How we change what others think, feel, believe and do

# Inferring cause

Explanations > Social Research > Drawing conclusions > Inferring cause

In social research the discovery of correlation, when two variables change at the same time is not proof of cause. Here are notes

## Hume's temporal regularity

18th century philosopher David Hume described three basic conditions that are necessary for cause and effect to be inferred:

1. Cause and effect must occur close together in time
2. The cause must occur before the effect
3. The effect should never occur without the cause occurring first

This is a good first attempt, but can be challenged:

1. There may be a significant delay between cause and effect. For example a person may be bitten by a mosquito and die some time later.
2. The cause and effect may occur so closely together in time it is impossible to measure the time difference between them.
3. Effects can be caused by multiple things. For example a person can die of things other than an insect bite.

A key issue here is that there can be multiple causes which have to occur in sequences or together for effects to happen. For example a mosquito bite does not kill the person directly -- it is the disease that does this, and it may only happen if the person is already weakened.

Another problem is that just because A follows B it need not happen next time. For example when a mosquito bites another person, they may not fall ill and die. This leads to experiments to determine the detail of causality.

## Mill's induction

In the 19th century, utilitarian John Stuart Mill adapted Hume's rules as follows:

1. Cause must precede effect.
2. Cause and effect must correlate (when one changes, the other also changes in a proportionate way).
3. All other explanations of the cause-effect relationship must be eliminated.

The third rule is typical of police methods, where elimination of suspects leads inexorably to the perpetrator of a crime (nobody else here could have done it, so it must be the butler!). Mill described three methods of inferring cause:

1. The method of agreement: the effect is present when the cause is present.
2. The method of difference: the effect is absent when the cause is absent.
3. The method of concomitant variation: When 1 and 2 are demonstrated, the case for causal connection is made stronger by eliminating other possible causes.

In other words, if you take a situation and change only the cause or do not change the cause, then the effect should correspondingly happen or not happen. This requires two experiments, one with the cause present and one without. What this calls for and which makes good science is the use of controlled experiments.

## Cook and Campbell's three criteria

Cook and Campbell (1979) propose three conditions that must be met before a cause-effect relation can be inferred:

1. Covariation: Changes in the assumed cause (X) are related to changes in the assumed effect (Y). Changing X results in a predictable change in Y.
2. Temporal Precedence: The assumed cause must occur before the asssumed effect.
3. No Plausible Alternative Explanations: The assumed cause must be the only reasonable explanation for changes in the measured assumed effect.

## Experimental structure

Experiments to prove cause and effect therefore need to:

• Vary the likely cause as the independent variable and measure the likely effect as the dependent variable.
• Monitor other factors that may influence the situation, particularly those that might have some causal effect.
• Control other factors as far as possible and monitor those that cannot be held steady.
• Have a separate 'control' experiment in which the cause is not present, but all other factors remain the same.
• Repeating the experiment a number of times to ensure the results are not random and that probabilistic causality can be assessed.
• Varying other factors across multiple experiments to determine whether the presence or absence of these factors is significant.

Cause-and-effect reasoning

Cook, T.D. and Campbell, D.T. (1979). Quasi-Experimentation: Design and Analysis for Field Settings. Rand McNally, Chicago, Illinois

And the big
paperback book

### And

Massive Content — Maximum Speed

Main sections: | Disciplines | Techniques | Principles | Explanations | Theories |

Other sections: | Blog! | Quotes | Guest articles | Analysis | Books | Help |

More pages: | Contact | Caveat | About | Students | Webmasters | Awards | Guestbook | Feedback | Sitemap | Changes |