Table 13.

Best 5 discriminant models for discriminating poor response to self-rated mental health from other ratings (fair to excellent).

Number of predictors and predictor names		AUCa	General accuracy (%)	Average accuracy (%)	F1 score
1
	Analytical thinking score	0.59	79.4	84.3	0.56
	Discrepancy	0.59	79.4	82.7	0.56
	Insight	0.53	77.6	53.5	0.49
	Present focus	0.51	79.4	75.9	0.55
	Causation	0.50	79.4	82	0.55
2
	Analytical thinking score+other words	0.66	79.4	88.4	0.55
	Discrepancy+drives	0.64	79.4	87.2	0.55
	Clout score+social processes	0.63	79.4	87.2	0.55
	Analytical thinking score+emotional tone score	0.62	79.4	85.9	0.56
	Analytical thinking score+positive emotions	0.62	79.4	85.8	0.56
3
	Positive emotions+discrepancy+personal concerns	0.70	81.2	89	0.63
	Analytical thinking score+other words+drives	0.69	79.4	86.5	0.55
	Analytical thinking score+clout score+other words	0.68	76.4	88.8	0.45
	Analytical thinking score+positive emotions+other words	0.67	79.4	86.1	0.56
	Analytical thinking score+positive emotions+discrepancy	0.66	80	88.6	0.59
4
	Analytical thinking score+other words+cognitive processes+drives	0.72	78.2	87.5	0.52
	Analytical thinking score+positive emotions+discrepancy+personal concerns	0.71	78.2	89.2	0.52
	Positive emotions+causation+discrepancy+personal concerns	0.70	77.6	90	0.49
	Analytical thinking score+clout score+first-person pronouns+positive emotions	0.70	80	88.8	0.59
	Analytical thinking score+positive emotions+discrepancy+drives	0.70	78.2	87.6	0.52
5
	Analytical thinking score+positive emotions+causation+discrepancy+drives	0.73	80.6	87.9	0.60
	Analytical thinking score+clout score+positive emotions+social processes+other words	0.72	80	88	0.59
	Analytical thinking score+emotional tone score+discrepancy+other words+personal concerns	0.72	78.8	87.9	0.54
	Analytical thinking score+positive emotions+discrepancy+other words+drives	0.72	80	87.5	0.59
	Clout score+positive emotions+discrepancy+social processes+personal concerns	0.72	81.2	89.7	0.63
21b	All predictors	0.45	34.7	32.1	0.30

^aAUC: Area under the Receiver Operating Characteristic curve.

^bThe best models with 6-20 predictors (total of 75 items) have lower AUC, general accuracy, average accuracy, and F1 score, thus are omitted.