Understanding population dynamics using predictive ability of models

Background/Question/Methods

Ecologists have debated the drivers of population fluctuations for more than a century. One of the key areas of disagreement is the relative importance of density dependent versus density independent factors. Density dependence implies that current abundance depends on historical abundance. However, researchers have generally assessed the relative importance of density dependence based on model fit rather than predictive performance. We developed four simple models, (1) mean abundance, (2) the linear trend, (3) logistic density dependence and (4) Gompertz density dependence, to predict population fluctuations. We used predictive performance on more than 16,000 populations from 14 data sets to compare predictive performance (i.e. the understanding captured by the four models).

Results/Conclusions

We found that the density dependence model only had unambiguously superior predictive performance in 5 of the 14 data sets and that in all 14 data sets the density dependence model made, on average, small or no improvements in predictive performance. We conclude that some, but not most, populations show evidence of density dependent regulation but, on average, the effects on population fluctuations are small.