Analyzing the Effectiveness of Farmers’ Practices from on-Farm Replicated Strip Trials.

On-farm replicated strip trials are commonly conducted by farmers to assess the effect of different agricultural practices on crop yields and economics. Typically, each trial includes two or three contrasted treatments replicated several times across fields. Yields are measured in each replicate by GPS enabled yield monitors. The analysis of yield data collected in such trials could provide growers with useful information. However, the statistical analysis of on-farm trials raises two important issues. First, the experimental designs may vary across trials, and include different types of treatments and different numbers of replicates. Second, on-farm trials usually cover contrasting environments corresponding to different soils, climates, and cropping systems. In this study, we used hierarchical Bayesian models (HBM) for analyzing yield data collected as part of on-farm trials managed by the Iowa Soybean Association On-Farm Network. Our approach is illustrated with two datasets to assess the effect of a fungicide on soybean yields and the effect of a soil-applied insecticide on corn yields. In 69 sites-years out of 207, the yield was significantly higher in soybean treated with fungicide than in untreated controls. The average yield ratio (treated/untreated soybean yield values) estimated over all trials was significantly higher than one, revealing a significant effect of fungicide. In seven site-years out of 36, the yield was higher in corn treated with insecticide than in untreated controls, but the effect of this treatment was not significant on average. In both case studies, the treatment effect was highly variable across trials, but we failed to find a covariate explaining this heterogeneity. Our results show that on-farm trial data can be analyzed using HBM and can help farmers and agronomists make management decisions.