Table 3

Model fitting criteria, effect selection tests and one goodness-of-fit metric for regression models with tension as dependent variable and the three timbral semantics as predictors. Two different models were created based on the profile pattern (multinomial logistic regression) and the maximum magnitude (ordinal regression). The Akaike information criterion (AIC) measures the model quality by balancing between goodness of fit and parsimony. The lower the values the better the quality. −2 log likelihood is a measure of the unexplained variance by the model therefore lower values are also desirable. A Chi-square test assesses the significance of each additional predictor by measuring whether the difference in −2 log likelihood resulting from its inclusion is significant. Finally, McFadden Pseudo R ² is a metric of the model’s goodness-of fit. As there is not a straightforward interpretation of this metric, its values should not be judged by the standards for a good fit in a regression analysis [24]. Although in general a larger value means that more variance is explained by the model, such metrics for logistic regression are most useful for comparing competing models for the same data.