SOME WILDLIFE CENSUS ESTIMATES BASED ON NON-NORMAL FREQUENCY DISTRIBUTIONS

Abstract

Wildlife census results often give very skewed distributions which make a poor fit to the normal distribution. The question arises whether we should use normal or non-normal methods in calculating a census estimate based on such data. In this study, four sets of sampling experiments were based on wildlife census results which made a good fit to the negative binomial or Poisson distributions. Confidence limits of the mean based on these distributions were similar in width to limits based on the normal distribution, and they contained the true population mean in an approximately equal proportion of cases. This held even at sample sizes which were below the level recommended for the normal approximation to be valid. It is concluded that the non-normal techniques provided no substantial improvement in estimating the confidence limits of the mean for a variety of positively skewed results.