Presidential Election Campaign: Surveys

From this point on in the 2016 Presidential election campaigns, we will be hearing or reading about more and more survey results from the media that ask respondents which of the (typically) two candidates the respondents plan to vote for in November. 

The results of such surveys are generally poor predictors of the election results, especially early in the campaigns, for two reasons, in particular.

1. What part of the population’s plans to vote are tallied by the surveys?

a. All potential voters, registered or not?

  

b. All registered voters at a specific point in time, which may change by election day?

c. Currently registered voters plus currently unregistered voters but who are “extremely” likely to register (which better reflects the actual electorate closer to election day) but may not , in fact, vote?  

2.  The survey percentages for each candidate that are commonly reported by the media are based on national random, but adequate, samples of the population. These samples are small but statistically adequate and, typically, only hundreds of voters.

a. Thus, such predictions are not “politically” useful in the sense that, as small national samples, they do not provide the basis for state electoral predictions.  Relatively small national samples are inexpensive because of the size of the survey.

b.  On the other hand, statistically significant state by state surveys  are necessarily larger and therefore more expensive. Such state surveys appear in the media only later in the campaign.