THE PROBLEM WITH POLLS

Frank, Mitch (2004) “The top line on polls” TIME, 11/1, p.
20.

You have a good poll if EVERY VOTER has an equal chance of being chosen. Most polls stop at around 600 which mean that 95% of the time, the results will be accurate with a sampling error of 4%. As an example, lets say that final-real results is Smith gets 70% of the vote and Jones gets 30%. With 4% error, Smith could get as high as 74% of the vote or down to 66%. Jones could get as high as 34% or go down to 26% of the vote. Since the race is not close, we don’t need any larger numbers. However, if the race is real close, 600 voters is not enough and 4% error is too risky to predict the election.

Usually, pollsters will use small numbers of voters throughout most of the election because the costs are so high to contact lots of people by phone. However, the LAST POLL taken right before the election when it is close will go as high as 5,000 in the sample with an error of less than 1%. At this time, ZOGBY appears to be the most accurate.

Pollsters continue to have problems. Here are some of them. They are:
1. Newly registered voters may not be counted.
2. Defining “likely” voters is troublesome.
3. Some groups do not have phones or use cell phones exclusively.
4. In the age of telemarketing, there is a high refusal rate.
5. Some groups are over-represented those that are underreported are then weighted so that the sample is more represented. Different pollster use different strategies. Zogby weights for political party. Most pollsters predicted that Bush would win the 2000 election. Zogby predicted Gore and was right, although the then vice president lost in the Electoral College.

At one time pollsters came to your door. That doesn’t work anymore, so we won’t go there.