It is too often thought that multiple regression overcomes the problem of isolating causes and allows us to weigh their importance to outcomes. Even careful writers often confuse the following ideas:

1. A (some “variable,” e.g., IQ) correlates with B (some other variable, e.g., income)

2. A “predicts” B

3. A “explains the variance” in B

4. A “explains” B

5. A “causes” B

We can handle 4 and 5 together. We can say 4, “A explains B” only if we can say “A causes B.” But, first, correlations do not establish causes. Causes are “mechanisms” which produce outcomes. We can have a correlation where there is no conceivable mechanism, e.g., the price of eggs in a Beijing market and the price of Microsoft on the New York Stock Exchange. Second, there are always many causes of any outcome. In order to make a fire, we need, in addition to some combustible material, a source of heat and oxygen. Absent any of these, no fire. So which is more important? We get a fire only if the right combination is present. (It takes a good deal more heat to ignite a vinyl fabric than it does to ignite cotton.) If we pick out a source of heat as “the cause,” that is because we assume the presence of oxygen and the combustible material. We forget about the oxygen and say, the spark “caused” the fire. (Weber called this “adequate causation,” the difference in the existing state which brought about the effect.) This is both convenient and unsurprising.