Recently Don Ludwig has shown that calculations of extinction probabilities
based on currently available data are often meaningless due to the large u
ncertainty accompanying the estimates. Here we address two questions posed
by his findings. Can one ever calculate extinction probabilities accurately
? If so, how much data would be necessary? Our analysis indicates that reli
able predictions of long-term extinction probabilities are likely to requir
e unattainable amounts of data. Analytic calculations based on diffusion ap
proximations indicate that reliable predictions of extinction probabilities
can be made only for short-term time horizons (10% to 20% as long as the p
eriod over which the population has been monitored). Simulation results for
unstructured and structured populations (three stage classes) agree with t
hese calculations.