Besides P budget and PUE by crop type, we also tried to collect data of all drivers for over 200 countries by crop type from 1961 to 2014. However, as the available data of most drivers are limited to a smaller set of countries (Table S13), we studied the drivers and their relationships with PUE in 113 major crop-producing and nutrient use countries (Text S8). These countries accounted for around 85% of the global total harvest area, 86% of total P in crop production, and 94% of total P fertilizer input in 2010 (average of 2005-2014).

We analyzed the relationships between PUE and all potential factors in and across these 113 major countries, using three methods:

1) fitting regression models with PUE as the dependent variable and the linear and quadratic terms of economic development as the independent variables, to test the Environmental Kuznets Curve (EKC) hypothesis (Dinda 2004, Zhang et al. 2015a) for individual country and across all 113 countries (Text S7);

2) studying the association between PUE and each drive for individual countries, by calculating Pearson’s correlation coefficient (Pradhan et al. 2017) (Text S10);

3) fitting constant intercept and fixed effects models with PUE as the dependent variable and other driver(s) as the independent variable(s) across all 113 countries (Text S11).

We examined the relationships both for individual countries and across countries to avoid the effects of the Simpsons Paradox (Scholz and Hirth 2015).