dY dVorce = ?

March 26, 2009 3:48 PM

Oxford Professor & Fellow of the Royal Society James Murray uses mathematical modelling to predict whether a marriage will survive or end in divorce, with 94% accuracy. His lecture to the Royal Society will be available for view on demand within two days.
posted by UbuRoivas (44 comments total)

The government should waive marriage licence fees for engaged couples who agree to take this test.
posted by No Robots at 3:53 PM on March 26, 2009

The gist seems to be that if a couple hates each other, they shouldn't get married. Science!
posted by stavrogin at 3:57 PM on March 26, 2009 [7 favorites]

Partners who showed affection, humour or happiness as they talked were given the maximum points, while those who displayed contempt or belligerence received the minimum.

But that's how I show affection! I'm paying for your love with my currency of sarcastic comments and snarky replies.

Mocking aside, I believe I've heard of his work before, and it makes sense. "If either the husband or the wife is consistently negative, then they are going to get a divorce."

Honey, I not-so-secretly loathe you. We'll be fine, I swear.
posted by filthy light thief at 3:57 PM on March 26, 2009

Wow, again science triumphs over the human heart! When will we learn? Would you guys just hurry up and mathematically model our tribulations and triumphs, our greatest satisfactions and our fondest joys? That way we can digest them and get on to the serious business of adding to the GDP.
posted by felix betachat at 4:01 PM on March 26, 2009

...couples...were asked to sit opposite each other in a room on their own and talk about a contentious issue, such as...relations with their in-laws.

"So, honey, describe to me in single words only the good things that come to mind about your mother."

"My mother? I'll tell you about my mother!"
posted by turgid dahlia at 4:01 PM on March 26, 2009 [7 favorites]

John Gottman's been at this for a while too. His work was mentioned in Gladwell's book Blink, which might be where other people have heard of this kind of thing.
posted by Science! at 4:03 PM on March 26, 2009

The gist seems to be that if a couple hates each other, they shouldn't get married. Science!

Dude isn't participating in this thread. Why bring his name into it?
posted by gman at 4:04 PM on March 26, 2009 [2 favorites]

"Depending on the group, some couples might as well get divorced right away," he said.

Myself, I believe that marriage is about the beautiful mystery of leaping headlong into the 50% chance that one day your entire being will want to do nothing more than punch your spouse in the neck.
posted by jimmythefish at 4:08 PM on March 26, 2009 [5 favorites]

If half of all marriages in the US end in divorce, and you flat-out guessed which ones would survive or not, couldn't you come close to 100% accuracy by pure chance?
posted by gottabefunky at 4:10 PM on March 26, 2009 [1 favorite]

I predict, with 94% accuracy, that this guy is not married.
posted by orme at 4:10 PM on March 26, 2009 [1 favorite]

If half of all marriages in the US end in divorce, and you flat-out guessed which ones would survive or not, couldn't you come close to 100% accuracy by pure chance?

No. You'd guess at 50% accuracy.
posted by jimmythefish at 4:12 PM on March 26, 2009 [12 favorites]

If half of all marriages in the US end in divorce, and you flat-out guessed which ones would survive or not, couldn't you come close to 100% accuracy by pure chance?

Huh?
posted by CRM114 at 4:13 PM on March 26, 2009 [7 favorites]

gman: "The gist seems to be that if a couple hates each other, they shouldn't get married. Science!

Dude isn't participating in this thread. Why bring his name into it?"

Ha ha, that in joke would have been funny and timely if I had posted right before you.
posted by Science! at 4:21 PM on March 26, 2009 [1 favorite]

Dude isn't participating in this thread. Why bring his name into it?

I've always secretly dreamed of getting a MeTa callout. I know, I know...
posted by stavrogin at 4:21 PM on March 26, 2009

*hadn't
posted by Science! at 4:21 PM on March 26, 2009

John Gottman's been at this for a while too.

I'm sure they're working together. Murray probably manages the mathematical modeling end of things while Gottman is in charge of the interviews and the coding system.

Jim Murray is an great scientist. He's had a really diverse career, and he's one of the pioneers of using advanced mathematics to study biological systems. His book is a standard.

