Balanced Designs for Triad Tests: Two Examples from English

An attempt was made to assess the reliability of balanced incomplete block design for semantic studies using triads tests. These designs allow for substantial reduction in the number of triads, making possible the extension of the triads technique to larger semantic domains. The investigation is pursued through two experiments. In the first, the semantic organization of nine vegetable names is instigated using several different incomplete designs (balanced & unbalanced) & the complete set of triads. Balanced incomplete block designs with two or more replications per pair of words produce data with high correlations to the complete triads data, but balanced designs with only one replication per pair are not reliable. Unbalanced designs are less reliable than balanced designs. In the second experiment, the semantic organization of fifteen English kin terms is investigated using a balanced block design with three replications per pair. In this experiment the results are compared to the results for complete triads on a subset of seven M terms plus cousin. Data from balanced block designs with two or more replications are found to be reliable. A model is formulated to explain why designs with only one replication per pair are unreliable. Examples of balanced block designs are appended. 6 Tables, 3 Figures, Appendix. Modified AA.