Confounds: Threats to Validity
There are various types of confounds:
Experimental Confounds:
Nuisance variable whose levels are correlated with the levels of the independent variable. For example, you want to see if people of different ethnic backgrounds will be more likely to buy your car (they don't know it catches fire).
You have them rate the attractiveness of the car. Did you control for income? Unfortunately, in the USA - discrimination has lead to different income levels. Thus, their car preference is confounded by income.
Order effects in repeated conditions:
If you test everybody in the same order of the independent variable conditions - at the end of the experiment, you might just be measuring boredom, fatigue or the subjects get better because of practice over times.
Differential Mortality:
Because of some characteristic of the independent variable subjects, drop out. Say you want to study the effects of violent TV on mood for over a period of a week. You set up a protocol where your subjects have to watch certain programs. It turns out that women dislike watching violent TV. If you don't notice this, your results would be confounded by gender leading to differential mortality. Your results would only be valid for men.
Demand Characteristics:
Subjects figure out the point of the study and try to confirm or disconfirm you results - depending how they feel or merely being in a study influences them.
Variants:
Positive self-presentation:
People want to be perceived as intelligent, moral, etc. and they may not respond in the typical fashion.
Hawthorne effect:
Based on studies at the Hawthorne Works of the Western Electric Company. This effect is generally defined as the problem in field Ss' knowledge that they are in an experiment modifies their behavior from what it would have been without the knowledge.
Experimenter effects:
Expectancy Bias: Experimenters' hypotheses may influence outcome of experiment
These are Classic
in studies of ESP. Fundamentally, the experimenter cues the subject on how to
respond. The cueing does not have to be conscious. Look up Clever
Hans, the mathematical problem solving horse, or analyses of sign language
using chimps or gorillas. The idea is that the experimenter may be unintentionally
influenced by the experimenter's expectations.
Solutions:
Use double-blind procedures when designing and running the experiment.
Bottom Line:
Confounding is an incredibly easy concept. In a study, you manipulate an independent variable in order to measure its effect on a dependent variable.
For your experiment to be valid, the only factor operating should be your independent variable. However, this can go awry and some other process or factor is influencing your study.
When this occurs, your study is confounded.
Confounding can occur when the experimenter accidentally manipulates the subjects in an unattended fashion, the subjects are influenced by merely being in an experiment and this is mistaken for a treatment effect, the groups are selected such that there is a bias between groups - and many other possible screwups.
Appropriate procedures exist to test for and to control these effects such that your experiment will be valid.
 |
