Introduction

Despite the recent explosion of activity in causal inference (drawing conclusions about causal connections in data), little research has been undertaken into how to perform causal inference in settings that have spatio-temporal relationships, such as urban traffic. The project aims to extend recent methodologies in causal inference to this setting including propensity score (estimating the effect of an action) and instrumental variable methods, with focus on the urban traffic setting.

Explaining the science

Causal inference is difficult in observational studies due to issues with selection bias and confounding (when certain variables influences both the dependent variable and independent variable causing spurious associations). Methods to offset these issues include considering conditional probability. The project aims to extend these methods to more general settings.

Project aims

The project aims to investigate the use of advanced causal inference techniques in application for datasets with spatio-temporal correlations. 

A major issue in the spatio-temporal setting is that recently developed advanced causal inference techniques rely on strong assumptions, including the assumption of no interference (i.e. that a treatment, or action, applied to one unit does not affect the outcome of another), and this is often violated in a spatio-temporal settings. The aim is to develop inference methodologies which account for this interference. 

Extending causal inference methodologies to a spatio-temporal setting would be useful in a wide variety of applications.

This project is part of the Data-centric engineering programme's Grand Challenge of 'Monitoring Complex Systems'.

Applications

Whilst the methodologies developed will be applicable in a wide range of applications, the transportation setting is of particular interest. The methods developed would enable the answering of questions such as how traffic interventions, such as the introduction of speed limit zones, improve safety where potential 'covariates' (related, independent variables such as road type and traffic intensity) vary over space and time.

Organisers

Researchers and collaborators

Funders