Re: weighted or non-weighted model?

> For ranges from 0 to100%, someone surely should have
> suggested Logistic. Logistic is also the natural shape for
> growth curves, though you seem to be describing un-growth
> or destruction. Zeros are a problem, though.

I don't have much experience with logistic regression. Do you think it is
right method for describing changes in cover? Can you suggest the formula of
model? I mean how should I transform plot covers to make them suitable for
logistic regression?

> I don't think a single analysis is going to satisfy
> everyone, so it might make sense to present results in
> more than one way. Is a decrease from 10% cover to 0%
> bigger than a decrease from 50% to 25%? - It depends
> on what the reader is interested in.

Good idea. In my area the problem is that high up in the mountains healthy
forest covers 10% of plot and around 50% in the lowest elevations. I tried
to use dependent variable change in cover as a percent of initial cover but
the problem is that in such situation correlation is weaker (R square around
0.3) model with dependant variable percent cover in 2000 has around 0.6.
Interesting fact is that distribution of standardized residuals looks
similar on map.

The scatter plots
> should be interesting. Perhaps you could scatter
> Pre versus Post, for one altitude, with different symbols
> for different cohorts of trees (assuming the same trees
> were 16 years older in 2000 than 1984).

Yes they are. When I plot initial cover versus final cover I can see
something which resembles cloud of linear dependance around line y=x and a
"rain" of plots with decrease in cover from y=x line.


tree cover in the end .x x x
| x .x x
| x .xx x
| x .x x x x
| x x . x
| x .x x
| x x .xx x
| x .x x x
| x . x x
| . x x x
| x .x x x
| x .x x x x
| . a a x x
| .x x x x x
|0 .x xxxxxxxxxxxxxxxxxx x x x start cover
|-------------------------------------------------------------------------


Plots located in y=x seem to suggest linear realtionship and the "rain" of
plots which decresase in cover causes heteroscedasticity.

Do you think that this impossible to remove heteroscedasticity is a sign
that I should look for different model then OLS?

Thank for you help

Best regards

Rafal


.