The thing about my way of teaching stats is that my cool, novel examples get old really fast. Like, pop culture is on a thirty-day cycle. Anyway, YEARS ago, 2014 to be exact, I blogged about illustrating bimodal data distributions using baby name data from the Social Security Administration. Specifically, when an older name comes back into vogue, it has a bimodal distribution when you plot time on the X axis. I stand by this example. However, all of the data illustrations come from 2014. When our traditionally aged college students were 6 or 7. Which makes this example feel a little stale. However, here is some data, courtesy of Redditor fevaah , showing old names that are back in vogue through 2025 ! And Eleanor, Violet, Nora, and Hazel are here, illustrating bimodal distributions for a new generation. If you used my old baby-name data, this is a quick, easy refresh for your lectures (or a fun new addition if you didn't). If this idea was help...
You all. Many people, but mostly Jenny Terry and Andy Fields, but also a number of my Twitter mutuals, collected a crap ton of data from statistics students worldwide . See: Here is the article describing the project . The data is embargoed until October 2024, but you can contact the corresponding authors if you would like early access. Also, they have tons and tons of documentation available at OSF . So you can come up with your own hypotheses and test them. Which is very, very generous.