Characteristics of a Good Hypothesis

From what we've covered up to now, you know that a hypothesis is much more than the “educated guess” that people often say it is, but we haven't looked into its properties in any detail.

2.I.1 Significance and Generality

A good hypothesis tackles a scientifically meaningful issue; an unsolved problem, a discrepancy in the data, an avenue of investigation made possible by some tech­nological advance. Of the infinite number of conceivable hypotheses, only a tiny minority are significant. A good hypothesis has generality, “reach” in David Deutsch's lexicon⁴⁰; it is not narrow and ad hoc, meaning that it tries for a compre­hensive perspective that rises above particular cases. For example, a hypothesis about Snowball's frenetic welcoming behavior would be scientifically insignifi­cant. A more significant hypothesis might try to account for the hyperactivity syn­drome that affects lonely dogs who suffer from sensory deprivation. A still more significant one might be that young mammals, including young humans, develop behavioral anomalies because of the lack of social stimulation. What constitutes scientific importance is a judgment call that can be tricky, especially since the significance of an investigation is not always immediately apparent. Few people would have anticipated that the biology of a bacterium living in the sulfurous hot springs of Yellowstone National Park could be significant, and yet an enzyme iso­lated from such bacteria is the basis of the most widely used technique in molec­ular biology,⁴¹ the polymerase chain reaction (PCR) that is used to analyze DNA rapidly and efficiently. The example of PCR illustrates that the significance of a hypothesis may become apparent only long after it has been advanced and tested.

2.1.2 Riskiness

A good hypothesis makes novel non-obvious predictions (i.e., risky ones).

The classic historical example of a non-obvious prediction came from Einstein's General Theory of Relativity, which predicted that the path of light from a distant star would curve significantly as it passed by a massive object such as another star. The Newtonian theories of gravity and space anticipated at most a minor bending of starlight, and Einstein's prediction seemed to come out of nowhere—but it was testable. It was also risky because so much was riding on it; if starlight did not bend a lot, Einstein's whole theory would have been in trouble. The British astrophysicist, Arthur Eddington, sailed to the island of Principe off the west coast of Africa to observe the light from distant stars that would become visible near the sun during a total eclipse. Normally, the glare of sunlight hides those stars, but they can be seen during the eclipse when the moon blocks the sun. By comparing the apparent position of the stars during the eclipse when their light passed close by the sun to their position at night, when the sun was on the other side of the earth, Eddington found that starlight did indeed follow a curved path as it passed by the Sun. His report caused a sensation and per­suaded physicists that Newton's model was incomplete⁴²; “just wrong” according to Richard Feynman.⁴³

We can contrast non- obvious predictions with obvious ones. For instance, Bob, the mechanic, would probably consider that, when the car wouldn't start and the radio and the lights didn't work, an obvious prediction would be that the electric power windows would not work either. Obvious predictions are not wrong; how­ever, they're not expected to yield breakthrough information.

2.1.3 Simplicity

A good hypothesis is the simplest explanation that encompasses the facts as they are currently understood. This is the rule (or law) of parsimony, also called Occam’s Razor, because it was stated by the Bishop of Occam (c. 1287-1347); the rule has been restated many times over the centuries.⁴⁴ The virtue of simplicity is ubiqui­tous and, according to the physicist and mathematician, Henri Poincare, ines­capable: “People will scoff at the rule of parsimony, but still draw a smooth curve through their data points... the smooth curve is an expression of the rule of parsimony. It is the simplest explanation for the data.”⁴⁵ Less obviously perhaps, parsimony also expresses bias, as we'll see in Chapter 11.

Here's an example of how parsimony can affect an investigation: when Bob found that the engine didn't start and the lights and radio didn't work, he rea­soned parsimoniously that the battery was probably dead; it could explain all of his observations. However, it is conceivable that the battery and the radio were both dead and the light switch broken; or that the battery was not dead but the switches to the ignition, radio, and lights all malfunctioned; or that the starter motor was defective, the radio switch broken, and the light bulbs burnt out; and so on and on, even for this mundane case. The alternatives, though conceivable, are complex and, because several parts would have to break down at the same time, they are less likely and less parsimonious than a battery failure. You do not want to be paying hourly labor rates to car mechanics who do not reason parsi­moniously when repairing your car.

