These are felt to cause some change in the dependent variables. They are manipulated by the researcher. Subjects may be presented different stimuli, placed in different settings, given different instructions, or affected directly in physical or chemical ways. What the experimenter does to the subject is called a treatment. Each independent variable takes on a certain number of levels. For example, in an experiment on the effect of a drug, the levels might be varying dosages.
These are felt to be affected by the independent variables. They are measured by the experimenter and are normally responses of the subjects.
To test a hypothesis, the experiment compares the behavior of subjects exposed to a treatment (the experimental group) with those not exposed (the control group)
A confounding variable is one which varies systematically with an independent variable. If a significant effect is found, then we do not know whether this is due to the independent variable or the confounding variable. For example, in an experiment on the effect of a drug, all subjects who are given the drug and know this may be affected by their belief that they have been given a potent drug.
The comparison is between different groups of subjects, normally a control group and a separate group for each level of the independent variable (treatment). Subjects are assigned randomly to groups.
The comparison is within subjects; that is, each subject is administered the control condition and each level of the treatment. Within-subject designs have the advantage that we don't have between-subject variance to contend with but we have to be careful to take into consideration the possible effects of the order in which the different conditions is presented. For example, if we want to know the relative effectiveness of two different techniques for teaching a foreign language, and we use the same students for each treatment, experience with the first technique may lead to better performance on the second than we would see with the second alone. An experiment can often factor out these effects by counterbalancing, that is, by varying the order of the treatments administered to the subjects.
There is a single independent variable. We are not interested in how variables interact.
There is more than one independent variable. The experiment provides a test of two kinds of effects.
These are just the separate effects of each of the independent variables. From the perspective of the main effects, the experiment is like running separate experiments for each independent variable.
These are effects that are due to particular combinations of the independent variables. That is, the effect of one independent variable changes across the levels of another independent variable. For example, an experiment on the effects of noise on frustration tested two separate noise variables, loudness and predictability. They found that loudness only mattered when the noise was unpredictable. Interactions can only be tested in factorial designs.
There are no independent variables; all variables are dependent. The goal is to be able to predict the value of one variable given another. For example, the variables might be SAT scores and college GPA. If a correlation is found, then knowing a student's SAT score would allow you to predict his or her GPA.
Both variables may be influenced by an unobserved third variable.
It may be impossible to determine which direction the causality goes in. Does aggressive behavior cause people to watch aggressive TV shows, or does watching aggressive TV shows cause aggressive behavior?
Take me back to the Rhythm and Cognition Home Page.
Last updated: 24 October 1995
Copyright 1995, The Trustees of Indiana University