Blog For Arizona

Predict This!

We have all become familiar with how influential polling has become in American politics. The 2000 Presidential election was plagued by whipsawing and inaccurate polling figures. The breakdown in the VNS exit polling system prompted the premature call of Florida by the Networks, first for Gore, then for Bush. The expectations engendered thereby helped to cover Bush's sharp elbows and distract from the blatant election fraud by members of JEB's Administration. Concern for the accuracy of polling results used by the media, and both an understanding and a healthy skepticism for polling results are crucial issues for our democracy.

The most common (because they are cheapest to conduct) and yet the most notoriously inaccurate methodology of commercial polling is automated auto-dialer polling. Essentially a computer selects numbers randomly (though it is not all that random), dials them, and presents the person who answers with recorded poll questions to which they respond with the phone's touchpad or by voice recognition. These polls are conducted cheaply by companies like SurveyUSA for local news broadcasts and newspapers. Because of the lower cost such polls are often purchased by campaigns seeking private polling and are often the only polls available for local elections. But sloppy sample framing and the complete lack of direct interaction with those polled cause experts to state that,"there is no sound theoretical basis for the way in which these surveys are conducted." Indeed, the Associated Press and many established political publications have standing policies against publishing the results of automated polls.

Several factors weigh against the accuracy of automated polls. They rely on landline telephonic contact between certain hours and on certain days. Dialing list generation is often farmed out and uses a selection of rural and urban exchanges with random generation of only the last four digits for dialing. As exchanges become unevenly populated with valid numbers due to an increasingly mobile population, the exchange selection balance can seriously skew the population framing. Over- or under-representation of certain geographic areas may result and easily skew an automated poll. As more households go entirely cellular and drop their home POTS line, the sample fails to include these demographic groups. Also excluded are a disproportionate number of minorities, low-income, and older Americans who do without a home phone. The hours and days of operation for dialers, generally between 6 and 9pm and all day on weekends, exclude yet more who are unlikely to be home during these hours because of work or recreation. The lack of human interaction means that people who have difficulty with English, or make mistakes in data entry, cannot be assisted. Minors and others ineligible to vote are less likely to be screened out, polluting the data. All of these factors tend to make automated polling inaccurate, unpredictable, irreproducible.

Arguably, no information is better than wildly inaccurate information. The cost of properly controlled and framed polling, though greater, is well worth the cost to news organizations who wish to inform their audiences, not mislead them, and to insulate themselves against charges of bias or lack of professionalism. A recent example illustrates the potential for confusion: during the run-up to the Australian national elections, automated polls published by different national newspapers indicated a cumulative 40 point whipsaw between the major parties in a span of only days- an obvious impossibility without major events intervening. Despite terrible results and professional and academic censure of automated polling, the market continues to grow. SurveyUSA boasts 50 local television stations as clients. That is a lot of potential havoc waiting to happen.

If you are concerned about the state of our democracy, and about the potential for electoral manipulation and misleading of the public these polls represent, demand that your local news outlets create a policy of not publishing automated polling results, or clearly label such polls and publish prominent disclaimers as to such poll's accuracy in addition to the standard Margin of Error and Confidence Level information.

posted by Michael Bryan @ 1/17/2004 08:07:00 PM