Is Marriage Good for You?
"When I was in my 20s and 30s, I knew I was supposed to get married..." Bella DePaulo, Tedx Talks.
On Saturday, I stumbled upon a Wall Street Journal article claiming that marriage makes you healthier and happier. My liberal instincts were immediately aroused. The Journal has a conservative slant, and praise for marriage sounds, well, conservative. But the article wasn't an opinion piece. It was describing a new study, one carried out by a team of researchers representing some pretty liberal institutions (Harvard and Stanford). I decided to have a look. Could it really be that married people are healthier and happier than single ones? That's a provocative claim.
As it turns out, what I stumbled upon is the latest salvo in a debate among experts who study marriage. I reached out to leading figures on both sides of the debate and they graciously responded. In the end, I took a side. And, along the way, I discovered a disturbing example of how statistics can be distorted by ideology. (Fair warning: This is an especially snarky newsletter.)
The debate
There's more than one debate about the benefits of marriage. The one I'll be focusing on boils down to this:
—Some experts hold that marriage tends to be a good thing, because married people are generally healthier and happier than unmarried people. There may be exceptions, according to this view, but marriage tends to be beneficial overall.
—Other experts hold that marriage is not inherently beneficial. Married people may flourish, but for many of them it's not marriage per se that makes the difference. Meanwhile, single people flourish too, and those who never marry may even end up slightly better off.
As I've framed it here, the debate isn't about whether marriage or singlehood is better. Rather, it's about whether marriage tends to be inherently beneficial or not.
You might think of this debate in probabilistic terms. If you're single and pondering marriage, some experts would say that your life will probably (if not necessarily) improve if you get married and stay that way. Other experts would say that what determines your future health and happiness depends on a lot of things – your friendships, your career, your lifestyle, etc. – and the mere act of marriage doesn't increase the probability of a better life later on.
The new study
This month, Global Epidemiology published a study by Dr. Ying Chen and colleagues that drew on a longitudinal database representing over 100,000 registered female nurses. The purpose of the study was to identify relationships between marital status and both physical as well as psychological well-being.
The main analyses focused on two sets of comparisons:
1. Nurses who got married for the first time were compared to nurses who never married. (First-Marriage group vs. Never-Married group.)
2. Nurses who were already in their first marriage and stayed married were compared to nurses who got divorced or separated. (Remained-Married group vs. Divorced/Separated group.)
These seem like sensible comparisons, but there's a catch. Marital status was defined for the time period 1989 through 1993. The outcomes – physical and psychological health – were recorded in 2015 and 2017.
Does that sound confusing? Bear with me for a moment, because this will be essential to evaluating the study.
The First-Marriage group consisted of 3,272 nurses who got married for the first time between 1989 and 1993.
The Never-Married group consisted of 8,558 nurses who remained single up through 1993.
The Remained-Married group consisted of 69,491 nurses who were in their first marriage and remained married between 1989 and 1993.
The Divorced/Separated group had been in their first marriage in 1989 but got divorced between 1989 and 1993.
In 2015 and 2017, health data for all groups was obtained from medical records, while psychological well-being was determined by self-report data provided by the nurses at those times.
In a moment I'll discuss why marital status was defined during a brief window of time nearly three decades prior to data collection on outcomes. Here are a few of the key findings that emerged:
—The First-Marriage group had less cardiovascular disease, greater psychological well-being, and less psychological distress than the Never-Married group.
—The Remained-Married group had less psychosocial distress and less cardiovascular disease than the Divorced/Separated group.
There were other findings too, but you get the idea. Getting married, and staying that way, seemed to promote better physical and mental health. The researchers don't mince words about this:
"[Our study] suggests that entering marriage is associated with better subsequent health and wellbeing outcomes...Likewise, this study also adds to the evidence that incident marital dissolution is related to multiple adverse health outcomes."
In fact, this is one of the most deeply flawed studies I've encountered in recent months. Ideological biases created fatal flaws in the way the data were reported and interpreted.
I realize how cranky and hostile that sounds. Let me take a moment to describe those fatal flaws, then I'll explain why I blame them on ideology.
Group membership was hopelessly imprecise
As I mentioned, marital status was identified on the basis of 1989–1993 data, nearly three decades before the outcomes were measured.
