ABSTRACT
Much of our knowledge of human behavior is based upon data obtained through the administration of multiple-choice tests to groups of subjects. This chapter examines some of the techniques which have been devised to assess a test's "reproducibility," "homogeneity," or internal consistency. The index of reproducibility is a function of the number of errors. The Guttman index of reproducibility meets the first criterion in that it has an absolute maximum of 100 percent for any test with more than one item, and the fourth criterion in that one can compute the index for each item as well as for the test as a whole. With short scales and wide spread in item difficulties, Guttman's figure of 90 percent may on occasion be very little higher than the minimum reproducibility of the scale. A measure of item reproducibility can be derived from the phi coefficient.
