ABSTRACT
In this chapter, the major approaches to reliability estimation will be discussed. In Chapter 3, we noticed that test reliability is equal to the correlation between a test and a parallel test. The moment of administration of the second test, however, is of crucial importance, as it may have an influence on error variance. If there is a long time interval between the administration of the first and the second test, the factor of time may play an important role-persons may change in the time between testing. On the other hand, when tests are administered consecutively, fatigue is to be expected to come into play. Therefore, with test administration in one session, it is advisable to split the persons into two groups, one of which is administered test
X
first, followed by test
Y
, and the other is given the two tests in reverse order. As the
parallel-test method
is not without its problems (see again Guttman, 1953), an alternative method for reliability estimation would be to administer the test twice. This method is the
test-retest method
. With a small time interval between test sessions, the risk is large that on the second test occasion persons remember their answers given on the first occasion. This would be a violation of the assumption of experimental independence. This violation would have a negative effect on the quality of the reliability estimate. With a larger time interval, persons might be changed on the characteristic of interest. Therefore, the test-retest method is useful only when a relatively stable characteristic is to be measured. The resulting reliability coefficient is called a stability coefficient for this reason.
