Statisticians working in all areas of science know that you can describe a
population's characteristics by  describing a sample drawn at random from that population.

Foresters, for example, might want to know the amount of lumber in a large tract of land. So measure a random sample of trees in that tract to  estimate the board feet for the entire area so that they don't have to measure millions of trees. Doctors take a sample of blood, because they know they can test blood for diseases without the drastic and fatal step of taking all the  patient's blood for the test!

Survey researchers use the same techniques, but they sample people and measure opinions, attitudes, beliefs, and self-reported behavior. The Minnesota Poll can - within certain limits of precision called the "margin of sampling error" - project the information from its samples to the population it  is measuring. The Poll routinely surveys Minnesotans, but it also conducts polls in Minneapolis or St. Paul, and it does national surveys on occasion.

                 The samples used to measure population characteristics have lots of things in
                 common with other types of sampling that aren't representative. Doing a poll
                 of one's friends, even hundreds of them, isn't a representative sample of the
                 general population. Neither is a bunch of people responding to an Internet
                 survey, or political parties calling voter list for candidates all have things in
                 common with a scientific survey. These lack at least one thing, perhaps more,
                 that are present in representative samples: the element of randomness.

                Deciding on the sample size

                 Before researchers can do a random-sample poll, they first must decide how
                 many interviews they will need. Each poll is different, but generally between
                 600 and 1,200 interviews are needed. The larger the sample size, the more
                 precise one can be with the sample's estimates of the population, and the
                 greater the ability one has to analyze smaller groups within the sample - and
                 population - such as men and women, or Democrats and Republicans. If the
                 Minnesota Poll needs to find and sample people who are relatively scarce in
                 the population, then it will need to interview a lot more than 1,000 to find
                 those people. The poll has surveyed such special populations as pastors (it
                 used a mail survey), people who are chronically sick, parents and teen in
                 household in which there's at least one teenager living full time, and evangelical
                 Christians.

                Deciding what kind of sample

                 Once researchers have decided on sample size, then researchers must design
                 the type of sample that would be best to measure the population. For a
                 statewide Minnesota Poll, researchers generally use what sampling
                 statisticians call a "stratified area probability sample, proportionate to strata
                 size." In practical terms, that means dividing the state into strata - in most
                 cases counties - then drawing a random sample of potential residential
                 telephone numbers from each county. It also could use a simple random
                 sample or other more complex sample designs, but this design is a
                 commonly-used one that works well.

                 When the Star Tribune does a statewide poll, its researchers first ascertain the
                 adult population of each county. Interviewers complete the number of
                 interviews in each county that is proportionate to its adult population. For
                 example, 26 percent of the state's adults live in Hennepin County, Minnesota's
                 most populous; consequently, in a sample of 1,000, a Minnesota Poll sample
                 would contain 260 interviews in Hennepin County. Thus, each county
                 represents a small, independent sample on its own. For a poll with a total
                 sample size of 800, there would be 208 interviews in Hennepin County.

                Drawing a sample of telephone numbers

                 Once each county's sample size is determined, researchers must draw a
                 random sample of telephone numbers for interviewers to call. How's that
                 done?

                 North America has 10-digit phone numbers, which have an area code, a
                 prefix, and a suffix. Here's what a phone number looks like:

                 zzz-nxx-abcd

                 Area code-prefix-suffix

                 In this example, the zzz is the area code and the nxx is the prefix. The suffix
                 also has special numbers - the thousands (a), hundreds (b), tens (c) and ones
                 (d). The database containing all the residential area codes and prefixes for all
                 telephone exchanges in the state, and researchers update it regularly to make
                 sure that area code changes and new telephone prefixes are included in its
                 samples. The poll's telephone number database contains only the banks of
                 thousands that are working.

                 First, the poll's survey specialist instructs the computer to draw enough
                 telephone numbers at random to complete the number of interviews needed in
                 each county. Next, the computer enumerates the area codes and telephone
                 prefixes in the county, and selects one area-code/prefix combination at
                 random. After that , it creates a three-digit random number to finish out the
                 phone number. This last step creates the "random-digit-dial" (RDD) telephone
                 number that interviewers actually can call. Finally, the computer takes all the
                 RDD telephone numbers and puts it in an electronic file.

