[MUSIC] Hi. Now I'm going to talk about cause and effect in observational studies in social science. I want to begin by reviewing the challenges to making claims about cause and effect in observational studies that we've talked about in a previous lecture. If we have some factor X, that we think or hypothesis effects Y, when we're studying it, we also have to keep in mind that there may be other factors that are also affecting Y. Again, we live embedded in a social world where we're interconnected. There's all sorts of influences on any kind of outcome that we might be interested in. Now if one of those outside influences is also affecting X or related to X that we hypothesize is the cause, then the relationship that we observe between X and Y, if we don't account for or control for that omitted variable, may be spurious. In an observational study, we might claim that X causes Y, when in fact there's a spurious relationship that's the result of the effective sum o,itted variable which we talked about previously on both X and Y. Another challenge that we have is that there are some times or some cases where changes in Y may actually lead back to changes in X, that we have reverse causality. With these difficulties in mind, I'd like to talk about when it is we actually can make a claim about cause and effect from an observational study. Again, it is important to keep in mind that it is very difficult to prove conclusively that we have cause and effect from observational studies alone. That said, there are several criteria we focus on in looking at a relationship in terms of assessing whether it might be evidence of cause and effect. One thing that we pay particular attention to in assessing a claim of cause and effect is the time ordering of the variables. We generally think it's more plausible to believe in cause and effect if the variable that we think of as the cause is fixed or measured at some earlier point in time then the effect. For example, if we see a relationship between education, and then health in later life, it's really plausible to believe that at least part of that relationship is a causal relationship. Because education is fixed early in life after certain point, typically when people are in their 20s, education doesn't change much and then it may affect health later. Another issue that we pay attention to is whether or not there's a proposed mechanism that links the cause and the effect. That is, can they go beyond a simple block box statement where we say there is a relationship here and it must be cause and effect, to actually laying out a pathway that connects cause and effect. In the case of education and health, it's plausible to suggest a pathway. For example, education changes people's behaviors. By being more educated, they may pay more attention to instructions from doctors and they may read information that changes their health related behaviors, which in turn leads to changes in their health. Finally, the design of the analysis or the study itself if done properly can help rule out the possibility that there are other factors that are actually related to the proposed cause that are also associated with the outcome of interest. Therefore, producing a spurious relationship between what appears to be cause and effect. There's several basic approaches that people take in observational studies to try to make a case that a relationship is cause and effect. One, the classic approach is a statistical analysis that includes control variables, controls, for factors that we think might be influencing both the variable that we think of as the cause and the variable that we have chosen to measure our proposed effect. This form of analysis helps us rule out the possibility that an observed relationship is a spurious relationship. More recently, especially in economics, people have been looking at opportunities to study natural or quasi-experiments in which events, disasters, for example, that occur in which the timing of the event and the people who experienced the event were actually unrelated to the preexisting characteristics of the people. By taking advantage of these situations, people can look at the association between the natural event, and then outcomes. Then because the occurrence of the event is unrelated to the characteristics of the study's subjects, if there is an association, it very likely is cause and effect. One typical example of this is that people who are trying to study the relationship of stress to health, a very complex relationship. They sometimes look at the effect of natural disasters on health where they can possibly claim that natural disasters are a source of stress for people that has nothing to do with preexisting characteristics of the people that are studying. We have natural disaster take place, people experience stress. We can look at their health outcomes and then compare it with people who did not experience that disaster but are otherwise similar to see if there is an effect of stress on health. Another approach is to make use of what are called instrumental variables. Economists are particularly enamored of this approach. It's essentially a statistical approach making use of regression-based techniques to try to Isolate variation in the proposed cause, the X variable that is independent or unrelated to other factors. And then see whether that variation is leading to changes in the outcome in the hypothesized way. Finally, we have matching approaches which involve taking subjects and trying to match them on various characteristics. Race, age, any number of other characteristics, and then comparing them according to whether or not they experience the outcome of interest for evidence about a causal relationship. Each of these is quite complex and so in the remaining modules, we're going to return to each of these and talk about them in more detail.
