Back in 2003, I reported (here, here and in more detail here) that in disparate cultures around the world, from the U.S. to Kenya and from Mexico to Vietnam, parents of daughters are more likely to get divorced. This phenomenon, discovered by the economists Gordon Dahl and Enrico Moretti, is based on a sample size over 3 million and is therefore surely no coincidence.
After seven years, psychologist Anita Kelly, writing in Psychology Today (which might want to consider changing its name to Psychology Yesterday) has penned a response. She accurately summarizes the original argument:
Dahl and Moretti have summarized attempts to explain their facts as follows: Sons may either improve the quality of married life or worsen the pain of divorce (perhaps by becoming more distraught when the father leaves). Landsburg chooses the former explanation based on the fact that parents, on average, prefer having boys over having girls.
Kelly offers an alternative explanation:
wives with daughters are less likely to stay with their husbands because they know that with a girl, they’ll never be lonely or without help. Thus, they may be less willing to tolerate any bad behaviors from their husbands (and less willing to stay married) because they don’t need their husbands as much. This idea could even explain why couples expecting a girl are less likely to marry: A woman carrying a girl anticipates that she won’t need a husband.
Well, maybe. But this still tends to founder on the fact that parents seem to prefer boys in the first place. We see this most dramatically in the fact that unmarried couples expecting a child are more likely to get married if the child is a boy. And parents of girls are quite a bit more likely to try for another child than parents of boys (this effect is seen in every culture). If girls were such a potential source of comfort, you might expect to see more parents striving for girls in the first place.
It’s true that adoptive parents show a preference for girls, but that, paradoxically, is evidence that parents in general prefer boys. When you’re adopting, your primary concern is to get a bright and healthy child. But if most parents prefer boys, then boys will tend to be put up for adoption only when there’s something seriously wrong with them, while girls will sometimes be put up for adoption simply for being girls. So adopting a girl (even if you’d really prefer a boy) can be a good strategy for avoiding the lemons.
In view of the apparent statistical preference for sons, it’s pretty hard to argue that daughters cause divorce by making divorce less painful. Hard to argue — but not impossible. In More Sex is Safer Sex, I offered the alternative theory that parents of boys are more likely to stick together because boys, more than girls, depend on wealth to attract mates, and therefore suffer more when their inheritances are diluted by expensive divorces. That would explain the divorce statistics, and would also explain why parents of boys are less likely to try for another child, without invoking a preference for boys.
Though Kelly herself accepts that daughters cause divorce, some commenters have been more skeptical, and not always unreasonably. The correlation between daughters and divorce is surely no coincidence (because of the enormous sample size). But that doesn’t necessarily make it causal. Maybe divorce somehow retroactively causes daughters, in the sense that less compatible parents are somehow more likely to produce daughters in the first place. Or maybe some third factor intervenes. For example, there’s some evidence that high status parents are more likely to produce boys and that stressed parents are more likely to produce girls. It’s not implausible that high status might tend to preserve marriages and/or that stress tends to, well, stress them.
The problem with those explanations is that, while they work in the right direction, they’re quantitatively implausible for reasons I explained in More Sex is Safer Sex. I’ll repeat my conclusion:
If stress and status can’t explain the numbers, we’re back where we began: Daughters really do cause divorce. When I reported as much in Slate, a distressing number of readers wrote to inform me that, evidence be damned, they would simply never believe that the children’s gender could be relevant to a divorce decision. My favorite of these came from a therapist in Iowa — it would
be inappropriate to mention her name, so let’s just call her “Bozo the Therapist” — who took me to task for subscribing to the “archaic notion” that children ever have anything to do with divorce. Unless she’s been practicing not in the state of Iowa but on the planet Iowa in some distant solar system, Bozo must win the prize for the least observant therapist in human history.
I’ve never understood why Fisher’s principle hasn’t countered son preference. If you have a daughter when females are relatively scarce, you should have more expected grandchildren than someone with a son. (China’s one child policy truncates your potential upside, but you at least have a higher chance of a single grandchild).
A bit like our empathy for distant strangers, there is a mis-match between our ideals, and how we actually think. Here in the fecund West, we say all children are equally valuable, possibly due to some application of the amnesiacs principle. In reality, our behaviour as a society seems to suggest that we actually prefer to have boys than girls. In some places and times, societies have been (are) quite content to openly admit this preference. I believe that the further you go back historically, the more the preference for boys would be openly acknowledged in the West. This suggest a progression towards a more equal preference, but perhaps we are not quite there yet. Maybe in a few decades, this effect will dwindle away to nothing.
