Sampling Overview R toy sampling Non-probability sampling - - PowerPoint PPT Presentation

Sampling Overview

• R toy sampling
• Non-probability sampling

Probability Methods (AKA random)

• Why is it be called “probability” sampling?
• What are we certain of?
• Everyone comfortable with
• Simple random sampling
• Stratified random sampling
• Disproportionate stratified random sampling
• Cluster sampling

How big of a sample?

• Probability Sampling:
• Power Calculations: If you know the size of the effect you expect and the variance of the

sample, you can calculate how large of a sample you need to detect that effect (with, for example, 95% confidence).

• Probability sampling is only useful with many cases
• Rule of thumb: at least 30
• So its not useful for in-depth study of a few cases
• We might still start out with random sample though, before we select cases

Nonresponse

• When we have random sample, but people refuse to participate.
• We can’t assume nonresponse is random.
• Often if it’s 20-30% nonresponse, we say it’s not random

Non-probability Methods (AKA non-random)

• Why might it be called “non-probability”?

Non-probability Methods (AKA non-random)

• Stratified random sampling and cluster sampling are still probability sampling methods
• Non-probability methods trade generalizability for logistics
• Probability sampling requires a list of all the units in our population
• Which we may not have
• What populations might it be hard for?
• What topics?

Availability Sampling

• AKA convenience sampling
• Magazine polls of their readership
• Polling people who walk by
• Amazon Turk
• What’s the population you can make an inference about?

Quota Sampling

• Quotas to match your sample to the population
• 50% women
• x% poor, x% rich, x% college educated, etc.
• Pros:
• Matches the population better
• Cons
• May not match the population on other dimensions
• Especially unobservable dimensions
• Requires detailed knowledge about the population

Purposive Sampling

• Targets key individuals with important knowledge
• e.g. Interview the diplomats who negotiated a particular treaty
• e.g. Interview the leaders of an important social movement

Snowball Sampling

• Each subject tells you who to talk to next
• Hard to reach groups.
• Respondent driven sampling
• Why is bias a particularly big problem here?

Sample quality

• Know what population it represents
• Unspecified is no good
• Know how cases were selected
• systemic, chance driven, haphazard
• Sample quality is determined by sample obtained, not by sampling method itself
• High nonresponse just as bad as haphazard
• Can’t be assumed to move across population
• Want to compare important variables across groups