3

I am using the R mediation package. The output of the mediate command is as follows:

Causal Mediation Analysis

Quasi-Bayesian Confidence Intervals


                      Estimate 95% CI Lower 95% CI Upper p-value    



ACME (control)           -0.001633    -0.002128         0.00  <2e-16 ***
ACME (treated)           -0.001251    -0.002040         0.00  <2e-16 ***
ADE (control)            -0.001082    -0.003008         0.00   0.296

ADE (treated)            -0.000700    -0.001937         0.00   0.296

Total Effect             -0.002333    -0.003881         0.00   0.022 *

Prop. Mediated (control)  0.677185     0.471999         1.66   0.022 *

Prop. Mediated (treated)  0.496793     0.178559         2.01   0.022 *

ACME (average)           -0.001442    -0.001940         0.00  <2e-16 ***
ADE (average)            -0.000891    -0.002488         0.00   0.296

Prop. Mediated (average)  0.586989     0.329850         1.85   0.022 *




Signif. codes:  0 ‘*’ 0.001 ‘’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1


Sample Size Used: 19727


Simulations: 1000

As you can see the proportion mediated upper 95% CI interval for the treated group is 2.01.

How should I interpret this proportion greater than 1?

kjetil b halvorsen
  • 77,844
robertspierre
  • 2,291

1 Answers1

0

In the documentation for the mediate function you can read this:

the "proportions mediated", or the size of the average causal mediation effects relative to the total effect.

So, a proportion mediated that is greater than 1 would indicate that the size of the mediation effect is larger than the total. This could happen if the directions of the main and mediated effect were opposite. This is equivalent to suppression in ordinary regression. If the direct effect is very small, then the proportion mediated could be far greater than 1.

Then the question of whether there is mediation at all becomes kind of philosophical. If there is no effect, is there anything to mediate?

Peter Flom
  • 119,535
  • 36
  • 175
  • 383
  • I didn't understand if the cause is (1) the total and mediated effect being in opposite direction or (2) the direct effect being very small or (3) both.... – robertspierre Jul 05 '21 at 19:11
  • It doesn't seem to give you the main effect or the mediated effect, so it's hard to tell. – Peter Flom Jul 05 '21 at 19:16
  • I don't understand your reply. It is telling me that, on average, 59% of the effect is mediated, that's how I interpret the "Prop. Mediated (average)". – robertspierre Jul 05 '21 at 21:04