Steve says: “Well, maybe. But this still tends to founder on the fact that parents seem to prefer boys in the first place.”
Having children has very complex motivations, and every parent will tell you it requires considerable sacrifice. Having a boy may require a bit more sacrifice, but the rewards are perhaps seen as greater. Having a girl could quite easily provide you with greater comfort and help. But as this is not the main reason for having children, the preference for boys could easily persist. I do not see that this suggested mechanism for the difference in divorce in any way founders on this fact.
So I guess it really was “the child’s fault” that the parents got divorced after all? Come on…
No one gets divorced without a stated reason. We could clear this up pretty quickly simply by asking the divorcees why they divorced instead of trying to stick together. If they tell you it was because they had more female children than male children, I’ll sign on to this theory.
Economics presents us with the best available framework to understand economic behavior. Non-economic behavior should not be analyzed in an economic framework.
Maybe Fathers believe that their sons need them more than their daughters do. Maybe mothers believe that also. Maybe the courts agree — assigning custody to the father 3% more often when the child is a son than a daughter.
So maybe fathers and mothers both are are willing tolerate more bad behaviour from their spouses (and/or work harder to be nice), given that the wife will usually get custody?
If there are other cultures where the father usually gets custody, we would expect to see this reverse.
(Just an idea. The difference in custody rates is similar to the difference in divorce rates, so it’s not a ridiculous idea, I don’t think).
@Ryan-the this argument isn’t based on an economics framework. It is simply a statistical outcome with all of the possible causal options laid out and analyzed.
thedifferentphil – Understood, but what I am suggesting is that there are as many causes for divorce as there are divorces, and that you can get closer to understanding divorce with one-on-one interviews than you can with a Best Linear Unbiased Estimator. Render unto Euclid that which belongs to Euclid…
@Ryan-I will put my money down on 3 million data points. What if the root cause of a divorce is not known by the couple? Or, even more likely, what if there are many overlapping causes? The most recent fight or fights or infidelities may be listed as the problem and other smaller effects not be brought up in an interview. And even if every weakness in a marriage is brought up, how do you weight the components of these to draw a conclusion?
Well, I’d suggest that you don’t actually learn anything by attempting to understand divorce in such a way. It seems almost silly to weight things like child gender, income, house size, geographical location, etc. when trying to explain divorce.
Divorce isn’t a collection of environmental factors, it’s a product of failed individual relationships. Any scientific process that obscures the nature of the relationship by drilling-down on environmental factors is, at best, misguided.
Every good statistician knows that when you include a lot of irrelevant explanatory variables your R-squared goes up. This represents not greater explanatory power, but a loss of degrees of freedom. Scientific analysis is only valid when the model has been designed correctly.
Child gender is a strong explanatory factor when it is important to the failure of the parents’ relationship, and irrelevant noise when the divorce was caused by some other reason. The most accurate thing anyone can say about child gender and divorce is that child gender matters only in those divorces in which child gender matters. There is nothing “universal” about a person’s relationship to a specific other person, and searching for universal causes for divorce is aggregating that which bears no fruit when aggregated.
“Child gender is a strong explanatory factor when it is important to the failure of the parents’ relationship, and irrelevant noise when the divorce was caused by some other reason”
No one is arguing with this point. The point is that when child gender is an issue, on average, male children lead to fewer divorces.
We may be talking past each other. :) I am inclined to say that it is a bit of a wild goose chace since, statistically, we are unable to differentiate between divorces in which child gender played a role and divorces in which child gender played no role. (Unless we interview the divorcees, in which case we’re just confirming a tautology. Either way, the statistical correlation either provides no new information, or no information at all.)
@Ryan,
I agree with thedifferentphil, a study using 3 million data points is hard to brush-off as mere correlation. This is just the identification of a phenomenon, it isn’t, as it stands now, a satisfactory explanation of divorce, and maybe not sufficient for predictive purposes.
Given that it is a *thing*, wouldn’t it be better to find out why this is the case? We might then have some way of dealing with the stressor(s) behind this and consequently we can better equip marriage counselors for couples to whom this is an unconscious factor.
What if they are the parents of a girl born on a Tuesday?
@Sam, @Ryan @thedifferentphil,
The cause of divorce, generally, is that the parties wish to get divorced, all things considered. Nobody really thinks that daughters *directly* cause divorce, only that they change one’s view of the prospect of divorce. So “daughters cause divorce” is a straw man – on its own it isn’t going to cause a happy couple to divorce.
