Towards Causal Inference for Spatio-Temporal Data: Conflict and Forest Loss in Colombia
share
In many data scientific problems, we are interested not only in modeling the behaviour of a system that is passively observed, but also in inferring how the system reacts to changes in the data generating mechanism. Given knowledge of the underlying causal structure, such behaviour can be estimated from purely observational data. To do so, one typically assumes that the causal structure of the data generating mechanism can be fully specified. Furthermore, many methods assume that data are generated as independent replications from that mechanism. Both of these assumptions are usually hard to justify in practice: datasets often have complex dependence structures, as is the case for spatio-temporal data, and the full causal structure between all involved variables is hardly known. Here, we present causal models that are adapted to the characteristics of spatio-temporal data, and which allow us to define and quantify causal effects despite incomplete causal background knowledge. We further introduce a simple approach for estimating causal effects, and a non-parametric hypothesis test for these effects being zero. The proposed methods do not rely on any distributional assumptions on the data, and allow for arbitrarily many latent confounders, given that these confounders do not vary across time (or, alternatively, they do not vary across space). Our theoretical findings are supported by simulations and code is available online. This work has been motivated by the following real-world question: how has the Colombian conflict influenced tropical forest loss? There is evidence for both enhancing and reducing impacts, but most literature analyzing this problem is not using formal causal methodology. When applying our method to data from 2000 to 2018, we find a reducing but insignificant causal effect of conflict on forest loss. Regionally, both enhancing and reducing effects can be identified.
READ FULL TEXT