Capturing headlines on a daily basis, a flood of polls and polling data paints a conflicting and very confusing picture of the 2012 Presidential election. Adding to the confusion, it seems like every poll is different and every news agency conveniently broadcasts polls that supports its most recent claims. Why are each of these polls “scientific” and yet so different? Just like manipulation of the unemployment rate discussed in part one of this series, you have to look behind the numbers to understand the meaning of presidential polls.
The entire point of polling data is to try to determine which candidate would win the election if it were held at the time the polling was conducted. Sounds pretty simple until you start looking at the variables that must be weighed to come up with an accurate poll result. Not surprisingly there are some polls that are actually intentionally inaccurate and are conducted for purposes other than as predictive of results. Any poll of all registered voters is definitionally a useless predictive poll. At any given presidential election no more that 55-65% of registered voters actually bother to show up and vote so needless to say a poll of all registered voters includes data from 35-45% of the people polled will not show up at the voting booth.
Only polls that focus on likely voters are of any use for predictive purposes, but even that is not the end of the inquiry. The trick is to create a model that is accurate to predict 1) how many voter will actually turn out to the election, and 2) which of the registered voters will be the ones that actually show up on election day. Each polling agency creates its own predictive model to identify the right mix of voters to poll or interview to come up with the most accurate result. To create these models the polling agencies turn to prior elections giving the most recent election turnout numbers the most credence. Needless to say, creating an accurate model is more art than science.
During this election season all of the polls except Rasmussen based their models largely on the 2008 election. In some ways this makes sense, the most recent human behavior is most often the best predictor of current behavior. Until the “Disaster In Denver” debate more than a week ago, all of the mainstream media were reporting a dominating lead by President Obama, especially in the key “swing” states. I think that all of these polls had a fatal flaw causing them to substantially over report support for President Obama.
As the November 2008 election approached, the Lehman Brothers collapse had virtually assured the election of Barack Obama. An electric excitement shot through the electorate as the possibility of the election of the first black president transformed into the probability of the election of the first black president. Those on the left who had openly mocked the pratfalls of President Bush were enthralled with the erudite and charismatic candidate Obama. Turnout among young and first time voters were at unprecedented levels during the 2008 election (youth turnout) as were voter turnout rates for blacks (black turnout). Both of these groups provided candidate Obama with almost monolithic support.
Flash forward four years to a floundering economy, the Middle East in chaos, and a federal budget deficit growing at an alarming rate and President Obama is facing a very different set of circumstances than he faced four years ago. Ironically, this backdrop is very relevant to polling models, not because it will affect how voters may or may not take these factors into consideration when responding to the polling questions but in determining how many people will come to the polls. Lacking the excitement and enthusiasm palpable four years ago, it is reasonable to predict that the youth and black voting populations will not support President Obama with the same intensity that they did four years ago with many staying home instead of voting. Conversely, President Obama’s “war on wealth” has energized a conservative community that was beaten and bedraggled four years ago. None of the polls have taken these factors into consideration and almost all of them, using the models from the 2008 election, have polled 15-23% more Democratic voters than Republican voters incorrectly tilting the field in President Obama’s favor.
In addition, almost all of the reporting agencies fail to explain the real and accurate meaning of the polls. Almost all of the new agencies will issue a report much like the following: “President Obama has a commanding three-point lead in Florida’ or “Governor Romney has taken a two point lead in Michigan.” Both of these reports are completely inaccurate because of the poll’s “margin of error.” As we have discussed above, a poll is created by getting input from a small number of the voters and projecting a result based upon that sampling. A poll’s “margin of error” takes into account the sample size and reflects the potential variation between polling results and predicted “real” results. For example, if a poll has a 3% margin of error and the poll shows that Romney (or Obama) leads 48-47% that poll result really means that it is still a toss-up because the results for the two candidates are “within the margin of error.” For a poll to really show a leader, the polling results have to be greater than the margin of error. In other words, if a poll has a 3% margin of error (probably the most common margin for error among presidential polls during this election), one candidate must be leading by more than 3% to have a real and statistically significant lead. Very few news agencies explain this because it would turn an exciting headline (Romney Leads) into an unclear and far less interesting although far more accurate headline (Romney Leads – but within the margin of error so you really can’t tell who is leading [not nearly as satisfying]).
Finally, polls can be used and abused by each campaign to create an impression of “momentum.” Momentum is really irrelevant in an election because unlike a sporting event, all of the action takes place in secret until the election is over. However, if a campaign can create the illusion of momentum than some voters can be persuaded to jump on the bandwagon and vote for the “leading” candidate thereby confirming the “reality” of what is really an illusion.