Planning a Study Deciding what and how to measure

Preview:

Citation preview

Planning a Study

Deciding what

and how to measure

Vocabulary

Measuring What?

Units

Experimental Units

Subjects

Participants

Various Variables

Explanatory (independent) variable

Response (dependent) variable

Confounding variable

Lurking variable

ExperimentSubjecting the sample to a

controlled treatment where one variable is altered.

The objects on which the treatment is imposed on are called experimental units (human subjects).

Explanatory variables are called factors and specific values of the explanatory variable are levels.

Designing a Good ExperimentRandomization--randomly assign subjects to treatment and

control groups

Control

Replication--consistency

“Differences in the response variable between groups, if enough to rule out natural chance variability, can then be attributed to the manipulation of the explanatory variable.” This will allow determination of cause and effect.

Randomization--Crucial“Researchers do experiments to reduce the likelihood that the

results will be affected by confounding variables and other sources of bias.”

Randomize Type of TreatmentRandomize Order of Treatment

Control GroupsControl group--receives standard treatment OR

Placebo (sham) group--receives no treatment

Single-Blind

Double-Blind

These control for UNKNOWN variability

Designing ControlBlock Design--”divide

units into homogeneous groups (called blocks) and each treatment is randomly assigned to one or more units in each block.”

Matched-Pair Design--”assigned either two matched individuals (identical twins) OR the same individual (repeated measure) to receive the two treatments”

This controls for KNOWN variability.

Quitting Smoking w/Nicotine PatchesRecruited 240 smokers (volunteers) at Mayo Clinic

from 3 large cities

Randomly assigned 22-mg nicotine

patch or placebo patch for 8 weeks.

All attended counseling before, during, and after.

Double-blind (neither volunteers nor nurses taking measurements knew type of patch)

After 8-wk (1 yr), 46% (27.5%) of nicotine patch group quit smoking and 20% (14.2%) of placebo group quit.

Observational StudyObserving the

behaviors of a sample from a population.

The observer does not impose active treatments on units/subjects.

Or using previously collected data to do statistical analysis.

Census--Observational StudyThe systematical

collection of data on the entire population.

When the population is large, it will become time consuming and expensive.

Sample Survey--Observational Study

A portion of the population is asked a question and the study is done based on their voluntary answers.

03-08-93 Newsweek announced “A Really Bad Hair Day: Researchers link baldness and heart attacks.” The article reported that “men with typical male pattern baldness…are anywhere from 30 to 300 percent more likely to suffer a heart attack than men with little or no hair loss at all.”

The report was based on an observational study conducted by researchers at Boston Univ. School of Medicine. They compared 665 men who had been admitted to the hospital with their 1st heart attack to 772 men in the same age group (21- to 54-years old) who had been admitted to the same hospital for other reasons.

Case Control Studies--Observational Study

“Cases” who have a specific attribute/condition are compared to “Controls” who don’t.

Efficiency Reduces potential confounding

variables Retrospective vs. Prospective

Characteristics of a well-designed and well-conducted

survey

Trained interviewers must be consistent with asking neutral, non-leading questions.

An unbiased sampling should represent the population of interest.

PopulationsPopulations

Random SelectionsRandom Selections

SamplesSamples

Sampling Methods

Simple Random Sample (SRS)

Stratified Random Sampling

Cluster Sampling

Systematic Sampling

Multi-Stage Sampling

Random Digit Dialing

Self-Selected Sample

Convenience Sample

“Quickie Polls”

Simple Random Sampling

From the entire population every possible grouping of specified size has same chance

of being selected.

Stratified RS vs Cluster S

1st divide population into groups (strata), then take a Simple Random Sample from each strata

1st divide population into groups (cluster), then randomly select some clusters and sample everyone in that cluster

Systematic Sampling & Random Digit Dialing

From a list, divide into consecutive segments (every 50 names), randomly choose starting point (21st entry), then sample at that same point in each segment (21, 71, 121, 171, …)

Sample that approximates a SRS of all households in US that have telephones with a specific exchange

(210-695-)

Multi-Stage Sampling

“survey designers might stratify population by region of country, then stratify by urban, suburban, or rural, then choose a random sample of communities within those strata. They would continue to divide communities into city blocks (fixed areas) as clusters, and sample from the selected clusters.”

Self-Selected Sample--radio station call-in

Convenience Sample--surveying folks in a mall who appear willing to talk to you

“Quickie Polls”--hastily designed, poorly pre-tested, one night survey sample for

evening news show

Sources of bias in surveysIf a selection process

consistently obtains values too high or too low, then BIAS exists.

Selection BiasNon-response BiasResponse Bias

Survey Questions

Unnecessary complexity to question

Misleading questionOrdering of questionsEnsuring confidentialityAnonymous survey

Gathering Data

Experimental Design

Observational Study

Recommended