A review of regression analysis methods: Establishing the quantitative relationships between subjective soundscape assessment and multiple factors

摘要

Soundscape research has long been exploring factors/indicators that would impact human perception, emotional assessment, behaviour, etc. Since soundscape is intrinsically a complex system, a wide variety of factors have been suggested by intensive studies and also ISO/TS 12913-2 standard, ranging from acoustics, psychoacoustics, sound source composition, to demographics, and other personal/social/cultural factors. To examine subsequently the quantitative relationships between such factors and impacts/effects of soundscape, data statistical methods are essential. A wide range of statistical methods have been commonly used in soundscape studies corresponding to different specific research tasks, e.g. correlation analysis, analysis of variance, factor analysis and cluster analysis. While soundscape is affected by not only any single factor but multiple factors simultaneously and interactively, this paper focuses on a statistical method, regression analysis, that can investigate multiple variables to study the relationships and predict soundscape effects from multiple possible factors. This paper provides a brief review of a set of regression analysis methods, which are used for analysing different types of variables, e.g. continuous, ordinal and nominal. It then exemplifies the methods with a number of regression models from previous soundscape studies, particularly on subjective soundscape assessments (namely soundscape descriptors, e.g. pleasantness, satisfaction, and comfort), which is a crucial part in soundscape research.