I last suggested that one might be able to show that teens can get sexually transmitted diseases without ever touching other humans. I do not mean that one can get a STD without some type of sexual contact. I mean that a social scientist might be able to get close to presenting a convincing case for something that is not possible. And, to be sure, I am exaggerating. But I exaggerate with a purpose, and it is not to be flip.
Recall what a social scientist means by a selection effect. There is a great deal of evidence of selection in numerous kinds of risk that people have in life, including in their love lives. Some of the most important selection effects are related to family history, poverty, and education. For example, people who cohabit with a number of different people before marriage are likely to have more trouble in marriage than those who do not, and they are also more likely to have selection factors like those I just listed. The interesting question is whether or not cohabiting with a number of people actually makes it more likely people will struggle in their marriages or if the other background risks simply lead to both the serial cohabitation and problems succeeding in marriage. Or both. You would not do too poorly in life to usually bet on “both” when dealing with questions like this.
My colleague Galena Rhoades and I talk about this a great deal because it affects what social scientists covey to others about what can and cannot change in their lives. I’ll get to that later after we have some fun with imaginary data.
Okay, now to sex. I know you have been waiting patiently. Assume you are a researcher studying sexually transmitted diseases in older teenagers. In fact, you have the most amazing data set in the history of your field. You have a sample where you know a stunning number of things. On the 25,000 people in your sample, you have measured these variables and more: family history, parental relationship quality, parental divorce (or, if parents ever married), the number of romantic relationship transitions each parent has had (and at what ages for the individuals in the sample), levels of parental supervision, personal insecurities, personality tendencies to seek stimulation and impulsivity, drug use history, alcohol use, school performance, the number of sexually active kids in each individuals’ school and neighborhood, physical health, everything else about the neighborhood including crime, stress, social integration/disintegration, resources, sex-and-stress-hormones like daily (I mean to be extreme here) blood levels of androgens and estrogen, oxytocin levels and reactivity to it, vasopressin levels, parents income, current household income, parents’ education levels, religious beliefs, religiousness, beliefs about sex, type and number of friends, specific genetic markers associated with sexual behavior and pair bonding (are you getting tired of this list yet?) . . . (Okay, I’ll stop). But, please assume I could have gone on for a while because I could have.
By the way, you have a huge budget. Huge. And a massive sample size. Massive. You have followed this sample from age 12 to 20. You and your assistants check in with each of the individuals, relentlessly, and you also have access to all kinds of records. And, you’ve not lost contact with too many people. It’s a dream data set (and it would be very expensive). You also know who has had sex, when, with whom, what type of sex, and under what circumstances. (Those GPS circuits in cell phones are really rather amazing.) You also know who has had, or continues to have, an STD. Let’s also assume that the data are really good with low measurement error; however, you also must assume that sexual behavior is pretty sensitive stuff to most people, and it’s a little fuzzier than most of the other measurements you have as to error. But still, you have very good information on sexual behavior.
It’s statistical time. Your first question is about selection. How much of the STD risk do all these variables, except for sexual behavior, explain? You crunch the data and find that you can explain 75% of the differences between individuals who get and do not get STDs. Pretty amazing (though, realize it’s harder to explain as much as a prediction, using the same variables, in a new data set—technical point you can ignore if you like). Now, you do another, similar analysis but you add the sexual behavior variables into the stew. Suppose the amount of variance explained goes up 80%. Wow. That’s a lot of explanation. You are happy because you will get this published and it’s actually kind of useful information.
Please keep in mind I’m making up an example, here. These are not real findings.
Why doesn’t the amount of variance explained shoot up a lot higher when you throw in the sexual behavior variables? In this case, it only goes up a bit higher because the selection variables explain so much, there is not a lot else to explain. Your big bunch of selection variables was so good at “predicting” who would get an STD that you hardly need to know whether or not, and when and how, people had sex. In fact, that whole having sex thing looks pretty inconsequential based on the analyses.This is an extreme example but the point I am making is that when it comes to come common behaviors and risks, there truly are a lot of background variables and factors that seem to determine the outcome.
Quiz. Does the fact that you have just shown that selection explains a ton about who gets STDs mean that teens can get STDs without ever having sexual contact? Of course not. While we are here, I want to make a couple of complicated points. If you want to skip ahead, just move on to “three points to ponder” below.
First, as a risk behavior becomes something more and more people do, it, ironically, becomes harder to detect it as mattering in risk outcomes. Extreme examples are extremely useful. If 93% of people do some behavior that is risky, the fact that you have almost no one to compare the 93% to makes it pretty hard to show that the behavior matters. Plus, the 7% will be quite unusual, making what you can conclude from the comparison of limited value. Let that sink in a bit.
