In order to select the final model between equally good predictor models we calculated a permutation P value associated with their prediction accuracy.
Prediction of glucagon and insulin profiles are equally good as the model fit, indicating a reduced long-term IOV of PK properties but also underlining the very good long term predictive power of the model when taking into consideration that the two visits were months apart.
Moreover, power function models are equally good as exponential models for that the two key model parameters (R 1 and k) have clear ecological meanings.
In our case, through minimizing the model size and the error, it will produce an array of equally good models that are not dominated by other models.
Since ER has a very broad transcriptional footprint and cell proliferation requires the coordinated expression of hundreds of genes, a very large number of nominally different but equally good prognostic models can be built from the same dataset.
The parameter functional relationships derived with catchment attributes produced equally good model results when applied to independent test catchments.
An equally good model was obtained if PC1 was dropped from the candidate variables, in which case the last year when the species had been sampled was selected at 5% level.
The positive aspects are clear from the examples where we use a help model that is roughly equally good as the tested model: then the ROC curves have a higher area under the curve, and because the model that should be rejected in the insulin signaling example gives a lower p-value.
It should be emphasized that the high accuracy of the model (3 4% prediction error on average for I-frames and P-frames) does not guarantee an equally good performance when the model is incorporated into the policer.
The initial analysis energy scale did not show an equally good fit to the model.
Our results suggest that the currently available relaxed-clock models suffer much less from the TDRP problem, and perhaps are equally good to the PL model, if calibrating nodes are dispersed in time (Fig. 4; bottom), consistent with the results from a previous study [ 32].
