Manipulate A or B

One way to produce such a reason is to manipulate A or B. If we vary factor A up and down, but B does not change at all, this finding provides some reason against possibility 1, since B would normally change along with A if A did cause B. Similarly, if we manipulate B up and down, but A does not vary at all, this result provides some reason against alternative 2 and for the hypothesis that that B does not cause A. Together these manipulations can reduce the live options to items 3 and 4.

Many scientific experiments work this way. When scientists first discov- ered the correlation between smoking and lung cancer, some cigarette man- ufacturers responded that lung cancer might cause the desire to smoke or there might be a third cause of both smoking and lung cancer that explains the correlation. Possibly, it was suggested, smoking relieves discomfort due to early lung cancer or due to a third factor that itself causes lung cancer. To test these hypotheses, scientists manipulated the amount of smoking by lab animals. When all other factors were held as constant as possible, but smok- ing was increased, lung cancer increased; and when smoking went down, lung cancer went down. These results would not have occurred if some third factor had caused both smoking and lung cancer but remained stable as smoking was manipulated. The findings would also have been different if incipient lung cancer caused smoking, but had remained constant as sci- entists manipulated smoking levels. Such experiments can, thus, help us rule out at least some of the options 1–4.

Direct manipulation like this is not always possible or ethically permis- sible. The data would probably be more reliable if the test subjects were human beings rather than lab animals, but that is not an ethical option. Perhaps more complicated statistical methods could produce more reliable results, but they often require large amounts and special kinds of data. Such data is, unfortunately, often unavailable.

In each of the following examples a strong correlation, either negative or posi- tive, holds between two sets of phenomena, A and B. Try to decide whether A is the cause of B, B is the cause of A, both are caused by some third factor, C, or the correlation is simply accidental.