There is an old joke about three pollsters out quail hunting. The first one aims and fires high. The second aims and fires low. The third one exclaims: "we got 'em!". This year more money is being spent on electoral polling than ever before yet the polls are providing more questionable results than ever. There are three good reasons to distrust all the polls.


Random sampling, which represents the logic of current polling methodology, has a good history of statistical accuracy when measuring issues. Social scientists have been using this type of polling data to gauge public opinion on a whole range of opinions about issues and ideology and have been doing it accurately for a number of decades. Unfortunately, this does not hold true for election polls because the overall population is unknown. A sample of 3000 people out of the 300 million population will provide fairly consistent results regardless of what 3000 people are polled (within the parameters accepted by social science). The problem with election polling is that the overall population is unknown. It is not 300 million. It is not the 121 million that voted in 2004 because new registrants are a huge unknown, both in their number and their likelihood of actually voting. The best methodologists can do is estimate the voter total (population) with about a 3% margin of error. When this margin of error is factored in to the 3-4% polling margin of error it is easy to see why polls can produce dramatically mixed results. A particular poll can easily be off by 9%. Even averaging all the polls will still provide a margin of error that makes any close election a mystery come election day.

The second reason why the polls may be less accurate than most would like is because of the increasing number of variables in who is being polled. Whether it is cell phone usage, the time of day the polls are taken, the role of race, the simple number of times people are being polled or the reluctance of people to reveal their true thinking in a more turbulent world, polling data is gradually becoming less accurate. This is being demonstrated by the wide range of results in all the numerous polls being conducted and is likely not to be corrected any time soon because of the third factor.

The third factor is polling costs. The population has grown by nearly 80 million in the past 20 years and voter turnout has also been on the increase in the past few elections. This requires pollsters to increase the number of their samples in order to maintain the same margin of error. Because of the increased costs of conducting polls (no one has tried using Indian call centers yet) the polling samples have not kept pace. It is common to find polling samples of 1200 given a 3.5% margin of error. This was true 20 years ago. It is no longer true now. However; pollsters still use these numbers when disseminating their results because publishing data with a 6-7% margin of error looks ridiculous. Because of the demographic changes pollsters should be increasing their sample size by about 30% in order to get back to the 3-3.5% margin of error they are reporting. Because of the costs this is not likely to occur and the end result is that more so than ever before, the election polls cannot be trusted.