The obvious problem with this approach is that by 2015, the marital status of many of the nurses would've changed. Some in the First-Marriage and Remained-Married group would've gotten divorced or separated. Some in the Never-Married group would've gotten married. Some in the Divorced-Separated group would've gotten remarried. Presumably in some cases multiple marriages/divorces/remarriages occurred.
Why did the researchers take this approach? The most charitable interpretation is that as epidemiologists, they wanted to learn something about the long-term impact of actions taken or not taken at a much earlier point in time. They do hint at this, and in their Methods section, they describe classification into groups as "exposure assessment"– meaning that from 1989 through 1993, the nurses were "exposed" to marriage or divorce. This would be fine if the study were about, say, the long-term effects of exposure to asbestos during childhood but never again. Clearly this isn't that kind of study, because marital status changes over time.
The researchers would probably respond that even though group membership was imprecise, the group differences they found are meaningful. Whatever else happened in nurses' lives, something about getting and/or staying married between 1989 and 1993 had benefits that were observable over two decades later. As we'll see, this doesn't save the study.
The findings are misleadingly reported
Here's a typical paragraph from the Results section:
"Among participants who were initially married, those who became divorced/separated reported substantially lower levels of social integration (β = -0.15, 95% CI = -0.19, -0.11), greater depression (RR = 1.23, 95% CI = 1.10, 1.37) and loneliness (β = 0.11, 95% CI = 0.08, 0.15), as compared to those remaining married."
This is the kind of language used throughout the article, even in the Discussion section. It's shockingly misleading.
You actually can't figure out how each group fared unless you consult a separate online supplement provided by the researchers. There you discover a huge discrepancy between the data and the way they presented it. I'll focus on one example.
Depression was scored on a 30-point scale. The mean for the nurses who got divorced or separated was 6.66. The mean for the nurses who stayed married was 5.97. That's a small difference. And, what the researchers don't tell you is that the scale they used, the CESD-10, has a cut-off value of 10 points. Any score above 10 is considered depressed. Clinically speaking, neither group is even close to being depressed, on average.
This is starting to seem ludicrous. We have two groups of nurses who complete the CESD-10 in 2017. They answer questions about how they were feeling during the prior week. Both groups are doing fine, on average, but one group scores about two-thirds of a point higher than the other group on a 30-point scale. The researchers then call that "substantially greater depression".
The only thing that's beginning to look substantial here is ideological bias. For the moment I just want to emphasize that I didn't cherrypick this example. The findings for loneliness and social integration, and for other group comparisons, are all like the ones for depression: The group differences are exceedingly small.
I don't know why, on one particular week in 2015 or 2017, tiny differences emerged between groups who were classified on the basis of their marital status nearly three decades earlier. But I will say this: It's inevitable that when tens of thousands of people are compared, group differences will emerge. In this particular study, these effects are essentially meaningless, for three reasons: The effects were tiny, the groups were not distinct, and the researchers failed to find group differences in many of their other comparisons.
Ideological bias
There's a fine line between ideological beliefs and ideological bias. The researchers' prior work points to an ideological commitment to marriage as a superior form of existence. There's nothing wrong with that. People are entitled to their beliefs. But in this particular study, ideological beliefs function as biases that distort the presentation of the data.
The first example of this distortion is one I've already discussed. Throughout the article, from abstract to discussion, the researchers describe the Never-Married and Divorced/Separated groups as having "greater psychological distress", "greater depression", etc. In fact, the researchers only looked at aggregate data (i.e., means) and, in the aggregate, none of the groups were distressed or depressed. Doubtless some individual nurses could be described that way, but there's no way of knowing whether there were more of these women in one group than in the other. In fact, the means are so similar for these and other variables that it would be mathematically unremarkable, using the example of depression, if the Remain-Married group turned out to include more women who exceeded the 10-point cut-off for depression than the Divorced-Separated group did. (By analogy, median annual income in New Mexico is slightly higher than in Alabama, but New Mexico also has a higher percentage of families below the federal poverty line. Would you say then that New Mexico is a richer state?)
This leads me to the second example, and the one I consider most damming. The researchers had access to lots of data they could've easily incorporated but elected not to. For instance, they could've looked at how many nurses in each group exceeded that 10-point cut-off for depression. If they wanted to claim that women who get divorced or separated show greater depression later on, they should've tallied up the number of nurses in each group that the CESD-10 labeled as depressed. They had the data, but they didn't use it. I suspect they didn't because the results wouldn't have supported their contentions, though I have no way of proving that.
Here's the most important data that wasn't included: The nurses' marital status was actually recorded every four years, not only in 1989 and 1993. Surprise! The researchers could've improved their groupings by making use of the more recent data. For example, a nurse who remained single from 1989 through 1993 was classified as Never-Married, but if it just so happens that she got married in 1995 and stayed married, she would've been a married person for 20 years by the time her outcome data was recorded in 2015. The researchers should've made note of the 1997 data, and switched this nurse to the Remain-Married group.
Why did the researchers rely on such out-of-date marital status data when more recent information was available? They gave two reasons:
"First, we sought to rule out confounding by prior marital history by examining first-time marriage.... Second, in this study we sought to understand the effects of the decision to become married."
The first reason is irrelevant; the second one doesn't make sense. A nurse who gets married for the first time in 1995 is entering a first-time marriage and can be classified accordingly. Instead, the researchers left this nurse in the Never-Married group. She did make a decision to become married. She just happened to do so after 1993.
If goal of the study is to understand the impact of marriage, changes in marital status over time do pose challenges for statistical analysis, but these challenges are manageable. At the risk of stereotyping, I assume that the researchers, being from Harvard and Stanford, would surely know what to do. Even simple approaches would work just fine. For example, you could compare a group of nurses who were in their first marriage by 1993 and stayed married through 2017, versus a group who'd never married by 2017, versus a group who got divorced/separated somewhere in between. With a database of over 100,000 nurses to begin with, the sizes of each group would be more than adequate. (If you're a stats person, you'll recognize missed opportunities here for difference-within-differences analyses.)
My assumption is that the researchers gathered data on marital status from 1989 and 1993 and stopped there, as opposed to including 1997, or 1997 and 2001, or even more recent data, because the most out-of-date data happened to yield the findings of greatest consistency with their ideological biases.
I'm making a strong allegation here – one that I have no way to prove directly – but notice that all of the concerns I've raised could've been easily addressed, because the researchers had access to all of the data. Detailed reportage would not have been necessary. Journal editors routinely allow brief, nonspecific summaries of analyses that fall outside the main analytic approach. Thus, the researchers could've simply included sentences like this in their article:
"Separate analyses showed that more participants in the Divorced/Separated group exceeded the 10-point cutoff for depression than participants in the Remained-Married group did."
"Separate analyses showed that our results were not affected by group definitions that incorporated marital status data from the time period 1989 to 1997" (or 1989 to 2001, or whatever).
You don't see those sentences because, I suspect, that's not what the researchers found. Instead, they cherry-picked findings that matched their ideological biases.
An author responds
I reached out to the corresponding author, Dr. Tyler VanderWeele, a professor of epidemiology at Harvard's School of Public Health and, among other things, director of the university's Human Flourishing Program. To be honest, I didn't share the full extent of my concerns with him. Rather, I focused on the lack of attention to committed relationships. My assumption is that in some cases, committed relationships may be essentially the same, both psychologically and materially, as marriages. By the same token, when people leave committed relationships, what can happen is tantamount to a divorce. My main question to Dr. VanderWeele was how the data might've been affected by lack of attention to these relationships, but I invited him, in effect, to comment on the broader issue of limitations in how marital status groups were defined. Here's the essential part of his reply:
"If one is looking at marriage (rather than just a committed relationship...) then someone who enters and then leaves a committed relationship without marrying is still properly classified as "never married." It could be interesting to look at committed relationships instead, but that would be a different study (and we didn't have data on that in this particular dataset). This was rather a study on marriage."
This is a cautious response. I do agree that marriages and committed relationships can differ (there's data on this), and that good science requires studying both rather than studying one and generalizing to the other. Unfortunately, Dr. VanderWeele didn't respond to the heart of my concern but rather, you might say, doubled down via his comment that "someone who enters and then leaves a committed relationship without marrying is still properly classified as 'not married'".
