Effects defined by shifting the mediator distribution, not the value

Natural effects need each person counterfactual mediator value under the exposure they did not get, which is not identified when something the exposure caused also confounds the mediator-outcome link. Interventional effects draw the mediator from its population distribution under the other exposure level instead. Weaker assumptions, and a question a policy could implement – at the cost of no longer decomposing an individual effect.

Usage

Intvse(Y, X, M, Cc = NULL, a = 1, astar = 0)

Arguments

Y

Outcome.

X

Exposure.

M

Mediator.

Cc

Optional covariates; read at their means.

a, astar

Exposure contrast levels.

Value

List with estimate, ide, iie, check, theta, beta, n.

Details

Formula: \(IDE = (th1 + th3(b0 + b1 a* + b2 primec))(a - a*)\), IIE = (th2 + th3 a) b1 (a - a*), summing to the total effect.

References

VanderWeele, T. J., Vansteelandt, S. & Robins, J. M. (2014). Epidemiology 25:300-306; VanderWeele (2014) Epidemiology 25:749-761.

Examples

Intvse(Y = c(1, 2, 3, 4, 5, 6, 7, 8), X = 5L, M = 5L)
#> $estimate
#> [1] 0
#> 
#> $ide
#> [1] 0
#> 
#> $iie
#> [1] 0
#> 
#> $check
#> [1] 0
#> 
#> $theta
#> [1] 1 0 0 0
#> 
#> $beta
#> [1] 5 0
#> 
#> $n
#> [1] 8
#> 
#> $method
#> [1] "Interventional direct and indirect effects"
#> 
#> attr(,"class")
#> [1] "morie_rich_result" "list"