Second, I’m using terms here like “risk behavior” a lot here, but doing so is complicated because one of the core issues in this discussion is whether or not a behavior associated with risk is truly risky behavior. In the case of a hookup with a stranger at a party, it’s pretty obvious that the behavior is risky, no matter what else is true. That’s the point of my STD example above. But here is a different example. Serial cohabitation is associated with later difficulties in marriage and/or family stability, and, of course, there is some selection involved (on average, it’s more likely for those growing up in a single parent home, those with economic disadvantage, etc.). A person who has those and other background factors is definitely at higher risk no matter if they cohabit with a number of people or not. Yet, isn’t it pretty obvious that a person with such background factors improves their odds if they do not cohabit before marriage or they cohabit only with one person, and only after having mutual plans to marry or some other really clear clarification about commitment?
I say this because by not having those extra cohabiting relationships, the person makes it less likely they will get stuck, at least for a while, in a difficult spot. That’s because constraints to remain together are greater when cohabiting, and that makes it harder to leave. If a single parent avoids extra cohabiting relationships, they also reduce the degree to which their children are exposed to significant attachments that end. Further, there is reason to believe that such a person could reduce the possibility of child maltreatment since the odds of that occurring are greater with live-in partners. Even with selection, a person making such adjustments in their life is changing here-and-now behavior that matters.
THREE POINTS TO PONDER: CAN ACTORS ACT?
Back to you and your amazing, fantasy data set and astounding evidence for selection. (Yes, researchers have fantasies about data sets. There, I’ve let the secret out. You might think you’ve heard about all the temptations on the web, but do you know there are some data sets out there on the web that anyone can access? I know, I should not tempt you.) So, now you have your findings and you present them to the world. You can very easily sound like you are saying that the actual sexual behavior of the teens didn’t matter much in producing risk. You’d be right in a way and way, way off in another way.
First, selection matters and it is everywhere. In fact, it’s closely related to what social psychologists call the fundamental attribution error. It has been repeatedly proven that we (all of us, really) tend to over attribute the causes of other peoples’ behavior to themselves and give far too little weight to their circumstances. (We happen to generously give ourselves credit for how circumstances affected what we did when we have done something wrong or poorly. Neat trick, that.) That’s why popular sayings like “walk a mile in their shoes” have real scientific oomph behind them. Environments matter, as does selection.
Second, social scientists have a difficult time figuring out how much of a risk effect to ascribe to selection and how much to experience. We can get close, but it’s hard to ever totally answer this question. One of the common ways to figure out if experience (versus selection) actually matters is to assess a massive amount of potential selection variables and see if experience still explains anything. This is an imperfect but credible and important approach to teasing selection and experience apart. This is one of the things we do in our research on cohabitation, particularly in one large sample we are following over time. We measure an amazing number of possible selection variables to “see” how much experience matters. There are other strategies one can pursue, as well.
Third, and a SUPER BIG SOCIAL SCIENCE DILEMMA: It’s not all that clear how researchers can actually measure whether or not people have the option to engage in, or not to engage in, a behavior associated with higher risk. You can measure if risky behavior happened. You can measure what background characteristics make it more likely that behavior will happen. But, you cannot easily measure if someone had a choice.There are some ways to get close to measuring choice, but they involve very creative experiments, and these are not the kind of data most debates about selection and experience revolve around. So, while it’s hard to demonstrate that people actually make choices, and, thereby, show they have choices to make, it’s pretty easy to get evidence for selection. But how to you even keep evidence of free-will and choice in your statistical model? This means social scientists can often sound like the individual's ability to decide or choose is not part of what matters. This is a profoundly important question (unless I’m missing something really obvious to another who will eventually point it out to me. That happens plenty often and I’m fine with it!).
An emphasis on selection can be motivated by good science and it can also be motivated by compassion and social justice concerns. This is because, in most things where it comes up, selection implies that individuals have disadvantages that contribute to how things turned out. But here’s the downside of so strongly emphasizing selection, as is truly, commonly done is social science. The misleading message carried in the DNA of selection is that you—the individual—can’t really do anything to control your odds of success in life. It’s out of your hands. And this is the greatest concern Galena Rhoades and I have identified in thinking deeply about this issue: while it’s important to take into account the hand someone has been dealt in life, it’s also important to look for ways to help that person play the hand they have as well as they can.It is important not to convey that they are helpless victims and cannot do anything that affects their own outcomes.
Very strong selection-based story lines tilt the whole board toward implying our behavior is determined somewhere other than in our decider circuits. Is that what most social scientists really have in mind when they emphasizing selection? I doubt it, but that is the implication of the message.
Sure, some people have selection factors that make it harder to choose, or even have access to, a lower risk pathway in life. But do we want to accept the idea that our here-and-now behaviors, in some of the most important areas of life, are out of our control? I choose not to believe this.
For the practical implications of how what someone believes in such matters can affect his or her love-life, see my post on gambling from last year. When you are up for reading more, that is.
Black Jack Or Roulette? You Choose