How does a telephone poll of 600 people accurately predict how 600,000 people will vote? The answer: random sample probability theory, computers, and a polling process.
For that New York city poll of 600, a random sampling computer program creates tens of thousands of telephone numbers of land line and cell phones with area codes from New York City, such as 212, 718, and 347. Then as many as 200 pollsters call the phone numbers trying to reach people.
Most numbers are no good: half the people are not home and many numbers are no longer active. Eventually some 1,200 people are interviewed. But another randomization is done on each call so that adults are reached, said Micheline Blum, the director of Baruch College Survey Research. After pollsters complete their questioning, some 600 likely voters are tallied. The results are compared to census data to ensure there is an overall match to numbers. Polls are generally accurate, though.
One recent pre-primary poll made a big mistake predicting John Liu would get no votes in Queens, his biggest base. The reason may be that polling Asians is expensive since more interviewers must be hired to speak many different Asian languages including Chinese, Korean and Urdu, Blum said.
Polls are an accurate statistical snapshot of how people feel at the moment, although they do have greater impact, according to political science professor Doug Muzzio.
Poll results can then influence media coverage and public perception, which then can influences the polls results.