When we do a California telephone survey, the Los Angeles Times Poll makes calls to a random sample of all phone numbers in the state. This technique is called "random-digit dialing" and is the best method for producing good predictive samples in a poll. Because of this, the people who answer our questions are in very similar proportions to that of the adult population of the state. For example, if 28% of the state's adult population is Latino, then our sample will have close to that proportion of Latinos. We look at several variables such as race, age, educational attainment and location of the household and use weights to make sure those proportions conform with data from the Census Bureau about California's adult population.
When the Times Poll conducts a pre-election survey such as the one published Aug. 23 and 24, we use a model we have developed to help us determine who is likely to turn out to vote. We call those people "likely voters." Since 50% to 70% of registered voters actually turn up at their precincts to cast ballots in any given race, this is an important part of our analysis. The "turnout model" is based on a person's interest in the election and their past voting behavior, among other things.
Once we've identified the smaller group of "likely voters" in our larger sample of all adults, it is quite possible that some important subgroups will contain too few respondents for us to feel comfortable reporting their opinions as a group. The Times Poll will not ordinarily report statistics about the opinions of any subgroup unless it contains at least 100 respondents.
In order to have, for example, enough Latino likely voters to be able to report their views, we call another telephone sample alongside the main one. This is called "oversampling." The second sample is another random sample of the state, but one in which we conduct interviews only with Latino respondents. The samples are then merged in such a way that the combined sample's demographic proportions again mirror the adult population of the state. By using these statistical techniques, we end up with both a random sample of the state and enough Latino likely voters to be able to report their opinions in the newspaper.
Random-digit dialing is a rigorous and expensive way to conduct election polling, but the Los Angeles Times Poll, like other reputable polling organizations, is committed to conducting the highest quality survey research. For this reason the Times Poll conducts its data collection using an in-house phone bank so that we can be assured that data collection is constantly monitored and conducted in the most professional and exacting manner possible.
There is a lot of cynicism in the general public when it comes to polls, and people are right to be skeptical. Many sets of published numbers call themselves polls when they are nothing like one. Some media outlets are not savvy enough or don't care enough to avoid reporting these non-polls, which adds to the general confusion. The Times reports the results of only those surveys that meet the guidelines of the American Assn. for Public Opinion Research (AAPOR).
Other sources of information
See the Times Poll's .
The AAPOR Web site is a good resource for information on industry standards.