Table 2.

Success rates of the most commonly measured relationship-specific constructs across datasets

Construct	No. of predictors tested		Percent of actor versions successful		Percent of partner versions successful		Overall success rate, %
	Predicting satisfaction	Predicting commitment	Predicting satisfaction, %	Predicting commitment, %	Predicting satisfaction, %	Predicting commitment, %
Perceived partner commitment	10	10	90	70	100	80	85
Intimacy	12	9	92	92	67	67	81
Appreciation	10	10	90	80	60	60	72
Love	17	17	88	53	76	65	71
Sexual satisfaction	20	13	90	75	54	54	71
Perceived partner satisfaction	11	9	91	64	78	44	70
Conflict	29	28	90	79	57	50	69
Perceived partner responsiveness	14	13	93	57	69	54	69
Trust	15	15	87	60	73	53	68
Investment	13	13	77	62	92	38	67
Support general	12	9	67	42	89	67	64
Capitalization	16	10	81	62	40	30	58
Normative attachment	13	13	69	38	69	54	58
Relationship length	54	41	59	67	44	56	57
Passion	14	13	64	50	54	46	54
Alternatives	12	12	58	33	67	50	52
Sexual frequency	11	8	73	36	25	50	47
Inclusion of the other in the self	24	23	54	33	65	35	47
Affection	10	7	50	50	29	43	44
Empathy	11	11	45	36	45	45	43
Intimate partner violence	26	17	27	62	47	35	43
Conflict strategies	23	15	52	30	27	27	36
Power	13	13	31	31	31	23	29
Relationship status	27	21	26	22	38	29	28
Cohabiting	15	14	27	20	29	36	28
Sacrifice motives	22	22	18	18	14	14	16
Children	32	23	16	6	4	13	10

Note: Success rate percentages can be interpreted as the strength of the variable relative to the other variables of this class, but it does not have any independent meaning or effect size. Random Forests do not specify the size or direction of the effect; only that the variable meaningfully contributes to the total variance explained in a given model. Some studies included multiple measures of the same construct, and thus the number of predictors tested can be higher than the total number of datasets. Boldfaced rows correspond to four constructs excluded from the primary models reported in the main text, because they are debatably indicators (not predictors) of relationship quality (1). The values for these four constructs derive from alternative models reported in SI Appendix.