Reading – Chapter 6 : de Vaus
Steps in Sampling Process
- Determine relevant population
- select appropriate sampling frame
- chose sampling method
- determine sample size
Sampling Concepts
- Population (Target population)
o universe to be sampled
- Census
- Sample
o subset of population
o best sample
§ representative of population
§ important characteristics are distributed similarly to how they are in population
§ can then make inferences about population
o selected sample
§ subset chosen to participate in survey
o achieved sample
· those members of selected sample who completed the questionnaire
o sample vs census
- sampling frame
o list of population elements
o sample is selected from frame
o quality issues with sampling frames : inclusion / exclusion, duplicates
- survey population
o population which includes subjects who can be contacted
Types of Sampling
- probability sampling
- non probability sampling
Probability Sampling
- equal or at least known chance of selection
- random selection
- types of probability sampling
o simple random sampling
§ requires good sampling frame
§ cost
o systematic sampling
§ periodicity
o stratified sampling
§ ensures representation from each strata
§ more complicated
§ strata must be identified and justified
o cluster sampling
§ cluster – naturally occurring unit à whereas in stratified sampling, researcher creates groups
§ only some clusters included, whereas all strata included.
§ multistage cluster sampling
Non Probability Sampling
- quota sampling
o produce representative samples without random selection
- purposive or judgment sampling
- snowball sampling
o word of mouth
- availability or convenience sampling
o convenience – group of individuals willing and able
o ease of access to researcher
o least likely to produce representative samples
Sampling Error
- results naturally from selecting sample, rather than entire population
- factors
o sample size
§ larger samples produce smaller sampling errors
o population variability
§ sampling error when population similar on characteristic being measured
Sample Size
- depends on degree of accuracy required for sample
o sampling error
§ amount of error we are willing to accept in our estimated value (ie + / - 2%)
§ confidence level – level of confidence we can have in our generalizations- ie 95% confidence level
o degree of diversity in population on key variables
§ variance
§ small samples – small increase in [absolute] sample size leads to substantial increase in accuracy.
· rule: halve sampling error, quadruple sample size
§ size of population is irrelevant for accuracy of sample
§ if sample broken into sub groups, degree of accuracy and variation within each group should determine sample size required for each group.
§ design specific
· systematic sampling à similar sampling errors
· stratified – smaller
· cluster à larger
§ will be errors based on other factors à calculation of precision based on sampling error alone misleading
Non Response
- response rate : actual respondents / eligible respondents
- issues
o reduction of sample size – results in non-representative samples
§ draw larger sample than what is needed
§ trained interviewers
§ reminders / follow-ups
§ incentives
§ graphically sophisticated surveys
o bias
§ differences between respondents and non respondents
· motivation
GW Comments / Questions
- what probability designs have unequal chance of selection
- KNNL – random effects model
o see separate notes
o I guess the difference between random effect Anova model and cluster sampling is what level is the subject of interest
§ example
· store level à sample of stores
· employee level à sampled stores and then sampled employees within stores
- variance with regard to sample size seems to be related to effect size
o sample size – split 50/50 vs 95 vs 5
- what is difference between sampling error and confidence level
o sampling error
§ differences between population and sample
§ sample is not perfectly representative
§ à standard error
§ despite random sampling techniques, still obtain poor samples à sampling error
No comments:
Post a Comment