I don’t know if it improves success conditional on dating someone; I could tell a story either way—being hot makes you more of a keeper, but then, maybe you’ll be quicker to leave. It’s a competitive market. Doesn’t seem too relevant. The point of the hotness part is that when I say ‘dating success is unpredictable’, that’s bracketing the obvious factor, visible to everyone. Obviously you can predict that a supermodel is probably going to get dates. It’s not interesting to say, “a 5 should probably not try to date a 10”. What is very surprising is, “here are a 5 and a 5; you can ask both of them as many questions as you like, of any sort, feed it into your fanciest statistical model if you need to, and you are still not going to outperform a coinflip in predicting how well their date, as far as we can tell from all our failed efforts to do so”. This is especially surprising because we do this all the time with many other apparently equally formidable social prediction tasks like predicting criminal recidivism or child abuse or suicide: the incremental prediction validity might not be great, but at least it’s not frigging negative variance! I’m not sure what other social psych thing seems to be so precisely null predictive.
this unpredictability sounds like something from another planet.
I agree that it’s very surprising and I’m still looking for a good explanation. The best I have so far is an analogy from evolutionary genetics theories of personality where personality is either stuck as, or possibly deliberately functioning as, a randomizing device (because personality is a hawk-dove model), and this leads to canalization of preferences for obviously desirable traits like health, but then additional preferences are completely inscrutable randomness.
Because if you’re on the tail then you need someone else on the tail (or at least skewed to your side) while if you’re in the “bulk” then most people (who are also in the bulk) will do?
I can’t disprove that but I see no reason to invoke it either. If there were small clusters or niches or heterogeneity like that, I would still expect approaches like random forests to have found them (that’s exactly what they are brilliant at). And I have already explained why there have to be many severe illusions about how effective any selection process is, because no one experiences adequate sample sizes nor do they ever see the counterfactual. As Bacon asked, “where are the sailors who prayed to the idols and did not return?”
Like Vanessa I’m confused about the tails story.
It is clear to me that people highly sort on class/education/iq when choosing a mate. You are saying this isn’t relevant—purely an artifact of who people hang out with? They could have just as easily found happiness outside their class? (Or is this somehow sneaked in by the hotness factor?)
It seems difficult to believe that this does not factor in to long-term compatibility.
Another: I strongly predict your chance of a successful long-term relationship will be much lower if you date somebody serious mental illness or substance abuse problems.
Perhaps you are only claiming there are no predictors for falling in love rather than long-term compatibility?
In that case, selecting on sensible deal-breakers becomes extra important: you can randomly at any moment fall in love with a completely terrible mate.
If love really is so random—the traditional control that family exerted over young people makes a lot more sense
The randomness theory explains assortative mating variables as a mix of desirability and selection effect. To expand on Kaj’s comment and move away from the STEM/nerd stuff, let’s consider the obvious one of politics. Variables like intelligence, income, or education are part of broadly-agreed upon desirability, OK, fine, so they don’t constitute individually-predictive or dyadic factors beyond the net desirability—but how can we explain Republican vs Democrat that way? Half the population thinks Republicans are better, and the other half Democrats, no net there. Some of these studies will do political-related questions as extremely obvious to ask about, and you do see large correlations between spouse’s (pre-relationship) party affiliation in the US population, but still, the study nulls are null.
Does that mean the studies are wrong somehow? Well, maybe a little: range restriction is a problem here. (I would definitely like more diverse studies here, even if the extent of range restriction in them ironically emphasizes my overall point about selection effects, and doesn’t undermine my other points too much.*) But the population correlates are probably more wrong. Because you can’t click with someone you haven’t met! As the joke goes, “don’t marry for money—spend all your time with rich people & marry for love”; if you choose to spend time only with Republicans because you’re a Republican, you will, of course, only find the Republican you click with, and not the equally numerous Democrats you would have clicked with. (How could it be otherwise? What, would you exude a pheromone that only your Democrats counterparts smell from miles away and snuffle their way to your address to ask you out on a date?) This mere proximity manufactures a population correlate, even though there is no individual correlate—nothing intrinsic about being a Republican makes that person more likely to click with you. The speed-dating gets a null on Rep vs Dem because it forces each side to encounter the other side proportionately much, revealing the proportionately-much click. This is the unseen counterfactual.
So, if you spend all of your time hanging out with STEM nerds or EAers, or if you are a celebrity which spends all their time hanging out with celebrities, or if you are an academic who—hard to believe, I know—spends much of their time interacting with other academics… Some of that is the desirability (high SES etc), and then the default for the rest is just “those are the people you meet, so that’s where you meet the people you can click with”.
