Part II Surveys Flashcards
What is a survey?
- Standardized, cross-sectional approach to data collection of info on individuals and households through questioning/measurement of systematically identified samples
4 steps of a survey
- Sample design
- measurement
- Survey operations (fieldwork)
- Statistical analyses and interpretation
What is sampling error and how does it compare to random error?
Error in the estimate due to the fact that only part of the population is included in the sample
- component of random error
Sample design has two aspects:
- Selection process (rules and operations by which some members of the pop are included in the sample)
- Estimation process (computing the sample statistics)
Elements or elementary units
Elements of a population are the units for which information is sought (i.e., individuals)
Sampling units
Contain the elements (e.g., households)
- Useful for when a list of the elements does not exist
- May contain only one element (element sampling) or several element (cluster sampling)
Sampling frame
List of target population from which the sample is actually drawn
Enumeration rule
Rule by which elementary units are to be associated with each enumeration
Observational units
Units from which the observations are obtained (i.e., respondents; e.g., mothers of children)
Sampling fraction
Probability with which the elements are sampled
Model sampling
Sampling based on broad assumptions about the distribution of the vx in the pop
- arbitrary and informal samples
- assume that the important characteristics are distributed either uniformly or randomly
e. g., Haphazard/Fortuitous sampling (whatever comes to hand)
e. g., Judgment sampling (expert’s choice for typical units)
Probability sampling
- Every element has a known and nonzero probability of being sampled
- statistical inferences to pop values can be made
Simple random sampling
Assign number, select n numbers by random process
- Requires a list of elements
- Difficult to assess subgroups
- Difficult for samples spread geographically
Equal probability of selection method sampling
- Elements have equal probabilities of being selected
e. g., 10% prob regardless of language
Non-EPSEM sampling: non-equal probability of selection
Systematic sampling
- Approximation of SRS, so widely used
- Random starting point on list, then select every nth unit
- Can be done as sample frame is constructed
Stratified sampling
- Divide sampling frame in subgroups
- Sampling done separately within each stratum
- Proportionate (same fraction in each stratum) vs disproportionate
Dis/Advantages of stratified sampling
Advantages (over SRS)
- Precision
- Estimates for each strata
- nothing to lose
Disadvantages
- Requires identification of every enumeration unit by stratum before sampling
- only for easily available criteria
Cluster sampling
- Elements included into larger sampling units (e.g., generally geographical areas)
- Decreases variability, decreases precision
- Easier to construct sampling frames
- Lower cost
Sample size adjustments
- For design effect (when the standard error is higher than for SRS)
- Expected response rate
- Proportion of eligibles
Conservative and less conservative approach to response rate?
Conservative: consider all refusals and non-contacted individuals as eligible
Less conservative: consider that the proportion of ineligible among refusals is the same as in the responders
3 types of nonresponse
- Refusals
- Non-reachable
- Selective non-response
Refusals
- Often impossible to know eligibility
- Selection bias if differential refusal
- If non-diff: decreased n, increased sampling error, increased standard error, decreased precision
Non-reachable
- Impossible to know eligibility
- Selection bias if differential
- If non-diff: decreased n, increased sampling error, increased standard error, decreased precision
(SAME AS REFUSALS)
Selective non-response
- Refusing to answer specific questions
- If large number missing, possibility of bias (25% over, exclude)
- Imputation methods
