Experimenter Bias occurs when the actions of the experimenter unknowingly bias
the outcome of an experiment thus confound
ing the results.
In one study by Rosenthal, experimenters were given two groups of rats, one they were told had been bred told run mazes quickly, the other bred to run mazes slowly. In actuality the rats are all of the same type and were randomly assigned to the groups. However the rats that the experimenters believed had been bred to run mazes quickly actually did solve mazes much more quickly than the rats in the "slow rat" group. Researchers discovered that the experimenters handled and played with the rats in the "fast" group much more than the "slow" rats, thus changing their performance.
Although not directly an example experimenter bias it is interesting to note that this effect has also been paralleled with school teachers. At the beginning of a school year researchers administered a battery of phoney intelligence tests to a class of elementary school students. The researchers randomly selected two groups of students and told the teacher that the first group had performed exceptionally well on the tests and would certainly excel in academics, while the other group had performed very poorly and would most likely not fare so well. Despite the fact that there was really no difference between the abilities of the two groups, at the end of the year the students who the teacher had been told would do very well were at the top of the class and the students in the other group, were at the bottom. Researchers concluded that the teachers expectations of how the students would perform biased how the teacher interacted with the students. Due to stronger ethical constraints it is unlikely that an experiment such as this would be allowed today.
Experimenter bias can be reduced in several ways. First because a human researcher can unknowingly and subtly convey response cues to a subject (such as the case of Clever Hans) one can attempt to mechanize the process as much as possible. For example a computer generated voice in a neutral tone could not accidentally convey any "extra" unintentional information to the participant.
Another method of reducing experimenter bias is through the use of a double blind study. In this situation a study will be run by one person who knows what is being tested, but the actual people doing the experimenting are left unaware. For example the department chair at a hospital may know who will receive an experimental drug and who will receive a placebo but the actual doctors administering the drugs will not.
ROSENTHAL, R. (1966) Experimenter effects in behavioural research. New York: Appleton-Century-Crofts.