Favorite datasets to use when teaching statistics

There are so many datasets available that it seems difficult to sort through them all and identify the best ones to use when teaching an introduction to statistics class. What are your favorite datasets to use when teaching statistics?

I want the students to have some interesting and fun datasets to work with when experimenting with data visualization, constructing confidence intervals, hypothesis testing, etc.

It's possible this question will be closed as being a duplicate of this question or for being opinion-based, but I'd like to know the specific datasets that you love to use when teaching. For example, DASL has been mentioned, but DASL is large -- which dataset within DASL do you like?

• I'm not a teacher and this isn't a dataset, but my favourite statistical reminder is that the vast majority of the population have more than the average number of legs. – timtfj Jan 24 at 0:34
• @timtfj: That statement is true if by average you are referring to the arithmetic mean, but it is false for the median or the mode. – Rory Daulton Jan 24 at 1:24
• @Rory I didn't want to labour it by spelling that out, because I felt it would spoil the effect of the statement. But that's one reason for it being a nice example—make the statement, ask whether it's true and if so why, ask whether it has to be true: and learn that "average" isn't a reliable term, that "obvious facts" about averages aren't necessarily true at all, and that it's crucial to define what you're actually using. – timtfj Jan 24 at 1:53
• Also note that no person has the average (arithmetic mean) number of legs. – Joel Reyes Noche Jan 24 at 2:51
• If you're in the US - or interested in US education statistics - then you might take a look at the Office for Civil Rights' Civil Rights Data Collection. – Benjamin Dickman Jan 24 at 4:22

19 public data sets, from Springborg blog, curated by T.J. DeGroat. Summaries and links for each in DeGroat's page.

1. United States Census Data
2. FBI Crime Data
3. CDC Cause of Death
4. Medicare Hospital Quality
5. SEER Cancer Incidence
6. Bureau of Labor Statistics
7. Bureau of Economic Analysis
8. IMF Economic Data
9. Dow Jones Weekly Returns
10. Data.gov.uk
11. Enron Emails
13. UNICEF
15. Wikipedia
16. Lending Club
17. Airbnb
18. Walmart
19. Yelp

I always enjoy bringing in Bortkiewicz’s data on the annual deaths by horse kicks in the Prussian Army from 1875-1894.This is what inspired the discovery of the Poisson distribution. Sometimes, students enjoy thinking about how to collect data about rare events:

How often does the anaesthetist has fallen asleep during surgery? And other medical “Never Events.”

The number of mutations in a given length of DNA.

The number of goals in a soccer game.

I think it depends on what you're trying to teach, the following are great for basic introductory stats:

1. Measures of Center: Income Data from US
2. Linear Regression: Advertising Dataset from Introduction to Statistical Learning here or Ames Housing here
3. Classification: Iris or Titanic Datasets, here and here

Kaggle.com, UCI Machine Learning Repo, and NYC Open Data are three fun collections of datasets where you'll find plenty more. These examples are pretty vanilla, but they're easy to find and don't require much preprocessing if any. Otherwise, learn some webscraping and start making your own!

• Thanks. Since Iris has only 50 examples of each of the three flower types, how many of the examples do you typically put into the training set and how many in the test set? The dataset seems almost too small to use as an example for classification. – eternalGoldenBraid Jan 31 at 22:46
• Well, usually I just use default 80/20. It’s from RA Fishers work, he was doing classification. You can get good performance here with a basic logistic regression or KNN. – jfkoehler Feb 1 at 4:19
• Thanks, good to know! – eternalGoldenBraid Feb 1 at 4:24
• The Iris data is great for non-ML examples. I use it for analysis of variance, Discriminant Analysis, clustering, simple distributions...it gives and gives and gives. – Sciolism Apparently Feb 7 at 20:46

I'm amazed that neither Gapminder nor the World Bank data sets have come up; they have tons of interesting development indicators, plus Gapminder comes with entertaining videos!