The strength of good causal arguments, and the power of such arguments to help us learn about the world, show the value of observation, especially when we collect observations carefully, evaluating their significance with the techniques described in this chapter.

But this stress on observation makes it easy to fall into an error about causation—not a fatal mistake when we're evaluating someone else's causal arguments, but certainly a misleading picture of human knowledge, and often enough an obstacle to the investigation of nature. We think of ourselves as a passive audience recording whatever regularities occur in front of us.

In many cases this assumption doesn't do any harm. Take one of the first links established between cancer and an environmental cause: In the late nineteenth century, London doctors observed a steep increase in cases of cancer of the scrotum. So many of the patients were chimney sweeps as to suggest a connection between the men's cancer and the soot they were constantly exposed to. Hardly a leap, when anyone seeing all the same patients would have had the same idea.

But things are rarely so straightforward, especially when we look for causation in populations. The time, trouble, and expense of running a responsible study mean that we can't go fishing for causal links without a clear, plausible, and justified hypothesis. Nature won't give an answer until we ask a question.

Where do these hypotheses come from? Suppose you want to study intelligence, as measured by IQ tests, to discover its causes. Where do you begin? Not with people's heights, because you have good reason not to expect intelligence to depend on height. What about parents' relative ages? Could it be that when two people are closer together in age, their children may score better on IQ tests than when the parents' ages are further apart? Maybe. A large and careful study will either confirm or refute the possibility. But where did the question come from? Someone has to think it up.

The plain fact is that effects don't come with possible causes written on them. Suppose fewer women in Guatemala have osteoporosis than U.S. women do. You start looking at the two groups' dietary calcium. But what if that hypothesis does not pan out? A wide range of cultural differences between the two countries gives you more possible causes than one researcher could hope to track down. You don't throw up your hands and call the matter inscrutable—such differences are bound to come from somewhere. But where do you find your questions?

Background information helps, and if you were a medical researcher you'd possess more and better background information. You would not set up studies of the two groups' different clothing or language, or of the forms of government they live under. You might look more generally at diet (i.e., more broadly than just with an eye to calcium), at childbearing patterns in the two countries, at the minerals that find their way into drinking water, or at the physical exertion typical of the cultures. If you noticed anything unusual you might formulate a hypothesis and then construct an effect-to-cause study.

So the questions are not stabs in the dark. But neither are they the results of just looking at the world. The questions that start off a causal investigation are made by the investigators, not found nestling among the evidence.

In one way this does not matter; in two ways it does. The contingency of the original questions—the fact that someone has to think them up; the fact that different people will think up different ones—does not diminish the truth or reliability of our causal conclusions. Sample size, attention to alternative causes, and so on, ensure that we can trust a statistically significant result. From the point of view of truth, it doesn't matter if someone dreamed the hypothesis, as long as evidence supports it.

However, you have to remain alert to the origins of causal hypotheses when you are engaged in looking for a cause. If you expect possible causes to shout at you from a mass of observations, you will be disappointed, and probably unsuccessful at finding a cause.

Also, the invention of hypotheses permits one kind of criticism that, while not denying the truth of the conclusion, does question the bias behind it. Take the number of studies linking pregnant women's smoking, drinking, and use of drugs to low birth weight and other medical problems in their children. Without denying the reliability of those studies, you might ask why there have been so few studies of men. What happens when a man has been smoking, drinking alcohol or coffee, or using drugs immediately before his child's conception? There may be more of a connection than we know; but the role of women has been studied far more extensively than that of men. You might well accuse investigators of bias in asking one sort of question and not the other.