Virtual StatsPD@Waite meeting

Every month, the professional development meetings of statisticians and data scientists at Waite, known as StatsPD@Waite, bring together specialists in various aspects of data sciences in agriculture from Waite, Roseworthy and Adelaide.

Please join us for the next StatsPD@Waite seminar where Dr Max Moldovan and Mr Aidan Moller, Biometry Hub, University of Adelaide will present on “Causal Learning in Agriculture Initiative: Which computational tools can work with causal methods?”.

Also please note that the StatsPD@Waite meetings are recorded. If you have a question to the speaker but had rather not be recorded, please send your question via chat during the meeting and it will be asked on your behalf.

Please email Beata Sznajder with questions or for details of the Zoom meeting.

Title: Causal Learning in Agriculture Initiative: Which computational tools can work with causal methods?

Presenter: Dr Max Moldovan and Mr Aidan Moller, Biometry Hub, University of Adelaide

While statistical modelling and machine learning have enhanced agricultural analytics by uncovering patterns for evidence-based predictions, they often fall short in revealing the true drivers behind observed outcomes. Causal inference addresses this gap by answering “what if” questions about experimental or observational interventions, such as the effect of fertiliser formulations or the effect of an irrigation system, to distinguish correlation from causation. This provides a missing translational link between explanations/predictions and on-farm solutions, enabling more robust, actionable insights that support decision making in agriculture.

Agricultural datasets pose unique challenges due to their longitudinal, hierarchical, and spatial structures, often involving repeated measures, multilevel interventions, and diverse data sources, with almost unavoidable missing observations. These complexities demand computational tools capable of handling such data specifics, while still being able to produce valid causal estimates.

We present a selection of computational tools suited for causal inference in agricultural analytics. These tools are evaluated for their capacity to handle mixed, hierarchical and longitudinal models while estimating causal parameters, such as average treatment effects (ATEs), with appropriate measures of uncertainty. This work underlies the paper currently in progress.