The homogeneity with respect to intelligibility of recorded word-recognition materials

In developing the PB-50 word lists, J. P. Egan suggested five developmental principles, two of which were "equal average difficulty" and an "equal range of difficulty" among the lists (page 963). Egan was satisfied that each of the 20 PB-50 lists had equivalent ranges of recognition performances and that the lists produced the same average performances. This was accomplished in preliminary studies that measured the recognition performance of each word and eliminated words that were always or never correct. In preparing for studies of interrupted words, we needed to know the range of difficulty inherent in the speaker specific NU-6 and Maryland CNC materials we planned to use when those words were not interrupted. There were only a few studies in the literature that touched on the range of difficulty characteristic of the word-recognition materials in common usage. The paucity of this information prompted this investigation whose scope broadened to include the CID W-22, Maryland CNC, NU-6, and PB-50 materials spoken by a variety of speakers.

The purpose was to evaluate the homogeneity with respect to intelligibility of the words that comprise several of the common word-recognition materials used in audiologic evaluations.

Both retrospective (10) and prospective (3) studies were involved. Data from six of the retrospective studies were from our labs. The prospective studies involved both listeners with normal hearing for pure tones and listeners with sensorineural hearing loss.

The sample sizes for the 13 data sets ranged from 24 to 1,030, with 24 the typical number for listeners with normal hearing.

The retrospective data were from published studies and archived data from our laboratories. The prospective studies involved presentation of the word-recognition materials to the listeners at a comfortable level. An item analysis was conducted on each data set with descriptive statistics used to characterize the data. Additionally, skewness coefficients were calculated on the distributions of word performances and the interquartile range was used to determine minor and major outliers within each set of 200 words and their component 50-word lists (300 words for the Maryland CNCs).

For listeners with normal hearing the majority of performances on the words within a 50-word list were better than the mean performance, which produced negatively skewed distributions with outlier performances in every list. For listeners with sensorineural hearing loss the performances on the words within a 50-word list were evenly distributed above and below the mean performance, which yielded essentially normal distributions with few outliers. There were a few words on which performances were better by the listeners with hearing loss.

Every list of word-recognition materials has a few words on which recognition performances are noticeably poorer than performances on the majority of the remaining words. If the intention of an experiment is to evaluate performance at the word level, then identifying these "outliers" becomes a necessity. Although not evaluated in this report, the implications for 25-word lists are they should be based on recognition-performance data and not compiled arbitrarily.