Cool dataset! Re: kids - having kids adds a lot of strain that can make relationships (& life) substantially harder, but people are more willing to stay in mediocre/bad relationships "for the kids". The kid burden might be combined with a survivorship effect - long relationships among ppl who have kids are worse on average, bc people without kids are less likely to stay in unsatisfying relationships for a long time.
My big pet peeve with this is that you have unequally-sized bins but are plotting them equidistant, or with equally-wide bars. Your first bin only covers 3 months, while your second-to-last covers 4 years. It should be 16 times as wide as the first bin. Plotting them on an equal-width histogram, or worse, a line chart with equally-spaced points, distorts the actual shape of the curves.
It seems highly plausible to me that relationships have fewer salient events as time goes on (i.e. the second 3 months might contain as many salient events as the fourth set of four years), and whether or not these plots look linear is a test of whether or not Aella got the 'natural binning' for relationships.
I don't agree, but still, if you're going to log-scale your time axis, then *actually* log-scale your time axis, don't use differently sized bins as a crude approximation of log scale.
Re: Porn question, my understanding/misogynistic stereotype? is that women are often more inclined towards written porn, raunchy books etc. than men, so your question which asks about 'Watching' porn may have underrated women's overall porn consumption
Loved reading this! Some comments about kids: the most difficult period is 0-2 yrs and then 2-4 yrs is still hard but not as fully consuming and sleep-deprived. Sense of self dips a ton at one kid, but goes very low after 2 kids. So yes, 2 kids is much harder. Spacing between kids matters too. (For the record I love my kids a lot and think it’s one of the best things I’ve ever done.)
Yeah this is what data generally shows when I have seen it, and definitely my experience and that of my network.
My wife and I have a great marriage, pretty much smooth sailing the whole way. Was strong the 5 years before kids, financially stable, decent jobs etc. Probably 18 months after the first kid (who was a terrible sleeper), was the only time in our relationship we ever had a fight that included the word divorce in it. I don't think we wee even 3% of the way there, but that the topic even came up was totally uncharacteristic and a sign of how compromised out relationship had become.
Her with ~5 months of increasingly uncomfortable pregnancy, and then 18 months of hard core sleep deprivation and hobby elimination. Me with 18 months of minor sleep deprivation, hobby elimination, some professional stress, and our sex life dropping to near zero (she isn't interested in sex unless she is well rested and in a good mood, whereas for me the more crappy I feel the more I need sex to cope).
All this was made a lot worse because we had kids away from family where there was no babysitting support and we were not so well off to afford regular sitting (all our sitting money was used to cover day-care being closed on certain work days).
Anyway, once the kid started sleeping through the night and we both got sleep our relationship generally, and sex life in particular, healed quickly. The second kid wasn't quite so bad because the second kid slept somewhat better, plus we had better methods of dealing and the professional stress for me was gone.
And now we are at almost 14 years of marriage and literally every year is the best year so far in terms of our relationship.
Kids are this crazy terrible disruption in your life, and yet somehow totally worth it. Hard to explain. Maybe imagine spending years writing a long and amazing and personally hugely fulfilling novel or art project. Not for commerce, but for yourself. Except it is way better than that even.
I wonder if any special treatment is needed here for long-distance relationships. Some questions explicitly ask for data in the last 6 months, which may change results compared to "the norm", however the people in the relationship imagine it. Though maybe that's exactly the point.
Thanks for putting up the full raw data. However, we need a "data dictionary" - some columns are not self-explanatory. For example, what is "codependent5" or "jealousy4" in this?
I feel like age is supposed to be a major confounder here: the bins with longer relationships are likely to be older (my own relationship is rather long and started rather early, but I'm an outlier among the people I know)
Your last point is something I was thinking a bunch about. All these relationships have worked out so far. I'd be interested in a similar survey based on "your last relationship, whether you're still in it or not". Or, alternatively, "your last relationship that you're no longer in". Both would be further illuminating.
> This specific type of question feels like unusually hard for me to think about - am I making sense? I’m trying to disentangle “of course you’re not going to have any 4-year pre-marriage dating periods in a 3-year relationship” from seeing relationships actually falling off the chart and only the successful ones left over.
If you assume age/start time of the relationship don't matter, then you can set up a dynamical model (see the sheet-each cell has ancestors, and strength of inheritance from them, determined by model parameters). In my simplest model, relationships survive to the next year 95% of the time, and get married 20% of the time each year. Then, by looking at how this distribution compares to the one you found, we can tune the numbers to find the best fit.
