Statistical Analysis of Cover Data
The appropriate statistical model for analyses to identify differences in cover between years or among sites depends on the method to determine cover that was followed, and decisions defining sample units made during preliminary planning.
If each point is considered an independent sample unit when point sampling to determine cover, the data will follow a binomial distribution, and analyses should be based on binomial statistics. In this situation, binomial confidence intervals are used to assess if two sample means are significantly different. The binomial confidence interval for a given cover value remains constant, according to sample size and the level of probability, and tables listing the width of confidence intervals have been developed for commonly used sample sizes (typically n=100 and n=200) and probability levels (typically p=0.1 or p=0.05). If the confidence intervals (for the correct sample size and probability level) for the sample means being compared overlap, it is concluded that these values are not significantly different.
In most methods of point sampling to determine cover, however, points are grouped in frames or transects before the raw cover data is derived, and the collected data tends toward a normal distribution. Methods involving line sampling to determine cover, sampling in quadrats to determine cover, or plotless sampling to determine cover also tend to generate normal distributions. In situations where cover data has been collected using broad intervals, such as the point frame method or the Daubenmire cover class method, angular transformations are useful to meet assumptions of a normal distribution. Under normal distributions, significant differences between two sample means are evaluated by considering the possibility that their respective confidence intervals overlap.
References and Further Reading
Greig-Smith, P. 1983. Quantitative plant ecology. University of California Press, Berkeley, CA. 3rd ed. pp 45-46.