I presented at the Australian Statistical Conference (ASC) in Perth (December 2025), sharing a comprehensive study on statistical strategies for the design and analysis of on-farm strip trials, bridging foundational experimental design principles with modern spatial modelling and simulation evaluations.
📋 Abstract
This study explores statistical strategies for the design and analysis of on-farm experiments (OFE), grounded in established principles of experimental design and spatial modelling. Through simulation studies, we assess various approaches, including design layouts combined with Geographically Weighted Regression (GWR) for continuous response variables and Linear Mixed Models (LMM) for categorical treatments.
Our results align with key findings in the literature, emphasising the importance of appropriate design and modelling choices for effectively capturing spatial heterogeneity. Additionally, we compare two trial types—large-strip trials and stacked replicated trials—and highlight the significance of data granularity, which informs data collection strategies for industry partners.
The findings advocate for appropriate trial designs for different purposes, and increased within-trial sampling, rather than reliance solely on average values, to better reflect local variability and enhance inference accuracy. This work underscores the critical role of tailored statistical methods in improving the reliability and practical applicability of OFE outcomes.
Authors
Zhanglong Cao, Julia Easton, Suman Rakshit, Barbara Kachigunda, Mark Gibberd
Downloads
🎯 Key Topics Covered
- On-farm strip trial design principles and frameworks
- Design layouts: Systematic vs. randomised arrangements
- Spatial analysis methods: GWR and LMM approaches
- Design comparison: Large-strip trials vs. stacked replicated trials
- Data granularity and within-trial sampling strategies
- Simulation evaluations comparing design and analysis combinations
- Spatial heterogeneity capture and characterisation
- Practical implementation and industry adoption strategies
💡 Highlights
A central finding was that design choice and data collection density significantly impact inference quality. Rather than relying on single average values per trial strip, increased within-trial sampling enables more nuanced characterisation of spatial patterns and provides growers with regionally-specific, actionable recommendations.
The simulation framework revealed that systematic design layouts combined with spatial analysis methods (GWR, LMM) outperformed conventional approaches across most scenarios, particularly in capturing treatment responses in heterogeneous field conditions. The comparison between large-strip and stacked replicated trials showed important trade-offs: large-strip designs suit farmers’ operational preferences, while replication improves inference but increases complexity.
We also demonstrated how communicating results in probabilistic, spatially-explicit formats—rather than single point estimates—enhances grower trust and uptake of recommendations.
🙏 Acknowledgments
Thanks to the Australian Statistical Conference organisers and the Perth host for providing a platform to share this applied research. Appreciation to all attendees for thoughtful questions, feedback, and the shared commitment to advancing statistical practice in agriculture.