Designing Studies

Population

entire group of people we interested in hearing about.

Sample

Subset of the population that we accually have data on

Anecdotal Data

Stories- may no be repetitive of the entire group

Available Data

already gathered-may not meet our needs

Gathered Data

Best way to get data- Census or Sample

Census

Survey the entire population- Very hard to do, but solid evidence

Convenience Sample

People are selected based upon ease of getting them

Voluntary Sample

People self selecting into the survey-often leads to strong feeling and bias

Simple Random Sample (SRS)

Individuals are selected in such a way that every possible group of X Individuals is equally likely to be chosen. Best way

Stratified Random Sample

Separate the population into groups based upon a relevant characteristic (Strata). Choose a SRS from each Strata. Combine for total sample

Cluster Random Sample

population is divvied into groups (that hopefully mirrors the entire population) number the groups, STS to chose some groups, then survey every member of the group chosen.

Multistage Random Sample

Repeated SRS EX: VA- SRS to find 5 counties- SRS 10 people from each county.

Probability Random Sample

Every subject has an equal chance of being chosen.

Systematic Random Sample

First choice is random, all subsequent chosen subjects are selected via a predetermined pattern.

Sample Frame

List of individuals from which the sample is accually taken. Ideally the sample frame matches the population.

Undercoverage

The sample frame misses part of the population

Nonresponse

the chosen subjects can’t be contacted or refuse to participate

Response Bias

People aren’t always fully truthful, they don’t want to look bad, gives the answer they think the survey wants.

Observational Study

Measure variables by only observing the subjects. Try to not influence anything.

Experiment

Imposed treatments on individuals in order to measure a response. Only way to establish cause and effect.

Lurking/ Confounding Variable

Variable, known or unknown, that has an impact on responses, but is not apart of the study.

Subjects/Experimental Units

Who/What gets treatments

Treatments

Conditions imposed on Subjects

Response Variable

Measured outcome

Placebo Effect

Untreated person incorrectly believes they are getting treatments and reports a change.

Blind/Double Blind

The subject don’t know what treatment they are getting or Neither the subject of evaluator knows who got what treatments.

What is needed for an experiment

Control and Randomization

Control

Comparative groups-balance lurking variables between groups.

Randomization

seek to minimize the impact of the lurking variables by creating roughly equal groups.

Replication

Apply treatments to many subjects to reduce variability in our outcome

Block Design

Group the subjects by a relevant characteristic then randomly assign the treatments

Matched Pairs

Block of two/data are not independents.