                 Conducting interviews

                 Telephone interviewing is hard work. One has to read questions exactly as the
                 researcher has written them, otherwise people who are interviewed (called
                 respondents) would hear different versions of the questions and might respond
                 to them differently. That would introduce bias into poll, and it would make
                 interpreting the responses difficult.

                 Consequently, good polling organizations take care to find, train and supervise
                 good interviewers. The Minnesota Poll uses interviewers at several market
                 research companies to conduct its polling, but the Market Solutions Group,
                 Inc., in Minneapolis, does most of the interviewing.

                 When it starts a new poll, researchers provide the interviewing company a
                 copy of the questionnaire, which programmers then load into a computer. This
                 computer (called a CATI system, for "computer-assisted telephone
                 interviewing") can dial the RDD telephone numbers for the interviewer and
                 supply the questions on the computer screen one at a time. It also records the
                 respondents' answers to the questions after the interview is completed.

                 But before that, interviewers are briefed on the questionnaire: Supervisors
                 point out when it will be done, what it's about and other things, and they go
                 over the questionnaire to show interviewers how questions should be read.
                 Interviewers practice with each other until they are familiar with the questions.
                 Interviewers also are provided scripts to read to respondents when they
                 encounter frequently asked questions. As a result of this and other training,
                 interviewers are ready to call when they go to their "CATI" station.

                 Interviewers are trained to do everything they can to keep response rates
                 high. They make appoints to call a respondent back if it's inconvenient to do
                 the interview at the time of the initial call. They leave messages on answering
                 machines identifying who they are and why they're calling; they even provide a
                 toll-free number for respondents to return the call. When respondents initially
                 refuse, a senior interviewer calls them back to give them another chance to be
                 included in the poll's sample. But if people don't want to be called back,
                 interviewers respect those wishes.

                Selecting respondents from within households

                 Interviewers call residential households at random, because randomness is the
                 basis of getting a representative sample of people. But randomness has to
                 apply to selecting the person to be interviewed once the household is included
                 in the sample.

                 The Minnesota Poll uses the "most-recent-birthday" technique to choose one
                 adult from each household to be interviewed. "Informants," someone in the
                 household who answers the phone, hear this script from the interviewer

                 "Hello, this is _________ calling for the Star Tribune Minnesota Poll. We
                 are not selling anything.Today we are asking Minnesotans some questions
                 about various issues . May I please speak with the person in this
                 household who had the most recent birthday and is 18 years of age or
                 older."

                 There are other ways to choose respondents, but the poll's researchers have
                 found that they are too intrusive, and result in too many people refusing to
                 conduct the interview. Consequently the Minnesota Poll has used this method
                 successfully for the past decade.

                Data analysis

                 Once the interviewing is complete, the interviewing company e-mails the data
                 set containing the data for all the respondents to the Star Tribune's polling unit.

                 The sample is weighted for age, gender and education, based on the 1996
                 Census estimates of the adult population. It also is weighted to take into
                 account factors contributing to unequal probability of selection, which are the
                 number of telephones going into the household and the number of adults in the
                 household. That way, the poll's estimates represent the opinions or attitudes of
                 the entire adult population - rather than for households. For election polls, the
                 data also are weighted for likelihood to vote.

                 How does weighting work? Let's look at an easy example - weighting only for
                 gender. We know that the adult population in Minnesota is about 50 percent
                 men and 50 percent women. But if a sample turns out to be 55 percent men
                 and 45 percent women, then it has to be weighted to make sure men count
                 for half of the responses and women count for half. That means, in this
                 example, men would have to be counted slightly less (0.91 to be precise) and
                 women would have to be counted more (1.11) so that the weighted data
                 would have half men and half women. (Do the math and see how it works:
                 0.91 * 0.55 = 50 percent men; 1.11 * 0.45 = 50 percent women.)

                 After the data are weighted, the analyst examines the results statistically and
                 writes a short report that he uses to brief reporters and editors about the key
                 findings. After the stories and graphics are written and proofed, the poll's
                 director and survey specialist scrutinize them again to make sure the numbers
                 and facts are right.

                Then you see it in the newspaper, and on startribune.com.

Disclaimer