STAT 100 Lecture Notes - Lecture 6: Confounding, Statistical Significance, Dependent And Independent Variables

• School: Simon Fraser University
• Department: Statistics
• Course: STAT 100
• Professor: Gamage Perera

Document Summary

A randomized comparative experiment will minimize the effect of lurking variables. Often we compare performance for two (or several) competing treatments. Sometimes we compare one treatment under consideration to a control. Control group in a drug study might get a placebo (fake treatment), or some previous standard drug. Major concept: subjects are assigned to each group at random. This way no lurking variable can systematically favor one group over another. Double-blind experiment: neither the experimenter nor the subject know which treatment the subject is receiving. Otherwise, for example: doctor might be tempted to assign sickest patients to treatment group and healthier patients to placebo group. Control: limit effects of lurking variables by making groups as equal as possible (except for the fact that they get different treatments). Randomization: use chance to assign subjects to treatments, to eliminate bias. Replication: use lots of subjects, to reduce variation due to chance in the results.