FoMind
Look at the numbers and you'll find that in months when ice-cream sales are higher, more people drown while swimming. So would selling less ice cream reduce drownings?
Obviously not. Both go up because of a shared cause: hot weather. When it's hot, people eat more ice cream and also go swimming more.
When two things move together, there are several possible reasons
When you see that A and B are correlated, consider at least these possibilities:
- A causes B: the one that comes to mind first.
- B causes A: the direction is reversed (see lesson 5.3).
- C causes both A and B: a hidden common cause, also called a confounder (see lesson 5.2).
- Selection bias: the sample you're looking at has been filtered (see lesson 5.3).
- Pure coincidence: with enough data, you can always find two things that happen to move together.
What this has to do with forecasting
Many forecasting reasons are really correlations in disguise: "Every time interest rates were cut, stocks went up," or "Students who use this app get better grades." If there's real causation behind it, it can help you predict the future. If it's a shared cause or a coincidence, the pattern can break as soon as circumstances change.
In the scatter plot below, two variables look strongly correlated. Add the hidden variable and split the data into groups: is the correlation still there within each group?
This is where the interactive demo “Hidden-variable scatter plot” lives: in the FoMind app you can drag the controls and see results update live.
Ask one more question when writing your reasoning
When your reason is "A and B always show up together," ask yourself: is there some C that could explain both? If you took A away, would B really change?
A correlation only tells you two things move together, not which one drives the other. Before forecasting from a correlation, ask what's really behind it.