Estimating competition coefficients from census data: A re-examination of the regression technique.

In this paper we assess the regression technique used to estimate competition coefficients (alpha) from census data for two species of rodents. We compare the results from this technique to those calculated from data collected in a field removal experiment conducted in coastal heathland in southeastern Australia. We examined the relationship between the competition coefficient and the population variance. Our results and analyses show that alpha depends heavily on the population variance, increasing the likelihood of producing a statistical artefact in some cases; they also indicate the need for a reappraisal of the conclusion that common species exert weak competitive effects while rare species exert strong effects. To overcome this artificial dependence we propose a procedure of normal standardization of species census data before calculating regression parameters. Our results show disagreements between the estimated alpha using regression and the observed cc calculated from the removal experiment, when the census data were not standardized. However, after the standardization procedure was applied to the census data, each estimated alpha matched the observed value well and the results were consistent with previously published results. This supports the validity of a standardized regression technique. We also discuss the inconsistency existing among different methods of factoring out habitat effects and recommend the use of competition coefficients calculated from the Residual PCA method as the best estimate for alpha from our data. The: results presented are stimulating and encouraging because they resurrect the regression technique and inspire future refinement and application.