I have no trouble believing that in the wide variety of the world, for some couples, whether their first child is a son would make enough difference to their view of the prospect of getting divorced versus remaining married, to affect the decision.
On the other hand, it needn’t be a large effect on each couple either – maybe it was a close decision – it only has to be enough to make a small difference. How much money would it take, (married tax benefits etc) to cause a 3% change in the divorce rate?
Does anybody know? If they do, we could put a number on this!
I am reasonably happy to accept that parents as a group prefer boys. It seems that from the evidence cited above parents with daughters get divorced more. Whilst it is consistent that the one causes the other, it could be some other factor, such as the caring qualities of daughters mentioned above.
What is the evidence from mixed families? Is divorce significantly reduced if you have 3 daughters and then a son compared to 4 daughters? Does a single son “innoculate” the marriage, so 1 son = 1 son plus any number of daughters?
“3 million data points is hard to brush-off as mere correlation”. Whether it can be described as “merely” or not, it is still correlation, albeit a very strong one. To establish causation I think we need to come up with a hypothesis and somehow test it.
We have 2 broadly competing hypotheses: boys improve quality of married life, or girls ease the pain of divorce. We have established that overall, couples seem to prefer boys. Steve has suggested that this strengthens the case that boys improve quality of marriage. I am not sure this is the case, since by and large people do not have children just to improve the quality of their marriage.
A few thoughts.
1. Are we comparing apples to apples? To what extent is the institution of divorce in one culture the same as the institution of divorce in another culture?
For example, Landsburg hypothesizes that sons my might benefit more than daughters if their parents avoid an expensive divorce. I wouldn’t be surprised to learn that the expense of divorce varies widely among cultures. I wouldn’t be surprised to learn that divorce often result in (or is the result of) growing disparities in status/wealth among the married partners, and that one party to a divorce ends up wealthier afterward. I wouldn’t be surprised to learn that in some societies the partner with the greater wealth/social status gains custody of the kids, and that divorce might result in increased wealth/social status for kids, but declines for one of the ex-spouses.
It’s my understanding that divorce was relatively uncommon in the US while the frontier was open. Instead, abandonment was common: A person (especially a young male person) would simply disappear and reappear in a new town in the west with a new identify. This technique provided a means for people to escape responsibilities for debts, kids, and other family members. No explanation in the change of divorce rates between then and now would be complete without an explanation of the changes in the experience of divorce, and the alternatives available.
Given the changes in divorce within a single nation, what conclusions can we draw about data aggregated from many nations?
2. That said, if I had to hazard a guess to explain the data, I think I’d go with the boy’s-enhance-the-quality-of-married-life thesis. That is, I’d guess that people tell themselves stories – dreams, if you will – about what having kids will be like. For better or worse, most stories have male protagonists. And to the extent that the story involves having kids that enhance your family’s social status, boys are often seen as the more active agent in such stories. The birth of a girl frustrates this story right at its inception.
(To be sure, I’m projecting here. I’m a boy, and I anticipated having a boy when I started having kids. I know my wife identified with having girls.)
3. Finally,
Ok, ok, I may have said something like that during a moment of frustration with my girls. But I don’t really think that. Honest.
We know that the global female-male split is more like 51-49, and this likely varies significantly by geographic location (not because geography determines gender, but rather due to natural random variation). That extra 1% translates into something like 67 million females, which is a lot more than the 3 million couples in the study.
I hope just by considering those numbers it is obvious to everyone that it doesn’t take very many of those “extra girls” to have an impact on study results like these. But if you’re the type who requires numbers to validate this, then if you assume a 50% divorce rate (which is the current divorce rate) and a 50-50 split between the female group and the male group, then when you crunch the numbers it requires less than 2000 extra couples in the “female” group to put the p-value of a chi-squared test above the 5% level of statistical significance.
2000 extra couples. That’s it. Surely none of you will suggest that 2000 extra couples is such a huge divergence of population distribution in a world of 6.6 billion people.
Now consider this article (http://www.sciencedaily.com/releases/2008/12/081211121835.htm) which explains that men with more brothers are more likely to father sons, men with more sisters are more likely to father daughters.
The point here is that what at first appears to be too unexpected to be coincidence can certainly very well be a coincidence when you consider the actual numbers involved and the fact that random variation is not only RANDOM, but it is also VARIATION. In other words, we can expect to see differences like these.
Absent of a coherent theory, they are just random demographic variations. Perhaps some of us are more comfortable with the variations than others. Remember, even in a perfectly random variable, we can expect to see outlandish outliers ~5% of the time, which is not so outlandish after all.
