02.03.2013 Views

Thinking and Deciding

Thinking and Deciding

Thinking and Deciding

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

CORRELATION, CAUSE, AND CONTINGENCY 185<br />

<strong>and</strong> then calculate the correlation by using the resulting numbers. (This yields a<br />

“φ (phi) coefficient.”) Another reasonable way to measure the association between<br />

two dichotomous variables is to subtract conditional probabilities. For example, we<br />

can subtract the conditional probability of blue eyes given blond hair from the conditional<br />

probability of blue eyes given nonblond hair. If the first conditional probability<br />

is higher than the other, knowing that a person has blond hair makes it more probable<br />

than otherwise that the person has blue eyes. If the two conditional probabilities are<br />

equal, then whether a person has blond hair tells us nothing about whether the person<br />

has blue eyes, <strong>and</strong> it makes sense to say that the two variables are unrelated. (We<br />

shall discuss more examples of this measure later in this chapter.)<br />

Correlations are sometimes confused, in scientific analysis <strong>and</strong> everyday reasoning,<br />

with causal relationships. It is important to realize that they are not the same.<br />

Establishing a correlation does not establish causation, though it often provides evidence<br />

about causation, because causation is one reason that correlations can exist.<br />

(To establish causation, other reasons must be ruled out.) For example, the use of<br />

marijuana <strong>and</strong> the use of heroin may be correlated, but it is not necessarily true that<br />

one causes the other. There may be some third factor, such as exposure to drug dealers<br />

or rebellious attitudes, that is a more likely cause of both. If this third factor<br />

explained the correlation, then stopping the use of marijuana would not necessarily<br />

reduce the use of heroin. When we ask whether use of marijuana causes use<br />

of heroin, we are asking what would happen to heroin use if marijuana use were<br />

changed. A correlation between marijuana use <strong>and</strong> heroin use provides evidence for<br />

some causal relationship, but it does not establish one with certainty. Often, a more<br />

conclusive way to find out about causal relationships is to do an experiment. For<br />

example, correlations between diets with high amounts of saturated fat <strong>and</strong> heart disease<br />

provide evidence that the former causes the latter. Experiments in which some<br />

people have been induced to eat less fat would show that the relationship is indeed a<br />

causal one, if heart attacks become less frequent in these people (<strong>and</strong> not in others).<br />

When we speak of correlations in a situation in which causal relationships are<br />

assumed, we can use the term contingency. A contingent relationship exists if one<br />

variable causally affects another. If smoking causes lung cancer, then lung cancer<br />

is, to some extent, contingent upon smoking. Our perception of contingency is an<br />

important determinant of our success at achieving our goals. Does Japan respond<br />

favorably when the United States threatens it with trade restrictions? If I drink three<br />

cups of coffee after dinner, will I be able to get to sleep tonight? If I try listening to<br />

people more, will they like me better? Of course, getting the answers to these kinds<br />

of questions does not always involve trial <strong>and</strong> error. There are ways of underst<strong>and</strong>ing<br />

such things that allow us to make reasonable guesses without any experimenting at<br />

all. We do not experiment, for example, in order to find out whether strong acids are<br />

really dangerous to drink. (Those who do, at any rate, are probably not around to<br />

argue with me.) Some of our learning, however, does involve observation.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!