But we can also complicate the model lots! Maybe marriages are more (or less) likely to last than unmarried relationships; let's tune those probabilities separately. Or maybe relationships have different failure probabilities depending on their age, independent of when they got married. Or maybe marriages lock-in their failure rate when they get married (the 'quick marriages fail more' model, rather than 'young relationships fail more').
But anyway, I think you want to build a model like this and then do some fitting and see how well they do. (You can eventually do fancy Bayesian stats here, but I think it helps to start out by playing around and doing the optimization by hand.)
Agree that explicit modeling of the relationship between age, relationship duration, and relationship extinction/censoring would improve the analysis.
Another piece is that young kids drive a selection bias that seems hard to account for: parents of toddlers don't have the time to fill random internet surveys. I'm definitely sympathetic to the argument that we shouldn't discount internet surveys just because they're drawn from convenience samples, but I suspect that this effect skews the picture for eg # children over relationship length -- it's probably much less true that '1/3 of relationships produce a child between years 10 and 14' and much more true that 'between years 10 and 14, childcare demands for 1/3 of relationships reduce to the level where they have the bandwidth to report children on internet surveys'.
One way around this issue would be to compare to a reference class taken from the census, though at that point we very likely would be running into issues with distributional differences between all Americans and the people who responded to the survey. Another way would be to ask explicitly for ages of children (and whether they're products of the current relationship), but that would add a lot of overhead and I don't recall it having been done. Probably some other ways to estimate, too.
Um, but also, to be clear, I'm very grateful that for the data collection and analysis that's been performed so far; my yammering about technical nitpicks shouldn't be taken to undermine that
One thing I'm going to speculate about - maybe the 7 year mark is when age sets in, and in hetero relationships, someone becomes less attractive.
While it's tempting to immediately jump to "ah! Straight men no longer like their wives when they stop looking 20!" I wanna change tack. With the advent of skincare, UV avoidance, good diet and exercise, women in their late 30s and 40s are looking better than ever.
I actually wanna posit that it's the men. I feel like late 30s/40s is often when the gut sets in, and if they were already fat, things start drooping unattractively. Also, the current crop of middle aged men have not been very sunsmart as kids, so they're all spotty and wrinkly (I suspect this aspect will improve as people who grew up more sun-aware age).
The oldest sexiest man alive ever voted by People is 59 (Sean Connelly). You notice that most 59 year old men look nothing like him? Conversely, I feel like most women look closer to an equivalent aged model/celebrity, because they are least try to look like that, and it makes the difference.
Go to a mall and people watch, and I feel like women tend to look better than their presumably same aged husbands.
The problem is, I suspect a lot of guys know that compared to 7 years ago, their aged appearance is a bit underwhelming. Hence men feeling more judged on average. Cheating doesn't have the same baggage - that other person has typically not seen him at his younger, hotter self. Having kids definitely accelerates this transition.
Solution.... I guess, much like the many attempts to expand what mainstream feminine sex appeal looks like, we could attempt to expand what mainstream masculine sex appeal looks like? And also somehow motivate men to put in the work to continue to be sexy. Appealing to your partner is, on the most part, absolutely not the reason why many women put in the effort to keep being hot - I feel it's most frequently self esteem with a pinch of peer pressure - there's a reason why shopping/makeup ads go "treat yourself", not "treat your husband".
(It could also be hair loss? Maybe that should be a survey question - attempt to match amount of balding with marital happiness)
The cheating thing blows my mind, it never was a surprise that ppl are often cheated on, what was such a surprise was *the frequency with which they never find out!*
I mean wow!
Then I continued to reflect on it - these are just relationships which are still surviving. I wonder how many relationships (lasting or not) people statistically are cheated on in. If 33% of men cheated in the relationship they are CURRENTLY in, I can’t even imagine how high that number would be as an average of all relationships! Maybe it’s pretty constant, but I’m just wondering.
Also curious to know data points such as “if your partner revealed to you that they cheated, how long did it take them to tell you?” And “if your partner never revealed to you but you found out anyway, how long did it take you to find out?”
These graphs could def use some standard deviations! I wonder how many of the lines that look stable over time are because they are averaging out a lot of variance between relationships vs almost all relationships at X years have very tightly clustered answers for this question.
That could be confounded quite a bit by retirement age people (who certainly do have lots of sex, but I am pretty sure at lower rates than younger people), and/or people with ill partners. Just a thought. But yes I doubt the sample was totally "representative" or people in general.
