Model selection results from cubic spline model fitted to female chamois body mass data.

<p>Models are distinguished by the functional forms of <i>α</i><sub>0</sub> and <i>β</i><sub>0</sub>. Specifically, we allowed <i>α</i><sub>0</sub> and <i>β</i><sub>0</sub> to be constant across years (<i>α</i><sub>0</sub>; <i>β</i><sub>0</sub>), vary linearly with year (<i>α</i><sub>0</sub>(y); <i>β</i><sub>0</sub>(y)), quadratically with year (<i>α</i><sub>0</sub>(y<sup>2</sup>); <i>β</i><sub>0</sub>(y<sup>2</sup>)) or linearly with population density (<i>α</i><sub>0</sub>(d); <i>β</i><sub>0</sub>(d)). Maximum log-likelihoods (LL) and ΔAICs are shown for each site. The most parsimonious models for each site are highlighted in bold (i.e. have a ΔAIC value that is ≤6 and lower than all simpler nested versions; see Richards <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0028002#pone.0028002-Richards1" target="_blank">[40]</a>). <i>n</i> is sample size for each site.</p>