Our life would be much easier and our policies would be much more amenable to directed action if we could infer a causal relationship between the variables of interests. But the inference of causality is much trickier than simply having a strong association between the variables. So what are the necessary conditions for causal inference? 19th-century philosopher John Stuart Mill formalized the existence of causal relationships if three required conditions are met. They are 1) covariation, 2) temporal ordering, and 3) ruling out plausible rival explanations for the observed association between the variables. Let’s look at each condition carefully.
Conditions for causal inference
Covariation – (causal inference condition 1)
This is the most straight forward conditions, which states that if A causes B, then A and B should covary or should be associated. That is, any change in A should produce a change in B.
Let’s take an example. If eating salmonella infected food causes people to fall sick, we expect to find more sick people in the area where infected food was consumed compared to other areas. If vaccinating ourselves against flu during the flu season can keep us healthy, we would expect vaccinated people to be healthier during the flu season compared to people who are not vaccinated against the flu.
Covariation or association is a necessary condition for causal inference but it alone cannot tell us the type of causal link that might exist between the two variables. Additional conditions need to be met to prove causality.
Temporal precedence (causal inference condition 2)
This condition extends the idea of causality further by stating that if A causes B, then A should come before B in time. That is, the cause should precede the effect, in time.
For example, we might see a strong association between heat exhaustion and dehydration and it makes logical sense if we claim that people who experience heat exhaustion for a long time might end up dehydrated. Or people experiencing heat exhaustion are more likely to get dehydrated. We can establish a temporal order between the cause and the effect. It is easy to see that heat exhaustion and dehydration strongly covary and that the direction is somewhat clear. Another example could be the relationship between the economic boom and increased income for individuals. We might expect the economic boom to result in increased income for individuals. It might be impossible to have increased income before an economic boom.
Knowing the sequence of events helps us to figure out the possible cause and the possible effect by ascertaining which comes first. But establishing covariation and temporal precedence is still not enough to establish causation. For example, we know that as children grow (in age) their reading capacity increases. We can establish covariation and we can also establish temporal precedence that, increase in age comes before an increased capacity to read. But would we say that age causes an increase in reading capacity? We wouldn’t! So we need to add the last necessary condition to the mix.
Ruling out plausible rival explanations (causal inference condition 3)
This might be one of the most difficult conditions to meet. This condition states that we should be able to rule out any reasonable or believable explanation that can cause the two events or variables to appear linked. This is a key criterion on which we judge social research based on the researcher’s ability to rule out rival explanations or limit rival explanations. The threat of competing explanations governs almost every aspect of research design and research methodology. There might be hidden factors that explain the relationship between variables and what appears unlikely to one researcher might appear quite likely to others.
Let’s look at an example. A study found that people who owned a home lived longer. Does this mean that homeownership increases one’s lifespan? Of course not. But if we think a little harder we can figure out that maybe affluence is the rival explanation here. More affluent people tend to be homeowners. They may have better health insurance and therefore may tend to live longer. That explains the spurious relationship between homeownership and longevity. Here’s another example of such a spurious relationship. Higher consumption of ice-cream is strongly associated with a higher likelihood of death by drowning. On its face, this assertion seems dubious. So what can explain this relationship even if we can prove association and temporal precedence? The hidden (sometimes referred to as lurking or confounding) variable here is the season or the hot weather. People tend to consume more ice-cream when it's hot and they also tend to be in the pool when it’s hot, thereby increasing the likelihood of death by drowning.
Identifying such spurious relations and the ability to identify rival explanations is a key factor governing research methodology. We refer to these variables as confounding variables or lurking variables and we call them threats. It is a threat to the validity of our claim. A threat can come from data collection if we cannot prove that our measures reflect our proposed cause and effect. Or a threat comes from the fact that social research often uses samples and it might limit the generalizability of the results to the whole population. The choice of research design (cross-sectional, longitudinal, experimental, quasi-experimental, panel design, etc.) also influence our ability to control rival explanations. And lastly, our ability to generalize results to different socio-cultural conditions, time, place, or how well our research represents the real world, all limit our ability to control rival explanations. There are four types of causations that we generally consider — reverse causation, reciprocal causation, indirect causation, and interactive causation.
Bibliography
Christensen, L. B., Johnson, B., & Turner, L. A. (2015). Research methods, design, and analysis (11th ed.). Essex: Pearson Education.
Dooley, D. (2001). Social research methods (4th ed.). Upper Saddle River, New Jersey: Pretence Hall.
Cite this article (APA)
Trivedi, C. (2020, November, 19). Causality or causal inference or conditions for causal inference. ConceptsHacked. https://conceptshacked.com/causal-inference/
