Thinking and Deciding

162 HYPOTHESIS TESTING
willing to make the correct inference from counterevidence) and that we challenge
our hypotheses actively rather than waiting passively for couterevidence to come
along (Horton, 1967; Popper, 1962).
Before we can test hypotheses, we must have them in mind. Hypothesis formulation
is the search for possibilities to test. A good hypothesis answers the original
question, can be tested by gathering additional evidence, and is consistent with the
evidence already at hand. A scientist must try to think, simultaneously, of good hypotheses
and good experiments to test them. Sometimes we collect evidence in the
hope that it will suggest a hypothesis — as Sherlock Holmes did so well — rather
than to test a hypothesis that we have in mind.
This chapter is concerned with the normative, descriptive, and prescriptive theory
of hypothesis testing. The theory of probability — coupled with a simple theory of
decision making — turns out to provide a good normative model. Departures from
this model suggest certain prescriptive heuristics: considering alternative hypotheses
and looking ahead by asking what we will do with the information we seek when we
get it.
Hypotheses in science
An example from medicine
A classic case of hypothesis testing in science, which illustrates how vital this process
can be, was the work of Ignaz Philipp Semmelweis on the causes of childbed
fever, or puerperal fever (Hempel, 1966), a disease that is now known to bear a large
responsibility for the high mortality rate for new mothers that was a fact of life, in
Europe and the United States, until the late nineteenth century. Semmelweis’s work
was done in the 1840s, before the acceptance of what we now call the “germ theory”
of disease and the use of antisepsis. The Vienna General Hospital had experienced
a very high rate of deaths from puerperal fever for several years (over 10% in some
years) among women who had just given birth in the First Maternity Division. The
prevalence of the disease in the Second Maternity Division was much lower (around
2%).
Semmelweis, a physician in the First Division, set out to discover the cause. He
excluded a number of hypotheses on the basis of evidence already at hand. For example,
he reasoned that the epidemic could not be the result of overcrowding, because
the First Division was less crowded than the Second. (Women tried to avoid it.)
Most other possible factors, such as diet and general care, were identical in the two
divisions. One difference was that in the First Division, deliveries were done with
the mothers on their backs during delivery, and in the Second, on their sides. Semmelweis
could not imagine how this could matter, but he ordered that all deliveries
be done with mothers on their sides — to no avail. Another hypothesis was that the
disease was psychologically transmitted by the priest, when he passed through the
wards of the First Division (with a distinctive bell to indicate his presence) in order