Nonmonotonic changes in species richness along ecological gradients are frequently observed in nature. While theories support both symmetric and skewed unimodal relationships, related studies usually fit second-order polynomials, which assume symmetric relationships. These studies often apply various transformations of the predictor variable to reduce the effects of outliers or to meet assumptions of normality. We studied whether predictor transformation affects the shape of the fitted curves. To test the effect of predictor transformation on the shape of the fitted curves, we re-analyzed the dataset of a highly-cited global analysis on the productivity–species richness relationship without performing any data transformations and contrasted the results with those of the original analyses that used log-transformed productivity data. We found that predictor variable transformation, which was used in the original paper, changed the shape of fitted curves in 32% of the sites as well as the shape of the global relationship compared to the use of untransformed data. Therefore, we propose the reconsideration of predictor transformation and suggest an alternative approach: the piecewise regression. We found that piecewise regression is robust against predictor variable transformation. It resulted in much fewer inconsistent shape categories between the transformed and untransformed cases compared to the original analyses (2 instead of 9). We suggest that studies applying untransformed and transformed predictors when studying the shape of species richness curves along gradients are not directly comparable. Using piecewise regression models may contribute toward resolving the ongoing debate on the change in species richness along ecological gradients in general, and the productivity-species richness relationship in particular.

在自然界中经常观察到沿生态梯度的物种丰富度的非单调变化。虽然理论支持对称和偏斜单峰关系，但相关研究通常适用于假设对称关系的二阶多项式。这些研究通常应用预测变量的各种变换来减少离群值的影响或满足正态性假设。我们研究了预测变量变换是否影响拟合曲线的形状。为了测试预测变量转换对拟合曲线形状的影响，我们重新分析了一个被高度引用的关于生产力-物种丰富度关系的全球分析的数据集，没有执行任何数据转换，并将结果与​​原始分析的结果进行对比使用对数转换的生产力数据。我们发现，与使用未转换数据相比，原始论文中使用的预测变量转换改变了 32% 站点的拟合曲线形状以及全局关系的形状。因此，我们建议重新考虑预测变量转换并提出一种替代方法：分段回归。我们发现分段回归对预测变量变换具有鲁棒性。与原始分析（2 个而不是 9 个）相比，它导致转换和未转换案例之间的不一致形状类别少得多。我们建议在研究沿梯度的物种丰富度曲线形状时应用未转换和转换预测因子的研究不能直接比较。