Very interesting ! I'd have very much been interested in seeing a bit more data to understand if the 7 years dip is child related or not.
The 2 graphs we have are inconclusive I think. We see people with kids have more difficulty in their relationships, but not really if that 7-year dip is related. Actually for these 2 questions I don't even see a dip. Whereas for most of the relationship health questions, there's a visible dip at 7 years.
I know the raw data is there, but I don't think I'm good enough at manipulating them to find the answer. Maybe for a follow-up email on the specific question "do kids fuck up your relationship right after you have them or are we all doomed to have a worse relationship post 7 years" ? :)
One day I'd like to open your data file and do my own slicing and dicing.
I'm curious where you got the fictional character ratings - why are Monica and Chandler closer than Morticia and Gomez? Why are there no actual romantic couples on the low end? You had to dig into obscure non-cannon fanfiction for Harry and Draco, who never got involved as far as most fans know?
On a practical level, I have one suggestion. Can some of the bar charts be expressed as a stacked bar to show percentiles or/and ranges instead of just averages? Do we, for example, have lots more different people interested in others after 15 years, or do we have a small number of people who are more acutely interested in others after 15 years?
👋 I started talking with you about this at the manifest after party. To analyze the "time until marriage" data correctly, I think you can:
* Pick two of your "time in relationship" groups to compare, for instance 5-6y and 9-12y.
* From the group that spent more time in a relationship, filter out the ones that got married more years into the relationship than the other group has had years in the relationship. For instance take the 9-12y group and filter out those that got married more than 6 years into the relationship.
* Now compute the average time before marriage of each group.
You can compare each group to one base group to get a plot, or compare each group to the one before it and plot the differences
Cool dataset! Re: kids - having kids adds a lot of strain that can make relationships (& life) substantially harder, but people are more willing to stay in mediocre/bad relationships "for the kids". The kid burden might be combined with a survivorship effect - long relationships among ppl who have kids are worse on average, bc people without kids are less likely to stay in unsatisfying relationships for a long time.
My big pet peeve with this is that you have unequally-sized bins but are plotting them equidistant, or with equally-wide bars. Your first bin only covers 3 months, while your second-to-last covers 4 years. It should be 16 times as wide as the first bin. Plotting them on an equal-width histogram, or worse, a line chart with equally-spaced points, distorts the actual shape of the curves.
It seems highly plausible to me that relationships have fewer salient events as time goes on (i.e. the second 3 months might contain as many salient events as the fourth set of four years), and whether or not these plots look linear is a test of whether or not Aella got the 'natural binning' for relationships.
Yes, non-linear binning is fine, it's the presentation that is misleading.
...why? It's like log-adjusting your axes, which seems fine to me.
Why would you log-scale a time axis?
When it's appropriate for the phenomenon in question?
Like, I think it makes sense for the Weibull distribution, which is maybe a good choice for modelling relationship failure.
I don't agree, but still, if you're going to log-scale your time axis, then *actually* log-scale your time axis, don't use differently sized bins as a crude approximation of log scale.
Re: Porn question, my understanding/misogynistic stereotype? is that women are often more inclined towards written porn, raunchy books etc. than men, so your question which asks about 'Watching' porn may have underrated women's overall porn consumption
I did the survey and it asked about porn and erotic content consumed for the purposes of arousal, which seemed a bit vague. What counts?
Loved reading this! Some comments about kids: the most difficult period is 0-2 yrs and then 2-4 yrs is still hard but not as fully consuming and sleep-deprived. Sense of self dips a ton at one kid, but goes very low after 2 kids. So yes, 2 kids is much harder. Spacing between kids matters too. (For the record I love my kids a lot and think it’s one of the best things I’ve ever done.)
Yeah this is what data generally shows when I have seen it, and definitely my experience and that of my network.
My wife and I have a great marriage, pretty much smooth sailing the whole way. Was strong the 5 years before kids, financially stable, decent jobs etc. Probably 18 months after the first kid (who was a terrible sleeper), was the only time in our relationship we ever had a fight that included the word divorce in it. I don't think we wee even 3% of the way there, but that the topic even came up was totally uncharacteristic and a sign of how compromised out relationship had become.
Her with ~5 months of increasingly uncomfortable pregnancy, and then 18 months of hard core sleep deprivation and hobby elimination. Me with 18 months of minor sleep deprivation, hobby elimination, some professional stress, and our sex life dropping to near zero (she isn't interested in sex unless she is well rested and in a good mood, whereas for me the more crappy I feel the more I need sex to cope).
