Unclear Causality
"Correlation does not imply causation."
Description
Even when we know the elements in a system, it can be hard to determine how they influence one another (i.e., the relationship links between them). There are numerous types of error that we can make when causality is unclear; this page only mentions a few of them.
After, Therefore Because
it's tempting to assume that if an action precedes an event, the action causes the event. For example, if you get a meal from a new restaurant and then feel ill the next day, it’s tempting to conclude that the restaurant gave you food poisoning. Maybe it did, or maybe you just picked up a virus from work.
This type of error can be caused by spurious correlation (i.e., a third factor that affects both of the observed events), or simply coincidence due to random timing.
Causal Oversimplification
Here is a quotation from an internet forum: 'I know two people who never smoked but they died of cancer, so smoking isn't a problem.' The assumption here is that cancer can only be caused by one thing. The medical view is that the risk of cancer is increased by many factors. To make matters even more complicated, any one factor may not cause cancer in a particular individual, but a combination of factors may do so.
This type of mistake is also known as 'fallacy of the single cause'.
A second problem with the quotation above is that it only considers two people. When many more people are considered, the effect of smoking becomes clearer (see statistical significance). Confirmation bias may also come into play: the quotation is from a smoker who would be attracted to evidence that supports his desire to believe that the habit is safe.
Reverse Causality
The relationship between physical exercise and mental health is unclear. People with depression tend to exercise less than others, but what causes what? There are reasons to believe that exercise increases mood, but there are also reasons to believe that a poor mental state reduces the desire to exercise. We know that the two elements are related, but we can't be sure about which one is the cause and which one is the effect. This is important in practice because we need to know the best way to improve the situation: should be encourage exercise to improve mental health, or should we improve mental health so that people exercise more?
Regardless of whether the chicken or the egg came first, once the system behaviour is established, it's possible that each element affects the other, resulting in a causal loop.
Spurious Correlation
Education as protection against adult mortality: A global systematic review and meta-analysis. Centre for Global Health Inequalities Research, 2024.
People with tertiary qualifications tend to live longer than people with less education. This tempts us to conclude that higher education causes longer life. However, the actual reason for the correlation is perhaps that educational opportunities and health support are both consequences of a common cause: socioeconomic status (i.e., wealth). Just because two elements change together doesn't necessarily mean that one influences the other. We shouldn't automatically conclude that giving people more education will cause them to live longer.
This type of fallacy is also known as 'joint effect'.
Examples
The elephant in the room is global warming. Large amounts of greenhouse gases have been released into the atmosphere in the last century or so, and the climate is getting warmer. Just because the two phenomena are correlated doesn't automatically mean that one caused the other. Sceptics are right to point that out.
In an online forum for pregnant women: “I’m 8 months pregnant, and my dentist told me that one of my teeth needs to be extracted. Is this going to affect my baby somehow? Could there be any sort of birth defects? I’m in full panic mode. TIA!”
Somebody responded: “A friend of mine had a tooth pulled out during her pregnancy. The baby was born without teeth.”
Upstate_Gooner_1972, r/Jokes.
Related Issues
This section describes similarities and differences between this issue and related problem factors and thinking traps.
Element and Relationship Scopes
If we're going to think of our problem as a system, that system needs to include all of the main elements and relationships that determine its behaviour. If it doesn't, we'll be blind to what is actually causing the behaviour we want to change, and the actions we take to attempt to change that behaviour are likely to be ineffective.
Time Scope
A long time scope can result in unclear causality because the effect can become noticeable much later than the change that caused it.
Tiny Impacts
Tiny impacts can result in unclear causality because a single impact, or even many impacts, may have no noticeable effect on system behaviour.
Intuition
Intuition gets a lot of things right, but it prefers simple explanations. So, when we subconsciously detect that the behaviour of two elements seems related, we assume a causal relationship between them, but we don't always avoid the issues described above. In addition, we may assume a causal relationship when the apparently-correlated behaviour is only a coincidence (which is the origin of many superstitions).
Complex Behaviour
Complex behaviour can result in unclear causality because the behaviour of related elements can be very different to each other.
Probability
When systems don't behave perfectly consistently but possess variability, it is more difficult to determine causality.
Related Engine Processes
Later, we will describe a thinking framework that comprises multiple processes. This section points forward to the processes that deal with this page's topic.
Priorities
Our priorities can affect the causal relationships that we're drawn to. For example, a professional educator is tempted to believe that education directly increases longevity and may avoid other theories that could explain the relationship between education and longevity. Self-awareness of the potential for such biases is important.
Scenario-Based Planning
If we aren't confident in our ability to determine causal relationships in a system, we could consider each possibility to be a separate 'scenario'. Each scenario represents one hypothesis about what causes what, and would have its own system structure (which would be devised in systems thinking).
Systems Thinking
Systems thinking requires assigning causal relationships between elements.
Sensitivity Analysis
Sensitivity analysis can be used to assess the importance of causal relationships. If we evaluate a relationship in our system model and find that it doesn't make much difference, erroneous inclusion of the relationship won't mess up our decision-making very much. However, sensitivity analysis can't help us if we fail to include a significant relationship because it can't assess anything that isn't present in the system model.
Adaptation
If we are unable to determine causality, the actions we take to fix our problem may not be effective. Adaptation involves looking for signs that our actions aren't working and, if such signs are found, repeating the decision-making process with different assumptions that are more consistent with the newly-observed behaviour. This can prompt us to consider different causal relationships.
Further Reading
Questionable Cause (Wikipedia).
To Do
Murdered Darlings
- READ MORE example: cricket review sound system (Snickometer): commentators assume that any sound that occurs when the ball passes close to the bat was caused by the ball touching the bat, even though they are well aware that similar audio indications can occur at other times or when the ball is nowhere near the bat (eg, due to other on-field sounds). The probability that the indication does not indicate an edge is wrongly assumed to be zero when bat and ball are very close. This is actually Bayesian: evidence of bat close to ball increases the likelihood that a sound was caused by a nick, but doesn't reduce the likelihood that the sound was caused by something else.
- Bayes Theorem can explain superstition: we notice coincidences but don't notice occasions when the correlation didn't happen.
After, Therefore Because: This type of error is formally called 'Post Hoc Ergo Propter Hoc' (latin).