This is boring and trivial, but it does have the implication for optimizing dating that you should not try to use simple partisan affiliation per se as a screen because it will just produce false negatives, and throw out people you should’ve tried a date with.
* to argue that ‘range restriction means that these studies are irrelevant to criticisms of dating-checklists because their measures have no useful variation’, you’d have to show that the dating-checklists don’t suffer from the same problem as much or worse. Which given the overlap in subject matter and the relative diversity of ‘a bunch of semi-random students at an ordinary university shanghaied by the researchers’ vs ‘the guys I recruit online via Facebook’, is… implausible.
From personal anecdote it seems pretty clear to me people have idiosyncratic preferences orthogonal to general desirability that are real.
Even spending a lot of time with people that don’t fit those preferences does not make them attractive.
I do take your points about the state of the research.
I can’t disprove that but I see no reason to invoke it either. If there were small clusters or niches or heterogeneity like that, I would still expect approaches like random forests to have found them (that’s exactly what they are brilliant at). And I have already explained why there have to be many severe illusions about how effective any selection process is, because no one experiences adequate sample sizes nor do they ever see the counterfactual.
On the one hand I agree that everything I know about this is anecdotal. On the other hand, am I really supposed to believe it’s a complete coincidence that my spouse is a STEM nerd who reads lots of sci/fantasy and is heavily involved with effective altruism[1]? The replication crisis etc left me with a feeling that when common sense and social science studies disagree, there’s a fair chance the studies are wrong. But, shrugs, what do I know...
Oh, and, another thought. There seems to be a pretty strong trend for celebrities to date other celebrities / show-business people. I think there’s also a trend for academics to date other academics. How come those random forests didn’t detect it?
And, I’m pretty sure there is a lot of selection based on ethnicity and religion (more than explainable by geography). Maybe you meant that this is just a “boring” part which we can assume given for the sake of the discussion?
A possible objection is: my spouse is like this because that’s how I’ve been selecting lovers. Okay, but the same is true about e.g. hotness. The aforementioned studies also weren’t causal interventions with control groups, I assume. (Plus, the selection pressure I applied wasn’t strong enough to explain the outcome.)
I think there’s also a trend for academics to date other academics. How come those random forests didn’t detect it?
Likely because all of the people in question were academics, or at least undergraduate ones, so there was no chance to detect differences in how they matched with non-academics:
Sample A consisted of 163 undergraduate students (81 women and 82 men; mean age = 19.6 years, SD = 1.0) who attended one of seven speed-dating events in 2005. Sample B consisted of 187 undergraduate students (93 women and 94 men; mean age = 19.6 years, SD = 1.2) who attended one of eight such events in 2007. Sample size was determined by the number of speed-dating events we were able to hold in 2005 and 2007 and the number of participants we were able to recruit for each event while maintaining an equal gender ratio. All participants, who were recruited via on-campus flyers and e-mails to participate in a speed-dating study, had the goal of meeting and potentially matching with opposite-sex participants. [...]
The present results were obtained with undergraduate samples; a more demographically diverse sample might exhibit matching by sociological factors such as age, socioeconomic status, cultural background, or religious background.
I don’t know if it improves success conditional on dating someone; I could tell a story either way—being hot makes you more of a keeper, but then, maybe you’ll be quicker to leave. It’s a competitive market. Doesn’t seem too relevant. The point of the hotness part is that when I say ‘dating success is unpredictable’, that’s bracketing the obvious factor, visible to everyone. Obviously you can predict that a supermodel is probably going to get dates. It’s not interesting to say, “a 5 should probably not try to date a 10”. What is very surprising is, “here are a 5 and a 5; you can ask both of them as many questions as you like, of any sort, feed it into your fanciest statistical model if you need to, and you are still not going to outperform a coinflip in predicting how well their date, as far as we can tell from all our failed efforts to do so”. This is especially surprising because we do this all the time with many other apparently equally formidable social prediction tasks like predicting criminal recidivism or child abuse or suicide: the incremental prediction validity might not be great, but at least it’s not frigging negative variance! I’m not sure what other social psych thing seems to be so precisely null predictive.
I agree that it’s very surprising and I’m still looking for a good explanation. The best I have so far is an analogy from evolutionary genetics theories of personality where personality is either stuck as, or possibly deliberately functioning as, a randomizing device (because personality is a hawk-dove model), and this leads to canalization of preferences for obviously desirable traits like health, but then additional preferences are completely inscrutable randomness.
I can’t disprove that but I see no reason to invoke it either. If there were small clusters or niches or heterogeneity like that, I would still expect approaches like random forests to have found them (that’s exactly what they are brilliant at). And I have already explained why there have to be many severe illusions about how effective any selection process is, because no one experiences adequate sample sizes nor do they ever see the counterfactual. As Bacon asked, “where are the sailors who prayed to the idols and did not return?”
