Hi Madhu,

Seems tricky!  I think you might be best served by using explicitly circular
methods for your circular data.  I don't really know anything about such
things (so why I'm responding to your post, I don't know!) but what little
information I've gathered leads me to believe that you could use "circular
statistics" to evaluate whether there's greater probability of finding a
plant species over certain degrees compared to others (e.g. SE is most
likely).  I think the idea would be to fit a circular probability density
function to your data, and see if it has a central tendency, and if so
where.

You could look at some R documentation to starting finding some ideas and
references:
http://cran.r-project.org/web/packages/circular/index.html

I think matlab also has some tools for doing this kind of thing.

I think either linear regression methods have some problems, and the take
home message is that for either you'd want/need to re-interpret your results
in light of the initial circular nature of the data.  E.g. if you just used
plain degrees (and 0 = north) and you found sp1 = 0.001*(x-180)^2...that's a
U shaped curve with peaks at 0 and 360...kind of strange to think of a
response like that in linear space, but makes perfect sense if you
re-connect (0 deg = 360 deg) the two ends--that plant likes north-facing
slopes.

Anyway, hopefully that's at least a little helpful.  Best of luck!
Andy



On Sun, Jul 4, 2010 at 7:04 PM, Madhu Srinivasan <[email protected]> wrote:

> Hi all,
> I am currently writing one of my dissertation chapters. I collected plant
> community data in a grassland ecosystem along with environmental variables
> .
> One of the questions I am addressing is: "How do the dominant grasses
> respond to the aspect?" Aspect was the direction that the slope faced. In
> the linked graphs 
> (http://sweb.uky.edu/~mpsrin2/aspect_fig.pdf<http://sweb.uky.edu/%7Empsrin2/aspect_fig.pdf>)
> I have
> displayed aspect in two ways: (1) in degrees as measured by compass
> bearings, Fig. 2,and  (2) converted to linear scale using:
>  A' = cos (45 - A) + 1, where A is the aspect in degrees (Beers et al,
> 1966,
> Journal of Forestry), Fig.3. The resulting index values range from 0 to 2
> (0
> = SW, 1 = SE and NW, 2 = NE), see Fig. 1. I have fitted regression lines
> after determining the appropriate fit. One of my concerns is: in Fig. 3,
> the
> hump around A'=1.5 could either correspond to N or E, as both take the
> value
> 1.5 (see Fig.1). So this index of aspect does not allow me to interpret if
> the plants are more abundant at N or at E facing slopes. Fig 2. allows me
> to
> distinguish data from the different aspects, and it is easier to explain;
> but the explanatory variable here is circular, and I am concerned whether
> it
> is correct to apply regressions on circular data. I showed these graphs to
> some of my colleagues and I got mixed responses. I would like to know which
> representation is more appropriate? Right now I am leaning towards Fig. 2,
> but I am concerned about the statistical appropriateness. I will include
> the
> explanatory Fig. 1 with either graph that I choose to finally use.
> Also, if there is a better way to display/ analyze this, please let me
> know.
>
> Thanks,
> Madhu Srinivasan
> Department of Biology
> University of Kentucky
>

Reply via email to