Examination of the Validity of Auditory Traits and Tests

In an unaided context, and in spite of obvious conceptual differences, direction and distance hearing, soft sounds hearing, and speech understanding in noise hearing are so strongly interrelated that accurate measures of a listener's particular abilities on one of these traits would be difficult to obtain, given prior knowledge of performance on at least one of the others. The tests that are commonly used to estimate a listener's hearing performance do not provide sufficiently accurate and unbiased information. Most auditory tests are heavily influenced by method-related factors that are not impacted by the trait expected to be measured by the test. Method-related variance in test scores is not related to the trait of interest, but might still provide useful information about the person being tested. If a tester must use a set of tests with small amounts of trait variance, he/she can obtain better estimates of a person's performance on the trait by: including tests that use very different measurement methods (e.g., psychophysical, self-report, and significant-other report) converting the scores from these tests into the same metric (e.g., standardized scores) and combining the scores into an overall score. Although this is informally done in most clinical situations, formal procedures and norms should be developed. In clinical trials, outcome measures with large amounts of method variance are less likely to identify treatment options that are indeed beneficial. Method variance is systematic (i.e., an individual's bias associated with a measurement method can be expected to be stable over time and consistently different from people with other method biases. Because of this, the method variance component of a person's test score cannot be expected to average out over repeated samples, nor can it be expected to be changed by a treatment designed to change performance on a trait. Observable differences between treatments are reduced by method variance in outcome measures. Acknowledgments The author thanks Robyn Cox and Will Shadish for their consultation and assistance with this project, and Ruth Bentler, Martyn Hyde, and an anonymous reviewer for their helpful comments on an earlier version of this manuscript. Chris Bentler and Jason Ruyle also deserve recognition for their help entering the simulation study data.