Breaking News: Scientists have discovered that people with yellow teeth are more likely to die of lung cancer. Patient-groups and dentists are now calling for tooth-whitening to be made freely available to everyone.”
Does anything about this statement strike you as illogical? Surely it is obvious. Having yellow teeth does not cause lung cancer – smoking causes both yellow teeth and lung cancer! Providing a tax-funded tooth-whitening service will be futile – banning smoking is the way to reduce deaths from lung cancer!
What is wrong here? Do we have a problem with mad scientists, misuse of statistics or manipulative journalists? Or all three?
Unfortunately, while we may believe that smoking causes both yellow teeth and lung cancer it is surprisingly difficult to prove it – even when sane scientists use the correct statistics and their results are accurately reported by trustworthy journalists. It is not easy to prove causality. So we just assume it.
We all do this many times every day – we infer causality from our experience of interacting with the real world – and it is our innate ability to do that which allows us to say that the opening statement does not feel right. And we do this effortlessly and unconsciously.
We then use our inferred-causality for three purposes. Firstly, we use it to explain how past actions led to the present situation. The chain of cause-and-effect. Secondly, we use it to create options in the present – our choices of actions. Thirdly, we use it to predict the outcome of our chosen action – we set our expectation and then compare the outcome with our prediction. If outcome is better than we expected then we feel good, if it is worse then we feel bad.
What we are doing naturally and effortlessly is called “causal modelling”. And it is an impressive skill. It is the skill needed to solve problems by designing ways around them.
Unfortunately – the ability to build and use a causal model does not guarantee that our model is a valid, complete or accurate representation of reality. Our model may be imperfect and we may not be aware of it. This raises two questions: “How could two people end up with different causal models when they are experiencing the same reality?” and “How do we prove if either is correct and if so, which it is?”
The issue here is that no two people can perceive reality exactly the same way – we each have an unique perspective – and it is an inevitable source of variation.
We also tend to assume that what-we-perceive-is-the-truth so if someone expresses a different view of reality then we habitually jump to the conclusion that they are “wrong” and we are “right”. This unconscious assumption of our own rightness extends to our causal models as well. If someone else believes a different explanation of how we got to where we are, what our choices are and what effect we might expect from a particular action then there is almost endless opportunity for disagreement!
Fortunately our different perceptions agree enough to create common ground which allows us to co-exist reasonably amicably. But, then we take the common ground for granted, it slips from our awareness, and we then magnify the molehills of disagreement into mountains of discontent. It is the way our caveman wetware works. It is part of the human condition.
So, if our goal is improvement, then we need to consider a more effective approach: which is to assume that all our causal models are approximate and that they are all works-in-progress. This implies that each of us has two challenges: first to develop a valid causal model by testing it against reality through experimentation; and second to assist the collective development of a common causal model by sharing our individual understanding through explanation and demonstration.
The problem we then encounter is that statistical analysis of historical data cannot answer questions of causality – it is necessary but it is not sufficient – and because it is insufficient it does not make common-sense. For example, there may well be a statistically significant association between “yellow teeth” and “lung cancer” and “premature death” but knowing those facts is not enough to help us create a valid cause-and-effect model that we then use to make wiser choices of more effective actions that cause us to live longer.
Learning how to make wiser choices that lead to better outcomes is what Improvement Science is all about – and we need more than statistics – we need to learn how to collectively create, test and employ causal models.
And that has another name – is called common sense.