Improving causal inference from unreplicated on‐farm strip trials with propensity score matching: Application to plant growth regulator effects in wheat

Summary

This paper proposes propensity score matching as a method to strengthen causal inference in unreplicated on-farm strip trials, which are commonly conducted by farmers but plagued by field variability and lack of randomisation. The authors implemented two-treatment plant growth regulator trials in growers' fields and used PSM with G-computation to estimate treatment effects, finding that traditional spatial models can exhibit spatial confounding artefacts. The methodology offers potential to improve the rigour and accessibility of farmer-conducted experiments without requiring prescription maps or formal replication.

Regional applicability

The methodology is directly applicable to United Kingdom arable farming, where farmer-led strip trials of inputs such as plant growth regulators are common practice. The approach addresses a genuine barrier to rigorous on-farm experimentation in UK cereal production, though the specific findings on spatial confounding and treatment effect magnitudes are site-dependent and would benefit from UK-based validation.

Key measures

Yield data, topographic covariates, soil properties, propensity scores, covariate balancing weights, treatment effect estimates with 95% confidence intervals, G-computation estimates

Outcomes reported

The study evaluated a propensity score matching (PSM) methodology to improve causal inference from unreplicated on-farm strip trials of plant growth regulators in winter wheat, comparing treatment effect estimates and confidence intervals against traditional spatial models. The analysis examined whether PSM could address confounding from field variability and lack of randomisation in farmer-conducted experiments.

Topic tags