Section 47 Contribution

Can we tell from our causal map what the contribution is of a certain intervention on a variable downstream of it?

Conversely, can we take a specific Difference on a downstream variable of interest and ask, what is the “contribution” of an upstream variable to that Difference?

It looks so easy, especially when we have a causal map to hand. In fact, as Pearl insists, we have to have a causal map to hand when we try to soft-calculate the contribution which C makes to E.

We have soft-specified the strength of the links and the values of the other exogenous variables, what could be easier than pressing a button and soft-calculating the contribution of the intervention? i.e. the value of outcome O is much higher in the scenario where the project is run than in the scenario where it is not run.

Information about contribution is probably number 1 on our client’s list when we are helping them construct a causal map.

But …

There are a lot of extreme difficulties outside the linear, quantitative comfort zone…

47.1 Why reporting correlation is not helpful

When calculating the downstream effect of an upstream variable in terms of the Difference made by it, we should be careful to avoid shortcuts like looking at a correlation coefficient. Correlation is a poor way to spot non-linear effects and in particular is bad at recognising necessary and sufficient conditions.

It’s worth noting that the formulation of the downstream effect of an upstream variable in terms of the Difference made by it is immune to this problem.

When examining the overall effect of a continuous variable on another, Differences are not enough. Need to have some kind of display.

[add]