Method
The StarLedger/EagletonRutgers Poll uses a randomdigit dial (RDD)
“probability” sampling design to select households for
participation in each of its quarterly surveys.
In lay terms, telephone numbers around the state are called
“blindly,” with due respect to their prevalence in the population.
The first “cut” is by area code. For example, if 30 percent of the
states 18+ population lives in the 201 area code, 30% of the sample is
drawn from that 201 code. The next cut is by county within that
parameter. For example, if 40 percent of those living in the 201 area
code are known (through census figures) to live in Bergen county, 12
percent of the StarLedger/EagletonRutgers Poll’s sample is chosen
from that county. [30% in (201) x 40% in Bergen = 12% of the statewide
adults 18 years of age and older.] This is done by sampling on
exchanges within area codes. Following this sampling procedure of area
codes and exchanges ensures a correct geographical representation of
residents of the state’s 21 counties.
The sampling of phone numbers has to this point been on the first 6
digits —the three digit “area code,” and the three digit
“exchange”— guaranteeing a proportional probability sample based
on county and townwithincounty distribution. The next four digits
making up the sample phone number are in effect randomly generated,
picking up listed and nonlisted number alike. Thus the RDD moniker
for “randomdigit dialing.” Respondents within the designated
households are also sampled, generally by means of a
“lastbirthday” method. Interviewers will ask to speak with the
person (18+) in the household who has had the most recent birthday.
This ensures a random selection of eligible respondents within the
households selected.
After the sample is selected and correct respondents interviewed, the
sample is “weighted” to correct for the fact that some respondents
are harder to reach than others. For example, less welleducated
respondents may be less comfortable talking on the telephone and may
“refuse” to be interviewed in greater proportion than
bettereducated respondents. At the end of the interviewing process,
the poll may have spoken with 40% of those who have a college degree,
when census figures tell us that only 25% of New Jerseyans have
attained this much education.
To make sure the sample is representative, each of those with a
college degree must be counted as .625 of a person by the computer
(40% x .625 = 25%) to make sure their prevalence in the sample
actually equals their prevalence in the population from which the
sample was selected. Those with less than a high school education may
be counted as more than 1.00 of a person by the computer. (Assume that
8 percent of the sample say they have less than a high school degree,
while the true population count is 12 percent. Each person in this
group must be counted as 1.5 for the sample to be truly
representative.) A common StarLeger/EagletonRutgers Poll weights on
age and education to adjust the sample to comport with current
censusprojected figures of the New Jersey adult population. This
weighting ensures that findings from a sample can be generalized to
the larger target population of all New Jersey adults.
All surveys are subject to “sampling error”—that is the expected
probable difference between interviewing everyone in a population, and
a scientifically selected sample of that population. Sampling error is
based on the size of the sample. The following chart shows the
relationship between sample size and sampling error.
By way of illustration, if the sample contains 400 respondents, the
associated sampling error is plus or minus 5 percentage points. [This
is true for percentages near 50% at a 95 percent confidence interval.]
If 50 percent of the sample percent were found to give the Governor a
positive job rating, one could be 95 percent confident that the true
value among all New Jersey adults in the population would be somewhere
between 45 and 55 percent (50 + 5), with 50 percent being the most
likely figure.
Sampling error does not take into account other sources of variation
inherent in public opinion studies, such as nonresponse, question
wording or context effects. The StarLedger/EagletonRutgers Poll
makes available its questionnaires online for those who wish to see
the questionnaire wording and sequencing. Response dispositions are
either available online or by request.