And please, Oxford is nice and all, but these guys are both professors at my alma mater.
posted by mr_roboto at 4:26 PM on March 26, 2009 [1 favorite]

Apparently if you have ever worn an "I'M WITH STUPID" t-shirt, you immediately lose 30 points.
posted by Blazecock Pileon at 4:28 PM on March 26, 2009

With this sort of study, I always have two questions:

- How many tweakable constants does the model have?
- Did you set the value of these constants before or after you looked at the data?

I have no knowledge of this study, but I do know that it's pretty easy to get a very predictive model if you base the model on the data you're trying to predict. This is essentially what people do when they correlate stock market performance with, say, hemlines. Post-hockery is very impressive, but means just about nothing.

The trick is using one set of data to generate the model, and then generating accurate predictions for a second whole set of data. But I'm not familiar at all with this research. Does anyone know how they tuned their model?
posted by goingonit at 4:35 PM on March 26, 2009 [2 favorites]

His calculations were based on 15-minute conversations between couples who were asked to sit opposite each other in a room on their own and talk about a contentious issue, such as money, sex or relations with their in-laws.

Professor Murray and his colleagues recorded the conversations and awarded each husband and wife positive or negative points depending on what was said.

Partners who showed affection, humour or happiness as they talked were given the maximum points, while those who displayed contempt or belligerence received the minimum.

So basically, couples who can't even fake nice for the scientists for 15 minutes are 100% likely to get divorced. I also wonder about the objectivity in these sorts of studies. Assigning a numerical value to contempt in conversation is not quite the same as reading a number off a dial. It is fairly easy to see individuals with in some cases probably very extensive experience watching couples communicate and being aware of the long term outcomes of their subjects. At what point are you projecting your experience-based intuition onto your results? Could the evaluation technique be taught to researchers with minimal experience in observing couples dynamics and maintain similar accuracy?

The other thing I always wonder about in this scenario is how the sample group is selected, whether it is possibly self-selected in a manner that increases its polarization (ideally you would just go out and grab married couples at random off the street if it wasn't for the stupid laws that impede science).

I have no doubt there is valuable and meaningful information in this research, but I'm more skeptical of the whole "secrets of relationships yield to math" aspect.
posted by nanojath at 4:45 PM on March 26, 2009 [2 favorites]

these guys are both professors at my alma mater

Maybe, but who ever heard of the University of Washington?

I mean, it's not one of the three Great Universities: Oxford, Cambridge, and Hull.
posted by UbuRoivas at 4:49 PM on March 26, 2009 [1 favorite]

jimmythefish: " I believe that marriage is about the beautiful mystery of leaping headlong into the 50% chance that one day your entire being will want to do nothing more than punch your spouse in the neck."

Norman Mailer - who had more than a little experience in this area - once said, "You never truly know a woman until you face her in court."
posted by Joe Beese at 4:50 PM on March 26, 2009 [2 favorites]

I also wonder about the objectivity in these sorts of studies. Assigning a numerical value to contempt in conversation is not quite the same as reading a number off a dial. It is fairly easy to see individuals with in some cases probably very extensive experience watching couples communicate and being aware of the long term outcomes of their subjects. At what point are you projecting your experience-based intuition onto your results?

Technique-wise, this is a major, major issue in observational psychological research. They work their asses off to develop objective systems for recording behavior and coding that behavior into quantitative data. I don't know what protocol is being used here, but you can bet that they've spent years or decades trying to get rid of this kind of bias.
posted by mr_roboto at 5:03 PM on March 26, 2009

I know I read about this in Blink, but I think I might have heard a radio interview about something similar. It involved looking at human faces during interviews in super-slow motion and cataloguing micro-facial-expressions. Things like a turn at the corner of the mouth, a furrowed brow, a flaring of the nostrils, which are often very quick (less than a second). Actually now that I think about it, it was to detect lying, and upon Googling I'm pretty sure it was this research by Mark Frank.

