Statistical Analysis of Density Data

Density measurements obtained by counting individual plants may follow a skewed or Poisson distribution, caused by a preponderance of low or zero values. The degree to which the data is skewed is influenced by sample unit size, because fewer individuals are typically found in small quadrats. Poisson distributions present a dilemma for conventional inferential statistical procedures, which involve the key assumption that the collected data fits a normal distribution. Careful consideration of sample unit shape and size (eg., adopting a design incorporating belt transects to determine density), helps to ensure that the collected data approaches a normal distribution. Alternatively, a design based on quadrats arranged as a group of subsamples to determine density is also more likely to generate data tending toward a normal distribution. Subjecting density values to a square-root transformation (which improves data symmetry) before statistical tests are conducted can also facilitate the assumption of normality.

Density measurements obtained by distance methods, such as the closest individual method, the point-center quarter method, or the wandering quarter method, are likely to approach a normal distribution because each value of the data set is derived from multiple readings. Providing this assumption is met, differences between sites or years can be determined by conventional inferential statistical procedures. If the confidence intervals for the two sample means being compared overlap, it is concluded that these values are not significantly different.

References and Further Reading

Greig-Smith, P. 1983. Quantitative plant ecology. Blackwell Scientific Publications, Oxford.3rd ed. pp. 33-36.