I recently developed and delivered a 50-minute lecture on the pitfalls of selection bias for NIH's Office of Behavioral and Social Sciences Research @NIHOBSSR.

It is now available for free online, and would appropriate for many undergraduate courses.

We start off with a puzzle. How can it be that every insurance company thinks they their new customers about $500 when they switch?

The fundamental subject of the lecture is, as promised, selection bias.

Selection bias occurs when the people who you sample differ systematically from the population at large, with respect to the question you are asking.

I draw upon a number of COVID examples, such as these dentists who failed to realize that the people who visited their practice during the NYC peak were not coming in for your average cleaning—and wrongly inferred that masks were causing tooth and gum damage.

And of course a lot of the instances are far more subtle. Being aware of selection bias is key for readers as well as researchers.

We consider selection bias broadly constructed, and explore the principle that where you look determines what you find.

Some of the twitterati make an appearance: https://twitter.com/statsepi/status/889466390054359040

I explain what Will Rogers has to do with cancer diagnosis.

Explain why country singers truly have achy breaky hearts.

(Yes, young 'uns, that's her dad.)

Resolve one of life's great mysteries, with help from @JSEllenberg.

And explore why universities claim to have average class sizes around 20 but their students rarely report rarely if ever actually being in classes that small.

In our book _Calling Bullshit_, @jevinwest and I devote an entire chapter to selection bias, and it's one of my favorite chapters of the book.

If you understand the concept, you can see through a lot of bad stats—without needing any technical expertise.

Enjoy!