Although you'd think it would be easy to grasp, the rule of parsimony is occa­sionally misunderstood or misrepresented. The rule does not say that the world is simple or that the simplest explanation is the most likely to be correct. It only says that, when you're trying to explain a phenomenon, you should first look at the simplest explanation for it, where “simplest” means the one with the fewest un­known variables, intermediate steps, or assumptions.

Probably most scientists nowadays prefer simplicity for practical or aesthetic, rather than philosophical, reasons. There are an infinite number of potential explanations for any phenomenon, and, if you have a fertile imagination, you can conjure up alternatives at every step and quickly be overwhelmed with elaborate possibilities. The rule of parsimony says you should formulate hypotheses in an orderly way, going from simple to complex, to minimize fruitless effort. Starting with the simplest explanation that can explain all of the relevant data, you only reluctantly resort to more complicated explanations when new results leave you no choice.

Historically, people have not always interpreted parsimony in this way. A de­vout conviction that the universe was governed by discoverable mathematical laws ordained by an all-powerful God convinced scientists such as Newton that the simplest explanation for a phenomenon would, in fact, be the True one. They believed that “God was a mathematician”⁴⁶ and thought that mathematical simplicity was a sure sign of the eternal Truth of the physical laws that they developed.

In contrast, Karl Popper likes simple hypotheses because they are the most open to falsification through testing. According to Popper, the more possibilities that a hypothesis forbids, the more powerful it is; and the simpler it is, the more it forbids.⁴⁷ Popper's example of the white swans was a model of simplicity: it forbids swans from having any color but white; therefore any non-white swan would have falsified it. Newton's laws of planetary motion were simple and strong, Ptolemy's system was complicated and weak.

There is an important caveat to the concept of parsimony. A hypothesis must explain all of the pertinent information. The key word is “pertinent,” and what information is pertinent depends on how you define your scientific question. Scientists are free to establish the limits of the problems that they are working on. They are not obliged to account for every scrap of data that they stumble across. Bob would be free to ignore a broken turn-signal lever, for instance, as a relation­ship between it and the engine's failure to start would be hard to imagine. It is also perfectly acceptable for scientists to break a big problem into manageable bits. If necessary they can set aside complexities that they can't deal with as long as that they are totally transparent about what they are doing. The scientific community can review their reasoning and criticize it if need be. However, selectively elimi­nating data without being open about it is not allowed.

In neurophysiology, the Nobel Prize-winning physiologist Bernard Katz and his colleague, Paul Fatt, made a fundamental discovery concerning the opera­tion of chemical synapses where a presynaptic cell meets a postsynaptic cell.⁴⁸ Initially, Fatt and Katz knew that a neurotransmitter was released from the presynaptic cell in tiny spritzes that occurred in distinct temporal patterns: ei­ther steady though irregular streams or sudden high-frequency bursts. Taken together, the two patterns formed a very confusing picture, so Fatt and Katz announced that they would focus on the steady though irregular streams and temporarily ignore the bursts. With this simplification, they were able to concen­trate on individual spritzes and eventually found that each one took place when one tiny presynaptic sac, a vesicle filled with neurotransmitter, suddenly dumped its entire contents into the synaptic space. And, sure enough, the bursts of release turned out to represent an entirely distinct process. To this day, the discoveries of Fatt and Katz remain the backbone of our knowledge of how chemical synapses work.

Indeed, if the boundaries of a problem are blurred you may face insurmount­able barriers to solving it. For example, the ancient Greek natural philosophers had great difficulty in thinking about motion, in part because their idea of “mo­tion” was so expansive: besides physical translocation of an object from one place to another, motion for them included the growth of trees, the aging of men, and the rusting of iron.⁴⁹ Since their notion of motion was for us some­thing more like “change” in a very broad sense, it is no wonder they couldn't make headway in understanding it. By the time of Newton, physicists had dis­pensed with the extraneous meanings and could think about motion as we normally do.

