Opinion polls play a major role in election campaigns although it is not clear whether they are used to measure voting intentions, or to move the vote. An early reliable indication of the election outcome could play a big role in changing voting behaviour, and misleading polls could be used for the same purpose. Just how reliable are opinion polls?

Andrew Heard, Associate Professor in SFU's Political Science Department, provides an excellent website with polling results and links to articles on polling. Heard provides a link to a paper on polling by Matthew Mendelsohn and Jason Brent at Queen's University. They offer a rough rule of thumb for calculating the often quoted margin of error for a poll. Their formula is equivalent to dividing one by the square root of the sample size for a 95% confidence interval. For example, if the sample size is 1,000, the square root of 1,000 is 31.6278, 1/31.6278 is 3.16% for the margin of error 19 times out of 20. This is only an approximation, but it is a lot simpler than the full formula, which can be found at the Free Dictionary's website. The full formula makes it clear that the error increases the greater the difference between the percentage measured in the sample and 50%.

It is important to note most polling organizations cite the margin of error for their overall poll, but not for the regional or other breakouts. Those smaller samples have much higher margins of error. A sample of 132, the sample size used in most BC polls up until now, would have a margin of error of approximately 8.7% 19 times out of 20. That large of an error makes the polling results for British Columbia consistent with statements by many who say it is a three way race here. On June 21, the Mustel Group released a BC poll with a sample size of 721. It had 10% undecided, and a margin of error of 3.6% 19 times out of 20. Province wide, it put the Conservatives at 36%, the NDP at 28% and the Liberals at 26%. It noted that on the issue of tax cuts, BC "voters prefer to maintain taxes and government services at the current level (49%), than reduce taxes and governmetn services (26%), or increases taxes and services (16%), with 9% undecided."

All the probability theory behind polling is worth nothing if the poll is poorly conducted or badly worded. Technology may be changing the accuracy of polling; call display and answering machines may result in distortions to the random nature of any sample. It is also the case that some people, especially in the heat of an election campaign, will not necessarily provide honest answers. A common sense way to judge polling accuracy is to look at as many polls as possible and to consider the changes between polls over the course of the campaign. Heard's website provides a table that makes that comparison easy at the national level. It is probably safe to say that so far on the national level over the course of the campaign, the Conservatives and the Liberals are close to each other but the Liberals have fallen while the Conservatives have shown gains; the NDP has been fairly steady. It is also safe to say that much more detailed information is necessary to make even a wild guess on the outcome of 308 seats.

The Laurier Institute for the Study of Public Opinion and Policy has a "model" for projecting seats using the results from several polls. On June 10th it predicted that the Liberals would win 12 seats in British Columbia but they reduced their projection to 10 seats on June 15. It would be hard to find anyone, other than a Liberal candidate or campaign manager, who could say that without breaking out laughing. Most people believe that the Liberals will end up with fewer than the 6 seats they held when the election was called.