Economists play an important role in the formation of social policies, be they economic or noneconomic, and in institutional design. To do policy analysis successfully, we have to know about causes.
Correlation is not causation. Correlation is the degree of linear association among variables. If a variable X is correlated with another variable Y, then Y is also correlated with X. But it is not true in general that if X causes Y, then Y causes X. so correlation cannot be causation.
Correlation plus time-ordering might be causation. But there are more potential reason for which an earlier variable X is correlated with a later variable Y than X being the cause of Y. One such reason is the existence if a third variable (or set of variables) Z such that Z causes both X and Y. variables Z that are common causes for two correlated variables X and Y are often called confounders, and correspondingly the problem that the correlation between X and Y may be explained not only by a direct causal relation between X and Y but also a structure in which a third factor Z causes both X and Y is called ‘problem of confounders' or the ‘third-factor problem'. Sometimes correlations arise form characteristics of variables that have nothing to do with their causal relations. That is, there may be non-causal reasons for the existence of a correlation. Another possibility are non-stationary time series, these will often be correlated whether or not they are causally related.
In brief, Hume thought that ‘X causes Y' is true whenever:
- X is universally associated with Y
- Y follows X in time
- X and Y are spatio-temporally contiguous (there are no time-wise or space-wise gaps in between X and Y).
Hume was what later has become to be called a ‘reductionist' about causation. What this means is that Hume held that causation is not among the fundamental building blocks of our world. Thus, if we say that ‘The Banks' irresponsible behaviour caused the financial crisis of the late 2000's what we really mean is that (a) irresponsible banking is universally associated with financial crises; (b) the banks' behaviour predated the crisis; and (c) there are no gaps between the banks' behaviour and the crisis.
As this example suggests, there are serious flaws in Hume's constant conjunction account. In fact, neither condition (a)-(c) is necessary, nor are the three together jointly sufficient for causation. First, most factors we regard as causes are not universally associated with their effects. Second, not all effects follow their causes. Third, causes may act at a temporal or spatial distance. Fourth, constant conjunction, temporal priority of the cause and contiguity are insufficient for causation.
A law is a statement about a generalization that holds universally. Strictly speaking, laws hold only under special conditions. Economists (and others) often refer to these conditions as ‘ceteris paribus' or ‘other things being equal'. For now, let us note that a tendency claim is a claim that describes a regularity, which holds in isolation, when disturbing factors are absent. Another way op putting it is to say that a tendency claim is a claim about a regularity that would hold if disturbing factors were absent.
Characteristics of Mill's tendencies:
- Describing a tendency means to describe what happens in isolation from disturbing factors.
- Tendencies are casual: a ‘power' that is ‘acting' with a certain ‘intensity'.
- Tendencies produce stable contributions to outcomes that persist even in the presence of disturbing causes.When disturbing factors are present, outcomes can still be explained when two conditions are met: the tendency laws of the disturbing factors are known; the law of composition is known.