Producing Data, Randomization, and Experimental Design Goals
Identify observational studies versus experiments
Design experiments to test hypotheses using appropriate randomization
Use the random number tables to assign subjects correctly to experimental groups
Define, use, and know the concepts behind all the new vocabulary words
Starting with a research question
Often can’t simply study the whole population
If you want to know the life expectancy for cancer patients you simply can’t identify all patients and then wait for them to die
Use a sample to draw conclusions about the whole Observational Study vs. Experiment
Observational study observes and measures variables of interest.
Experiment imposes a treatment in order to observe outcome
New Terminology
Population Sample
Voluntary response sample
Convenience sampling
Bias
Simple random sample (SRS) Use Random Numbers to Generate SRS
Label all the individuals in a population with numerical labels.
Use random number table (or statistical package) to choose individuals randomly.
Example: To divide 100 students into two groups of 50 label them 00 to 99 and go through the table starting at a random line until the first 50 have been chosen for a group. Other Sampling Designs
Probability sample
Individuals chosen with some given probability
Stratified random sample
Population divided into strata and individuals chosen at random from each strata
Multistage random sample
Sample chosen in a number of stages Problems
Undercoverage: some groups have no chance of being sampled; as in phone polling
Nonresponse: individuals chosen cannot or will not participate
Response bias: people may lie about illegal or embarrassing behavior; may respond to the questioner
Wording of questions may effect the outcome Designing Experiments
Experimental units or subjects
Treatment: experimental condition imposed
Factors: explanatory variables
Level: value of a given factor Example: Television ads Comparative Experiments
Compare 2 or more groups
Use a control group to eliminate confounding and placebo effect
A randomized comparative experiment uses comparisons between two (or more) groups and randomization of subjects into treatment groups. Design of a randomized comparative experiment Principles of Experimental Design
Control effects of lurking variables via comparison of several treatments
Randomization to assign units to treatment
Replication of experiment on many units to reduce chance variation