Currently points and regression line are plotted using the means of 20 iterations (to obtain a 90% confidence interval).
The error bars are not very readable. Instead, it might be better just to plot all iterations (with smaller dots). Then draw a confidence interval around the regression line as an area.
Ideally the error bars would represent the uncertainty in tree-building, not sampling; however, since NJ only gives one answer, we don't know the tree uncertainty.
In addition, the boundary problem may be tested (and solved) by varying the grid offset. This would then result in samples under different conditions which should be plotted separately (or on the same plot using different colours or symbols).
Undecided as to whether it is valid to consider relationship between TD and PD using different grid offsets at the same time. Seems a bit like trying to draw a correlation from unlike data; however, if the areas are the same, surely they reflect the same experiment?
Currently points and regression line are plotted using the means of 20 iterations (to obtain a 90% confidence interval).
The error bars are not very readable. Instead, it might be better just to plot all iterations (with smaller dots). Then draw a confidence interval around the regression line as an area.
Ideally the error bars would represent the uncertainty in tree-building, not sampling; however, since NJ only gives one answer, we don't know the tree uncertainty.
In addition, the boundary problem may be tested (and solved) by varying the grid offset. This would then result in samples under different conditions which should be plotted separately (or on the same plot using different colours or symbols).
Undecided as to whether it is valid to consider relationship between TD and PD using different grid offsets at the same time. Seems a bit like trying to draw a correlation from unlike data; however, if the areas are the same, surely they reflect the same experiment?