Transcript Document

2.2: Sampling methods (pp. 17 – 20)
• Probability sampling: methods that can specify
the probability that a given sample will be
selected.
• Randomization: a technique for insuring that
any member of a population has an equal
chance of appearing in a sample.
– With randomization, sample statistics will on average
have the same values as the population parameters.
• Simple random sample: each possible sample
of a given size has the same likelihood of being
selected.
How to select a simple random sample
• 1. list all the subjects in a population
• 2. assign a number to each subject
• 3. pick numbers from a list of random
numbers
• 4. put the corresponding subjects in the
sample.
Cost and feasibility can be problems,
especially if the population is large. OK, for
people in households or students in
Non-probability sampling (pp. 20 – 21)
• Non-probability sampling methods cannot specify
the probability that a given sample will be selected.
– Example: snowball sampling methods (Edin and Lein)
• Why use such methods?
– They are often inexpensive
– They can provide information about groups that are
difficult to sample or require great trust or will get lengthy
unstructured interviews.
– Some social variables and their relationships are
universal, which makes sampling method irrelevant!
• This is assumed for many psychology studies and medical
studies.
Common research designs (pp. 21 – 22)
• Experimental design
– Subjects are randomly assigned to treatments (=variables)
by the researcher
– Causal inferences are stronger
– Random sampling from the population less important
– Usually laboratory (exc. Moving to Opportunity, MTO)
• Observational design (e.g., surveys)
–
–
–
–
Subjects are not randomly assigned to variables
Random sampling is important.
Selection bias
Causal inferences are compromised.
Natural Experiments
Observational studies (esp. surveys) where
respondents’ values on a causal variable are
plausibly random.
Examples:
• Military draft lottery
• Births in last half of year
• Indian panchayats headed by women
• Parity 3 birth after same sex or opposite sex
2.3: Sampling and non-sampling variability (pp. 22 – 24)
We ideally like sample statistics to be as close as
possible to population parameters, but several factors
can cause variability:
•
Sampling error: the difference between a sample statistic
and its population parameter.
• Random sampling allows us to estimate the typical size of the
sampling error.
•
Non-sampling error: comes from other sources, can be
systematically biased, and is difficult to estimate.
• Examples of nonsampling error include undercoverage,
nonresponse, question wording (e.g., response bias), question
order.
2.4: probability sampling methods (pp. 25 – 28)
• Systematic random sample:
–
• Stratified random sample:
–
• Cluster sampling:
–
• Multistage sampling:
–
2.4: probability sampling methods (pp. 25 – 28)
• Systematic random sample:
– pick a random case from the first k cases of a sample;
select every kth case after that one
• Stratified random sample:
–
• Cluster sampling:
–
• Multistage sampling:
–
2.4: probability sampling methods (pp. 25 – 28)
• Systematic random sample:
– pick a random case from the first k cases of a sample;
select every kth case after that one
• Stratified random sample:
– divide a population into groups, then select a simple
random sample from each stratum
• Cluster sampling:
–
• Multistage sampling:
–
2.4: probability sampling methods (pp. 25 – 28)
• Systematic random sample:
– pick a random case from the first k cases of a sample;
select every kth case after that one
• Stratified random sample:
– divide a population into groups, then select a simple
random sample from each stratum
• Cluster sampling:
– divide the population into groups called clusters or
primary sampling units (PSUs); take a random sample of
the clusters
• Multistage sampling:
–
2.4: probability sampling methods (pp. 25 – 28)
• Systematic random sample:
– pick a random case from the first k cases of a sample;
select every kth case after that one
• Stratified random sample:
– divide a population into groups, then select a simple
random sample from each stratum
• Cluster sampling:
– divide the population into groups called clusters or
primary sampling units (PSUs); take a random sample of
the clusters
• Multistage sampling:
– several levels of nested clusters, often including both
stratified and cluster sampling techniques
Examples of sampling in typical surveys
• National Longitudinal Survey of Youth (NLSY)
– 12,686 men and women ages 14-22 in 1979.
– includes a multistage sample designed to be nationally
representative.
– includes oversamples of hispanic women and men, black
nonhispanic women and men, poor white women and men,
plus military subsamples, along with sampling weights.
– A reinterview every two years loses some respondents
(nonrandomly) to attrition.
• Current Population Survey:
http://www.census.gov/prod/2000pubs/tp63.pdf,
section 14, especially Table 14-5 for DEFF