Plots of Δelpd_loo of two models (broken and power law) as a function of both scatter (x-axis) and threshold set (y-axis) in terms of percentage of excluded data. Different figures represent different cases of break point assumption; from the upper left to bottom right panel x_c = 2, 4, 6, 8. The orange and white contours trace a Δelpd_loo of ten and four, respectively. The horizontal dashed line is placed at 40%, while the two vertical lines are at one scatter of 0.20 and 0.30.