See also Polling Methodology
Polling is the act of putting one or more questions to a representative group of people and extrapolating the results to represent a larger population. For example, one might poll 1,003 American citizens on their views about chocolate ice cream, asking each of them "Do you like chocolate ice cream?" If you received 530 "yes" responses, you would conclude that "53% of all Americans like chocolate ice cream".
There are two main questions posed by making such a statement.
- How can we be sure that the views of 1,003 people reflect the views of the larger population?
- How accurate is the figure in the statement?
The answer to both questions lies in examining the methodology of a poll. Most polling firms follow a statistical model which gives much of the answer to both questions. First, the polling firm identifies the universe they wish to target. This could be, as in the example above, "Americans". In political polling, the universe is often "all voters", "registered voters" or "likely voters". Second, the polling firm then picks a random sample of this universe and polls them, the model provides a calculation of the margin of error. In other words, the mathematical model guarantees that the sample reflects the whole universe and gives an estimate of the maximum possible error created by the sampling method.
While this is the method used by the vast majority of polling firms, there are some notable exceptions. For example, some polls are "self-selecting", meaning that anyone who wishes to answer can do so. Virtually all internet polls, such as the polls on CNN's main page, are "self-selecting". These polls obviously do not create a random sample of the overall population and cannot be used as a statistically accurate reflection of the general population. Indeed, these types of polls can be easily manipulated (see Freep, Torture Wolf!).
In addition, there are other parts of the methodology that can introduce error into the results of the poll. Almost all polling firms do some weighting or other massaging of the raw data according to their models. These may include a closer matching of demographic data or adjustments meant to mitigate factors that reduce the effectiveness of the polling firm's randomization schemes. These methods are usually proprietary and kept secret. Error can also be introduced through the wording of the questions. Poorly worded and/or misleading questions can easily yield inaccurate or, in some cases, utterly useless results. For example, polling Americans with the question "Is Bill Clinton a crook, a thief, or a liar?" is unlikely to render a realistic perspective on American attitudes toward Bill Clinton. Likewise, "Do you support the President in his efforts to defeat terrorism?" will likely get a different response from "Do you think George W. Bush has taken the right approach toward making the world safer?" The most extreme case of biased phrasing of questions is found in push polls, in which assertions about a candidate or issue are given in the form of questions, under the guise of seeking opinions on these "facts".
- American Research Group
- Democracy Corps
- Fako & Associates, Political Polling (D)
- Netroots Research, Strategy and Analysis
- Opinion Dynamics
- Program on International Policy Attitudes (PIPA)
- Gallup Organization
- Rasmussen Reports
- Polling Report, Inc. - collection of polls
- Professor Pollkatz's Pool of Polls
- Pew Research Center For People & The Press
- Pew Internet and American Life Project
- Program on International Policy Attitudes
- 20 Questions a Journalist should Ask about a poll before reporting it.
- A Consumer's Guide to the Polls by Slate