2: Sampling and Study Design

individuals

also known as units or cases; objects on which a measurement is made

population

collection of ALL observations of interest (hard to get, can be very large)

sample

SUBSET of observations from the population (smaller, easier to get, used to infer)

parameter

numeric characteristic of a POPULATON (often unknown)

statistics

numeric characteristic of a SAMPLE

sampling bias

consequence of how the sample data was obtained; usually results from non-random sampling

selection bias

selection procedure introduces a bias (faulty selection)

measurement error bias

data recorded incorrectly; also can be from faulty machine or non-reliable input (intentional or not)

non-response bias

individuals are unwilling/unable to respond (ie. ‘prefer not to answer, unable to participate (no wifi, not tech, savvy))

sampling methods with randomization

simple random sampling

sampling method with NO randomization

voluntary response sampling, convenience sampling

simple random sampling

entire population has EQUAL change of being selected

stratified random sampling

randomly select from GROUPS of the population; groups have a common characteristic (strata); select some from each group

cluster sampling

randomly select CLUSTERS and sample EVERYONE in selected cluster; works well for demo-/geo-graphic challenges

voluntary response sampling

individual SELF-SELECT to be in sample; tends to be biased

convenience sampling

sample what’s EASIEST to obtain; might exclude individuals or those sampled may biased

variable

a characteristic of interest; examples: weight, height, age, etc.

response variable

the variable of interest in the study

explanatory variable

variable that explains or possibly causes chain in the response variable

observational study

variables measured/observed, but NO conditions imposed by researchers

designed experiment

variables measured/observed UNDER conditions imposed by researchers

lurking variables

variables that affect BOTH explanatory and response variables; may explain a relationship between an E/R variable (unmeasured)

confounding variables

multiple variables that may affect response variable; it can be a lurking variable and be measured/known