All this was made a lot worse because we had kids away from family where there was no babysitting support and we were not so well off to afford regular sitting (all our sitting money was used to cover day-care being closed on certain work days).
Anyway, once the kid started sleeping through the night and we both got sleep our relationship generally, and sex life in particular, healed quickly. The second kid wasn't quite so bad because the second kid slept somewhat better, plus we had better methods of dealing and the professional stress for me was gone.
And now we are at almost 14 years of marriage and literally every year is the best year so far in terms of our relationship.
Kids are this crazy terrible disruption in your life, and yet somehow totally worth it. Hard to explain. Maybe imagine spending years writing a long and amazing and personally hugely fulfilling novel or art project. Not for commerce, but for yourself. Except it is way better than that even.
I wonder if any special treatment is needed here for long-distance relationships. Some questions explicitly ask for data in the last 6 months, which may change results compared to "the norm", however the people in the relationship imagine it. Though maybe that's exactly the point.
Hi Aella.
Thanks for putting up the full raw data. However, we need a "data dictionary" - some columns are not self-explanatory. For example, what is "codependent5" or "jealousy4" in this?
Agreed. I'd also love a copy of the original survey questions, and the mapping for how you gave out the relationship rankings at the end.
I feel like age is supposed to be a major confounder here: the bins with longer relationships are likely to be older (my own relationship is rather long and started rather early, but I'm an outlier among the people I know)
Your last point is something I was thinking a bunch about. All these relationships have worked out so far. I'd be interested in a similar survey based on "your last relationship, whether you're still in it or not". Or, alternatively, "your last relationship that you're no longer in". Both would be further illuminating.
> This specific type of question feels like unusually hard for me to think about - am I making sense? I’m trying to disentangle “of course you’re not going to have any 4-year pre-marriage dating periods in a 3-year relationship” from seeing relationships actually falling off the chart and only the successful ones left over.
I think you want to look at this data in another dimension. I tried to make a real example to show you, but I couldn't find those columns in your public data, so I'll use fictional data. (See some examples here: https://docs.google.com/spreadsheets/d/1fL2FOslsScCTuWSJI-UQaP3Ycil7JA605JoxEtYxVpA/edit?usp=sharing )
If you assume age/start time of the relationship don't matter, then you can set up a dynamical model (see the sheet-each cell has ancestors, and strength of inheritance from them, determined by model parameters). In my simplest model, relationships survive to the next year 95% of the time, and get married 20% of the time each year. Then, by looking at how this distribution compares to the one you found, we can tune the numbers to find the best fit.
But we can also complicate the model lots! Maybe marriages are more (or less) likely to last than unmarried relationships; let's tune those probabilities separately. Or maybe relationships have different failure probabilities depending on their age, independent of when they got married. Or maybe marriages lock-in their failure rate when they get married (the 'quick marriages fail more' model, rather than 'young relationships fail more').
But anyway, I think you want to build a model like this and then do some fitting and see how well they do. (You can eventually do fancy Bayesian stats here, but I think it helps to start out by playing around and doing the optimization by hand.)
Agree that explicit modeling of the relationship between age, relationship duration, and relationship extinction/censoring would improve the analysis.
Another piece is that young kids drive a selection bias that seems hard to account for: parents of toddlers don't have the time to fill random internet surveys. I'm definitely sympathetic to the argument that we shouldn't discount internet surveys just because they're drawn from convenience samples, but I suspect that this effect skews the picture for eg # children over relationship length -- it's probably much less true that '1/3 of relationships produce a child between years 10 and 14' and much more true that 'between years 10 and 14, childcare demands for 1/3 of relationships reduce to the level where they have the bandwidth to report children on internet surveys'.
One way around this issue would be to compare to a reference class taken from the census, though at that point we very likely would be running into issues with distributional differences between all Americans and the people who responded to the survey. Another way would be to ask explicitly for ages of children (and whether they're products of the current relationship), but that would add a lot of overhead and I don't recall it having been done. Probably some other ways to estimate, too.
Um, but also, to be clear, I'm very grateful that for the data collection and analysis that's been performed so far; my yammering about technical nitpicks shouldn't be taken to undermine that
One thing I'm going to speculate about - maybe the 7 year mark is when age sets in, and in hetero relationships, someone becomes less attractive.
While it's tempting to immediately jump to "ah! Straight men no longer like their wives when they stop looking 20!" I wanna change tack. With the advent of skincare, UV avoidance, good diet and exercise, women in their late 30s and 40s are looking better than ever.
