Sampling Methods Stats

Convenience sample

Easiest to reach —> bias (don’t represent pop)

Voluntary response sample

Individuals choose to join —> bias (don’t represent pop)

SRS

Without replacement, all possible samples of the same size have same chance of being chosen —> representative of pop

Stratified random sample

Divide pop into groups (characteristics that may affect response), SRS from each strata —> more precise

Cluster sample

Divide pop into groups near each other, randomly select some of the clusters —> save time/money

Systematic sample

Select a value k, select a value from 1-k, and sample every kth individual after —> easier than some sampling methods unless a pattern

Undercoverage

Some members of pop less likely to/cannot be chosen

Nonresponse

Can’t be contacted/refuse to answer

Confounding variable

Effects on a response variable indistinguishable

Pro vs retro obs vs exp

Pro: track individuals, retro: past data, exp: assign treatment

Purpose of Comparison in experiment

Compare two or more treatments

Purpose of random assignment in experiment

Create roughly equivalent groups before imposing treatment

Control purpose experiment

Avoid confounding, reduce variation, making it easier to decide if treatment effective

Replication purpose experiment

Effects of treatment distinguishable from chance differences

Randomized block design

Blocks with similar traits that would affect response —> responses compared and combined —> easier to see effects of treatment

Matched pairs design

Subject receives treatment in random order —> control for differences between individuals

Inference about population requires:

Individuals in study randomly selected from population

Inference about cause and effect:

Well designed experiment, random assignment to treatment