Recent simulations suggest that ecologists can enhance the predictive ability of models by designing experiments that maximise the number of levels of an experimental factor by sacrificing replication. Here, I describe how these simulations were based on a faulty metric of prediction success and reinforce the importance of replication.
Keywords: Ecological models; experimental design; general linear model; polynomial contrasts; prediction success; replication; statistics; trend analysis.
© 2019 John Wiley & Sons Ltd/CNRS.