It’s certainly no reason to go blaming female children for divorce. That’s going too far, and most importantly, it isn’t substantiated by random variations that are not actually uncommon.
*Correction to the above: 2000 extra observations in the ( female children + divorced ) group will push the chi-square test into “significant” territory, not 2000 extra in all couples with female children. Actual number is 1687.5, if you assume the other 3 groups stay constant at 750,000 each. This is a back-of-the-envelope calculation, but it should suffice for the purposes of an illustration, anyway.
The point is that I don’t believe the variation is incompatible with random chance.
@Ryan,
Obviously, no child would consciously try to break their parents up and no parent would say “Well, Mommy and Daddy aren’t living together anymore because we really wanted you to be a boy”. Asking parents why they got divorced doesn’t factor unconscious motives for divorce.
True, correlation doesn’t imply causation, but a strong correlation should make you ask a few questions. I’ve never been married and I have no children so I’m not an expert, but I would imagine that raising daughters and raising sons is a much different process. Maybe going to your son’s little league games creates more family bonding than going to your daughter’s ballet recital (yes, I realize that girls play sports too and boys can do ballet, but on the whole, the problem is probably with gender roles to being with). Maybe buying Barbie dolls makes men want to cheat and, thus, cause more divorce. Granted, these are all fairly dumb reasons, but it’s probably going to be a number of small reasons, rather than one large reason.
Out of curiosity, where do childless couples fall in the data?
Am I missing something here? Are daughters more loyal to their mothers than sons?
In any event, just because a variable is statistically significant doesn’t mean it’s deserving of a profound or satisfying explanation; spurious relationships simply exist, which is something of a letdown for economists who believe data is the fabric of the universe.
Ryan – are you serious – of course 2000 extra couples isn’t much in a population of 6 billion, but it is quite a lot in a *sample* of 3 million. Enough, in fact, to be statistically significant. Are you actually questioning the entire concept of statistical significance? Or just hoping that by saying enough numbers your posts will sound impressive enough for people to think you have a point?
Ryan, that’s like saying “Precise scales can measure that gold is heavier than silver. But precise scales can measure differences of .0000001%. Therefore, scales cannot be used to conclude that there is a substantial difference between their weights.”
2000 couples is not a measure of how small the difference is. It’s a measure of how small the difference could be, and yet still be detectable.
I agree completely with Ryan (first comment). Economics, the study of how to allocate scarce resources, should only be applied to situations where you have scarce resources that need allocating.
Of course, unless you don’t value anything, you are faced with scarcity so it really applies to anything on God’s green earth requiring a decision. Seeing as you can only be married to one person at a time in this country, I’d consider the slot of marriage partner to be quite scarce.
@ Ryan,
Gender birth ratios are beside the point when you’re performing a between groups test.
@Pete,
I wouldn’t consider this economics, more statistics/probability.
Stephen has filed this under empirical puzzles.
But in terms of economics, what we’re seeing here is something that’s happening on the margin, not systematically to each and every son-less father.
Paper found here (2008 version): http://www.econ.berkeley.edu/~moretti/sons.pdf
John – 2,000 is 0.07% of the population of 3,000,000. Large enough to affect the outcome of a chi-squared test, but otherwise trivial. My point is that what at first seems like a statistical impossibility is actually not so outlandish when you consider normal random variations. When you consider the fact that gender breakdowns AND the divorce rate are subject to random variation, you realize that variation like this isn’t statistically unlikely enough to justify implied causation. If I told you “we witnessed 2000 statistical abnormalities in a population of 3 million,” you’d say big deal, and you’d be right. But for some reason if I make the **EQUIVALENT** statement that in a population of 3 million, we saw significant variation in the female + divorced group at the 5% level of confidence, everyone says, “There MUST be causation!!!” That’s my point.
Jeffrey – No, what I’m saying is that normal random variation accounts for the observed result; and by way of example, it one requires about 1700 couples in a population of 3,000,000 to reach such conclusions.
Sam – Gender birth ratios are incredibly important when your sample selection is based on the gender of people being born.
When we say some data analysis shows that something may cause something else, what do we really mean? Obviously in this case, I don’t think anyone is saying that if you have a girl, you will always get divorced if you’re married. That’s obviusly not what we mean by causation here. We mean that the data show that (at least) there are some people out there who have gotten divorced who would not have had they not had girls and there are people who would have gotten divorced had they had girls and not boys. Am I wrong?
