Reverse Causation and Selection Bias

The data shows that people who are hospitalized are far more likely to die within a year than people who aren't. Are hospitals dangerous?

Of course not. Seriously ill people are the ones who end up in hospital; the hospital doesn't make them ill. That's reverse causation: you think A causes B, but actually B causes A.

Reverse causation

It's everywhere:

• "Older people who exercise regularly are healthier": partly because healthier people are able to keep exercising.
• "Employees who work a lot of overtime have worse performance": maybe low performers need overtime to catch up.
• "Restaurants that spend heavily on delivery-app ads sell more": maybe restaurants that sell more can afford the ads.

When judging direction, ask: could the outcome have come first and then triggered the "cause"?

Selection bias

The other trap is that the sample you're looking at has already been filtered.

"Lots of college dropouts become billionaires": you only hear about the few who succeeded. The thousands who dropped out and failed don't make the news. Seeing only the "survivors" is called survivorship bias, one kind of selection bias.

Restaurant reviews work the same way: people who bother to write reviews tend to be either delighted or furious, so the reviews don't represent the typical customer.

Using this in forecasts

• When you cite success stories, ask: where are the failures, and how many are there?
• When your data comes from a particular source, ask: who did this source filter out?
• When you believe A causes B, ask: could B be causing A?