The trick behind the research was that you don't need to be able to read people, only to identify the expressions, which you can then map to emotions simply by doing a big enough controlled study in which subjects display the emotions you want to predict. As a consequence of hours spent on the research, the researchers could eventually spot the key indicators in real time, without having to resort to video. This meant that they can tell if you're lying simply by watching you speak, with very high accuracy. The research got quite a bit of attention from the security and intelligence sector.
posted by PercussivePaul at 5:06 PM on March 26, 2009

Yep, Murray and Gottman are long-time collaborators. Here's what I wrote about them a few years ago in Slate.
posted by escabeche at 5:09 PM on March 26, 2009 [1 favorite]

It seemed that people either started in a mean-spirited way, a critical way, started talking about a disagreement, started talking about a problem as just a symptom of their partner's inadequate character, which made their partner defensive and escalated the conflict, and people started getting mean and insulting to one another. That predicted the relationship was going to fall apart. 96% of the time the way the conflict discussion started in the first 3 minutes determined how it would go for the rest of the discussion.

...

Psycho-physiology is an important part of this research. It's something that Bob Levenson brought to the search initially, and then I got trained in psycho-physiology as well. And the reason we're interested in what was happening in the body is that there's an intimate connection between what's happening to the autonomic nervous system and what happening in the brain, and how well people can take in information — how well they can just process information — for example, just being able to listen to your partner — that is much harder when your heart rate is above the intrinsic rate of the heart, which is around a hundred to a hundred and five beats a minute for most people with a healthy heart.

At that point we know, from Loren Rowling's work, that people start secreting adrenalin, and then they get into a state of diffuse physiological arousal (or DPA) , so their heart is beating faster, it's contracting harder, the arteries start getting constricted, blood is drawn away from the periphery into the trunk, the blood supply shuts down to the gut and the kidney, and all kinds of other things are happening — people are sweating, and things are happening in the brain that create a tunnel vision, one in which they perceive everything as a threat and they react as if they have been put in great danger by this conversation.

I think a lot of you are giving this the short shrift. Yes, judging whether a comment is harsh or contemptuous is by nature subjective. On the other hand, this is interesting insight and it makes a fair amount of sense. Just because it requires a subjective human observer doesn't take away its usefulness, either as a tool for relationship counselors or for the participants in a relationship themselves

Read the interview
posted by crayz at 5:13 PM on March 26, 2009 [1 favorite]

Does anyone know how they tuned their model?

I didn't have a whole lot of luck finding the published paper.

I'm assuming that if it were published, it'd steal the thunder from the lecture, and that it will therfore be published at a later date...?
posted by UbuRoivas at 5:15 PM on March 26, 2009

I'm not sure why people are resisting this finding. 96% accuracy is pretty high and if I remember correctly from Blink, the subjects were chosen pretty randomly. A lot of us, including married but 'in love' folks, are blind to our own and other's faults and behaviors. I bet most of these people don't think they're going to get divorced. I think most of of us think our snarkiness is charming but it may be more meaningful than we think.
posted by shoesietart at 5:44 PM on March 26, 2009 [1 favorite]

It will be published in Cosmopolitan soon.
posted by Xoebe at 6:04 PM on March 26, 2009

I'm not sure why people are resisting this finding.

The inevitability of decay and death is not a romantic notion. People like to think their relationships will have permanence.
posted by Blazecock Pileon at 6:06 PM on March 26, 2009

I didn't have a whole lot of luck finding the published paper.

I would guess from Googling that a large part of the talk is pulled from Chapter 5 of his textbook Mathematical Biology: Modelling the Dynamics of Marital Interaction: Divorce Prediction and Marriage Repair
posted by PercussivePaul at 6:22 PM on March 26, 2009

If half of all marriages in the US end in divorce, and you flat-out guessed which ones would survive or not, couldn't you come close to 100% accuracy by pure chance?

I think I know the answer to this one: At least 50% of the time, yes.
posted by From Bklyn at 7:16 PM on March 26, 2009

dY dVorce = ?

I just wanted to point out that that's the best post title ever.
posted by Alex404 at 7:45 PM on March 26, 2009

I read about this in Blink too. And though I really didn't care for the book at all, it made me realize how incredibly hideously contemptuous I am. Of everything. There is so vanishingly little I won't roll my eyes at. I felt quite badly after that and immediately apologized to my partner and told him that now that I'm aware of it I am going to work toward doing it less, and hopefully not at all.

