Analysis of Plant Monitoring

Analysis of Results from Method 1 (Plant Community Monitoring)

For most purposes, results must be analysed for the trend of each individual species, or the amount of bare ground, etc., but ‘species diversity’ is best measured by the average number of species per plot (‘richness’). For all these measures, the agreement, or lack of it, between the replicate plots in the same compartment will give confidence in the trends. If these agree well in magnitude and direction the trend is confirmed. The more plots studied in a given compartment the smaller will be the trends that can be detected with this method, but the extra return per unit effort falls off significantly with an increased number of plots.

The Domin scale (see methods) approximately ranks the abundance of each species each year, and in the typical situation where there are rather few plots studied, it does not meet the requirements of parametric statistical testing (an approximately normal distribution and no great dependence of the variance on the mean).

Analyses of the data show that the ‘pairing’ (measuring the same plots each year) results in a significant improvement in the precision of the measurement of change (the median value of the correlation between values obtained in the same plots in two years is around 0.4).

The ‘Binomial signs test 1 ‘ is used to test any pair of years, as illustrated below:

Binomial signs test for: Leucanthemum vulgare at Bedfont Lakes, 1993-4
Domin scores
Plots19931994DifferenceSign
10000
22311
30000
40000
53300
60000
35211
83411
93411
103411
112311
123300
Positive:6
Negative:0
Sum:6
Two-tailed binomial probability = 0.031

The test uses the direction of the change in each plot between the two years. In this case all changes were either zero or positive. The zeros do not take part in the test, but the positive and negative signs are summed separately (to give 6 and 0 respectively). Then the binomial probability of an outcome at least as extreme as the smaller of these two totals (0 in this case) is determined. For this you need also to know the grand total of signs (6 in this case). Some spreadsheets will provide a function to calculate this probability, or you may use Siegel’s table D (Siegel, 1956).

For more than two year’s worth of data, Friedman two-way analysis of variance (Siegel, 1956) can be used:

Friedman test for: Leucanthemum vulgare at Bedfont Lakes 1993-4
Domin scores
199319941995199319941995
Plots
1001113
2233122
3001113
4000111
5354111
6000111
7354132
8345123
9346123
10345123
11233122
12332221
N12Ri132025
k3Ri squared169400625
Chi-squared = 55
(for 2 degrees of freedom)
P = 0.000

Where ‘N’ = number of plots, k = number of years and Ri the sum of the ranks in each year.

The Domin scores for each plot in the years concerned (here just 3 years) are ranked. The chi-squared statistic is given by :

Chi-sq = 12/(Nk (k 1) Sum (Ri2)-3N (k 1) for k-1 degrees of freedom.

The probability can be found in a text book table, or some spreadsheets provide the appropriate function. This test is believed to be approximately as powerful as the parametric analysis of variance (Siegel, 1956).

N.B. Confirmed trends can be related to particular targets. Trends may be in terms of rarity, interest, ecology and attractiveness of species, or an approach to a particular community composition. Where such aims can be made to depend upon a set of species, parametric tests may be more suitable (eg. for testing whether the species richness has changed between years, or whether the average Domin score of a set of species has increased or decreased). This is because of the action of the central limit theorem, which tells us that the standard error of a distribution that is not normal tends towards normal as the sample size increases.

Use of Results From Other Two Methods

Use of the results from the other two methods will vary depending upon the aims of the survey. In some cases the aim may simply be to use the mapped locations for conservation purposes but in others analysis of population changes may be appropriate.

Storage of Data in Recorder Format

The data can be held in databases, such as Recorder and GIS. Where data is held in Recorder format it is important that the plot numbers/locations and measures of abundance are clearly shown, to enable comparisons to be made. Where plots are sufficiently far apart, a correct unique grid reference can be allocated to each plot to make analysis more straightforward. If plots are too close together to permit this to be done, the plot number should be recorded in the ‘Comments’ field.

References

Siegel, S. 1956. Nonparametric statistics for the behavioural sciences. McGraw-Hill.