Table 5 Mediation analysis

From: Behavioral savings sessions increase the pursuit of solar products among refugees in Uganda

Mediators

Total effect

Mediator on outcome

Mediated effect

Confidence interval

Proportion mediated

Contact solar company

Knowledge

 

0.10**

0.02**

0.003

0.030

31%

Trust

 

−0.00

0.00

−0.005

0.006

1%

VSLA support

 

0.03***

0.03***

0.016

0.048

58%

Aspiration

 

0.03

0.00

0.000

0.004

3%

Self-efficacy

 

−0.02

0.00

−0.001

0.008

 

Direct effect

0

     

Total

0.05

 

0.05***

0.032

0.077

100%

Solar savings goal

Knowledge

 

0.18***

0.10***

0.070

0.131

30%

Trust

 

0.10***

0.02***

0.008

0.026

5%

VSLA support

 

0.04**

0.03**

0.001

0.066

10%

Aspiration

 

0.25***

0.01**

0.000

0.023

3%

Self-efficacy

 

−0.04*

0.00

−0.004

0.006

 

Direct effect

0.17

     

Total

0.34

 

0.16***

0.119

0.207

49%

Solar savings with VSLA

Knowledge

 

1826.4***

~692.2

   

Trust

 

525.3**

~106.9

   

VSLA support

 

69.7**

~281.9

   

Aspiration

 

2305.1***

~176.8

   

Self-efficacy

 

−582.4***

    

Direct effect

1240

     

Total

1888

     
  1. The table presents the results of the mediation analysis for the three key significant outcomes of interest (contacting a solar company, having a solar goal, and solar savings with the VSLA). OLS regressions, Tobit regressions, and seemingly unrelated regression analyses (in panels 1 and 2) are used to generate outcomes. Column 2 presents the total effect of treatment on the outcome of interest: by combining the impact of treatment on outcome, controlling for mediators (direct effect), plus the impact of treatment on outcome through the mediators (indirect effect). Column 3 presents the impacts of the mediators on the outcomes in a regression, including the treatment assignment dummy. Column 4 presents the mediated effect by combining the impact of the treatment on the mediator and the impact of the mediator on the outcome. Asterisks denote a statistically significant mediation effect at the 1% ***, 5% **, or 10% * levels, calculated using bootstrapped standard errors. Column 5 displays bias-corrected and accelerated (BCa) confidence intervals (95%) for the mediated effect. Column 6 presents the proportion mediated by dividing the mediated effect by the total effect.
Back to article page