I couldn't believe what a dick I am. And that somebody would tolerate it! Nobody deserves that.
posted by birdie birdington at 8:07 PM on March 26, 2009

Ha ha, that in joke would have been funny and timely if I had posted right before you.

Timing!
posted by gman at 8:25 PM on March 26, 2009

I think I know the answer to this one: At least 50% of the time, yes.
Um, no, not even close to 50 % of the time.
posted by peacheater at 8:29 PM on March 26, 2009 [1 favorite]

I'm not sure why people are resisting this finding.

In the book linked above, Murray devotes a very interesting footnote to this phenomenon:

Among most people, particularly biophysical scientists, there is considerable skepticism expressed when it is proposed to try to use mathematical modeling in the psychological arena. Even when such an endeavor has been shown to be extremely useful as, for example, in the case of Zeeman (1977) in his seminal work on anorexia, the prejudice remains. Initially the research here was no exception. Interestingly, during the original discussions and meetings, without exception all of the mathematicians involved were initially skeptical (as was I). Also, without exception everyone involved became totally convinced in a very short time as o is relevance and practical use. Perhaps no one likes to believe that their emotions and reactions can be so starkly predicted with such simple mathematical models.

posted by Llama-Lime at 8:36 PM on March 26, 2009

I think I know the answer to this one: At least 50% of the time, yes.

It all depends on the setup. If he were just guessing, we'd expect him to guess each couple correctly 50% of the time, since that's the approximate rate of divorce. It's like guessing the outcome of a coin-flip. If he repeated the process for, say, 100 couples, the number of correct guesses would have a binomial (100,.5) distribution. The chance of getting at least 94 out of 100 correct would be something like 10^(-21).
posted by albrecht at 8:51 PM on March 26, 2009 [1 favorite]

Earlier.
posted by daksya at 9:29 PM on March 26, 2009

I don't think it's all that unlikely that you could get 94% accuracy from interviews and scoring. Don't you sometimes look at a couple and think that there's no way they'll last? What if you actually got some practice at assessing couples, and had assistance from a data set of previous divorces/happy relationships?

I had one of those relationships with lots of sarcasm and eye-rolling, and it didn't end well. And my friends were kind enough to only say "we told you so" once - they knew right from the start how it would be, tried to tell me, then let me do my own thing. If they'd had some data to back them up, well, I probably would have still hung around for the hot sex. But I bet I'd also have given up on the guy about a year earlier than I did.
posted by harriet vane at 1:26 AM on March 27, 2009

If half of all marriages in the US end in divorce, and you flat-out guessed which ones would survive or not, couldn't you come close to 100% accuracy by pure chance?

I think I know the answer to this one: At least 50% of the time, yes.

I thought he was saying that, with the application of Murray's special divorce-calculating math, and then some judicious guessing on top of that to pick up the slack for the last 6% the formula can't account for, wouldn't he be able to reach 100%? My answer was, stated as intentionally nebulously as the question in order to highlight the fact that I could be talking about damn near anything implying I am really shitty at math, which I am, that he would be 100% right, 50% of the time. Or, more concisely, "No."
posted by From Bklyn at 1:49 AM on March 27, 2009 [1 favorite]

As a scientific exercise, it would be interesting to see a variant of this research applied to arranged marriages to see if the two participants were even remotely compatible.
posted by MuffinMan at 5:51 AM on March 27, 2009

He looks so much different in the match.com ads...

The information in the article is so limited, it is hard to know what to make of this. I would like to see the comparison of his model's predictions and the predictions of the interviewers without the models.
posted by batou_ at 6:41 AM on March 27, 2009

As a consequence of hours spent on the research, the researchers could eventually spot the key indicators in real time, without having to resort to video. This meant that they can tell if you're lying simply by watching you speak, with very high accuracy.

So you're telling me that Tim Roth in LIE TO ME isn't complete bullshit?
posted by Justinian at 11:11 AM on March 27, 2009

They could have acheived 100% accuracy if they factored in ƒ(ȹ/t) {frequency of oral over time}
posted by MiltonRandKalman at 2:56 PM on March 27, 2009