Experimental Design

Careful and detailed plan of an experiment.

In simple psychological experiments, one characteristic—the independent variable—is manipulated by the experimenter to enable the study of its effects on another characteristic—the dependent variable. In many experiments, the independent variable is a characteristic that can either be present or absent. In these cases, one group of subjects represent the experiment group, where the independent variable characteristic exists. The other group of subjects represent the control group, where the independent variable is absent.

The validity of psychological research relies on sound procedures in which the experimental manipulation of an independent variable can be seen as the sole reason for the differences in behavior in two groups. Research has shown, however, that an experimenter can unknowingly affect the outcome of a study by influencing the behavior of the research participants.

When the goal of an experiment is more complicated, the experimenter must design a test that will test the effects of more than one variable. These are called multivariate experiments, and their design requires sophisticated understanding of statistics and careful planning of the variable manipulations.

When the actual experiment is conducted, subjects are selected according to specifications of the independent and dependent variables. People who participate as research subjects often want to be helpful as possible and can be very sensitive to the subtle cues on the part of the experimenter. As a result, the person may use a small smile or a frown by the experimenter as a cue for future behavior. The subject may be as unaware of this condition, known as experimenter bias, as the experimenter.

Experimenter bias is not limited to research with people. Studies have shown that animals (e.g., laboratory rats) may act differently depending on the expectations of the experimenter. For example, when experimenters expected rats to learn a maze-running task quickly, the rats tended to do so; on the other hand, animals expected not to learn quickly showed slower learning. This difference in learning resulted even when the animals were actually very similar; the experimenter's expectations seemed to play a causal role in producing the differences.

Some of the studies that have examined experimenter bias have been criticized because those studies may have had methodological flaws. Nonetheless, most researchers agree that they need to control for the experimenter bias. Some strategies for reducing such bias include automation of research procedures. In this way, an experimenter cannot provide cues to the participant because the procedure is mechanical. Computer-directed experiments can be very useful in reducing this bias.

Another means of eliminating experimenter bias if to create a double-blind procedure in which neither the subject nor the experimenter knows which condition the subject is in. In this way, the experimenter is not able to influence the subject to act in a particular way because the researcher does not know what to expect from that subject.

The results of experiments can also be influenced by characteristics of an experimenter, such as sex, race, euthanasic or other personal factors. As such, a subject might act in an unnatural way not because of any behavior on the part of the experimenter, but because of the subject's own biases.