Who gets your vote in 2012?

Barack Obama (BO)	Mike Huckabee (MH)
Email:

<< Previous | Next >>

Sep. 9, 2007

Various Sampling Methods Used by Polling Organizations

There are many different kinds of polls; some are scientific while others are not. Scientific polls are generally much more reliable than nonsceintific ones because error analysics can be applied to them. Nonsceintifc polls include straw polls and other related polls in which response to the polls are voluntary. An extreme example of a nonscientific poll was a voluntary magazine poll conducted by the Literary Digest in 1936 in order to try to predict the winner of the Roosevelt vs. Landon presidential election. The poll was mailed out to 10 million subscribers, 2.4 million of which responded. The poll predicted that Roosevelt would receive just 43% of the popular votes among the two major candidates in that election, wherease he actually received 62%! This shows how wrong a voluntary poll can be.

In scientific polls, the sample being polled is chosen at random from the population in question. Some random polls are purely random, meaning that there is an equal chance of anyone in the population being polled. On the other hand, some polls are not purely random in this sense. Some scientific polls take into account demographics of the population being polled, such as location, race, and gender. The advantage this has over a purely random poll is that it may be possible to reduce the margin of error for a given sample size by taking demographics into account. To see this, consider an extreme example. Suppose 100% of voters living in the eastern United States support candidate X, while 100% of those in the western United States support candidate Y. Assume that the eastern and western United States contain the same number of registered voters. If we perform a purely random survey of say 400 American registered voters to determine the percentage supproting candidate X, we are likely to obtain close to the true value of 50%, but the value we obtain will probably not be exactly 50%, due to the fact that we will probably not pick exatly 200 people form the eastern United States for our survey. In fact, the margin of error for this poll turns out to be 5%, meaning that there is a 95% chance that the percentage of reginstered voters polled who support candidate X, i.e. those living in the eastern part of the country, will be between 45% and 55%. On the other hand, if we instead insist on polling exactly 200 people from the east and 200 from the west, we known in this case that we will obtain exactly 50% in support of candidate X, with a margin of error of zero.

Although the poll described above is an extreme hypothetical example which would never occur in practice, the effect of demographics is nevertheless significant and can in practice reduce the margin of error. For instance, in conducting a national poll of Democratic vs. Republican candidates, it makes sense to select the correct percentage of voters from urban areas, since they are much more likely to vote Democrat than those from rural areas.