- This event has passed.
National Institute for Applied Statistics Research Australia Seminar Series
October 10 @ 1:30 pm - 2:30 pm
Rune Christiansen (firstname.lastname@example.org)
Department of Mathematical Sciences
University of Copenhagen
Using space as an anchor: causality meets spatial data
In statistical causality, we are interested not only in modeling the behaviour
of a system that is passively observed, but also how the system reacts
to changes in the data generating mechanism. Controlled executions of
such interventions (randomized trials) are viewed as the gold standard for
learning causal models, but are often practically impossible to conduct.
When controlled interventions are infeasible, it has been suggested
to exploit natural variations resulting from external forces on the system.
Such scenarios are common in spatial statistics: local conditions in which the
system of interest is imbedded are rarely constant across large spatial domains,
and strong heterogeneity is thus a typical characteristic of spatial data sets.
In this talk, I present ideas on how the two fields could benefit each other.
In particular, I discuss how spatial heterogeneity might be used to infer
causal relationships (via the instrumental variables approach), and introduce
a nonlinear version of anchor regression, which is an an estimation procedure
that encourages regression models to be (more) spatially invariant.
No prior knowledge of causality is required.