Detrended Correspondence Analysis - DECORANA
|Top Previous Next|
Detrended Correspondence Analysis was devised by Hill (1979) as an attempt to improve upon reciprocal averaging (RA). Two problems that occur with reciprocal averaging are termed the 'arch effect' and 'end point compression'. When the first and second axes produced by RA are plotted, it is often observed that the points are arranged in an arch, because of the quadratic relationship between the axes, rather than because of any relationship between the samples. DECORANA removes this arch by a technique termed detrending. The tendency for points at each end of the first axis to be closer together than those in the middle is removed by segmenting the axis and expanding the terminal segments and compressing those towards the centre. Whereas RA scales the axes between 0 and 100 in relation to the magnitude of the eigenvalue, DECORANA scales in units of average standard deviation of species turnover. Therefore a change of 50% in species composition occurs in about 1 standard deviation.
Samples or species which are very different in their composition or distribution from others in the data set present problems for DECORANA and in general such outliers should be removed from the data before the method is applied.
When DECORANA is selected from the ordination drop-down menu, the Setup for DECORANA window is displayed. In many cases, the default settings can be used by simply clicking OK.
The options are described in turn below:
Downweight rare - Select this option if the influence of rare species is to be reduced. If selected, the abundances of species rarer than the frequency of the commonest species divided by 5 are down-weighted in proportion to their frequency.
Number of axis rescalings - Input an integer number; the default of 4 should generally be used.
Rescaling threshold - Axes shorter than this value will not be re-scaled. The default is zero.
Number of segments - This is the number of segments the axis is divided into for re-scaling. Input an integer number, the default of 26 should generally be used.
Output for DECORANA is presented under a number of tabbed components that can each be viewed by clicking on the tab. These are described in turn below: