In a recent Washington Post article—originally headlined "The Best Way to End Violence Against Women? Stop Taking Lovers and Get Married"—writers Brad Wilcox and Robin Wilson argue that getting married protects a woman from violence and rape. They support this claim using data from recent studies of violence against women. But the data don't show that marriage makes women safer. And even if they did, that would be of little use to individual women thinking about marriage.
Marriage may be many wonderful things, but the article seems to suggest that women should get married for their own protection. Aside from insulting both women who do want to get married (should they just try harder?) and women who don't want to get married (I assume they have their reasons) the data don't support this claim. Violence within marriages is known to be significantly under-reported, and it's not possible to determine the causal relationship between marriage and safety using the data we have.
The best available data on the relationship between violence and marriage comes from the National Crime Victimization Survey. Wilcox and Wilson reproduce this graph from a November 2012 report:
/media/img/posts/2014/06/Screen_Shot_2014_06_17_at_3.56.04_PM/original.png)
All kinds of violence against women fell a lot in the 1990s, along with most other violent crime. But single mothers are still disproportionately victims. Single women without children are victimized a lot less than single mothers, and married women a lot less than anyone else. So when Wilcox and Wilson write "married women are noticeably safer" they're telling the truth, right?
Not necessarily. First we have to ask where the data comes from and how it was collected. The National Crime Victimization Survey (NCVS) is conducted by interviewers who enter randomly selected households and ask people about their experiences. This method has the advantage of counting crimes that were never reported to the police, but only if the person being interviewed is comfortable talking about them at home. A recent report of of the National Research Council discusses this problem with the NCVS in detail:
Because most rapes and sexual assaults are committed by individuals whom the victim knows, respondents may be reluctant to disclose their victimization during an interview that takes place in the home within earshot of other family members. The training for NCVS interviewers does not stress privacy, and even if adequate training were provided, the nature of the survey—a general-purpose criminal victimization survey—means that interviewers very rarely get positive responses on questions of rape and sexual assault.
The report concludes that the NCVS is "likely undercounting incidences of rape and sexual assault," and notes that other surveys consistently report higher numbers. Since we'd expect women to be more likely to talk about domestic violence if their partner isn't around, we should expect that the NCVS under-reports violence against married women specifically. This means that the difference between the married and unmarried rates of violence is at least in part an illusion, an artifact of the survey method. We just don't know how much of this difference is real.
But let's suppose that these numbers are accurate, or that the general trend of less violence against married women still holds. Wilcox and Wilson still have a second big problem: the confusion between correlation and cause.
They are careful with their language, writing "married women are noticeably safer" which is both perfectly true and completely misleading. We may read "married women are safer" but we probably understand it more like "married women are safer because they are married." Not only are our minds always looking for causes, but the rest of the article reinforces this notion. It puts an idea like this in your head:
/media/img/posts/2014/06/married_implies_safe-1/original.png)
And really, what's so unreasonable about that? A good marriage is something that should feel safe—and as the authors point out, it's not unreasonable to imagine that a good man will protect his wife from harm.
Yet all we know is that there is a correlation between marriage and a lower (reported) rate of violence against women, and any correlation goes both ways. So let's try it this way instead: "safer women are married." If that sounds weird, it's probably because you were thinking of "married women are safer" as a causal statement.
"Safer women are married" says the exact same thing as far as the data is concerned, and it's reasonable to imagine that the causality actually goes this way: women who have never been a victim of violence might be much more likely to get married and less likely to get divorced. Or there could be a third factor that influences both safety and marriage. A woman who is already in a safe situation for some other reason—perhaps she is financially secure—might be more willing and able to get married. In fact, income does seem to be part of the story: Marriage rates are lower and falling faster among lower income families.
/media/img/posts/2014/06/alternate_marriage_causality/original.png)
Wilcox and Wilson know that these alternate explanations are possible. They write,
For women, part of the story is about what social scientists call a “selection effect,” namely, women in healthy, safe relationships are more likely to select into marriage, and women in unhealthy, unsafe relationships often lack the power to demand marriage or the desire to marry. Of course, women in high conflict marriages are more likely to select into divorce.
Yet the rest of the article is written like they know for sure that marriage causes safety; that's the headline. They repeat the same correlation/causation language fudge several times, writing "Women are also safer in married homes" and "What’s going on here? Why are women safer when married?" They could have written "why are women married when safer?" because it's just as true, but they didn't.
If we're going to read cause into the correlations, we might as well look at the chart above and conclude that not having children makes women safer. It's certainly strongly correlated. But the article never proposes not having kids. (My guess is that this correlation actually means that women who have children find it harder to leave an abusive relationship.)
No matter how hard we try, comparing rates of violence between married and unmarried women just can't tell us how for sure a woman's choice affects her safety. That's because this sort of survey can't tell us which of the above diagrams shows the true causal structure. Most likely it's really some combination of all the above causes, and others too. This is a question that can only be answered by further research, both quantitative (data from different survey designs, like this one) and qualitative (the stories of individual women).
So what can we actually learn from this data? Still quite a lot! Knowing such correlations can help to focus society's resources. Single women, especially single mothers, probably really are at higher risk for violence. They're not necessarily at higher risk because they are single. But if you have only a limited budget for preventative programs it makes sense to concentrate on higher risk women, guided by the data that tells you where they might be.
But even if the data were unambiguous, even if we knew for sure that marriage protects women, Wilcox and Wilson's argument makes no sense as life advice.
While women in general may benefit from knowledge of the correlation between marriage and violence, an individual woman can do little or nothing with this information. These statistics hide enormous diversity in people and situations; they only say what happens on average. The difference in risk that this data shows is irrelevant compared to differences in your specific circumstance: marrying an abusive boyfriend will not make you safer. Even if your prospective husband is safe, minor risk reduction is a terrible consolation for marrying someone you don't actually like.
Anyone who makes major life decisions without considering their personal situation has got bigger problems than bad data analysis. But this seems to be be exactly what Wilcox and Wilson are suggesting.