The Carolina Poll

1
The Carolina Poll

• A statewide, professional quality telephone
  survey conducted by the School of Journalism and
  others for more than 20 years.

2
What polls can measure

• Narrow behavior choices (like voting)
• Thoughtless opinion
• Salient issues (support/oppose)
• Simple evaluations (good/bad)
• Recent, specific behavior
• Reasons for behavior (sometimes)

3
What you can do with polls

• Estimate population parameters.
• Locate and describe certain kinds of individuals.
• Explain individual or social behavior.
• Test hypotheses about how social system operates.
  
• Measure influence of the media.

4
Principles of polls

• A relatively small, carefully drawn sample can
  come close to a very large population.
• The error associated with the sample estimate can
  be measured.

5
Logic of sampling

• We draw random samples of large populations and
  measure interesting characteristics or
  statistics.
• We use the sample statistics to estimate
  parameters of the large population.
• We can calculate the probability of error
  associated with our estimates.

6
The famous sampling error
1.96
p q
n
Where p percent q 100 p n sample size
7
Probability applied

• A random survey of 600 found that 40 support
  Issue A.
• Our best guess of the entire population is that
  40 support Issue A. But probably not.
• We are 95 certain that 36-44 of the population
  support Issue A.
• How do we come up with that?

8
Calculations

• p 40 q (100-p) or 6 n600.
• p q 40 60 2400
• 2400/600 4
• v4 2
• 1.962 4
• Population estimate equals sample figure 4.

9
Really important ideas

• Accuracy of sample has nothing to do with size of
  population from which it is drawn.
• Accuracy does depend on
• Size of sample and variance within in.
• Biases or non-random exclusion of elements of
  the population.

10
Most important ideas

• Key to good surveys is getting a good sample, not
  a big one.
• And asking questions that people can and are
  willing to answer.

11
Questions you should not ask

• What is the sampling error? (You know how to
  calculate it)
• Was it a good sample? (Is it good enough for your
  purposes?)

12
Questions you should ask

• Does the sample match, more or less, the
  population?
• Do weaknesses of the sample affect the
  conclusions you want to draw?
• Can people realistically answer the questions?
  Are they willing to?

13
Good poll stories

• Present results. Get reaction.
• Explain opinion describe or locate subgroups.
• Test argument or hypothesis.
• Confront assertions.

14
A typical story

• Almost half (45) of North Carolina residents
  believe the country is heading in the wrong
  direction, a sharp increase from one in five
  (20) in the spring, according to the UNC-CH
  School of Journalism and Mass Communication.
• Reaction or details
• Technical stuff

15
The technical stuff

• The figure is based on telephone interviews with
  a representative sample of 512 adults in the
  state in early November. In 95 out of 100
  surveys of this size, results are within 4 of
  the figures obtained by interviewing all adults
  in the state.

16
Other good story ideas

• Show change over time by comparing results with
  previous surveys.
• Compare results with state-wide or national
  surveys.
• Who supports/opposes policies. Is it race or
  gender issue? Partisan issue?

17
For help with analysis

• See Joseph Kim in the School of Journalism
  Research Center (Room 340)

18
Other resources

• Odum Institute on campus (www2.irss.unc.edu)
• Pew Center for People and the Press
  (www.people-press.org)
• Univ Maryland website for analysis of General
  Social Survey (www.webuse.umd.edu)