Comparison of logistic-regression based methods for simple mediation analysis with a dichotomous outcome variable

Table 4 Application of the three compared methods to the real-life data examples

		Multiple regression and SEM				Potential outcomes^a
		Crude	y-standardization	Full-standardization	Standardized logistic solution^b
Situation 1 M continuous Y dichotomous	Total effect (c)	−0.20	−0.11	−0.20	−0.21	−0.21
	a coefficient^d	0.67	0.67	0.67	0.67	0.67
	b coefficient	−0.04	−0.02	−0.23	−0.04	−0.04
	Direct effect (c’)	−0.19	−0.10	−0.18	−0.19	−0.19
	Indirect effect					−0.03
	ab	−0.03	−0.01	−0.15	−0.03
	c-c’	−0.02	−0.01	−0.02	−0.02
	Proportion mediated					0.12
	ab/(ab + c’)	0.12	0.12	0.47	0.12
	ab/c	0.13	0.12	0.77	0.12
	1-(c’/c)	0.09	0.11	0.11	0.11
Situation 2 M dichotomous Y dichotomous	Total effect (c)	−0.20	−0.11	−0.20	NA	−0.21
	a coefficient	0.11	0.06	0.11	NA	0.11
	b coefficient	−0.60	−0.32	−0.16	NA	−0.60
	Direct effect (c’)	−0.19	−0.10	−0.19	NA	−0.19
	Indirect effect					−0.01
	ab	−0.06	−0.02	−0.02	NA
	c-c’	−0.01	−0.01	−0.01	NA
	Proportion mediated					0.07
	ab/(ab + c’)	0.25	0.16	0.09	NA
	ab/c	0.32	0.17	0.09	NA
	1-(c’/c)	0.06	0.07	0.07	NA

Abbreviations: SEM structural equation modeling, M mediator variable, Y outcome variable, NA not available
^aThe output of the potential outcomes framework contains odds ratios, the coefficients in the table are log transformed to make the coefficients comparable to the coefficients yielded by multiple regression and SEM
^bThe standardized logistic solution cannot be applied to mediation models with a dichotomous mediator variable
^dThe a coefficient is based on linear regression

ISSN: 1471-2288