Table 2.

Multi-linear regression results of childhood adversities/socioeconomic variables that are associated with resilience (for overall cohort, n=1713).

Variable	Model 1 (R2=14%)			Model 2 (R2=17%)			Model 3 (R2=23%)
	Estimate	SE	P-value	Estimate	SE	P-value	Estimate	SE	P-value
Intercept	88.31	6.22	<0.0001	106.87	6.185	<0.0001	106.50	6.163	<0.0001
Childhood economic adversity
0	Reference group			–			Reference group
1	−1.22	0.606	0.0452				−0.76	0.594	0.2010
2	−4.34	0.823	<0.0001				−3.35	0.809	<0.0001
3	−1.58	1.170	0.1764				−0.20	1.152	0.8678
Childhood social adversity
0	Reference group			–			Reference group
1	1.56	0.752	0.0390				1.38	0.736	0.0612
2	0.01	1.061	0.9957				0.68	1.040	0.5133
3	−1.14	1.732	0.5098				−0.71	1.691	0.6782
Childhood physical abuse
Yes	Reference group			–			–
No	−1.89	0.832	0.0229	–
Income sufficient
Yes	–			Reference group			Reference group
No	–			−4.19	0.657	<0.0001	−4.04	0.660	<0.0001
Social support provided by friends
No friends				−1.52	0.741	0.0401	−1.34	0.745	0.0733
low support	–			−1.07	0.655	0.109	−0.95	0.654	0.1487
High support				Reference group			Reference group
Social support provided by children (no children vs low support)
No children	–			−2.17	0.918	0.0182	−2.16	0.914	0.0185
Low support	–			−2.05	0.619	0.001	−1.97	0.621	0.0016
High support	–			Reference group			Reference group
Social support provided by partner
No partner	–			0.38	0.630	0.5429	0.38	0.629	0.5414
low support	–			−0.52	0.695	0.4557	−0.49 −0.71	0.693	0.4764
High support	–			Reference group			Reference group
Social Engagement	–			1.23	0.329	0.0002	1.16	0.329	0.0004

^* Model 1: childhood economic and social adversities only; Model 2: current socioeconomic status only; model 3: both models 1 and 2 together. All three regression models accounted for the following confounders: age, sex, and sites. Scores: social and economic: 0(none); 1(one); 2(two); 3 (three adversities) for site-specific analyses. Friends, children and partner social support were categorized into 3 subgroups: no friend/partner/ children, low, and high support based on site=specific quartiles. Significant p-values are bolded. Regression procedure (proc reg) in SAS uses list-wise deletion and this explains the extra missing cases in the regression models.