COMM 88 Lecture Notes - Lecture 8: Simple Random Sample, Cluster Sampling, Nth Metal
Comm 88 Lecture 8
Lecture 8 - April 26, 2018
Sampling - How do we select participants (or other units) for a study?
•Sample - a subset of the target population (who//what you want to report about)
•Ex: target pops: voters, Facebook users, married couples, juries, football fans, business
owners, etc.
•Or: TV shows, magazine ads, blog posts, etc.
Representative (probability) sampling
•Sample should be a “miniature” version of the target population
•Allows you to generalize results to that pop
•Key is random selection - everyone in pop has equal chance of being included in sample
•How representative is it?
•Will always be “sampling error”
•Sample data will be slightly different from pop because of chance alone (aka “random” error)
•Statistically, this is known as the “margin of error”
•Ex: national poll N ≅ 1000 → ± 3%
•↑sample size, ↓margin of error
•Representative sampling techniques
•Simple random sampling - select elements randomly from population
•Listed pops: random #s table, phones: random-digit dialing
•Systematic random sampling - from a list of the population, take random starting point, then
select every “nth” element until cycle through entire list
•Similar results as SRS
•Close to the right proportions, but with some sampling error
•Stratified sampling - for getting pop proportions even more accurate
•Divide pop into subsets (“strata”) of a particular variable (usually demographics: sex, race,
pol party)
•Select randomly from each strata to get right proportions of the pop
•Need prior knowledge of pop proportions
•Increases representativeness because reduces sampling error (for the stratified variable)
•But more costly and time consuming
•Multistage cluster sampling
•Useful for populations not listed as individuals
•First randomly sample groupings (“clusters”), then randomly sample individual elements with
each cluster
•Reduces costs but sampling error at each stage
•For all 4 kinds:
•Will always have sampling error
•But can generalize findings to the larger target population (assuming done properly)
•What should you avoid: systematic error (sampling bias)
•Systematically over- or under-represent certain segments of pop
•Caused by: wrong sampling frame, low response rate (response bias), improper weighting
Non-representative sampling
•Cannot generalize results, can only make conclusions about participants in sample
•Typical of experiments and qualitative research