Thinking and Deciding

166 HYPOTHESIS TESTING
they are never finally “proved,” because they can always be refuted in the future (just
as Newton’s theory was partly refuted by the success of Einstein’s).
Popper’s theory set the agenda for reflection about the prescriptive theory of science.
Subsequent philosophical writers have criticized Popper’s argument, qualifying
it in important respects. “Bold conjectures,” they point out, are often wrong,
especially if the theory that predicts certain observations is unlikely to be true. Popper’s
theory does better as an account of successful theory formulation in hindsight
than as a prescription for scientific practice. The procedure he advises is not very
practical: It is a little like telling a scientist, “Take a wild guess, and be right!”
John Platt (1964) suggested that scientists should try to play the game “Twenty
Questions” 3 with nature using a more conservative strategy, which he called “strong
inference.” Rather than making a bold (unlikely) conjecture, the scientist should
divide the possible hypotheses about some phenomenon roughly in half and then try
to rule out one of the halves. When we play Twenty Questions, we usually begin
with some question like “Is it alive?” If the answer is positive, we then ask, “Is it
animate?” Each question divides the possibilities roughly in half. Likewise, in real
science, we might ask whether a disease is transmitted by an organism. If the answer
is yes, we might ask whether the organism is bacterial, and so on. Platt argues
that this method, which he called “strong inference,” is more efficient than asking
“boldly” whether the disease is caused by a spirochete (when there is no reason to
think that it is).
Another difficulty with Popper’s theory is that it assumes that hypotheses can be
falsified. Platt’s theory has this assumption as well, for it assumes that one half of
the hypotheses or the other will be eliminated by a good question. We have already
seen that this assumption may be too idealistic to serve as a prescriptive theory for
scientific practice. It is rarely, if ever, true that a scientific theory can be refuted
by any one observation or experiment. Usually it is possible — with more or less
plausibility — to find some reason why the experiment was not a good test of the
hypothesis. In some cases, like the case of rinsing the hands in a basin of water, the
experiment may truly be a poor one.
Imre Lakatos (1978) attempts to answer this criticism by arguing that most scientific
theories have a “core” of crucial claims, along with a “periphery” of claims
that can be changed as needed. A particular hypothesis involves both core and peripheral
claims. If the hypothesis is rejected by an experiment, we can reject the
peripheral claim and keep the core. For example, he says, the core of the Ptolemaic
theory of astronomy was the claim that the sun and the planets revolve around the
earth. Over the years, many peripheral claims (about “epicycles,” or orbits within
orbits) had been added, subtracted, and modified, in order to explain the fact that
the planets seemed at times to reverse direction relative to the stars. The core of the
Ptolemaic theory was protected by these modifications. When Copernicus showed
3 In this game, one person thinks of something (a person, object, or animal, for example), and the other
tries to discover what it is by asking no more than twenty questions, to which the answers given are only
yes or no. The “scientist” may guess the word at any time, but if she is wrong, she must give up for that
word.