Statistical tests:
Due to the fact in all indexes the value of one species - no matter how large it is in size - is 0 or 1, their use has been limited . We have chosen to use mainly three groups of indexes: the Shannon-Winer (by individual numbers), the modified Simpson index (by proportion cover) and Shannon-Winer (by percentage/proportional cover). These indexes were found to be the most appropriate for linear comparisons as they grow in number with higher diversity /individual number. Shannon-Winer index ("SW") for example is known for being sensitive to species richness but not to sample size. This can be demonstrated by looking at values of a quadrat with and without the soft coral data (it's relatively similar - Chart 1). We used the Modified Simpson index ("smi") because it tends to increase with the increase of diversity and numbers. It is also almost identical to the Hill index. By values the index tends to be sensitive to high number of individuals and that reflects in the fluctuation of values between with and without soft corals.
Regression tests and multiple regressions of the 5 main elements affecting coral growth against various indexes (and amongst themselves) were plotted.
Multiple regression table
All corals Stony corals only Independent variable P-value
P-value Intercept 0.297439 0.760119 Light strength (1-5) 0.948876
0.411069 Sedimentation (0-1) 0.46987 0.552837 Disturbance
(0-1) 0.13641 0.155539 Surface complexity (0-2) 0.901323
0.638406 Material (0-1) 0.478065 0.801744 Significance
F 0.525754 0.656052
Judging by the P values (non are >0.05 = not significant) we can't accept any of the independent variables to be significantly responsible for attachment pattern. The significance F value in both data sets is high suggesting that the probability that these data sets are randomly obtained is high too.
In all regression tests following the above (we ran 5,4,3 and 2 independent variables tests) there was only a small degree of fluctuation and for all tests we had to accept the null hypothesis. These tests would be valid only if the variables are statistically independent amongst themselves. Assuming indexes values are indeed dependent on these factors, we are forced to look at every independent variable separately and try to understand what are the combinations or independent variables which influence the quadrats individually. However, Chi2 tests were run to avoid having to calculate each variable independence (from the other variables). The index values ware divided into high and low value (> or < then 1) and reflect coverage and individual counts. This also means that effectively this division represents all the index sets we calculated. It also solves the "problematic" 0 values in many index sets.
Test of independence df P Index values X Depth 1 1.269 Index values X light intensity 1 0 Index values X Disturbance 1 4.444 Index values X Sedimentation 1 0.625 Index values X Surface complexity 1 3.402 Index values X surface material 1 0.625
So where does all that leaves us??
| ||
|
|
|