I actually wanna posit that it's the men. I feel like late 30s/40s is often when the gut sets in, and if they were already fat, things start drooping unattractively. Also, the current crop of middle aged men have not been very sunsmart as kids, so they're all spotty and wrinkly (I suspect this aspect will improve as people who grew up more sun-aware age).
The oldest sexiest man alive ever voted by People is 59 (Sean Connelly). You notice that most 59 year old men look nothing like him? Conversely, I feel like most women look closer to an equivalent aged model/celebrity, because they are least try to look like that, and it makes the difference.
Go to a mall and people watch, and I feel like women tend to look better than their presumably same aged husbands.
The problem is, I suspect a lot of guys know that compared to 7 years ago, their aged appearance is a bit underwhelming. Hence men feeling more judged on average. Cheating doesn't have the same baggage - that other person has typically not seen him at his younger, hotter self. Having kids definitely accelerates this transition.
Solution.... I guess, much like the many attempts to expand what mainstream feminine sex appeal looks like, we could attempt to expand what mainstream masculine sex appeal looks like? And also somehow motivate men to put in the work to continue to be sexy. Appealing to your partner is, on the most part, absolutely not the reason why many women put in the effort to keep being hot - I feel it's most frequently self esteem with a pinch of peer pressure - there's a reason why shopping/makeup ads go "treat yourself", not "treat your husband".
(It could also be hair loss? Maybe that should be a survey question - attempt to match amount of balding with marital happiness)
Wonderful, wonderful data! Thank you Aella!
The cheating thing blows my mind, it never was a surprise that ppl are often cheated on, what was such a surprise was *the frequency with which they never find out!*
I mean wow!
Then I continued to reflect on it - these are just relationships which are still surviving. I wonder how many relationships (lasting or not) people statistically are cheated on in. If 33% of men cheated in the relationship they are CURRENTLY in, I can’t even imagine how high that number would be as an average of all relationships! Maybe it’s pretty constant, but I’m just wondering.
Also curious to know data points such as “if your partner revealed to you that they cheated, how long did it take them to tell you?” And “if your partner never revealed to you but you found out anyway, how long did it take you to find out?”
Some juicy data points available here 😂
Thanks again aella!
These graphs could def use some standard deviations! I wonder how many of the lines that look stable over time are because they are averaging out a lot of variance between relationships vs almost all relationships at X years have very tightly clustered answers for this question.
This can't possibly be right. For instance, "47% of married couples have sex less than once a week."
Your numbers are SIGNIFICANTLY off that...
Source: https://www.webmd.com/sex-relationships/what-to-know-how-often-married-couples-have-sex
That could be confounded quite a bit by retirement age people (who certainly do have lots of sex, but I am pretty sure at lower rates than younger people), and/or people with ill partners. Just a thought. But yes I doubt the sample was totally "representative" or people in general.
Very interesting ! I'd have very much been interested in seeing a bit more data to understand if the 7 years dip is child related or not.
The 2 graphs we have are inconclusive I think. We see people with kids have more difficulty in their relationships, but not really if that 7-year dip is related. Actually for these 2 questions I don't even see a dip. Whereas for most of the relationship health questions, there's a visible dip at 7 years.
I know the raw data is there, but I don't think I'm good enough at manipulating them to find the answer. Maybe for a follow-up email on the specific question "do kids fuck up your relationship right after you have them or are we all doomed to have a worse relationship post 7 years" ? :)
This is really impressive!
One day I'd like to open your data file and do my own slicing and dicing.
I'm curious where you got the fictional character ratings - why are Monica and Chandler closer than Morticia and Gomez? Why are there no actual romantic couples on the low end? You had to dig into obscure non-cannon fanfiction for Harry and Draco, who never got involved as far as most fans know?
On a practical level, I have one suggestion. Can some of the bar charts be expressed as a stacked bar to show percentiles or/and ranges instead of just averages? Do we, for example, have lots more different people interested in others after 15 years, or do we have a small number of people who are more acutely interested in others after 15 years?
👋 I started talking with you about this at the manifest after party. To analyze the "time until marriage" data correctly, I think you can:
* Pick two of your "time in relationship" groups to compare, for instance 5-6y and 9-12y.
* From the group that spent more time in a relationship, filter out the ones that got married more years into the relationship than the other group has had years in the relationship. For instance take the 9-12y group and filter out those that got married more than 6 years into the relationship.
* Now compute the average time before marriage of each group.
You can compare each group to one base group to get a plot, or compare each group to the one before it and plot the differences