5 ms·
Beware of the sample size issue. The Midwest has lots of low-population counties, for which a single cancer case plus or minus makes a great difference in the i
by hikarudo 7y ago
Beware of the sample size issue. The Midwest has lots of low-population counties, for which a single cancer case plus or minus makes a great difference in the incidence rate.
See also: Bayesian Data Analysis, Gelman et al., 2nd edition, section 2.8.
- Splendor 7y agoWyoming and Idaho should have the same problem but appear pretty blue.
- ceejayoz 7y agoThat's the plus or minus part.
- airstrike 7y agoThank you for the reference. Imagine where we would be if everyone provided sources to their statements like that...
- Misdicorl 7y agoBut we should expect the plus or minus to be fairly evenly distributed if there weren't an underlying reason besides randomness and small sample sizes. That would lead to regions with lots of blue/red/yellow speckling. Instead we see fairly homogeneous regions of red and blue with transition zones between them. This definitely doesn't look like a noisy undersampled data set