2.I.4 Specificity

A good hypothesis is specific, or restricted, meaning that it is exclusionary; it not only explains a phenomenon, but it also rules out other explanations, which makes it precise and informative. Suppose a biochemist hypothesizes that a particular enzyme, say a mutant form of protein kinase C, causes a tumor to grow, and she tests the hypothesis by altering the amount of the mutant en­zyme while measuring tumor growth. If the amount of enzyme affected tumor growth as her hypothesis predicted, she could reasonably conclude that the mutant enzyme was a factor. A stronger hypothesis would be that, not only is mutant protein kinase C activity enzyme a factor, but that other similar protein kinases are not. A still stronger hypothesis would be that mutant protein kinase C is a necessary and sufficient condition for growth of the tumor: that you have to have the mutant for the tumor to grow and that having the mutant guaran­tees that it will grow. Specific and powerful hypotheses make it possible to do rigorous tests that exclude numerous alternative hypotheses. The conclusions that you can draw from narrow hypotheses are not necessarily wrong but are wishy-washy when compared to the ones you can get from testing restricted hypotheses.

It is worth noting that, as a practical matter, it is often best to begin an in­vestigation with a coarser, less powerful hypothesis so as not to miss making a discovery because your focus is too narrow. In the tumor biology example, you might start with the general hypothesis that a protein kinase of some sort is in­volved in tumor growth and do a nonspecific test, perhaps with a treatment that would affect all such enzymes. If you find no evidence that a protein kinase of any kind was important, your specific candidate, mutant protein kinase C, almost certainly would not be. Hence, you might want to start with a “dirty” drug (i.e., a nonselective one) that will block all kinases. Initially, casting a wide net with a nonselective test can be helpful because it can effectively eliminate many hypoth­eses at once.

2.1.5 Constraint

Finally, according to David Deutsch,⁵⁰ a good hypothesis is “hard to vary.” All of the details of a good explanation are vital; change one of them and the ex­planation no longer explains what it was supposed to. With a bad explanation, however, you can change many details and still be able to concoct an explana­tion that will work. Deutsch's favorite illustration involves the accounts of the annual earthly seasons given by ancient mythologies and by modern astronomy. In the myths, the actions of a god or gods cause the transition from summer to fall. You can freely substitute one god for another, alter his or her circumstances, motivations, powers, or objectives and generate an explanation for the seasonal progression that is as good as the one you started with. In contrast, change any detail of the modern theory, say the degree of tilt of the earth's axis of rotation with respect to its orbital plane around the sun, and the astronomical explana­tion is ruined. It is a good explanation that is hard to vary, and the myths are bad ones. Of course, the modern astronomer has a lot more data to work with than the Greek mythologists did, and indeed the “hard to vary” criterion implies that a good explanation is more tightly constrained by data than a poor one.

2.I.6 Falsifiability in Practice

If a proposition is not falsifiable, then it is not a hypothesis at all. Still, it is hard to judge a hypothesis that you can only test in principle to be as good as one that you can test in practice. There may be both theoretical and practical barriers in the way of actually testing a hypothesis. If your hypothetical explanation of Snowball's hyperactive behavior is that he is deprived of canine society during the day, then it is certainly falsifiable; but if you will be evicted if you get another dog, then it may well remain testable only in principle. Like other criteria for good hypotheses, this one is not an absolute requirement. Provocative, though presently untestable, hypotheses inhabit fields from psychology to theoretical physics. A few will lead to major new developments, so we shouldn't write off apparently untestable hypotheses altogether. Still, it's probably best to aim for achievable results, and that means you want practically testable hypotheses.

Up to now, I've been discussing the hypothesis as if it were a public intellectual product, an object with specific properties that we could evaluate. Yet, scientists don't always make their hypotheses overt. Their hypotheses exist in an implied state and we, readers and science-consumers, have to work out what they are. This tendency toward leaving hypotheses unstated makes scientific communi­cation more difficult than it needs to be. Implicit hypotheses are pervasive, and they are significant for deeper reasons than communication alone.

2.