In the wake of the 2012 Presidential Election, I thought it would be interesting to review recent history and the performance of polling aggregates in the modern era of widespread polling. In so doing, I've gathered poll results from the final week, respectively, before each of the last four Presidential elections. For the last three elections, RealClearPolitics has a nice sample which aggregates all polls in which the entire poll sample was conducted within the last seven days before the election. When a single pollster polled multiple times in that period, only that pollster's final pre-election poll is used in the sample. For 2000, there was no such RealClearPolitics sample, so I created my own using PollingReport. In that case, lacking the information on when each poll sample began, I simply used the final poll from each pollster that concluded its polling within five days of the election. Bear in mind that what national polls directly predict the outcome of the popular vote.
2000
Polling aggregate: Bush +2.1%
Election Day Result: Gore +0.5%
Polling aggregate bias:
+2.6% Republican
2004
Polling aggregate: Bush +1.5%
Election Day Result: Bush +1.5%
Polling aggregate bias:
none
2008
Polling aggregate: Obama +7.3%
Election Day Result: Obama +7.6%
Polling aggregate bias:
+0.3% Republican
2012
Polling aggregate: Obama +0.7%
Election Day Result: Obama +3.3%
Polling aggregate bias:
+2.6% Republican
Average Presidential Election polling aggregate bias, 2000 thru 2012:
+1.4% Republican
Sources:
WH2000 Trend
RealClear Politics - Polls
RealClearPolitics - Election 2008 - General Election: McCain vs. Obama
RealClearPolitics - Election 2012 - General Election: Romney vs. Obama
Results:
In three of the previous four Presidential elections, the polling aggregate shows a bias in favor of the Republican candidate. In the fourth instance (2004), the pollingaggregate nailed the final outcome perfectly. In one of the three results showing a Republican bias, the bias was very small, a mere 0.3%. This is well within that margin of error even for the enormous sample of many aggregated polls. The other two results, with years 2000 and 2012 each showing a bias of 2.6% in favor of the Republican candidate, are more significant.
What this survey destroys is the meme that polls have a liberal bias. They don't. Every year, many conservatives manage to convince themselves that the polls have a systematic bias towards Democratic candidates. But the actual numbers show otherwise.
Note #1:
When I use the word 'bias' here I use it in a statistical sense, not in the sense that the takers of these polls are either overtly or subconsciously favoring Republicans.
Note # 2::
2.6% is the current margin for President Obama in the 2012 election. Given that most of the precincts that have not yet reported are from heavily Democratic states, it is very possible his margin may tick up another tenth of a percent or more. With that in mind, the current margin suffices for the purposes of this post.