Using paired comparisons to create the semantic construct of frequency

Using a paired comparison data collection procedure and a one-faceted Rasch model, a stable construct of frequency is derived using 43 non-numeric quantitative descriptors, such as never, almost never, rarely, sometimes, often, etc. This construct is acontextual in that only the words or phrases were compared. The raters were not supplied with any particular context. To make the paired comparison survey more manageable certain assumptions were made that would result in a sparse, but connected 43 x 43 data matrix. Three hundred ninety-six pairs were selected and each pair was assigned to one of two forms. The quality control procedures to detect pair-order effects, fatigue effects, aberrant raters, model misfit, etc. are discussed. The results were mapped onto a continuum, which seems to represent the common understanding of frequency using non-numeric quantitative descriptors. These results have implications for how many different strata of frequency people can reliably differentiate in spoken language. This study can also serve as a primer for using paired comparision data collection procedures with the Rasch model and the methodology can easily be applied to other similar semantic continua. This study used a small sample and the pool of raters was not very diverse, therefore future studies should further consider those issues.