Understanding causal pathways within health systems policy evaluation through mediation analysis: an application to payment for performance (P4P) in Tanzania

Table 9 P4P Average causal mediation effect (ACME) and direct effects, sensitivity analysis using Imai et al. (2010) approach

	Mediation analysis			Sensitivity results
	Mean	[95% Conf.	Interval]
Mediators for delivery in health facility
Oxytocin stock-out
ACME	0.017	0.006	0.028
Direct effect	0.059	0.021	0.095	Correlation at which ACME = 0	−0.0439
Total effect	0.077	0.041	0.109
% of tot eff mediated	0.226	0.159	0.424
Health worker kindness
ACME	0.039	0.031	0.048
Direct effect	0.042	0.007	0.075	Correlation at which ACME = 0	0.2085
Total effect	0.081	0.045	0.116
% of tot eff mediated	0.482	0.337	0.864
Percentage of women who paid for delivery in HF
ACME	0.041	0.032	0.051
Direct effect	0.040	0.004	0.072	Correlation at which ACME = 0	−0.2326
Total effect	0.081	0.045	0.113
% of tot eff mediated	0.511	0.364	0.911
Mediators for delivery in public HF
Oxytocin stock-out
ACME	0.019	0.008	0.031
Direct effect	0.037	0.000	0.077	Correlation at which ACME = 0	−0.0437
Total effect	0.057	0.021	0.096
% of tot eff mediated	0.342	0.202	0.933
Health worker kindness
ACME	0.033	0.026	0.042
Direct effect	0.031	−0.008	0.068	Correlation at which ACME = 0	0.161
Total effect	0.065	0.024	0.102
% of tot eff mediated	0.516	0.325	1.374
Percentage of women who paid for delivery in HF
ACME	0.051	0.039	0.062
Direct effect	0.014	−0.025	0.050	Correlation at which ACME = 0	−0.259
Total effect	0.064	0.025	0.100
% of tot eff mediated	0.791	0.506	2.051
Mediators for IPT2 uptake
Health workers received supervision visit in last 90 days
ACME	0.016	0.002	0.030
Direct effect	0.102	0.038	0.161	Correlation at which ACME = 0	−0.0343
Total effect	0.117	0.056	0.174
% of tot eff mediated	0.133	0.089	0.280

ISSN: 1748-5908