Sources of error in extinction risk estimates with population count data

Simple time series population counts are among the most common kinds of monitoring data used to make management decisions, but recent studies suggest that problems with accurately estimating variances in growth rate for data sets less than 20 years in length greatly limit their usefulness for detecting trends.  We used the diffusion approximation method to analyze time series of different lengths drawn from more than 150 long-term data sets from the Global Population Dynamics Database.  We explored two main questions: 1) Are poor estimates of variance a more important source of error in projecting extinction risk for short-term data sets than poor estimates of mean growth rate?; and 2) Are poor estimates of variance due primarily to chance over or under sampling of extreme years?  For data sets from 10-20 years in length, the long-term variance was more often under than over estimated, but more of the extreme estimation errors were over-estimates.  Poor estimates of long-term observed growth rates posed a substantial problem even for 30 year time series, with errors in the range of 100% typical.  This uncertainty about mean growth rate was the major source of error in predicting extinction risk, suggesting that poor estimates of variance are not necessarily the most important problem in using shorter data sets.  Overestimation of variance in short-term data sets appeared to occur primarily because of chance sampling of extreme years.  These results suggest some potential strategies for more effectively using population count data in conservation.