Sources of variability in speechreading sentences: a generalizability analysis

Area of Science:

• Audiology
• Speech-Language Pathology
• Psychometrics

Background:

• Speechreading, or lipreading, is a critical skill for individuals with hearing impairments.
• Understanding sources of variability in speechreading performance is essential for developing reliable assessments.
• Generalizability theory provides a framework for quantifying variance components in measurement.

Purpose of the Study:

• To apply Generalizability Theory to quantify variance in speechreading sentences.
• To identify the primary sources of variability (subject, talker, sentence) in speechreading.
• To predict generalizability as a function of test length.

Main Methods:

• Utilized Generalizability Theory (Cronbach et al., 1972).
• Analyzed variance in speechreading 100 CID Everyday Sentences (Davis & Silverman, 1970) by male and female talkers.
• Presented stimuli to 104 normal-hearing subjects via videodisc recordings.

Main Results:

• Sentence materials accounted for the largest proportion of systematic variance (26.3%).
• Speechreader (10.5%) and talker (4.9%) also contributed significantly to variability.
• Residual error constituted the majority of variance (51.2%).
• Predicted generalizability coefficients of .70, .92, and .96 for 10-, 50-, and 100-item lists, respectively.

Conclusions:

• Sentence characteristics are the most influential factor in speechreading performance variability.
• Increasing test length substantially enhances the generalizability of speechreading measures.
• These findings inform the design of more reliable speechreading assessments.