From: hal@finney.org
Date: Sun Nov 12 2000 - 16:34:37 MST
Robin forwards:
> I thought some of you might find this amusing as an example of how
>badly the media presents even simple statistics and how such poor
>analysis can be important. On CNN, Salon, Fox Television News and
>elsewhere a big deal has been made of a graph (attached) showing that
>Pat Buchanan had far more votes in Palm Beach County than in any other
>Florida county. Need I explain the error any further? :) What follows
>is an II press release on the issue and two graphs which tell the story.
It is amazing that two different presentations of the same data can lead
to completely different interpretations.
Harvey presented us with Buchanan's vote by county. If we sort it by
percentage, we can see that the Palm Beach vote is not too unusual:
Calhoun* 1.74% 90 / 5174
Liberty* 1.62% 39 / 2410
Gulf* 1.16% 71 / 6144
Washington* 1.10% 88 / 8021
Holmes* 1.03% 76 / 7395
Union* 0.97% 37 / 3826
Baker* 0.90% 73 / 8154
Suwannee* 0.87% 108 / 12441
Palm Beach* 0.79% 3407 / 432286
Bradford* 0.75% 65 / 8673
Franklin* 0.71% 33 / 4644
Walton* 0.66% 120 / 18318
Jackson* 0.63% 102 / 16300
Dixie* 0.62% 29 / 4666
Santa Rosa* 0.62% 311 / 50319
Hamilton* 0.58% 23 / 3964
Putnam* 0.56% 148 / 26222
... [lower percentages truncated]
However the L.A. times had a different chart the other day, a scatter
chart showing the total Buchanan vote against the total county vote.
We can approximate this by sorting Harvey's figures by county size:
Miami-Dade* 0.09% 560 / 625362
Broward* 0.14% 788 / 573396
Palm Beach* 0.79% 3407 / 432286
Pinellas* 0.25% 1013 / 398469
Hillsborough* 0.24% 847 / 360295
Orange* 0.16% 446 / 280125
Duval* 0.25% 652 / 264636
Brevard* 0.26% 570 / 218395
Lee* 0.17% 305 / 184377
Volusia* 0.27% 496 / 183256
Polk* 0.32% 532 / 168486
Sarasota* 0.19% 305 / 160942
Pasco* 0.40% 570 / 142731
Seminole* 0.14% 194 / 137634
Escambia* 0.43% 502 / 116648
Manatee* 0.25% 271 / 110221
Leon* 0.27% 282 / 103113
... [smaller counties truncated]
On the Times chart, the total Buchanan vote was an extreme outlier.
Most of his votes could have been fit pretty closely to a linear
regression (not necessarily passing through the origin). On this
basis we would have expected about 800-1000 votes in Palm Beach county.
Instead we saw three to four times that number. It was totally out of
line with the rest of the figures.
So, which analysis is correct? Are the Buchanan votes reasonable, as
might be suspected if we look only at the percentage levels compared to
other counties? Or are they unreasonable, as appears when we look at
them compared just to other populous counties?
The moral is that while the networks may have been guilty of simplistic
analysis by focusing solely on vote totals, it may also be wrong to
focus solely on percentages.
Hal
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:31:55 MST