chapter 4 · 2015. 1. 26. · chapter 4 . 4.3 what are good and poor ways to experiment? elements...

Post on 02-Mar-2021

3 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

GATHERING DATA

Chapter 4

4.3 What are Good and Poor Ways to Experiment?

Elements of an Experiment

¨  Experimental units: Subjects

¨  Treatment: Conditions imposed on subjects

¨  Explanatory variable: Defines groups and treatments

¨  Response variable: Outcome

www.clinplus.com

Experiments

¨  Impose treatments on subjects to observe responses

¨  Goal: compare effects of treatments on response

¨  Randomized experiments – subjects randomly assigned to treatments

www.privatemdlabs.com

Placebo effect

Placebo – fake treatment; sugar pill

Placebo effect –improving not from real treatment but from belief that he or she should improve

www.scientificamerican.com

3 Components of a Good Experiment

¨  Control or Comparison Group ¨  Randomization ¨  Replication

pages.cs.wisc.edu

Principle 1: Control or Comparison Group

¨  Helps analyze effectiveness of primary treatment

¨  Placebo removes lurking variables

¨  Control group gets placebo ¤ Clinical trials may

compare new treatment with existing

www.redicecreations.com

¨  Experiments should compare treatments rather than effect of single treatment

¨  Example: 400 volunteers asked to quit smoking with some taking and some not taking antidepressant

Control or Comparison Group

www.gpaulmckinneyjr.com

Principle 2: Randomization

1.  Eliminates bias from researcher assigning subjects

2.  Balances groups on known and lurking variables

colorectal.surgery.ucsf.edu

Principle 3: Replication

1.  Reduces difference due to ordinary variation or chance

2.  Increases chance that results show true difference

www.printercopierblog.com

Blinding the Experiment

¨  Blind – subjects unaware of which treatment used

¨  Double-Blind Experiment - Neither subjects nor investigators know which treatment ¤ Controls bias from

respondent and experimenter

gridskipper.com

Statistically Significant Difference – Observed difference is larger than expected from chance

Statistical Significance

Generalizing Results

¨  Goal of Experimentation – Analyze association between treatment and response for entire population

¨  Generalize only to population represented by study

¨  Page 180 #34, 40

static.howstuffworks.com

4.4 Other Ways to Conduct Experimental and Observational Studies

Sample Surveys: Random Sampling Designs

¨  Alternative to experiments 1.  Simple Random

Sampling 2.  Cluster Random

Sampling 3.  Stratified

Random Sampling

codetechnology.files.wordpress.com

Cluster Random Sample

1.  Divide population into large number of clusters, such as city blocks

2.  Select simple random sample of clusters

3.  Use all subjects in clusters as sample

Advantages ¤ Sampling

frame unavailable

¤ Cost Disadvantage

¤ Need larger sample size for same reliability

Cluster Random Sample

reactorfire.files.wordpress.com

Stratified Random Sample

1.  Divide the population into groups, strata

2.  Select SRS from each strata

3.  Combine samples from each for total sample

www.nedarc.org

Advantage ¨  Ensures stratum

representation Disadvantage ¨  Need sampling

frame and to which stratum each subject belongs

Stratified Random Sample www.smh.com.au

Comparing Random Sampling Methods

Types of Observational Studies

1.  Sample Survey: current

2.  Retrospective Study: past

3.  Prospective Study: future

¨  Cause not proven, but studies can support beliefs

Retrospective Case-Control Study

www.lisisoft.com

Studying sunlight exposure and multiple sclerosis connection…

Retrospective

Cases – have MS Controls or don’t

Explanatory variable – low sun or not

Prospective Case-Control Study

www.dubaiinternetmarketing.com

Studying effects of

vegetarian diet on heart

disease…

Prospective

Cases – have heart disease Controls or don’t

Explanatory variable – vegetarian or not

Multifactor Experiments

¨  Single experiment analyzes two or more factors

¨  Learn more since combinations may affect response

biobreak.files.wordpress.com

Matched Pairs Design

¨  Subjects are somehow matched ¤ Husband/wife, two plots in

same field, etc. ¤ Same individual – crossover

design ¨  Randomly assign or randomize

order of treatments ¨  Reduces effects of lurking

variables

www.marshallhammondwedding.info

Randomized Block Design

¨  Block – subjects with common characteristics ¨  Randomized Block Design, RBD – within each block,

randomly assign to treatments

top related