Reliability of procedures used for scaling loudness

In this study, 16 normally-hearing listeners judged the loudness of 1000-Hz sinusoids using magnitude estimation (ME), magnitude production (MP), and categorical loudness scaling (CLS). Listeners in each of four groups completed the loudness scaling tasks in a different sequence on the first visit (ME, MP, CLS; MP, ME, CLS; CLS, ME, MP; CLS, MP, ME), and the order was reversed on the second visit. This design made it possible to compare the reliability of estimates of the slope of the loudness function across procedures in the same listeners. The ME data were well fitted by an inflected exponential (INEX) function, but a modified power law was used to obtain slope estimates for both ME and MP. ME and CLS were more reliable than MP. CLS results were consistent across groups, but ME and MP results differed across groups in a way that suggested influence of experience with CLS. Although CLS results were the most reproducible, they do not provide direct information about the slope of the loudness function because the numbers assigned to CLS categories are arbitrary. This problem can be corrected by using data from the other procedures to assign numbers that are proportional to loudness. [Supported by NIH]