We compiled a table of all unique spellings given in response to each of the 46 words that were produced correctly by both talkers. Each unique response was assigned a classification which indicated whether (a) the entire response was a phonetic match to the stimulus item (word-correct), (b) the onset consonant or cluster of the response matched that of the stimulus item (onset-correct), or (c) the offset consonant or cluster of the response matched that of the stimulus item (offset-correct). Three judges examined the classification items in the table to arrive at a consensus. For the present paper, only word-correct scores will be considered (onset and offset data were found to provide similar, although weaker, results).
The dependent measure for the repeated measures analysis of variance was the difference in percentage of words (N=46) correct between the unaltered and F0-shifted tokens. The factorial design for the analysis was thus a 2 (TALKERS) by 2 (DIRECTION of F0 shift) by 2 (processing METHOD) design nested within listeners. An additional analysis of variance was conducted with the raw percentage correct scores to provide information about absolute word identification rates for both the original tokens and the F0-shifted tokens.