The Generalizability of Class Means

Recently, a number of researchers have given serious consideration to the problem of estimating reliability when the unit of analysis is a class mean or some other aggregate score for a set of persons. Haney (1974a, 1974b) has reviewed important aspects of the relevant literature. The study of this topic has been motivated by the analysis of data from several different sources. In particular, large scale evaluations, such as those, undertaken for Head Start (see Smith & Bissell, 1970) and Follow Through (see Abt Associates, 1974), frequently require estimates of reliability when class is the unit of analysis. Similar issues arise in the study of course evaluation questionnaires (See Kane, Gillmore, & Crooks, 1976), and studies of school effectiveness and accountability (see Dyer, Linn, & Patton, 1969; Marco, 1974). Using concepts from classical reliability theory, Shaycoft (1962), Wiley (1970), and Thrash and Porter (Note 1) have de-