EFFECTS OF VIOLATING ITEM CONSTRUCTION PRINCIPLES

Two forms of a social studies achievement test were constructed with half the items for each form containing a cue, grammar, or length fault. Faults were found to make the items easier, which was supported by confidence intervals for the differences. However, validity coefficients with achievement and intelligence criteria, as well as the reliability coefficients, were virtually unchanged. The results agreed with those of Dunn and Goldstein (1959), even though the methodology differed. A suggested measure of test-wiseness for groups is presented.