2.3 The Hypothetico-Deductive Method

When the young Isaac Newton was toying with prisms in 1666 he noticed that they dispersed a white beam of sunlight into all the colors of the rainbow. If the colored light from a part of this rainbow was passed through a second prism it did not result in a new rainbow, only in further dispersion of the colored beam. The experiment is described schematically in Figure 2.2. Newton concluded that the prism did not add the rainbow colors to the light. Maybe the white light of a sunbeam already contained all the colors and all the prism did was to spread them out over the wall opposite his window? As a consequence, he thought, re-combining the colors should produce white light [4]. He set up an experiment to do just this and, as most readers are probably aware, his thought was confirmed.

Figure 2.2 Schematic representation of Newton's experimentum crucis. The second prism does not produce rainbow colors, as the first one does. Newton concluded that prisms do not add color to white light, but the colors are constituents of white light.

This is an example of working out the consequence of an hypothesis by logical deduction and testing it by an experiment (although Newton himself would not agree that this is the full story of what he did [4]). An hypothesis can be seen as a preliminary attempt to explain something – a theoretical statement without much observational support. We shall now discuss a method based on this procedure, where scientific inquiry begins with theory instead of observation. According to it, a scientist faced with a seemingly inexplicable phenomenon should begin by proposing speculative theories, or hypotheses, about its cause. These can then be tested rigorously against observation, for example by controlled experiments, to see if the consequences that follow from them correspond with reality. Working out consequences from premises is called deduction, speculative theories are called hypotheses and, accordingly, the method is called hypothetico-deductive. My description of the method closely follows how Chalmers describes falsificationism [3], which is a variant of the method proposed by Karl Popper. Most working scientists seem to apply some of the method's central elements, not least since they often discuss hypotheses and generally agree that hypotheses must be testable in order to be scientifically useful. In the words of Popper, hypotheses must be falsifiable.

If an hypothesis is formulated in a way that makes it theoretically impossible to disprove, it is not falsifiable. For example, a recruitment firm that throws half of the applications for a job into the wastebasket with the motivation “we don't want to recruit unlucky people” bases its decision on a non-falsifiable hypothesis. It is non-falsifiable because people whose applications end up in the wastebasket are, by definition, unlucky. Why would their applications otherwise be thrown away? It is a bit like saying that all sides of an equilateral triangle have equal length. If they did not it would not be an equilateral triangle – the statement is logically impossible to contradict. Non-falsifiable hypotheses are statements that are true whatever properties the world may have. Such hypotheses tell us nothing about the world and are, therefore, of no use in science.

According to the hypothetico-deductivist, research begins with an hypothesis and proceeds by testing it against observation. If our observations contradict the hypothesis we discard it and look for a new one. If the hypothesis is supported by our observations we keep it, but continue testing it under various conditions. The following classical example from Hempel [5] gives a flavor of how it could work.

Example 2.1: Semmelweis and the solution to childbed fever Ignaz Semmelweis was a Hungarian physician who worked at Vienna General Hospital in the mid-1800s. He noted that a relatively large portion of the women who delivered their babies at the first maternity division contracted a serious disease called childbed fever. About 7–11% died from it. The second maternity division at the same hospital accommodated about as many women as the first, but only 2–3% of them died from childbed fever. The theory that diseases are caused by microorganisms would not be widely accepted until the late 1800s, so Semmelweis considered various other explanations. He quickly dismissed some of them as they were contradicted by what he already knew. One of these was the then common idea that diseases were caused by “epidemic influences”, attributed to “atmospheric-cosmic-telluric changes”, affecting whole districts. If this were so, how could the first division be plagued by the disease while the second was relatively spared? He also noted that women who gave birth in the street had a much lower risk of contracting childbed fever, although they were later admitted to the first division. These were often women who lived far from the hospital and were not able to arrive in time after going into labor. A street birth just outside the hospital should not decrease the risk for epidemic influences.

Could the problem be due to overcrowding? Semmelweis saw that it was not. The second division tended to be more crowded, partly because the patients desperately wanted to avoid the notorious first division. Instead, he turned to psychological factors: The priest, bearing the last sacrament to dying women, had to pass five wards to reach the sick-room of the first division. He was accompanied by an assistant ringing a bell. Presumably, this display would be terrifying to the other patients. Fear could explain the fever, because in the second division the priest could access the sick-room directly, without passing other patients. To test his idea Semmelweis persuaded the priest to take a roundabout route and enter the sick-room unobserved and silently, without ringing the bell. Unfortunately, this had no effect on the mortality in the first division.

Semmelweis also noticed another difference. In the first division the women were delivered lying on their backs, while in the second division they lay on their sides during birth. He decided to try the lateral position in the first division but, again, without result.

In 1847 he got a decisive clue to the solution. During an autopsy, a colleague of his received a puncture wound in the finger from a scalpel. Later, the colleague showed the same symptoms as the women that had contracted childbed fever and, eventually, he died. As previously mentioned, the role of microorganisms was not yet recognized but Semmelweis started to suspect that some kind of “cadavic matter” caused the disease. In that case, the physicians and their students would carry infectious material to the wards directly from dissections in the autopsy room. The difference in mortality between the divisions could then be explained by the patients in the second division being attended by to midwives, whose training did not include dissections. To test his hypothesis he required all medical students to wash their hands in a solution of chlorinated lime, which he assumed would destroy the infectious material chemically, before examining the patients. As a result, in 1848 the mortality in childbed fever decreased to 1.27% in the first division. This was fully comparable with the 1.33% in the second division that year. The hypothesis was further supported by the lower mortality among women giving street birth: mothers who arrived with babies in their arms were rarely examined after admission and, thereby, escaped infection.

We see how Semmelweis proceeded by first trying to understand the cause of the disease and then comparing his ideas to the information at hand, or to information obtained by specific experiments. Unsuccessful ideas were discarded and new ones proposed until the problem was solved. We may also note that, although his idea about “cadavic matter” improved the understanding of the disease, it was still far from the more modern, microbiological theory. Although it may be tempting for modern readers to exchange the term with “bacteria”, Semmelweis did not understand that living organisms caused the disease. It is again clear that observational support does not prove an hypothesis to be true. The hypothetico-deductivist must abandon the claim that theories can be proved by observation, at least in a strictly logical sense. Still, hypotheses that resist falsification in a wide variety of tests tend to become established theories with time, as the support for them grows ever stronger. Even though they are not proven true, we may still acknowledge that they are the best theories available, since they are supported by the most evidence.

Finally, hypotheses should not only be falsifiable, they should even be as falsifiable as possible. It is difficult to contradict a vague statement. For science to progress there cannot be any doubt whether an observation supports an hypothesis or not. A good hypothesis is falsifiable just because it makes precise assertions about the world. It becomes more falsifiable the more wide-ranging the claims it makes, since the number of observations that could potentially contradict it increases. For instance, Newton's theory for the movement of the planets around the sun is more falsifiable than Kepler's older theory. Newton's theory consists of his three laws of motion and the law of gravity, stating that all pairs of objects in the universe attract each other with a force that varies inversely with the square of their separation. Whereas Kepler's laws of planetary motion only apply to the planets, Newton's theory is more general: it describes planetary motion as well as a large number of phenomena, including falling bodies, the motion of pendulums, and the relationship of the tides to the positions of the sun and moon. This means that there are more opportunities to make observations that contradict Newton's theory. In other words, the set of potential falsifiers of Kepler's theory is a subset of the potential falsifiers of Newton's theory, which then becomes more falsifiable.