A Monte Carlo study of tests on data originating from quadrat sampling. I: Data from a Poisson distribution

Area of Science:

• Ecology
• Statistics
• Computational Biology

Background:

• Comparing means of Poisson distributions is crucial in ecological studies, especially when analyzing small population counts from quadrat sampling.
• Existing statistical tests often involve data transformations or log-linear models, but their performance with small means is not well understood.

Purpose of the Study:

• To evaluate the significance levels and statistical power of various tests for comparing Poisson distribution means.
• To specifically assess test performance when dealing with small mean values, typical in ecological count data.

Main Methods:

• Computer simulations were employed to rigorously examine the behavior of different statistical tests.
• Two primary testing approaches were considered: log-linear models (deviance-based tests) and analysis of variance (ANOVA) on transformed data (logarithmic, square root).

Main Results:

• For very small Poisson means, deviance-based tests demonstrated superior performance compared to analysis of variance tests on transformed data.
• No single analysis of variance transformation consistently outperformed others for small means.
• Standard analysis of variance on untransformed data yielded good results for larger mean values.

Conclusions:

• Deviance-based tests are recommended for comparing Poisson means when dealing with small count data, common in ecological research.
• For larger Poisson means, traditional analysis of variance on untransformed data is a robust and effective method.