1.4 Introduction to causality

This object introduces the concept of causation and the causal path diagram as a context which provides a framework for the exploration of the many different factors that may contribute to observed patterns of disease.

Patterns of disease over time or space often suggest associations between disease rates and other factors with similar distributions. In December 1952 London experienced five days of thick smog brought about by a natural temperature inversion and high levels of pollution from coal and coke fires. That week experienced dramatic increases in death rates from respiratory complaints (Bell et al., 2004), provoking widespread interest in the relationship between atmospheric quality and respiratory illness, and resulting in significant legislative changes. The investigation of apparent associations between health events and other factors leads us to question whether the observed relationship is genuine or coincidental, positive or negative, occurs at all times and in all situations. Specifically, we are concerned to identify whether or not it is causal – does the associated factor actually in some way cause the health outcome in which we are interested? We may consider a relationship to be causal if a change in the level of the independent variable (conventionally X) produces a change in the level of the dependent variable (Y). Our discussion is facilitated by introduction of the causal path diagram. In most geographical contexts, such relationships are explored using aggregated datasets, but here we shall illustrate the factors involved with the aid of an individual example.

We shall here be primarily concerned with the causes of non-communicable diseases. It should be apparent from this presentation that this approach to explaining health events is probabilistic and multicausal: effects are not completely determined by any cause, nor are they the result of a single causal factor. Instead causal factors may be related to many effects and an effect may be influenced by many causal factors. This differs from some historical biomedical perspectives that sought to identify the single causal organism or mechanism for each disease, often in the hope that this would lead to a single specific treatment regime.

Jones and Moon (1987) summarise five further characteristics that we should look for in an observed association between variables if the hypothesis of causality is to be confirmed. These should be interpreted with care, as all are amenable to disguise if extraneous, suppressor or distorter variables are present.

1. Strength of relationship: although the strength of relationship is not necessarily an indication of causality, we would expect a strong relationship to be less readily disguised by extraneous factors.
2. Dose-response relationship: if the relationship is causal, we would expect an increasing amount of the causal variable to be reflected in an increasing amount of the response variable.
3. Consistency: the relationship between the variables should operate in the same way at different times and places – it is much less likely that extraneous factors would disguise the relationship in the same way in many diverse circumstances.
4. Coherence with biological plausibility: the nature of the relationship should be consistent with known biological processes. The difficulty with this criterion is that it does not allow for the observation of causal associations to lead to new biological understanding, which does happen on rare occasions.
5. Specificity: ideally, we would hope to see that the relationship operates only between the variables in question, although in reality this is very rare.

Consider how each of these factors might be interpreted in the context of our example of exposure to sunlight and skin irritation.

---

---