Thank you for your scholarship, gwern!
Like Vanessa I’m confused about the tails story. It is clear to me that people highly sort on class/education/iq when choosing a mate. You are saying this isn’t relevant—purely an artifact of who people hang out with? They could have just as easily found happiness outside their class? (Or is this somehow sneaked in by the hotness factor?) It seems difficult to believe that this does not factor in to long-term compatibility.
Another: I strongly predict your chance of a successful long-term relationship will be much lower if you date somebody serious mental illness or substance abuse problems.
Perhaps you are only claiming there are no predictors for falling in love rather than long-term compatibility?
In that case, selecting on sensible deal-breakers becomes extra important: you can randomly at any moment fall in love with a completely terrible mate.
If love really is so random—the traditional control that family exerted over young people makes a lot more sense
The randomness theory explains assortative mating variables as a mix of desirability and selection effect. To expand on Kaj’s comment and move away from the STEM/nerd stuff, let’s consider the obvious one of politics. Variables like intelligence, income, or education are part of broadly-agreed upon desirability, OK, fine, so they don’t constitute individually-predictive or dyadic factors beyond the net desirability—but how can we explain Republican vs Democrat that way? Half the population thinks Republicans are better, and the other half Democrats, no net there. Some of these studies will do political-related questions as extremely obvious to ask about, and you do see large correlations between spouse’s (pre-relationship) party affiliation in the US population, but still, the study nulls are null.
Does that mean the studies are wrong somehow? Well, maybe a little: range restriction is a problem here. (I would definitely like more diverse studies here, even if the extent of range restriction in them ironically emphasizes my overall point about selection effects, and doesn’t undermine my other points too much.*) But the population correlates are probably more wrong. Because you can’t click with someone you haven’t met! As the joke goes, “don’t marry for money—spend all your time with rich people & marry for love”; if you choose to spend time only with Republicans because you’re a Republican, you will, of course, only find the Republican you click with, and not the equally numerous Democrats you would have clicked with. (How could it be otherwise? What, would you exude a pheromone that only your Democrats counterparts smell from miles away and snuffle their way to your address to ask you out on a date?) This mere proximity manufactures a population correlate, even though there is no individual correlate—nothing intrinsic about being a Republican makes that person more likely to click with you. The speed-dating gets a null on Rep vs Dem because it forces each side to encounter the other side proportionately much, revealing the proportionately-much click. This is the unseen counterfactual.
So, if you spend all of your time hanging out with STEM nerds or EAers, or if you are a celebrity which spends all their time hanging out with celebrities, or if you are an academic who—hard to believe, I know—spends much of their time interacting with other academics… Some of that is the desirability (high SES etc), and then the default for the rest is just “those are the people you meet, so that’s where you meet the people you can click with”.
This is boring and trivial, but it does have the implication for optimizing dating that you should not try to use simple partisan affiliation per se as a screen because it will just produce false negatives, and throw out people you should’ve tried a date with.
* to argue that ‘range restriction means that these studies are irrelevant to criticisms of dating-checklists because their measures have no useful variation’, you’d have to show that the dating-checklists don’t suffer from the same problem as much or worse. Which given the overlap in subject matter and the relative diversity of ‘a bunch of semi-random students at an ordinary university shanghaied by the researchers’ vs ‘the guys I recruit online via Facebook’, is… implausible.
From personal anecdote it seems pretty clear to me people have idiosyncratic preferences orthogonal to general desirability that are real. Even spending a lot of time with people that don’t fit those preferences does not make them attractive.
I do take your points about the state of the research.
On the one hand I agree that everything I know about this is anecdotal. On the other hand, am I really supposed to believe it’s a complete coincidence that my spouse is a STEM nerd who reads lots of sci/fantasy and is heavily involved with effective altruism[1]? The replication crisis etc left me with a feeling that when common sense and social science studies disagree, there’s a fair chance the studies are wrong. But, shrugs, what do I know...
Oh, and, another thought. There seems to be a pretty strong trend for celebrities to date other celebrities / show-business people. I think there’s also a trend for academics to date other academics. How come those random forests didn’t detect it?
And, I’m pretty sure there is a lot of selection based on ethnicity and religion (more than explainable by geography). Maybe you meant that this is just a “boring” part which we can assume given for the sake of the discussion?
A possible objection is: my spouse is like this because that’s how I’ve been selecting lovers. Okay, but the same is true about e.g. hotness. The aforementioned studies also weren’t causal interventions with control groups, I assume. (Plus, the selection pressure I applied wasn’t strong enough to explain the outcome.)
Likely because all of the people in question were academics, or at least undergraduate ones, so there was no chance to detect differences in how they matched with non-academics: