Correlation Predicts, Causality Corrects

Toothpaste taught me about prediction.

“Dad, does toothpaste help me fall asleep?” wondered my son.

“What do you mean?” I asked.

“You always brush my teeth before bed, so I thought maybe toothpaste helps me sleep,” he replied.

Something in my head clicked.

Correlation and causation are easily confused. Whilst toothpaste and sleeping are correlated - one usually follows the other - toothpaste doesn’t cause sleepiness.

Ackoff elaborates:

“In one large city, it was discovered that people who live in neighbourhoods with a great deal of soot-fall were more likely to get tuberculosis than those who live in neighbourhoods with less soot-fall.

Some researchers erroneously concluded that soot was a producer of tuberculosis. Subsequent research showed that it was neither necessary nor sufficient for tuberculosis. It showed that the more soot-fall in a neighbourhood, the lower the cost of housing in that neighbourhood tended to be. The lower the cost of housing, the poorer its inhabitants tended to be. The poorer people were, the more critically deficient their diets were. Dietary deficiencies were found to be a producer of tuberculosis, not soot.”

So, whilst soot-fall and tuberculosis were correlated - they usually occurred together, there was no causal link - soot didn’t cause tuberculosis.

This has two interesting consequences:

Firstly, correlation enables prediction.

Observing greater soot-fall in a neighbourhood was a convenient method of predicting a greater likelihood of tuberculosis.

Secondly, causation (and not correlation) enables intervention.

Reducing soot-fall in a neighbourhood would not reduce tuberculosis. Soot didn’t cause tuberculosis, but only indicated a poorer neighbourhood where critically-deficient diets were likely. Improving diets was needed to reduce tuberculosis.

In organisations, it’s easy to confuse causation with correlation. We may believe staff sitting idle is the cause of slow project delivery, and encourage them to pick up more work.

However, whilst slow delivery and idle staff may be correlated and occur together, both may be caused by something deeper - such as insufficient test environments. Keeping staff busy won’t help.

So, an understanding of causality is needed before we can intervene successfully.

The takeaway? Whilst correlation enables prediction, an understanding of causation is needed for intervention. And, keep calm and carry on brushing.

This article was inspired by Ackoff’s Differences That Make a Difference.