The use of multiple regression as a data-analytic tool is examined for the cases of balanced and unbalanced designs. The utility of this method for testing specific contrasts, both orthogonal and nonorthogonal is discussed and some interpretive cautions are examined. (Author)

A slightly modified Flanders interaction coding system was used to code teacher verbal behaviors with reference to the ethnicity of the student to whom each behavior was directed. A number of statistically significant, substantial and pedagogically important disparities in teacher behaviors were directed toward Mexican American and Anglo students.…