Expected Reliability as a Function of Choices Per Item

BY making several reasonable assumptions, it is possible to estimate the reliability that should be expected in an objective test as a function of the number of choices per item. These estimates for a 100 item objective test are presented in Table 1. They indicate an appreciable increase in reliability when the number of choices per item is increased from two to three; a smaller increase in going to four choice items; and still smaller increases beyond that point. They indicate, too, that a 100 item true-false test yielding a reliability coefficient of .74 is a reasonably good test, as true-false tests go, whereas a reasonably good 100 item test made up of four-choice items ought to yield a reliability coefficient of .86. The same assumptions can be used to answer a different question: how many two-choice, three-choice, etc. items must be included in