Polls Apart

We are now nearing the six-month mark until the 2012 election, and polls are giving President Obama a slight edge in the head-to-head match-ups.  However, historically, does this slight edge mean anything for predicting the final outcome?

To find the answer, I first gathered and averaged polling data for the period between April 1 and May 10 of the election year.  The table, below, lists those results and highlights the candidate who went on to win.

Barack Obama	John McCain	Margin
45.71	45.14	0.56
George Bush	John Kerry
45	42.43	2.57
George Bush	Al Gore
47.5	42	5.5
Bill Clinton	Bob Dole
54.17	40.67	13.5
Bill Clinton	George Bush
40.5	47.75	-7.25
George Bush	Michael Dukakis
42.75	44.5	-1.75
Ronald Reagan	Walter Mondale
51	39	12
Ronald Reagan	Jimmy Carter
41.67	46	-4.33
Jimmy Carter	Gerald Ford
47.5	40	7.5

Judging by this chart, candidates who were up (by any margin) six months prior to the election ended up winning the election 67% of the time (though, technically George W. Bush lost the popular vote in November 2000, putting the success rate at 56%).  Nonetheless, the sample size is so small and the margin of polling so sporadic that this percentage really has no predictive value.

Thus, I decided to look at both the six-month polling and the final result in hope that there could be some correlation between the two.  Again, nothing.  If anything, the results were even less extrapolative.

Barack Obama	John McCain	Margin	Election Margin	Difference
45.71	45.14	0.57	7.2	6.63
George Bush	John Kerry
45	42.43	2.57	2.4	-0.17
George Bush	Al Gore
47.5	42	5.5	-0.5	-6
Bill Clinton	Bob Dole
54.17	40.67	13.5	8.5	-5
Bill Clinton	George Bush
40.5	47.75	-7.25	5.5	12.75
George Bush	Michael Dukakis
42.75	44.5	-1.75	7.7	9.45
Ronald Reagan	Walter Mondale
51	39	12	18.2	6.2
Ronald Reagan	Jimmy Carter
41.67	46	-4.33	9.7	14.03
Jimmy Carter	Gerald Ford
47.5	40	7.5	2.1	-5.4

Put graphically,

Again, none of this data is helpful.  One main takeaway is that the least square fit (LSF) for the second graph is at least parallel; meaning the rate of the margin of the eventual win over time is at least similar to that of poll deviations.  A better look at this trend is exemplified if we take the absolute value of the poll deviations, giving the graph below:

Nevertheless, the data points are so sporadic that the LSF should be seen as more random than fixed. 

My next idea was to see if perhaps polls were more accurate for one party to take office or if there was an incumbent trend.  Unfortunately, there is neither. 

Finally, a note on the overall accuracy of polls; when averaging the margin of difference between the polls and the election results, polls were 3.61% off (for the eventual winner).  When taking the average of the difference margin for years with correct and incorrect polling, the polls were off 0.45% and 7.56%, respectively.  Again, this is not saying much because the data points are so diverse.

In conclusion, neither President Obama nor Mitt Romney should read too far into these current polls.  Sure, there are benefits to being ahead in the polls early in the process.   A candidate can use that information to leverage better fundraising in an inevitability strategy.  Still, the candidate is in no way inevitable.