ABSTRACT
While dimensional models play a key role in emotion psychology, no consensus has been reached about their number and nature. The current study sheds a new light on this central issue by examining linear and non-linear relationships among the dimensions in the cognitive emotion structure. The meaning of 80 emotion terms was evaluated on 68 features representing appraisals, action tendencies, bodily reactions, expressions, and subjective experiences by 213 English-speaking US, 156 French-speaking Swiss, and 194 Indonesian-speaking Indonesian students. In a two-dimensional valence and arousal representation, neither linear nor non-linear relationships were observed. In a four-dimensional valence, power, arousal, and novelty representation, both linear (e.g. a positive relationship between valence and power) and non-linear (e.g. a strong positive correlation between valence and power found only for positively valenced emotion terms) relationships were observed. Joy- and sadness-related emotion terms where about as well represented by the two- than by the four-dimensional representation. However, especially anger- and surprise-related terms were only adequately represented by the four-dimensional representation. These findings were generalisable across the three languages. Even though a two-dimensional structure fits the data well in general, four dimensions are needed to sufficiently represent the cognitive structure of the whole gamut of human emotions.
Acknowledgements
We thank Efrata Sri Rezeki Kristina who collected the Indonesian data as well as Anastasia R. Hendrarini, Ignatius Y. P. Putra, Johana E. P. Hadiyono, Lisa Ratriana, Nanang Suprayogi, Ranta D. Suryaratri, Rena Latifa and Susi Fitri for their support of the research process in Indonesia.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 Preliminary results of the French-speaking sample only looking at the four-dimensional representation with orthogonal dimensions has been reported in Gillioz et al. (Citation2016).
2 The uneven distribution in terms of gender, with women substantially outweighing men is a limitation of the current study. However, previous research using a similarity rating approach between emotion terms found no significant gender difference in an adolescent, a student, and a community sample with respect to the cognitive structure of emotions (Veirman & Fontaine, Citation2015). Apparently women and men are quite alike when it comes to cognitively representing the emotion domain. Gender differences are probably more likely to be found in the proneness to experience and express certain emotions.
3 In addition, exploratory investigations of higher-dimensional structures were performed. However, these higher-order dimensions were difficult to interpret and had very few features that exclusively defined the respective dimensions.
4 Since all four dimensions are bipolar dimensions with a meaningful zero point (not positive, not negative; not strong, not weak; not relaxed, not aroused; not expected, not novel), doing separate analyses for terms that score below or above the zero point on each dimension has intuitive appeal. An additional regression analysis has been executed with the coordinates on the arousal dimension being the criterion and the coordinates on the valence dimension (in step 1) and the squared coordinates on the valence dimension (in step 2) being the predictors. The squared coordinates in the second step, however, did not add any information [F(1,237)=1.321, p=.252, ΔR²=.005], which confirms the absence of a V-shaped relationship.
5 For this computation, the feature scores have first been standardized across emotion terms. This computation is not to be confounded with the communalities of the features in the principal component structure.