A shared structure for emotion experiences from narratives, videos, and everyday life

Summary Our knowledge of the diversity and psychological organization of emotion experiences is based primarily on studies that used a single type of stimulus with an often limited set of rating scales and analyses. Here we take a comprehensive data-driven approach. We surveyed 1,000+ participants on a diverse set of ratings of emotion experiences to a validated set of ca. 150 text narratives, a validated set of ca. 1,000 videos, and over 10,000 personal experiences sampled longitudinally in everyday life, permitting a unique comparison. All three types of emotion experiences were characterized by similar dimensional spaces that included valence and arousal, as well as dimensions related to generalizability. Emotion experiences were distributed along continuous gradients, with no clear clusters even for the so-called basic emotions. Individual differences in personality traits were associated with differences in everyday emotion experiences but not with emotions evoked by narratives or videos.


Fig. S7.
Temporal trend across waves for (a-g) covid-related metrics at the US national level (a: daily covid cases, b: daily covid deaths, c: percent change in mobility from baseline, d: total number of states with mandatory stay at home restrictions, e: percent of population reporting always wearing a mask when leaving home, f: unemployment rate, and g: anti-racism protest counts), and (h) 'negative affect' factor in real-life emotions, Related to Figure 4. Data sources: https://www.healthdata.org/covid/data-downloadsand https://coviddynamic.caltech.edu/data-sharing.0.0 0.3 0.6 0.9 0.0 0.3 0.6 0.9 0.0 0.3 0.6 0.9 0.0 0.3 0.6 0.9 0.0 0.3 0.6 0.9 0.0 0.3 0.6 0.9 0.0 0.3 0.6 0.9 0.0 0.  Table S1.Definition of 28 rating scales, with the description of the two ends of the scales and grade level required to understand the definitions, Related to STAR Methods and Figure 1.

Fig. S2 .
Fig. S2.Parameter selection for UMAP, Related to STAR Methods.The means (dots) and standard deviations (error bars, n = 10 iterations) of trustworthiness (upper) and rank correlation of pairwise distances (lower) of the training data and testing data (left to right) for different sizes of the local neighborhood for (a) emotions evoked by narratives, (b) emotions evoked by videos, and (c) real-life emotions.

Fig. S4 .
Fig. S4.Representational structures across stimulus domains, Related to Figure 3. Correlation matrices across 18 shared scales for (a) emotions evoked by narratives, (b) emotions evoked by videos, and (c) real-life emotions.All matrices were sorted using hierarchical clustering applied to real-life emotions for easier comparison.

Fig. S5 .
Fig. S5.Dimensionality analysis with cross-validation, Related to STAR Methods and Figure 4.The means (dots) and standard deviations (error bars, n = 20 iterations) of explained variance from the EFA on training data on the left, and root mean square error of approximation (RMSEA) fit index from the CFA on testing data on the right for (a) emotions evoked by narratives, (b) emotions evoked by videos, and (c) real-life emotions.

Fig. S8 .Fig. S9 .
Fig. S8.Individual differences in 'negative affect' factor in real-life emotion experiences, Related to Figure 4. Empirical cumulative distribution function for different groups partitioned based on (a) gender, (b) geographic regions, and (c) political parties.

Fig. S13 .Fig. S14 .
Fig. S13.Recovery of intended categories, Related to Figure 6.(a) Contingency matrix between the 20 discovered categories (columns) and the ones intended (rows) for emotions evoked by narratives, and (b) Contingency matrix between the 30 discovered categories (columns) and the ones intended (rows) for emotions evoked by videos.

Fig. S15 .
Fig. S15.Individual differences in real-life emotion experiences (corrected for baseline rating biases from narrative/video ratings), Related to Figure 7. Pairwise Pearson's correlations (upper) between ratings and demographic and psychological variables, and Welch's t-test (lower) for means of ratings of different groups (divided based on sex, education and political party; t-statistics for males -females, high -low education, and republicans -democrats respectively).Raw results with p < 0.05 are colored (otherwise masked) and Bonferroni corrected results with p < 0.05 are annotated with values.

Fig. S16 .Fig. S17 .
Fig. S16.Dimensionality reduction results, Related to STAR Methods and Figure 4.The percentage of explained variance by EFA and PCA for (a) emotions evoked by narratives, (b) emotions evoked by videos, and (c) real-life emotions.The means (points) and standard deviations (bars) of the explained variance (10 iterations) on the training data and testing data from autoencoders with various numbers of units in the hidden layer for (d) emotions evoked by narratives, © emotions evoked by videos, and (f) reallife emotions (colors indicate different configurations of activation functions in the encoder and decoder layers).The percentage of explained covariance by PPCA for (g) emotions evoked by narratives, (h) emotions evoked by videos.

Fig. S18 .
Fig. S18.Histograms of pairwise Euclidean distances in the original high dimensional spaces for (a) emotions evoked by narratives, (b) emotions evoked by videos, and (c) real-life emotions, Related to Figure 5 and Figure 6.

Table S2 .
Demographic characteristics (means and standard deviations) of all measures for the final sample after exclusion, Related to STAR Methods and Figure1.