A practical handbook for anyone who wants to use statistics without embarrassing themselves. My biggest takeaways, in short:
- Know your sample well: who's in it, what you need it for, and how much you can generalise it. Just because it's not a random sample doesn't mean it's not useful.
- Know your data well: definitions, units, quality, distribution, missing values. What does a one-unit increase in poverty mean? Is it a lot? Which definition of poverty does it follow?
- Don't underestimate simple statistics - know precisely what they're saying and what they're for
- It's both important and difficult to determine if an effect is large or small, important or worth ignoring. This won't come from statistical significance but from the context and your audience.
- Just because there is a true causal effect doesn't mean it explains all of the phenomenon. People stay home when it's cold, but that doesn't mean people stayed home in 2020 because it was cold. Sanity check yourself when talking about causality.
- Know your audience and present your findings accordingly - for laypeople: simplify, simplify, simplify.
- Be kind to future you and your fact-checker and document your analysis well