusions beyond that a surprising number of people dislike their spouse’s identical twin or at least aren’t especially attracted to them.

What would it mean if people weren’t very attracted to their spouse’s identical twin? Maybe that attraction is contextual - if you know someone’s not available and it would cause lots of problems if you expressed it, you don’t show it? But that wouldn’t explain the many people who have affairs or otherwise fall in love with inconvenient people. Or maybe that attraction is very path-dependent - if you see someone at the exact right moment when your brain is primed for attraction, you feel attracted to them, and then that gets locked in, regardless of their appearance, behavior, or characteristics? Or does it mean that attraction is based on such fine-grained characteristics that even identical twins share them only at a very basic level?

Finally, this isn’t a group of studies exactly, but you ought to be able to compare all the different partners of an individual - either a serial monogamist who’s dated many people throughout their life, or a polyamorist who dates many people at the same time. How similar are these partners - presumably chosen to satisfy the same person’s partner selection function - on legible criteria?

My impression, based just on thinking about people I know, is that they’re pretty similar on a lot of measures of class/education/sorting, on sociosexuality, emotional maturity, values, and what they want out of life; lower on things like agreeableness, anxiety, conscientiousness, specific interests, and specific details of their appearance3.

Can We Reconcile Scientific And Common-Sensical Evidence About Partner Preferences?

I don’t think any of the studies above present strong evidence that people don’t or can’t sort based on age, class, education, IQ, income, appearance, broadly-defined values, attractiveness, or long-term life plans. There’s plenty of evidence people do this, and the studies either use populations pre-sorted for some of these variables (eg undergrads at a specific institution), evaluations that discourage investigating these variables (eg three-minute speed dates), or both.

The studies do seem to provide evidence that people don’t heavily select on clearly defined psychological traits like agreeableness, and a few of them suggest people don’t select on interests like basketball (although they do find a general factor of outdoorsiness, and maybe individual sports are just too weak to show up).

Except for the twin study, all of these focus on initial attraction, not on which relationships work and survive. They don’t rule out a situation where the “initial spark” of romantic attraction is random, but people with similar interests and personalities are more likely to stay together. Maybe everyone in these studies is very stupid (cf. they’re mostly undergrads), and they all feel attraction to random unsuitable people at speed dating events, but in real life, selecting people who are long-term compatible with you is the way to go.

None of these studies rule out something about writing, or broad gestalt impressions of people, being a strong screening tool. In fact, they almost demand it. Let’s go back to those sample profiles from earlier:

Hiiiiii! I’m Cindy, 29 yo! My favorite things are listening to Taylor Swift ( <3 Taylor 4-ever!) and going out with my friends, maybe I go out a little too much lol. I want a man who treats me like a princess and isn’t afraid of a girl who knows what she wants lol. Good taste in bars and clubs is a must. If you can’t handle me at my worst, you don’t deserve me at my best.

Hi, I’m Larisa. You could say I’m kind of a go-getter. After graduating second in my class at Brown, I was featured in “Twenty Young People To Watch” in TALYNT Magazine. Since then, I’ve started my own eco-conscious footwear line, with branches in five states (soon to be six). I’m looking for someone who moves as fast as I do, a relationship where we motivate and complement each other. Someone who can enjoy a working vacation in Bali, or going skydiving in the Italian Alps (see attached picture). I know there are quality men out there, so book a spot on my calendar if you want to get coffee sometime.

I’m Hana. I’m a grad student in economics, studying how poor countries develop infrastructure. I’ve been kind of obsessed with it lately. In order to motivate me to do my chores, I name the rooms of my house after underdeveloped countries and tell myself things like “Kitchenya has a food import-export imbalance, you need to buy more rice right now”. I promise I can think about other things. Sometimes I play around with AI art or try degenerate crypto betting schemes that I always lose money on. Looking for someone who will help mesolve the gender imbalance and fertility crisis in Bedroomrundi go on a few dates with me and see what develops (no pun intended).

How do you translate these to single-dimensional scores on a psychological exam? Aren’t they more like the impression you would get after talking to someone for a few minutes at a speed dating event? Doesn’t that mean that studies showing psychological exam scores are worse than speed dating events don’t disprove their use? Is all this talk equating “describable preferences” and “objective preferences” a red herring, since dating docs express describable subjective preferences?

Finally, how do studies looking at the general population transfer to the specific population of people who think they can do this? If 90% of people will just go for the hottest person they can find, and 10% of people look through dating docs very carefully, you shouldn’t tell the 10% they’re wrong because studies show that on average people only care about attractiveness.

Today there are lots of options for people who only care about attractiveness. The most popular dating apps, like Tinder, almost push you into that mode. I don’t know if their designers were going off of research suggesting that nothing else mattered. If they were, I think they should give the research a second look. If not, I think that leaves a hole for someone else to fill. Until someone does so at scale, dating docs are a good first-pass solution.

  1. Others have pointed out that this is the same as a shidduch (matchmaking) resume in Orthodox Judaism and other traditional cultures, and so passes the Cultural Evolution Test.

  2. Technically the statistic was that “only 4% of marriages are between Democrats and Republicans”, but I think if we assume most people are one or the other then this is equivalent.

  3. I’m being vague here because I and most of my friends are rationalists and mostly date rationalists and that already sorts heavily on a lot of things, and I don’t have good intuitions for what would happen without that filter.