23. Experimental Science & The Mill's Methods
Summary
Building on the previous episode's introduction to scientific method, this video examines experimental techniques for identifying causes, developed by the 19th-century British philosopher John Stuart Mill. Recall that experimental science is distinct from historical science: it deals with observable, present events, since experimentation is necessarily observable and present. When it comes to finding causes, experimentation has proven extraordinarily powerful, advancing science further in the last 500 years than in all preceding millennia.
The Inductive Experimental Method
The inductivist experimental method forms and tests hypotheses to discover causal connections. Scientists try to determine both what causes an effect and what does not. Mill, an ethicist, logician, and political philosopher, outlined several inductive procedures that let researchers home in on probable causes and rule out non-causal factors. These became known as Mill's Methods, and there are five: the method of agreement, the method of difference, the joint method, the method of concomitant variation, and the method of residues.
Several methods come in positive and negative versions. Positive tests, at best, yield only probable results: they aim to show what the cause is, but some unnoticed factor might always be at work. Negative tests provide virtually certain results: they show for sure what the cause is not.
Key Concepts
An effect is anything that begins to exist as the result of some cause; a cause is what produces an effect. Although the definitions are circular, it helps to think of cause as that which produces and effect as that which is produced. Something can be both a cause and an effect, just not at the same time and in the same way (the law of non-contradiction), because these are relative terms.
An antecedent factor is an event or thing that precedes the effect under consideration; it should not be confused with the cause but rather tested as a possible cause. A concomitant factor happens at the same time as the cause but is not itself the cause, and is easily mistaken for it. A necessary condition must hold for an effect to occur; a sufficient condition is one that guarantees the effect.
Method of Agreement
Also called common thread reasoning. The positive version seeks a sufficient condition: the single antecedent factor common to all situations where the effect occurs is probably the cause. For example, across three chemistry trials with factors (A, B, C), (D, C, F), and (F, G, C), the reaction occurs each time, and the only common factor is C, so C is probably the cause, though only highly probable.
The negative version finds what the cause is not: no antecedent factor is the cause in whose absence the effect still occurs. If trials with (A, B, C), (A, B, C), and just (B, C) all produce the effect, then A can be eliminated.
Method of Difference
Also called relevant difference reasoning. The positive version seeks a necessary condition: in otherwise identical situations, the antecedent factor unique to the situation where the effect occurs is probably the cause. If (A, B, C, D) produces an effect but (B, C, D) does not, then A is likely the cause. A useful mnemonic: in agreement the results agree (the effect occurs in each instance), while in difference the effect differs between instances.
The negative version states that no antecedent factor can be the cause in whose presence the effect fails to occur. If (A, B, C, D) yields an effect but (A, E, F, G) does not, A can be ruled out. One caveat: A might be the cause only in combination with another factor, making it necessary but not sufficient. This is exactly why a single method should not be used in isolation.
Joint Method
The third method simply combines the methods of agreement and difference. It is a strategy of cross-checking: when results from different methods point to the same conclusion, the probability of accuracy increases.
Method of Concomitant Variation
If the occurrence of a phenomenon varies with a particular element, that element is probably causally related to the phenomenon. When a possible cause and effect vary together (the effect grows as the factor increases and shrinks as it decreases), you may have found the cause. Di Donato cautions, however, that correlation is not causation. In a drug trial, increasing drug B might track with a patient's remission, yet the real cause could be an allergic reaction that triggers a cancer-fighting hormone, or many other drugs might produce the same benefit. We must be careful with our conclusions.
Method of Residues
Also known as the process of elimination: the antecedent factor that remains after all others have been eliminated as probable causes is probably the cause. This is essentially the same as inference to the best explanation.
The next episode turns to historical science and common mistakes in causal reasoning.