Table 2. Correlations between the prediction performance of the random forest models using different features and self-reported SWL.

Feature set	r	p value	RMSE
Baseline (1)	0.001	0.97	1.37
LIWC (13)	0.29	1.3e -15	1.32
selected LDA (117)	0.33	< 2.2e -16	1.32
selected LDA + sentiment (120)	0.34	< 2.2e -16	1.31
selected LDA + selected LIWC + sentiment (133)	0.36	< 2.2e -16	1.30

The baseline model uses the median of the self-reported SWL with variation as feature. Root mean square error (RMSE) is relative to a range of SWL scores from the full dataset of 1.2 to 6.8. Numbers within brackets in the ‘feature set’ column are numbers of features in those sets.