Estimating the magnitude of anxiety from speech.

Abstract

Listen

Translate article

The main purpose of this investigation was to determine how much, if any, information is lost towards estimating the amount of anxiety in speech by using only content or lexical variables in speech as compared with the measures of anxiety obtained when vocal properties as well as content variables are employed. A set of twelve free-associative five-minute tape recordings of speech produced by twelve different people were rated with respect to the magnitude of the anxiety by 16 judges (11 psychiatrists and 5 psychologists) following the anxiety rating scale of the Overall and Gorham Brief Psychiatric Rating Scale. The judges first rated the typescripts of the verbal samples and one month later the judges again estimated the magnitude of the anxiety in these speech samples by listening to the tape recordings of the verbal samples while reading the typescripts. Since the problem of quantitative assessment of anxiety involves, necessarily, the problem of validity of the construct being measured, and validity in turn depends in part on reliability of measurement, reliability of ratings of anxiety was calculated. Raters were less able to differentiate the verbal samples, differed considerably more in their average ratings, and had greater error variance when using the typescripts alone than when using typescript plus sound recording. The product-moment correlation between these two sets of anxiety scores, calculated from the covariance and variance of the sums of the ratings over the 16 judges, was 0.846. Correcting this correlation for attenuation, due to the imperfect reliabilities of the separate anxiety ratings under the two conditions, gave an estimated correlation of 0.913. Consequently, 16.6 percent of a universe of anxiety scores derived from typescripts plus sound recordings would not be predictable from anxiety scores derived from typescripts alone. Product-moment correlations were also obtained between anxiety scores derived from a content analysis method applied to only the typescripts of the five-minute verbal samples (the Gottschalk-Gleser method) and the two sets of anxiety ratings of the verbal samples obtained by the Overall-Gorham scale. These correlations were as follows: (1) anxiety rating by Overall-Gorham scale, using typescript plus sound recording, and Gottschalk-Gleser anxiety scores = 0.86; (2) anxiety rating by Overall-Gorham scale, using typescript alone, and Gottschalk-Gleser anxiety scores = 0.78. These findings indicate that there is negligible information lost with respect to assessing anxiety, when using either the Overall-Gorham or Gottschalk-Gleser anxiety scales, if typescripts alone are used or typescripts plus sound recordings are used. Moreover, the findings tend to support a theory of the redundancy of lexical and vocal factors rather than an additive theory in the expression and communication of the intensity of affects in speech.