Transcript SURVEY RESEARCH

Survey Research
AD700
13 October, 2004
College of Advancing Studies
Brendan Rapple
This presentation owes much to the American
Statistical Association brochure series on survey
research:
http://www.amstat.org/sections/srms/whatsurvey.html

Little Cards on Restaurant Tables:
Was the service good?

Telephone:
Is president doing a good job?
The most popular programs on public radio?

Census:
How many bathrooms do you have?

Magazines:
How is your romantic life?

Market Researchers:
Brand X or Brand Y?
How many drinks last week?
Surveys Provide Important Knowledge

Economists, psychologists, health professionals,
political scientists, and sociologists conduct surveys to
study such topics as:





Income and expenditure patterns among households;
Roots of ethnic or racial prejudice;
Implications of health problems on people’s lives;
Voting behavior;
Effects on family life of women working outside the home, etc.
Specific Purpose Essential

Objectives of a survey should be as
 Specific
 Clear-cut
 Unambiguous as possible

"Men's Health Practices" is a very nebulous topic.

Better:
How often do African-American males aged 40-49 visit the dentist?
Steps in Conducting A Survey

Define precise purpose

Specify population

Specify appropriate sample

How to administer survey?

Draft of survey instrument

Pretest it

Revise it

Administer survey to sample

Analyze, write it up, and communicate the results

Use results meaningfully
Decide on Mode of Data Collection

Mail

Telephone

In Person Interview

Computer
Pre-Testing

Critical for identifying questionnaire problems.

Main problems revolve about:
 Question content, e.g. confusion with overall meaning of question as
well as misinterpretation of individual terms or concepts
 Formatting, e.g. problems with how to skip or navigate from question
to question may result in missing data and frustration for both
interviewers and respondents.
Population

Individuals

Larger units, e.g. families
Sometimes Difficult to Specify
Population

e.g. female faculty members at BC: do we include
part-time profs?
Samples

Must be representative of population.

Are the distributions of attributes, opinions, and beliefs in the
sample the same as in the population?

You want to be able to make inferences about the population as a
whole based on what you find to be true of the sample.

Often difficult to find representative sample.

Always a danger of sampling error or bias.
Quality of Sample Important

The quality of the sample – whether it is up-to-date and
complete – is probably the dominant feature for
ensuring adequate coverage of the desired population
to be surveyed.
Variability

Variability is large, then sample should be large

Converse also true
2 Barrels of Apples

Barrel A (low variability) -- all apples about 3 ins.
in diameter (range 3.1 to 2.9 ins.)

Barrel B (high variability) -- apples range from 2
to 5 ins. in diameter

Picking 3 apples from Barrel B might give result
well below (above) average.
Size of Sample Isn't Everything

Large numbers do not, in and of themselves, increase
the representativeness of a sample.

Most professional survey conductors hold that a
moderate sample size is enough statistically and
operationally.
Whole Population and Sample
Sometimes the Same
Example:
– Small companies in the paper recycling industry in LA.
–
Unit of Analysis: a company
–
You define "small company" as a private co. with turnover of
less than $2,000,000 per annum
–
Research shows that there are 34 relevant companies
–
Therefore, manageable to use ALL in sample
N.B.
Results will only relate to small paper recycling
companies in LA -- difficult to generalize about
other types of company in other parts of the
country.
Population Often Not Feasible Due to Size

•
Welfare Recipients
•
Mentally Ill
•
Prison Inmates
Often essential to survey a REPRESENTATIVE
sample.
Early Studies of Gay Men

Sampling frame composed of men, patients of
therapists participating in research

But most gay men were not patients of therapists
Representative Sample

EXAMPLE--Success of unwed teenage mothers in
raising children?

To be representative, sample must contain same
proportion of unwed teenage mothers at
--each age level
--each educational level
--each socio-economic status
in community
Lists May Be Very Exclusive
EXAMPLE
Undocumented Aliens
--We know that many live in LA
--But relying on Govt. lists may be useless
Suppose You Have a “Population,”
e.g.

all registered voters in your county

all Mercedes owners in the state

all soccer players in your school district who drive green mopeds
THEN YOU SELECT SAMPLE IN SUCH A WAY THAT EVERY
NAME ON THE LIST HAS AN EQUAL CHANCE OF BEING
INCLUDED IN THE SAMPLE
Random Sample
Random =
Purposeful & methodical
Not reflect biases of researcher
Everyone has equal & independent chance
of being selected
Random Sample

Once selected it cannot be chosen again (like lottery
winners)
 Example: 500 part-time students in Advancing Studies
 Sample of 20% is required
 Assign each student a number from 1 to 500
 Randomly select 100 numbers (by computer or by table of
random numbers)
Systematic Random Sampling
Example 1.
 2,000 in sampling frame and you want a sample of 200, then you
might select every 10th name
Example 2.
 500 part-time students in Advancing Studies
 Sample of 20% is required
--Randomly Select a Number from 1 to 5
--Select Every 5th Person
--002, 007, 012. 017, 022, and up to 497.
Possible Problem:

Staff in govt. agency may be listed unit by unit

Each unit has 9 line-level workers and 1 supervisor.

The supervisor is the 10th person on the list.

It’s a survey of 20% -- every 5th person is selected.

If first no. selected is 1, 2, 3, or 4 then no supervisor will be
selected, though they comprise 10% of population.

If first number selected is 5, then supervisors will be greatly
overrepresented.
Thus, possibility of bias due to periodicity or patterns.
Stratified Sampling
Population: 2,000 (800 females; 1,200 males)
Sample required:
200
If gender is an important variable in your survey, then
both females and males should be included in
appropriate numbers, that is, in proportions that
correspond to their presence in the population.
Strategy:
Treat both sexes as separate populations and take 10%
sample from each.
OR
Make sure that all females are listed first and then take
every tenth name.
Either way you will end up with 80 females and 120 males
Convenience Sampling
Cluster Sampling
Often difficult to list all members of target population
and select the sample from among them
e.g. 1)
Population of American high school
students
2)
Population of U.S. postal delivery workers
3)
Adult population of Atlanta
Possible Strategies

Population of American high school students
choose 50 schools randomly from entire list and include all
students in those schools in the sample.

Population of U.S. postal delivery workers
choose 100 post offices randomly from all 50 States and include
all deliverers in those post offices in the sample.

Adult population of Atlanta
Randomly choose sample of 50 blocks from a city map and
then poll all adults living on those blocks.
Potential Problems
Confidentiality

Confidentiality of data supplied by respondents is of prime
concern to all reputable survey organizations.

Strategies:
 Using only number codes to link the respondent to a questionnaire.
 Refusing to divulge names and addresses of survey respondents.
 Omitting the names and addresses of survey respondents from
computer files used for analysis
 Presenting statistical tabulations by broad enough categories so that
individual respondents cannot be singled out.
Reporting

Important that individual respondents are not identified in
reporting survey findings.

All of the survey’s results should be presented in totally
anonymous summaries, such as statistical tables and charts.
Volunteers



Volunteers usually have characteristics that differentiate
them from the larger population.
The fact that they volunteer makes them different from
persons in the population who do not volunteer.
They tend (but not in all circumstances):
– to be better educated
– have higher social class
– to be more intelligent
– have greater need for social approval
– to be more sociable
– more unconventional (especially when volunteering for
studies of sex behavior)
– less authoritarian
– less conforming
Volunteers (Cont.)

Jews more likely than Protestants.

Protestants more likely than Catholics.

Females more likely than males.
Volunteers -- Example

TV programs asking viewers to vote.
– people call who are most committed to issue.
– “stuffing of ballots” by multiple calls.
– Time of day is important – who’s available?
Margin of Error

Error margin of 1,000 randomly chosen individuals is said
to be 3.1%.

Thus, if a random sample of 1,000 indicates that 59% will
vote for Bush, the actual number could range from 55.9% to
62.1%.
In the 1984 election, the Gallup Survey (using 3,456
individual responses) missed by just +0.2 of 1% when it
predicted that Ronald Reagan would win by 59.0%.