11/15/00 2:35 p.m.
The Lack of Evidence in Palm Beach
A field day for misleading statistics.

By Patrick L. Anderson, managing director of Anderson Economic Group,
& former deputy secretary of state for the State of Michigan

eldom has a single chart had such power.

Much of the outrage over the voting in Palm Beach County stems from the misuse of a single chart. That chart—shown on network television, and recounted in newspapers and talk shows — displayed the number of votes for Pat Buchanan in each of Florida's 67 counties. As we all know, Buchanan's 3000-plus vote total in that county was much higher than his totals in other counties, lending credence to the theory that many of those votes should have been Al Gore's.

The ballot may have been confusing, and some votes probably were cast incorrectly. Even one miscast vote is a serious matter. But the data just don't support the contention that there are 1,000 or 2,000 "lost" Al Gore votes in Palm Beach.

The common portrayal of the data is reproduced below. As anyone can see, it sure looks like something went wrong in Palm Beach County.

But this chart makes use of two classic techniques for misleading readers: It distorts the scale of the chart, and it eliminates relevant information.

The scale of the chart — the range of values that can be portrayed — has been magnified 100 times in order to make Buchanan's vote look impressive. In fact, he received much less than 1 percent of the total vote, and almost placed fifth overall in the presidential voting.

The relevant information omitted from the chart includes the amount of votes that other candidates received, as well as the different number of votes in each county. Without this information, the reader could be duped into believing that Buchanan had a landslide of support — in a county where he received less than eight-tenths of 1 percent of the vote, and where even Ralph Nader beat him by a 60 percent margin.

To show the data fairly, we should put all the candidates on the chart. The problem with that is that Buchanan's support almost disappears. When your top competitor receives over 260,000 votes in a county, your 3,000 is not only not much to cheer about, it's also not very visible on the chart.

To understand what was going on, take a look instead at a scatter plot showing the share of votes in each county for Bush, and those for Buchanan. First notice that Buchanan got very few votes, period. In only five counties (Calhoun, Gulf, Holmes, Liberty, and Washington) did Buchanan manage to scrape together even 1 percent of the vote.

The plot is mostly interesting for what it doesn't show. It doesn't show that Palm Beach is unusual for having high Buchanan votes; it shows Palm Beach has an unusually low number of Bush votes. Finally, by being scattered around the chart, the data are saying that there is little systematic relationship between Bush and Buchanan voters.

Now, again using proper technique, compare Bush vote with Gore vote:

Looking at this chart explains an awful lot about the election. As a rule of thumb, those counties that didn't vote for Gore voted for Bush, and vice versa. Palm Beach County, highlighted in the chart, fits exactly the overall pattern.

There certainly isn't any evidence of a sharp drop in Gore support; indeed, the main "unusual" aspect of Palm Beach County is the large number of Gore voters.

This Florida presidential election will provide fodder for a generation of statistics books, not because the math is hard, but because misreading the data was just too easy. While those voters who believe they miscast their ballots are understandably angry, the data simply do not show a significant discrepancy in the Gore vote in Palm Beach County.

For more information on the statistics presented in this article, including additional charts, see the Anderson Economic Group web site.