Unit 3 Part 2 Surveys, Experiments, and Simulations Experiments
Why Experiment? The Purpose of Experiments We need better information to guide decisions. Experiments allow us to look more closely at causation. Can remove the impact of lurking and confounding variables.
Lurking Variables Ice Cream Sales (x) The Purpose of Experiments In this classic example it appears there is a causatory relationship between the explanatory variable (x) and response variable (y). In reality there is a 3rd variable (z) which is “lurking” and impacting both x and y. Ice Cream Sales (x) Drowning Deaths (y)
Lurking Variables Ice Cream Sales (x) The Purpose of Experiments In this classic example it appears there is a causatory relationship between the explanatory variable (x) and response variable (y). In reality there is a 3rd variable (z) which is “lurking” and impacting both x and y. Ice Cream Sales (x) Drowning Deaths (y) This would be a great example of how correlation doesn’t always imply causation. Temperature (z) Try to think of an example of a lurking variable scenario!
Confounding Variables The Purpose of Experiments Confounding Variables The most important thing to remember is that in the case of confounding variables, there actually is a causatory relationship between x and y. However! There are other variables that also impact y (but not x). Weight of a Vehicle (x) Gas Mileage (y)
Confounding Variables The Purpose of Experiments Confounding Variables The most important thing to remember is that in the case of confounding variables, there actually is a causatory relationship between x and y. However! There are other variables that also impact y (but not x). Weight of a Vehicle (x) Gas Mileage (y) Number of Cylinders (1) Try to think of an example of a confounding variable scenario! Style of Driving (2)
Why Experiment? How? The Purpose of Experiments We need better information to guide decisions. Experiments allow us to look more closely at causation. Can remove the impact of lurking and confounding variables. How?
Experimental Subjects Experimental Design Experimental Subjects Experimental units are the experiment participants (not to be confused with the experimenters). Human experimental units are often referred to as subjects. Treatment The treatment applied will vary across treatment groups. Often, there are multiple factors involved in treatments. Response The response is simply the measurable change in the response variable.
Simple Design Example Experimental Design Experimental Units or Subjects Treatment Applied Observed Response Population of Interest Group A: SRS of 40 individuals Nothing Measured Response: Heart Attack Rate Observed Rate: 11% Individuals 40 to 60 years of age. Group B: SRS of 40 individuals 100mg of Aspirin Daily Measured Response: Heart Attack Rate Observed Rate: 6%
Control Placebo Experimental Design Experiments are designed around the concept of control That is, an attempt is made to control the impacts of possible lurking or confounding variables. Placebo One of the most important confounding variables in experimentations involving human beings is the placebo effect. That is, simply knowing a treatment is being applied can have an impact on the response variable. We can control for the placebo effect by applying a placebo.
Placebo Design Example Experimental Design Placebo Design Example Experimental Units or Subjects Treatment Applied Observed Response Population of Interest Group A: SRS of 40 individuals 0mg of Aspirin Daily Measured Response: Heart Attack Rate Observed Rate: 11% Individuals 40 to 60 years of age. Group B: SRS of 40 individuals Placebo Daily Measured Response: Heart Attack Rate Observed Rate: 8% Group B: SRS of 40 individuals 100mg of Aspirin Daily Measured Response: Heart Attack Rate Observed Rate: 6%
Random Assignment Experimental Design We have discussed Simple Random Samples twice – with Observational Studies and with Experiments. But we still need to talk about how this is accomplished and practice it.
More Complex Experimental Design Multiple Factors Blind Double Blind Block Design Matched Pairs Design