The only thing that's "proper" about this is that committed relationships and marriages are, legally speaking, different things. This comment sidesteps two important issues:
First, what impacts people most about a close relationship may not be its legal status but rather their attitudes and feelings about it. If you've been romantically involved with someone for seven years and then break up, it may not matter whether you were married or just partnered – the impact of parting may be the same. Studies focusing exclusively on marriage and divorce may be misleading without data on peoples' other committed relationships, and without including people whose only experience is with committed relationships other than marriage. (Pew data and other sources suggest that at any given time in recent years, more American adults are in committed relationships than married.)
Second, as I keep saying, it's not "proper" to classify someone into any marital status category based on what happened nearly three decades prior to obtaining data on the effects of that category.
Other evidence
A last-ditch effort to saving this study might be to say that the findings are consistent with other evidence that marriage is good for you.
This approach won't work, because some studies fail to demonstrate overall benefits of marriage. I'm not an expert, so I will just say that the literature is mixed, and so are experts' interpretations of it. Thus, the current study gains no credibility by replicating established effects, because those effects aren't established.
There's a deeper issue too. Studies that do show benefits of marriage are often inherently flawed. This has been pointed out by Dr. Bella DePaulo, an Academic Affiliate in the Department of Brain and Behavioral Sciences at UC Santa Barbara and easily the most prominent, influential advocate for living a single life.
Dr. DePaulo has noted that in studies showing that married people fare better than unmarried ones, the comparison being made is unfair, methodologically speaking, because some people in the unmarried group are divorced. Why is that problematic? Because it stacks the deck. Researchers who conclude that marriage is beneficial are including, in the unmarried group, people who tried marriage but found that it didn't work for them. As a result, the benefits of marriage get overstated. Sure, the married group turned out well, but that's largely because the researchers focused on the subset of people whose marriages lasted, not on the more inclusive group of people who at some point chose to get married.
If researchers want to claim that getting marriage is a good thing, they'd be better off comparing married people to those who never married. (This is what the Harvard/Stanford researchers tried to do – they even claimed that their study addresses Dr. DePaulo's criticism – but, as I've noted, the classification of groups is hopelessly flawed.)
Broadly speaking, being divorced is not the same as having never married. As Dr. DePaulo put it in an email to me:
".,,people who get divorced or widowed typically have worse outcomes – at least at first. I don’t think that’s about becoming single per se, but becoming single after having been married. Lifelong single people often do better than divorced or widowed people."
Conclusion
Marriage can be a good thing. So can a committed relationship. So can the single life. And, so can a life that ends up having been divided across these experiences.
In my case, I was married, it didn't work out, and I'm happily single now.
In Dr. DePaulo's case, she describes herself as Single at Heart, a term she coined to describe what, for her and for others, is their most "authentic, meaningful, and fulfilling life."
In my opinion, people can be many ways at heart, sometimes in the space of a single lifetime.
As for the data, I've come to the conclusion that marriage is not inherently beneficial. I'm not saying it's a bad thing. I'm simply saying that marriage per se, on the whole, is different things for different people, and when you look at aggregate data (i.e., entire groups of people) there's no clear basis for saying that married people are generally better off. It depends, it depends, it depends.
At some point, we reach the limits of what statistics can tell us. One of the questions posed to nurses was simply "How happy were you (last week)?" Apart from the fact that the question was decades removed from their initial decision about marriage, it comes nowhere near the granularity of how it feels to be married, committed, single, etc. You know what I mean. When you're single you enjoy a sense of freedom at times, and at times you feel lonely. When you're married or in a committed relationship you enjoy the other person at times – and at times you feel lonely. You feel a lot of other things too.
What we ought to do is to reject data that tells us how to live, when that data turns out to be ideology in disguise. If you eat walnuts, do it because you like them and nutritional data suggests they're healthy, not because ancient Greeks noticed that they look like brains and theorized that they must therefore be good for your brain. I appreciate Dr. DePaulo's tireless critiques of the ideological bias she calls "singlism" – i.e., prejudice and discrimination against people who choose to live single. As she put it in her email to me:
"[N]o study can account for people’s wishes. So, even if it were generally true that single people who marry get healthier or happier (it’s not), that still would not mean it would be true for everyone."
Although I believe committed partnerships and marriages are great for some people (especially when they work at them), I see no reason to downgrade the value of other modes of living.
Thanks for reading!