This study investigates the relationship between extinction rate and temporal variability in population size for 174 populations of seven orb spider species on 106 subtropical islands. Evidence for all population time series with at least two nonzero values supports either a negative (inverse) relationship or no relationship at all; this result is robust over a wide variety of statistical treatments. This perhaps unexpected result can be interpreted mainly as reflecting the inability of populations frequently going extinct ever to achieve large enough population sizes to attain even large relative variances (such as coefficient of variation), as well as reflecting the very erratic trajectories of boom-and-bust-type populations that seldom reach resource carrying capacities. Removal of zero counts from time series is also probably involved, and skewness of population-size values may have an effect as well. One type of data preparation, however-that including zero values of the population series and using arithmetic means and coefficients of variation-shows the opposite trend. Measures of temporal variability, coefficients of variation or standard deviations of log population sizes, are usually positively related to mean population size; the second derivative of this relationship is usually negative. Relation of extinction rate to mean population size is in all cases negative and highly significantly so; the second derivative of this relationship is usually positive. The latter relation is usually stronger than those involving variability, but taking into account relations with means in multiple regression still leaves a fair number of significantly negative relations of extinction rate to variability. Populations showing low temporal variabilities are mainly of two kinds: small populations with moderate to high extinction rates and large populations with low extinction rates. Thus, when only series with mean population size greater than or equal to eight are analyzed, the relation between extinction rate and temporal variability is, if anything, mildly positive. The entire analysis illustrates the great sensitivity of conclusions to seemingly biologically arbitrary conventions with respect to selection and transformation of data.
