Poll articles do not have to be boring accounts, solely listing percentages in favor or against any measure or politician. Every poll article must remain neutral, be reported in context of current events, include the most important findings of interest to readers, translate statistics clearly (e.g. one out of two respondents said ...) and organize the significant findings for reader comprehension. Bullets help outline key statistics. The lead of any poll story should summarize the most important findings and attribute the source of the poll. As the inverted-pyramid style story continues, reporters discuss findings and technical details. A good poll story will compare results from previous polls that asked the same questions of the public. Reporters ask political observers to comment on poll results, which provides context to the numbers and explains why certain demographics voted the way they did. Each poll can be treated as just another source.
But that's not the end of writing a good poll article; help your
audience more by making your article meet AAPOR standards.
The American Association for Public Opinion Research's Standards
of Minimal Disclosure specifying the information that pollsters should include
when they release poll results to the public.
The AAPOR Standards obligate members of the organization to adhere all of
items listed when they report any part of their research to the general
public. This precludes partial release.
 Who sponsored the survey. Obviously we would look differently on a poll sponsored by the National Rifle Association than one sponsored by Citizens for Handgun Control. Even though a given poll may be scrupulous in its methodology, the sponsoring organization still decides which questions to pose and thus can shape what gets reported. Also, when results are not favorable to the sponsoring organization, that organization can decline to reveal them at all. Who conducted the poll. Established pollsters put their reputations on the line when they report results, and we're familiar with many of them. If we haven't heard of the polling organization, we have a right to be skeptical. The exact wording of the questions asked. If there's a single piece of information that we need to evaluate poll results, this is it. Pollsters know that even minor variations in the phrasing of a question can alter results significantly, and highly loaded questions, by several percentage points. If we don't know the wording of questions, we can't really interpret the results. Text of any preceding instruction or explanation to the interviewer or respondents. The context in which a question is asked certainly can affect its answer. If for example, respondents are told: "As you know, Congressman Smith has been linked to a Washington D.C. call girl network," before we asked if we're planning to vote for him, it we may change our answers. Of course, reputable pollsters don't do anything that crude, but the principle is important. A definition of the population under study. To a pollster the population is the group from which the sample is being drawn. Of course, it makes a difference if the populations is "adults over 18," or "adults over 21; or "registered voters" or "likely voters." A description of the sampling frame used to identify the population. The sampling frame refers to the list or procedure from which the sample was drawn. Examples would be voter registration lists, or telephone numbers derived by random digit dialing. Information on eligibility criteria and screening procedures. To generalize to a specific population, pollsters often need to select respondents. For example, it may make a difference if a respondent is designated a likely voter on the basis of a single question, "Do you plan to vote," or on the basis of a battery of questions. The method by which respondents were selected by the researcher, or whether the respondents were entirely self-selected. Volunteer samples such as those done by phone-in surveys can easily be biased. They can be generalized only to the population of people who decided to call in and we don't know how that populations compares to others. There are documented cases in which people made multiple phone calls to alter poll results. Size of samples. Although they can provide qualitative insights, interviews conducted with only a few people can't provide percentage estimates that we can count on. But samples of a few hundred can provide such good percentage estimates. Well drawn samples of moderate sizes give pollsters access to mathematical tools that allow them to estimate how close the estimates are. A discussion of the precision of the findings, including, if appropriate, estimates of sampling error. Sampling error is the degree to which percentage estimates could be expected to be different if the researchers had drawn a different sample than the one under consideration. Usually these errors are expresses something like, "We can be 95 percent certain that the reported figures this sample are accurate within plus or minus 3 percentage points." Some commentators have argued that such phrasing is too technical for most people to understand and implies more precision than survey research can really claim. A description of any weighting or estimation procedures used. Occasionally researchers will observe that their samples do not match known population patterns and will weight or adjust the poll results to compensate. For example, a sample might have a larger proportion of retired people than the population so the pollster would weight the responses of non-retired people more heavily. While such adjustments can provide more accurate figures, they can also cause problems. Which results are based on parts of the sample. When results on subsamples are reported, numbers can become quite small. There have been cases reported in which statements like "Three quarters of the Native Americans in the sample reported ... " was actually referring to only four respondents. Method, location and dates of data collection. The method of data gathering — telephone, face-to-face, or mail — doesn't make a lot of difference in poll results, but it should be reported. The location of interviews can make a large difference. For example, we expect response on questions concerning social mores to be more conservative in the South than in the Northeast. Dates of data collection — particularly in elections — can make an important difference. We certainly need to know when interviews were conducted to assess the value of an election poll.
(Image used here under Educational Fair Use).
If you don't understand something in this Web note, please e-mail
Dr. Sitton.
©Ronald W. Sitton 2009 Revised 102209 — http://www.uamont.edu/FacultyWeb/sitton/crz/edtn/aaporstand.html |
