Unit 3: Data Collection

Simple Random Sample (SRS)

Simple Random Sample (SRS)

1. Label Individuals: assign #s/write names on slips

2. Randomize: random # generator (no repeats/unique)/put the names in a hat-shuffle

3. Select the group corresponding to these #s…

experimental units

what/who the treatments are imposed on (the students are assigned to eat ice cream)

treatments

what is done (or not done) to experimental units

Samples that lead to bias

convenience sample, voluntary response

Stratified Random Sample

splits population into strata (groups) and then creates a simple random sample from each strata (sample some from all groups); each starts has individuals with shared attributes or characteristics (homogenous groups)

Cluster Sample

sample all from same groups

Systematic Random Sample

choose a random starting point, use equal intervals

Undercoverage Bias

when some members of a population cannot or are less likely to be included in a sample

Nonresponse Bias

When an individual is part of a sample but chooses not to respond or they cannot be reached

Response Bias

pattern of inaccurate results (wording of question, interviewer, lying)

A sampling method produces the best estimates if there is (representative sample)

low bias and low variability

Census

collection of data from every individual or element in an entire population, as opposed to just a sample (better for small populations)

observational study

no treatments are imposed (put into effect); can’t determine causation, only show association

experimental

treatments are imposed, allows us to show causation

blinding

when subjects (single blind) and/or experinentors (double blind) don’t know about treatments

Placebo Effect

when a fake treatment works

Replication

Having many participants or volunteers in your experiment. This reduces the impact of differences from person to person.

Random Assignment

helps account for confounding variables through randomly choosing which participants are put in a group (shows causation)

Different samples

yield different estimates

larger samples

produce more accurate estimates

random sample and no random assignment

allows us to generalize our conclusions to the population that was sampled (observational)

Random Assignment

allows us to say a treatment causes change in the response (experiment)

statistically significant

when results from a study are too unusual to have occurred purely by chance (if the probability of something happens is less than 5%)

Types of experimental design

completely randomized, randomized block, matched pairs

Sampling Methods

SRS, stratified, systematic, cluster, convenience

Completely Randomized Design

subjects are randomly assigned to different treatment groups. This randomization ensures that each participant has an equal chance of being assigned to any group, eliminating biases and allowing for a straightforward comparison of treatments.

Randomized Block Design

subjects are divided into groups (or blocks) based on a specific characteristic (such as age, gender, or skill level) that may affect the outcome. Within each block, subjects are randomly assigned to different treatment groups. This design helps control for variability within blocks, leading to more precise estimates of treatment effects.

Matched Pairs Design

subjects are paired based on similar characteristics. Each pair is then assigned to different treatments. This design is particularly useful when the sample size is small, and it helps control for individual variability by comparing outcomes within pairs. (twin)

random assignment but not random sampling

you can show causation, but only within the specific sample tested—not the whole population

How to answer: what is the potential source of bias?

Name the bias (nonresponse, response, undercoverage) and say if it underestimates or overestimates