Owen Yang
On reflection, the word ‘reverse causation’ is slightly misleading. It does sound a little like time travelling: changing the past to reverse what has been done.
Reverse causation is a relative term
When we say reverse causation we usually mean that the cause and effect is opposite to what is initially thought or hypothesised. It is most commonly seen in a cross-sectional study or an observational study. When we find older individuals who participate in social groups are less likely to have dementia, we might get excited and say participation in social groups prevents future dementia. At this point we need to calm down and realise that it could also be those who have dementia do not participate social groups.
So the cause and effect can reverse to either direction; hence ‘reverse causation.’ If you can explain clearly, it is probably wise not to use the word ‘reverse causation.’
Can future reversely cause the past?
In a prospective study where participating in social group occurs before dementia, say 10 years before, we need not to be naive, and should not conclude that causation cannot be interpreted reversely in time. In this example, dementia is a long process. We gradually lose our memory and cognitive function, and would only be diagnosed to have dementia when our cognitive function declines below an arbitrary criteria, and when we have a chance and reason to see a doctor to get diagnosed. Therefore, process of dementia may have already started unnoticed at the time when social group participation is assessed in that prospective study.
A lot of time we need to be careful when speaking to a pure theorists, say someone specialised in out-of-context causal methods. When taken by the face value, dementia that occurs after social group participation cannot cause social group participation. However, it is important to know that we are not speaking in pedantic logic languages, and this needs both parties to understand that there is a clinical question to be answered.
Reverse causation is essentially a confounding issue
So what happens here are two issues to be clarified. First, when we say reverse causation, is actually confounding. The fundamental form of confounding in a cohort study, or a prospective study, is that there is a confounder (or confounding factor) that is highly correlated with both our exposure and outcome. Without appropriately taking the confounder into account, we erroneously attribute outcomes to an exposure, whilst it was the confounder that caused the outcomes.
The second issue is that cognitive function at baseline and cognitive function in the future can be seen as two, and should not always be confused as one. It is the cognitive function at the baseline that is the potential confounding factor that, without being taken into account, may let us erroneously attribute future cognitive function to social group participation at baseline. Here you can see there are actually three factors here: social group participation, cognitive function at baseline, and cognitive function in the future.
Here we have it. So reverse causation is a relative term depending on what our hypothesis of the causal direction.
What do you think that we can do to avoid incorrect interpretation of results due to reverse causation? This is perhaps for another article to discuss, but let me know what you have done previously or what you can think of.
