Wednesday, October 26, 2011

History Repeats Itself (or Not)


Yesterday, Mitt Romney gave reporters this oft-repeated pearl of conventional wisdom: "Can't win the White House without winning Ohio", he told them. Is this true?

Here's a comparison: the NFC team almost always wins the pre-Super Bowl coin toss. Since 1994, the AFC team has only won twice; in fact, between 1998 and 2006 the NFC went totally undefeated in this event (odds 1 : 512).



AFC
NFC
2006
lost
won
2005
lost
won
2004
lost
won
2003
lost
won
2002
lost
won
2001
lost
won
2000
lost
won
1999
lost
won
1998
lost
won

Does the above table tell us that only NFC teams can win coin tosses? Of course not. That's ridiculous, but no more so than a lot of other things...

2008
365_173
FL
OH
2004
286_251
FL
OH
2000
271_266
FL
OH
1996
379_159
FL
OH

Above: results of past four presidential elections by electoral votes. In all cases, winner took FL and OH.

We hear it time and time again: correlation is not causation, and yet we generally heed the warning only so far -- a few concurrent data points can be chalked up to coincidence, sure, but too many and one begins to suspect a pattern. In the coin toss example, we make a special consideration -- however uncanny the degree of correlation, we know that coin tossing is gambling so it must be one of those rare coincidences -- but how rare are such coincidences?

Prior knowledge of the random nature of coin tosses prevents one from imagining that there might be any causal relationship in this particular scenario, but what about scenarios where the characters are unfamiliar? Just how often is noise mistaken for meaningful data?

1992
395_143
FL
OH

Move the baseline back one cycle and the premise that FL is necessary to win the presidency is proven false. The fact that the winner of every election since 1964 has taken OH merely demonstrates correlation (think 12 consecutive coin tosses coming up heads) augmented by the fact that since the winner generally takes the most states, one of them is likely to end up being OH. To test the statement's veracity one must address how often, if ever, the victor would have won without FL and/or OH:


+ FL, OH
- FL
- OH
- FL, OH
2008
365_173
338_200
345_193
318_220
2004
286_251
259_278
266_271
239_298
2000
271_266
246_291
250_287
225_312
1996
379_159
354_184
358_180
333_205
1992
395_143
370_168
374_164
349_189

Above: results of past five presidential elections by electoral votes (in bold letters) including hypothetical scenarios in which FL and/or OH are given to the losing candidate.

Were we to extend the table back to 1920, we would find that with the single exception of George W. Bush, giving FL and OH to the other candidate would fail to alter the outcome of the election. So what is the origin of this peculiar and totally false belief held by Romney and so many others? Most likely the lens of recent history: it seems this piece of lore made its way into the collective largely due to the resemblance of Kerry's narrow defeat in 2004 to Gore's in 2000, at the hands of FL and OH, respectively.

---

Update (10/29): "We can't win without Florida"