First,
the Survey Center contracts a company to put together a list of random numbers in New Hampshire using a process called random digit dialing. According to Smith, "every household telephone number in the state has the chance of being selected."
Problems arise however because some people do not have a landline phone:
Smith said these people make up such a small part of the voting population, about 3 percent, that it has very little effect on polling outcomes. He added that this is particularly true in New Hampshire where cell phone reception is spotty to non-existent in some regions. Also, people who only use a cell phones usually are people in the demographics that vote the least, like young people and people in transit.
By federal law pollsters cannot call cellphones. This is a problem that will only get larger. Smith downplays it now, but he doesn't really have any choice but to downplay it because there are not any other options.
After weeding out disconnected and business numbers, the staff begins calling. If a person is not home, Smith said the center calls the number up to eight times trying to reach someone. If the center does identify a possible subject at the number, the staff will attempt to reach that person eight more times.
"We really work hard to get the randomly selected households," Smith said. "We don't just go for what's easy."
Because statistics show that certain segments of the population, like women and senior citizens, are more likely to answer the phone, Smith said the center asks to speak with the person with the last birthday in the household. This way, Smith said, the staff can get the most random sample possible.
After going through all this trouble to follow true random selection, you don't have to get a huge sample for the poll to be accurate:
The Survey Center usually interviews about 300 to 400 voters, depending upon how much the client is willing to spend. This takes three to five days. Smith said he's found that after 100 or so responses, the poll results don't tend to change that much.
But that's not the end.
He then weights the polling sample depending upon the demographics of those who answered. For instance, about 60 percent of Democratic primary voters are women, so the answers from women voters might get a slightly higher weight in that poll. Geographic location of the subjects' homes and past polls also factor into the weighting process.
Weighting has always troubled me. It seems to presume that our voting choices are determined by demographic characteristics. It reflects the built-in conservatism of polls, by smoothing out the result with past results. After going through so much trouble to get a random sample, the proportion of the sample is then made to be "correct" according to the census demographic account. Since it smooths out different poll results to decrease the differential between them, it makes polling look more scientific, because the only check we have on polls to see if they are correct is other polls. There is no real measurement to see if a poll is baloney or not, except election day. And only the polls taken right before election day can accurately predict that result.
This brings us to the last topic of interest in the article, how to weed out non-voters.
Some polling organizations may use all adult residents or all registered voters as a sample, but Smith said he has found using these sample groups doesn't always reflect an accurate picture of the voting population and he prefers to find the likely voters.
"We're interested in the people who are actually going to show up," he said. To that end, Smith said he has developed an effective method to determine likely voters.
By asking questions in a certain order, Smith said he can get a better picture of who is just saying they will vote and who is more likely to show.
First, the polling staff asks if the person is registered and under what party. If a person is not registered, the chances they will vote go down slightly, but because residents can register on Election Day, this answer does not exclude anyone.
Next, the subject is asked how interested he or she is in the election. If a person is not registered, but is very interested, that person is more likely to register later and ultimately vote, said Smith. If the person is not interested however, the chances that person will vote again goes down.
Finally, the polling staff asks the subject if he or she is going to vote. Smith said he's found that if you ask people initially, ‘are you going to vote?', they almost always say yes. However, if someone says he is not interested in the election and isn't registered, chances are he will admit he wasn't planning to vote either, said Smith."We want to encourage people to tell us they are not going to vote," Smith said. "[The method] gives people a context to drop out."
No comments:
Post a Comment