A COMPARATIVE STUDY ON DATA PROCESSING METHODS FOR NOISE ANNOYANCE RATING

Subjective rating-scale experiments are fundamental for noise annoyance assessment in psychoacoustic research; however, divergent data processing methodologies may affect result consistency. In order to evaluate their distinct effects on individual/group reliability and statistical outcomes in annoyance ratings, this study systematically compares four kinds of methods for data processing — misjudgment analysis (MA), inter-participant correlation analysis (IPCA), combined multistep analysis (CMA), and interquartile range (IQR). Using triplicate datasets from range hood noise assessments, we implemented each method's exclusion criteria. The reliability impacts were assessed through intraclass correlation coefficients (ICC(2,1) for individual consistency and ICC(2,k) for group consensus). Processing effects on annoyance metrics were evaluated using paired statistical comparisons. Results demonstrated that all methods improved individual reliability (ICC(2,1): 0.78 to 0.83), while group reliability remained near-perfect (ICC(2,k): 0.995–0.996), confirming stable group consensus and highlighting that data processing primarily addresses individual variability. IPCA, CMA, and IQR significantly reduced inter-participant variance per sound sample (p<0.001), whereas MA inconsistently affected this metric. Mean annoyance values exhibited negligible practical differences after processing (Cohen’s d<0.02, corrected for inter-observation correlation). These findings provide practical guidance for researchers selecting and applying data processing methods in subjective evaluation experiments.