before getting scolded for submitting a (non-)bug report:
when using the 'symbols' function for plotting boxplot data (i.e. using
'boxplots' symbols), I noted that the x/y-position of the symbols is
associated with the center of the box.
while this is obviously natural for a usual plotting symbol (say a
circle or a rectangle), it is probably never desired if one uses the
'boxplots' symbols: looking at such a plot, probably everyone will
assume that y-position is that of the median within the box (in other
words: that one can read off the values of the medians from the y-axis).
the center of the box in a 'boxplots' symbol is not a very special
point, generally. in short: I think 'symbols' with 'boxplots'
should
behave as 'boxplot' does with respect to the y-position. (this of course
does make sense only, if you presume, that the user actually _did_
specify the median as the y-coordinate in the 'symbols' call -- but that
seems the only natural choice in this case, right?)
the current behaviour is counter-intuitive, I believe, if the
distributions are asymmetrical (and the median is not centered in it's
box). (I even think, that such plots are misinterpreted easily: think
what happens if the median lies very near one of the hinges in one box
and is centered in another one within the same plot and the medians are
actually the same)
in short: I think the 'boxplots' should not be centered at the specified
y-coordinates but rather drawn with a y-coordinate of
y + bxh * (0.5 - bxm)
where bxh and bxm are the second (box height) and fifths (median as
fraction of box height) column of the 'boxplots'
matrix. in this way, the median position is identical to the specified
y-coordinate.
if such a change 'breaks' current usage, I would propose a further
argument to 'symbols' to control the positioning behaviour (box center
vs. median drawn at specified y-coordinate).
at the very least, I think, the manpage should explicitely state that
all symbols (including boxplots) are positioned with their geometrical
center at the specified coordinates and that this might not be what the
user wants for 'boxplots' (OK: I was looking at my data for quite some
time until I noticed the source of my trouble. I would have appreciated
such a remark :-)).
regards,
joerg
