Presentation is loading. Please wait.

Presentation is loading. Please wait.

Chapter 9 Designing Experiments

Similar presentations


Presentation on theme: "Chapter 9 Designing Experiments"— Presentation transcript:

1 Chapter 9 Designing Experiments
Observational Study Observes individuals and measures variables of interest but does NOT attempt to influence the responses. Experiment Deliberately imposes some treatment on individuals in order to observe their responses. vs. “To see how nature responds to a change, we must actually impose the change”… Vocabulary of Experiments: Experimental Units - Individuals on which an experiment is performed. Humans are called ‘subjects’

2 Vocabulary of Experiments (cont’d):
Treatment - A specific experimental condition applied to the units. Factors - The explanatory variables in an experiment. Level - A specific value or amount of, a factor. ex: How many treatments? Drug experiment: 1 dosage of 1 drug = 1 level of 1 factor = 1 treatment. 3 dosage levels of 1 drug = 3 levels of 1 factor = 3 treatments. 2 dosage levels of 2 drugs = 2 levels of 2 factors = 4 treatments. Advantages of Experiments over Observational Studies Allows for the study of a specific treatments effects. Allows for control of lurking variables. Allows for the study of combined factors effects.

3 Comparative Experiments
The design of an experiment first describes: The response variable or variables The factors (explanatory variables) The specific treatments ex: Treatment ---> Observation Or: Observation > Treatment ---->Observation 2 Problem: Experiments conducted in the field or with living subjects are often confounded. (lack of control / placebo effect) Solution: Compare treatments. Lurking variables operate on both groups Only difference is treatment (or absence of it). Control Group: A group of individuals / subjects given a fake treatment.

4 Enables control of the effects of lurking variables
1st principle of statistical design of experiments Completely Randomized Experiments Problem: How to assign treatments to experimental units Systematic differences between groups of exp. units is a possible source of bias in comparative experiments. Solution: “Randomization” Groups don’t depend on any characteristic of the exp. units Groups don’t depend on the judgment / preference of the experimenter. 2nd principle of statistical design of experiments.

5 ex: Nutritional product test comparing weight gain of lab rats.
30 rats for the experiment = 2 groups of 15 Control group gets standard diet / experimental group gets new diet - all other conditions are equal. 1 factor (Diet) 2 levels (Std. diet / Experimental diet) Diet is explanatory variable / weight gain is response variable Pick rats for groups by random allocation: label rats 01 to 30, use Table B or computer to pick control group of 15. The rest go to experimental group. Design: Random Allocation Group 1 15 Rats Group 2 15 Rats New Diet Std. Diet (control) Compare Weight Gain

6 ex: Consumer electric consumption experiment
example 3.12, pg 205 Completely Randomized Design All experimental units are allocated at random to all treatments Tips: Make as many groups as there are treatments Display treatments, response variable, # of units in each group Should try to have groups of equal size Logic of Experimental Design Randomization: Groups of experimental units that should be similar in all respects before treatments are applied. Comparative Design: influences other than the treatment(s) operate equally on all groups. Differences in the response variable are therefore only attributable to the treatment(s).

7 Statistical Significance
An observed effect too large to attribute plausibly to chance is called “statistically significant”. Principles of Experimental Design Control of lurking variables. Randomization: Use of chance to assign subjects to treatments. Replication: Repeating each treatment on a large enough number of exp units or subjects to allow the systematic effects of the treatments to be seen. Double-Blind Experiment Neither the subjects nor the people who have contact with them know which treatment they are receiving.

8 EXPERIMENTATION WARNING LABEL
Hidden Bias Experimenters must take great care to deal with all exp. units or subjects in exactly the same way so that the treatment is the only difference present. Lack of Realism Most serious potential weakness of experimentation. The subjects, treatments or setting may not realistically duplicate the conditions we really want to study. Can limit the application of conclusions to settings of wider interest. (Generalization)


Download ppt "Chapter 9 Designing Experiments"

Similar presentations


Ads by Google