Population Variability and Extinction Risk.

Population models generally predict increased extinction risk (ER) with increased population variability (  PV  ), yet some empirical tests have provided contradictory findings. We resolve this conflict by attributing negative measured relationships to a statistical artifact that arises because PV tends to be underestimated for populations with short persistence. Such populations do not go extinct quickly as a consequence of low intrinsic variability; instead, the measured variability is low because they go extinct so quickly. Consequently, any underlying positive relationship between PV and ER tends to be obscured. We conducted a series of analyses to evaluate this claim. Simulations showed that negative measured relationships are to be expected, despite an underlying positive relationship. Simulations also identified properties of data, minimizing this bias and thereby permitting meaningful analysis. Experimental data on laboratory populations of a bruchid beetle (Callosobruchus maculatus) supported the simulation results. Likewise, with an appropriate statistical approach (Cox regression on untransformed data), reanalysis of a controversial data set on British island bird populations revealed a significant positive association between PV and ER (p = 0.03). Finally, a similar analysis of time series for naturally regulated animal populations revealed a positive association between PV and quasiextinction risk (p < 0.01). Without exception, our simulation results, experimental findings, reanalysis of published data, and analysis of quasiextinction risk all contradict previous reports of negative or equivocal relationships. Valid analysis of meaningful data provides strong evidence that increased population variability leads to increased extinction risk.