marginally significantP = 0.048
This decision was based on two lines of evidence: The three-breakpoint model showed signs of overfitting with a higher 10-fold cross-validation error (CV = 0.070) compared to the one-breakpoint model (CV = 0.066), and the first breakpoint in the two-breakpoint model was located at an extreme end of the data gradient (AI ≈ 0.05), and the P -value for the change in slope at this breakpoint was marginally significant ( P = 0.048).