Abstract While relative yield is widely used for its comparability, normalization can cause significant information loss. This study reframes yield metric selection as a model evaluation problem to determine the most accurate representation of crop response. We evaluated seven yield metrics with three agronomic models, comparing estimates of a critical soil phosphorus (P) threshold for robustness and plausibility against a regional guideline of 30 mg kg − 1 . Models using raw yield consistently provided the best fit, with threshold estimates (28.9–34.6 mg kg − 1 ) closely aligning with the benchmark. Conversely, normalized metrics like relative yield performed poorly, generating highly variable and biased estimates. These findings establish that raw yield is the most statistically robust and economically relevant metric for quantifying crop response, holding important implications for nutrient management.
Sun et al. (2026) studied this question.