AIC charts give incorrect relative AIC value

[beazors]

I am seeing that the relative AIC charts should graph the change in AIC value relative to that of the lowest knot number tested - for example, in the fitGAM article, the middle right panel (relative AIC) is described as "the change in AIC relative to the average AIC at the lowest knot number (here, this is 3 knots, as can also be seen from the plot since the relative AIC equals 1" and the example shows a monotonic drop in average AIC plot, leading to a monotonically decreasing relative AIC plot. However, in this post and in my own data below, I am seeing what appears to be erroneous relative AIC values.

Notice how the average AIC plot shows monotonically decreasing values, whereas the relative AIC is up and down, which makes no sense if this is plotting the AIC relative to the lowest knot value.

[koenvandenberge]

Hi @beazors

The relative AIC in the third panel is actually the average of the gene-specific relative AIC's. It is not the relative AIC of the averages. This is why the two plots don't match and are not expected to.

Best, Koen