LESSON: Experimental Design

Design Methods

In order for an experiment to produce meaningful and unbiased results, they should be designed and executed carefully. To begin with, let's learn about a few design methods for testing our treatment on different types of groups. 

In this video you will learn about: Completely Randomized Design, Randomized Block Design (Blocks), and Matched Pair Design.

Control Groups

In good experiment design, a researcher will divide the population sample into a control group and an experimental group using one of the methods we just learned about. How do we proceed from here? How do we ensure that the results we obtain from the control group and the experimental group are unbiased?

In this video we will learn about: placebos, the placebo effect, control groups and double-blind experiments. 

Types of Variables

In statistical studies, a variable is something that is liable to vary or change and they are usually the main focus of experiment. 

There are two types of variables: quantitative variables and qualitative variables. Quantitative variables are numerical and represent something that can be counted (height, age, # of siblings, etc). Qualitative variables represent names, categories or labels (yes/no, breed of dog, smoker/nonsmoker, etc). While it's important to recognize this distinction, we will not be making this distinction moving forward. 

Explanatory & Response Variables

The main focus of an experiment are the two variables called the explanatory variable and the response variable. You may have also heard these variables called the independent and dependent variables in past math or psychology classes. However, in statistics courses we don't usually those terms because the independent variable doesn't necessarily have to be independent. 

Here are the definitions of these variables. We will go into more examples later.

Let's go through a few examples so you can get the hang of distinguishing these two types of variables. 

Confounding Variables

Experiment results can be ruined in many different ways. Being able to control as much of the experiment as possible is an important consideration in experimental design. One such way an experiment can be ruined is by not taking into account confounding variables. A confounding variable occurs when a researcher can't distinguish between the effects of different factors in an experiment. 

For example, suppose a professor at the University of Utah experiments with a new attendance policy. He let's the class know that their individual grades will drop by one percentage point for every class they miss. Well, Salt Lake City experiences a very mild winter with little snow fall, making it easier for students to physically be able to make it to class. The professor fails to consider this and concludes that his new attendance policy has done the trick and decreased absences. The effects of the attendance policy and the weather have been confounded.

Watch the video below to see another explanation of confounding variables in experiments. Remember that independent variables = explanatory variables and dependent variables = response variables, which we learned about previously.