ABSTRACT
This book has thus far been concerned only with data structures in which each of the Y observations has come from a different sampling unit or “ subject.” There is another class of research designs in which each of the n subjects gives rise to a Y observation under each of c (> 1) different conditions (C). The data layout has n rows, c columns, and hence a total of N = nc Y observations. Such repeated measurement designs are frequently used in psychological experimentation. The statistical issues are much the same for matched-subject designs: instead of a single subject being observed under the c conditions, each row contains a set of c subjects that are homogeneous with regard to one or more variable(s) that are believed to be correlated with Y, but extraneous to the purpose of the experiment as would be the case, for example, for litter mates. Such matching (or blocking) serves to control this extraneous variance in comparisons involving C. Thus, whether a row represents a single subject or a “ c-tuple” 1 of matched subjects, variance in Y due to systematic between-row differences can be identified and removed with salutary effects on precision and statistical power in studying the effects of C.
