Pairwise listener preferences in hearing aid evaluation.

The relationships among pairwise judgments of the quality of connected discourse, pairwise judgments of the relative intelligibility of discourse, and measured intelligibility on a nonsense-syllable test were evaluated under identical conditions of primary talker and competitive babble. Stimuli were processed by eight hearing aids and presented in a repeated-measures design to 12 listeners with sensorineural hearing loss. Results revealed moderately high test-retest reliability for all three experimental conditions. Overall, a noteworthy positive relationship was evident between relative intelligibility judgments and measured phonemic identification, although this relationship varied considerably among individual listeners. The correspondence between quality judgments and relative intelligibility judgments was substantially lower, while the relationship between judgments of quality and phonemic identification scores was negligible. Findings demonstrate the potential importance of instructional set in producing valid judgments of the relative intelligibility of aided speech. The determination of measurement error inherent in pairwise preference data is discussed from the viewpoint of a probabilistic model encompassing a binomial distribution. It is concluded that experimental optimization of the agreement between relative intelligibility judgments and measured intelligibility performance of individual listeners will be required before the pairwise comparison technique can be considered a viable alternative to traditional hearing aid evaluation procedures.