Discussion Notes: Causality

As scientists we are interested in the concept of cause. We want to know if a particular treatment caused a particular outcome. A well controlled, randomized, laboratory experiment may allow us to determine cause. In quasi-experimental and correlational research there is so much happening that is outside of our control that there is a real issue of how to determine the cause of an effect. How do we determine cause? Let's look at how some philosophers have thought about the concept of causality.

We humans easily attribute cause to events that exhibit contiguity (nearness or contact), an unbroken series, or temporal precedence. Our stubborn and strong predisposition to infer cause is a product of biological evolution. It may or may not be adaptive for our species. Some of the causes that we see will reflect passive relationships, they will not be the cause.

David Hume (1711-1776) was a positivist. He wanted to infer causality based on observed high correlations between events. Today, every introductory psychology student is taught to reject the notion that a correlation between events, even a very high correlation, is proof of causation. We are teaching them to reject Hume's positivist position.

Here are Hume's three conditions for inferring cause:

Billiard ball examples.

Hume denied the conceptual status of unobservable phenomena. All you can do is observe relationships. He would argue that "Try as you will, you can find no relationship between one billiard ball's motion and the motion of another ball it strikes other than coincidence in space and time." If you see cause in that relationship, then it is a problem in psychology and not logic.

Necessary and Sufficient Causes

For the essentialist philosophers cause is the constellation of variables that, when taken together, are both necessary and sufficient for the cause to occur. They reject as causes those factors which are known to bring about effects sometimes, but not always.

For example, what is the causal relationship between experiencing a severe automobile accident and the development of PTSD? Is such an experience a sufficient condition to produce PTSD, it is a necessary condition, or is it both a necessary and sufficient condition?  

Necessary and sufficient can be defined by a set of 2 x 2 tables comparing whether or not the cause was present and whether or not the effect was obtained. Consider 1,000 observations of both cause and effect.  The tables below indicate the presence of any observations in each cell. (Note that that the following tables are hypothetical, they do not represent actual data for the automobile accident scenario.)

A cause is necessary and sufficient if the effect is never present when the cause is absent and if the effect is always present when the cause is present.

A cause is necessary, but not sufficient, if the effect never occurs when the cause is not present, but sometimes does not occur when the effect is present.

A cause is sufficient, but not necessary, if the effect is always present when the cause is present, but the effect sometimes occurs when the cause is not present.

If you find any occurrences outside of the main diagonal of the necessary and sufficient table, then you have not found the cause. You need to look at a more detailed level of analysis in order to determine cause. Essentials typically search for the ultimate cause, typically at the micromediational level. They are reductionists.

Using this strict definition of necessary and sufficient what is the causal relationship between experiencing a severe automobile accident and the development of PTSD? Here is what the causal table would look like.

The Time Interval Between Cause and Effect

Think of the time interval between the cause and the effect. If the time interval is long, then there can be other things occurring between the "cause" and the effect. In fact, they require that the two variables be simultaneously related.

For example, suppose you have a wire running from a battery through a variable resistor and back to the battery. You measure the amount of current flowing through the wire by a voltmeter. When you turn the knob on the variable resistor the current measured by voltmeter simultaneously changes. The change in current did not follow the change in resistance, it occurred simultaneously with it.

Can you be an essentialist given the particular problem you are studying?

Social sciences can rarely specify the critical interval between the cause A and its effect B. The inoculation theory of persuasion in social psychology is an example of a delayed treatment effect. In medicine a person is "inoculated" with a weakened form of the disease. With time the body builds up immunities against the disease so that at a future date exposure to the full blown disease will not cause the inoculated individual to be infected with the disease. In the inoculation theory of persuasion you first present weak arguments against a cultural truism (e.g., why it is bad to brush your teeth every day). Then later later present the full blown persuasive communication against the cultural truism and measure attitude change (resistance to the disease/communication). If you present the full blown communication too soon (within a few hours), the person will not be resistant to that communication. If you wait for a couple of days, then the person will be more resistant to the communication. It is as if he has built up psychological antibodies against that communication. How can you be an essentialist and study this phenomenon?

Three Conditions for Inferring Cause:

The third condition is the most difficult to satisfy.

Three Methods for Eliminating Other Explanations:

The concept of a control group is implicit in this set of conditions.

Concomitant variation may not often occur in nature, we cannot say that night causes the day because the day always follows the night. Varying the conditions ourselves gives us a means of eliminating alternative explanations.

The use of controls in experimental research helps us to infer cause. Controls help us to rule out alternative explanations. There are three ways in which controls are used in research:

1. Control of the situation.

Laboratory experiments are run in controlled environments where extraneous causal influences are held to a minimum. For example, in research that measures psychophysiological responses, e.g., skin conductance and heart rate, it is important to control for room temperature and extraneous noises.

2. Control of the presentation of the treatment.

This is normally accomplished by random assignment to treatment (Mill's method of agreement) and control conditions (Mill's method of differences).

3. Control for particular threats to valid inference by design features or statistical analyses.

For example, you might control for a participant's knowledge about the experimenter's hypothesis by deliberately giving the participants a false hypothesis.

Analysis of covariance is sometimes used to statistically control for critical differences between control and experimental participants that exist prior to the treatment For example, in a study of learning preexisting differences in intelligence between the control and experimental participants might account for differences in learning. If intelligence was measured, the preexisting differences could be held constant by the use of an analysis of covariance.

"Popper (1902- ) has been the most explicit and systematic in recognizing the necessity of basing knowledge on ruling out alternative explanations of the phenomena so as to remain, the researcher hopes, with only a single explanation."

Popper is a "falsificationalist." He claims that we can prove that a theory is false by finding evidence that contradicts the theory. But that we can not confirm a theory, there may one day be another theory which supersedes the theory, or there may be one day a falsification of the theory.

But note that not any discordant observation can falsify a theory. Popper emphasizes multiple validation criteria that permit one theory to be preferred over another.

The logical positivists were "confirmationalists." They believe that data could be found which would confirm a theory.

Both believe that data can be collected which are relevant to scientific theories. And that the data can (a) confirm the prediction, (b) disconfirm the prediction, or (c) be ambiguous.

For Popper, "corroboration gives only the comfort that the theory has been tested and survived the test, that even after the most impressive corroborations of predictions, it has only achieved the status of " not yet disconfirmed." "Not yet disconfirmed" is not the same as "being true."

"The only process available for establishing a scientific theory is one of eliminating plausible rival hypotheses."

Theories are falsified by a "combination of alternative theory and discordant facts." Without the alternative theory a set of discordant facts is likely to be ignored.

Poppers work implies:

1. "a logical stress on falsifying causal propositions under test and on giving the status of 'not yet disproved' to data patters that corroborate a particular causal hypothesis but do not rule out all plausible alternative causal hypotheses."

2. "a need to collect data which will confront the causal propositions under test, recognizing that convincing data based refutations rely on multiple disconfirmations" ... "data from any one disconfirmation is not "objective" it is not free of all theoretical assumptions."

3. "a need to collect data which confronts causal propositions by putting them in competition with other plausible causal propositions." ... winnowing out weaker propositions

The method of strong inference (Platt, 1964) focuses attention on possible alternative explanations of an observation. It is a method that potentially allows us to discriminate between alternative explanations.

1. Devise alternative hypotheses.
2. Devise crucial experiment.
3. Carry out the experiment so as to get a clean result.
1'. Recycle the procedure.

If you know of any studies in psychology that exemplify this approach please bring them to class.

Platt, J. R. (1964). Strong inference. Science, 146, 347-353.