Objective: This study examined the statistical properties of normal air-conduction thresholds obtained with automated and manual audiometry to test the hypothesis that thresholds are normally distributed and to examine the distributions for evidence of bias in manual testing.
Design: Four databases were mined for normal thresholds. One contained audiograms obtained with an automated method. The other three were obtained with manual audiometry. Frequency distributions were examined for four test frequencies (250, 500, 1000, and 2000 Hz).
Study sample: The analysis is based on 317 569 threshold determinations of 80 547 subjects from four clinical databases.
Results: Frequency distributions of thresholds obtained with automated audiometry are normal in form. Corrected for age, the mean thresholds are within 1.5 dB of reference equivalent threshold sound pressure levels. Frequency distributions of thresholds obtained by manual audiometry are shifted toward higher thresholds. Two of the three datasets obtained by manual audiometry are positively skewed.
Conclusions: The positive shift and skew of the manual audiometry data may result from tester bias. The striking scarcity of thresholds below 0 dB HL suggests that audiologists place less importance on identifying low thresholds than they do for higher-level thresholds. We refer to this as the Good enough bias and suggest that it may be responsible for differences in distributions of thresholds obtained by automated and manual audiometry.
Keywords: Audiometry; air conduction; automated audiometry; bias; hearing; hearing test; normal hearing; pure-tone thresholds; threshold.