EMPIRICAL METHODS FOR DIFFERENTIAL ASSIGNMENT TO ALTERNATIVE CURRICULA (APTITUDE-TREATMENT INTERACTION

Prior research on the use of formal decision rules in the assignment of students to alternative educational treatments has focused on determining an optimal decision rule, and the value of instructional adaptation using that decision rule. In deriving their results, previous researchers have assumed all relevant information about the probability distributions of the measure used in the decision rule and the criterion variable are known. This dissertation examines using data to determine a decision rule and to evaluate whether instructional adaptation to individual differences based on using that decision rule has sufficient instructional benefit to be implemented.;The optimal decision rule in a particular situation depends upon the joint probability distribution, in each instructional treatment, of the measure used in the decision rule and the criterion variable. In applied settings, only non-optimal decision rules using data from samples are available. Many types of such non-optimal decision rules based on data (referred to as empirical decision rules) can be used. Monte Carlo methods are used to compare two major types of empirical decision rules that have been previously proposed. Conditions under which each type of empirical decision rule gives better performance are presented.;The effect on empirical decision rules of non-random assignment to treatments is considered. A Monte Carlo study indicated that non-random assignment to treatments, in many cases, does not have a significant negative effect on empirical decision rules relative to random assignment. Non-random assignment based on a variable not used in the decision rule is found to perform only slightly better than non-random assignment based on the variable to be used in the decision rule.;Using data to determine whether an instructional adaptation to individual differences has sufficient instructional benefit as compared with no instructional adaptation to individual differences is discussed. A method for making such determinations is presented and compared to a commonly used approach.