Top-N tips for talking with non-Data Scientists
Neal Lathia

Neal, I’m not sure whether it’s directly related, but I differentiate between mechanistic and statistical models. Mechanistic models being an explanation of the underlying drivers (e.g. laws in hard science — or the wildest speculative views of how something works). Statistical models being numerical relationships between measurements. Mechanistic models can be validated to a given probability of correctness, by statistically sane tests. Statistical models can also indicate possible mechanisms.

The conceptual models you describe often require a bit of sharpening if they’re to be used by >1 person.

The specific danger that I encounter is that a statistical model has been created with insufficient rigour, leading to a misunderstanding of how a phenomenon works. The biggest non-obvious issue is that the numbers going into the statistical model don’t represent what anyone thinks they do.

What do you think?

Show your support

Clapping shows how much you appreciated Tim Coote’s story.