Matched Latin Squares: A Nifty Solution to a Tricky Design Problem.

The effects of various discourse variables on the psycholinguistic processing of sentences within paragraphs by 36 competent adult readers were investigated. Eye movement measures were used as the dependent variables reflecting psycholinguistic processing. When all of the constraints within which the study had to be performed were examined, it became obvious that none of the traditional research designs would be appropriate. Hence, a new experimental design, Latin square matched with other Latin squares, was conceived which met the constraints of the study. This design has some features which make it very useful for research in many different fields of education. Furthermore, from a statistical analysis point of view, this design has several desirable characteristics: it allows the researcher to test a larger set of hypotheses than is possible under other designs employing the same number of individuals, and it allows the researcher to gain sufficient power for the hypothesis tests while using only a small number of individuals. A description of the study and its constraints, an explanation of why the traditional designs cannot be used, and an introduction to the conceived design are given. The statistical analysis techniques to be employed on data collected using this design are then explained in detail. (Author/PN)