Also…in my comment above, you can accept the second to last sentence and still think of examples why the sex per se of the child may not be the driving force of these divorces.
“I told you “we witnessed 2000 statistical abnormalities in a population of 3 million,” you’d say big deal”
Actually, what I’d say is, “oh, is that statistically significant?” because I know full well that my intuition is not very good at deciding whether or not a given amount of variation is down to random chance or evidence of an actual difference. That’s exactly why people invented things like tests of statistical significance in the first place.
The whole point of having statistics as a discipline is so that we don’t have to make subjective decisions about whether getting 400 heads in 1000 tosses of a coin is good evidence that the coin is biased – we have objective, rigorous test for doing that for us.
“what I’m saying is that normal random variation accounts for the observed result; and by way of example, it one requires about 1700 couples in a population of 3,000,000 to reach such conclusions.”
To paraphrase a quote (which I think is from Steven Landsburg): Telling a statistician that he needs to account for random variation is like telling a chemist that he needs to clean his test-tubes – that’s exactly what statistics is for!
John,
The statistician’s ability to choose the appropriate tools is vital to determining the validity of the results. Just because someone uses a statistical model to explain something doesn’t mean the model has been used correctly. You can’t just hide behind the WORD “statistics,” you actually have to apply some statistical knowledge to the design of the model in the first place.
What I mean by this is that 1,700 is not a significant level of binary variation in a population of 3,000,000 at the 5% confidence level, even though it is signficant variation in a chi-squared test of exactly the same data, at the 5% confidence level.
Once you realize that as a statistician, then the question you have to ask is, “Which of these two models actually reflects the reality in question? Which model is valid? Are either of them valid?”
As per my original comment, I would suggest that no such economic models are valid for psychological phenomenon like relationship failure, for THEORETICAL reasons.
To highlight this fact, I provided one simple example that illustrates that two STATISTICAL tests against the same set of data, testing roughly the same thing, come to remarkably different results.
So, to sum up, model design matters more than the fact that someone came up with an equation to predict divorce. Divorce isn’t math. You can’t just add things up. And it’s really easy to trick ourselves into concluding false positives if we don’t spend a little time at the model design stage of the analysis.
One hypothetical example where you’d see having a girl associated with a higher rate of divorce would be where some genetic factor not only made it more likely for you to produce girls but also more likely for you to have some personality disposition that would make you more likely to get divorced. After all, we know genetics influences both your personality and your likelihood to produce a girl or boy. The fact that these individuals would be more likely to get divorced AND have girls would of course not mean that the girls were driving the divorces. Is this far fetched? Maybe…but it’s possible.
Makes a lot of sense for sons to save marriages. In the East, failure to provide sons was a reason to take a second or third or twentieth wife or concubine. This translated into divorce in the modern era.
I guess a dad might feel there’s some point to sticking around so his son learns how to be a man but there is no similar incentive for daughters.
Daughter mother pairs may find it more easy to gain a new male in the household- who might prey upon the daughter as she reaches puberty.
In any case there are two optimal strategies for daughters viz be a total slut and breed till you drop, for which stress and hardship is good because it yields earlier menses and promiscuity. The other, adopted by middle class parents, is investment in education and grooming and ‘virtue’ (the hijab for example) and payment of dowry (very big amongst Hindus) so as to get a shot at hypergamy.
If a couple aren’t getting along really well, or if they aren’t going to end up in the middle class, it makes sense to break up the family so that the daughter turns into a slut with a string of kids by different dads so that, hopefully, at least one hits the genetic jackpot.
How depressing! Yet if any part of our behaviour is genetically conditioned it must be stuff to do with sex and bonding- the most intimate and ‘personal’ parts of our lives. Jeez! I can see why, as Pascal said ‘there will always be more monks than Reason.’
I don’t read all of the responses… but I find it interesting, that people who response to the article are all MEN.
Do you have any explanation to this phenomenon?
To me the main at least *potential* flaw here is the assumption that the odds for any given couple of having a boy are girl is a coin flip/random.
We know this isn’t true. There are men that carry more male sperm than female sperm and thus more likely to produce a male child and vice versa. And for second children we know that longer it takes to produce a second the more likely it is to be male. And in war torn countries in general the overall odds shift to the entire population. There are plenty of families who produce many sons or many daughters because the man’s sperm is more populated with a certain sex.
If it were truly a coin flip, if it were truly 50/50, then I think there would be something to this.
But what we don’t know is what else is going on with these people who are more likely to produce girls/boys. In any given the odds are different and not 50/50.