WHY THE REF'S AVERAGES ARE BETTER
Only one poll can throw off a poll average significantly. Other poll averages possess five flaws that allow biased polls to distort the average.
- Other Averages Include Extreme Outlier Polls
- Other Averages Include Polls that Offer No Indication of their Reliability
Allow me to explain . . .
FIRST, and most importantly, other poll averages usually include polls of all adults in their averages. Political polls by definition should not include the responses of all adults. Political polls should survey only those who either intend to or might vote. Polls of likely voters represent the most accurate surveys, although polls of registered voters also tend to more accurately reflect the mood of the electorate than simple polls of all adults.
When a poll average includes all adults the numbers tend to skew dramatically in one direction or another as compared to polls of registered or likely voters. Some polling organizations seem to poll all adults when the organization desires a certain outcome. At polticalref.com we have no interest in advancing the various causes of polling or media organizations and as a result, we exclude polls of all adults. We will report the numbers to you because these polls accurately depict the mood of all Americans, but on political matters only polls of potential voters should matter.
SECOND, politicalref.com does not include extreme outlier polls that have a demonstrably and deeply flawed polling sample. Examples of such polls include a heavy oversampling of either party, Independents, or a demographic group. On occasion a polling organization will attempt to make a splash by releasing a poll showing a big shift in a race that is really the result of a heavy oversampling of one of these groups. We rarely exclude a poll for this reason. The oversampling must be demonstrated either through the internals or by an objective and well-reasoned article.
THIRD, politicalref.com does not include polls in its averages that offer no indication of their reliability. Polls usually offer their sample and margin of error as indications of their reliability. The sample indicates the number and type of respondents surveyed by the pollster. The larger the sample the more reliable the poll. Also, if a sample includes all adults the poll is less reliable than one of registered or likely voters. A margin of error, or amount of random sampling error in a survey, indicates the level of reliability of a poll. The larger the margin of error, the less reliable the poll. If a pollster refuses to offer a margin of error or a sample one has no ability to determine the reliability of that poll. One might argue that we should give reputable polling organizations the benefit of the doubt, but such an approach defeats the purpose of this website. Very often bias within the media affects coverage because of this trust the source mentality. We at politicalref.com are dedicated to ferreting out media bias.
FOURTH, the Ref's Poll Averages are superior to other averages for a number of reasons. Other websites that compile poll numbers make several crucial mistakes. They often leave out the margins of error of a poll. While they may report the number and type of respondent polled they often exclude the margin of error, or a number that represents the amount of random sampling error in a survey.
The larger the margin of error, the less confidence one should possess that the numbers in the survey represent the truth of the matter addressed by the survey. If a poll average fails to show the margin of error one cannot assess as easily, or at all, the reliability of the average. At politcalref.com we provide the margins of error of each poll averaged and an average of all margins of error included in the overall average.
FIFTH, politcalref.com does not include Internet polls. Partisan interests can, and seemingly do, manipulate the results of "scientific" Internet polling. Internet polling, in this context, does not refer to questions posted on the Internet to which anyone can respond. Rather, Internet polls in this context refer to those polls that attempt to control the sample. These polls ask respondents to identify their party identification, a series of questions designed to determine their likelihood of voting, and other questions also asked in telephone polling.
The problem is not with the methodology, but with the security features designed to prevent individuals from being counted multiple times. Internet polling usually only requires a unique email address to substantiate that a respondent has not already participated in the poll. The polling organization will invite respondents to participate in a poll through email. Polling organizations might ask for a zip code or an address to prevent redundant participation, but a respondent attempting to manipulate the results can provide false information to defeat this security measure.
As of now, polling organizations using the Internet have not perfected the process. Perhaps once they require proof via a mailing requiring a response to the address provided by the respondent, the process will be more legitimate. But if an organization goes to these lengths the efficiency offered by the Internet does not exist.
The Ref