Too many points on xy scatter plot!

  • Thread starter Thread starter Steve
  • Start date Start date
S

Steve

I have been searching for a way to limit the number of
symbols in an xy plot. When I have a couple of columns of
several hundred points that I want to plot on the same
chart, I need a way to differentiate them in black and
white. I can do thick and thin, but if I want to use
symbols, they all overlap to make a thick line. Dashes
overlap to make a thin line.

Is there a way to only put a symbol every tenth point or
something similar?

Thanks!
Steve
 
Debra Dalgleish said:
Jon Peltier has instructions for this in his Charting FAQ article:

http://pubs.logicalexpressions.com/Pub0009/LPMArticle.asp?ID=209#jon024
Click to expand...
IMHO, following Edward Tufte, the purpose of a graph is to display
information and what does plotting the points contribute when you have
several hundred (less than half the resolution of the screen, I hope)? The
same information is present in a non-smoothed XY plot without points. Again
for myself, unless I am using a smoothed curve with a limited number of
points or plotting error bars, I never plot the actual points.

That being said, I can see the utility of plotting a graph with a fraction
of the available points for illustrative purposes.
 
James -

IMHO, plotting smoothed curves can distort a chart, and it ranks up
there with 3D effects as techniques I avoid. If you plot a polynomial
or other fitted line to data, and present it as such, that is not a
distortion. Showing all the points in addition to the fitted curve can
help illustrate the variability in the data. Depending on the data and
on the proper use of markers, showing several hundred points on the
chart can improve the information content of the chart. Small markers
(2-pt squares) or unfilled ones like crosses work pretty well. Other
ways to indicate variability include judicious use of error bars, or
drawing confidence bands around the fitted curve.

Showing every Nth point only helps to identify different lines in a
chart, helpful if you only have a B&W printer. I don't think you should
show a marker at every Nth point without having the line connect every
point, to retain a sense of the overall distribution.

Following Edward Tufte (I'm a big fan of his, BTW), you should not use
smoothed curves unless there's an underlying reality to the curve. Oh,
I said that. And also, you should probably not eliminate data if the
data adds to the transfer of information about the data (i.e., its
distribution). Hmm, I said that too, not in so many words.

- Jon
 
Jon Peltier said:
James -

IMHO, plotting smoothed curves can distort a chart, and it ranks up
there with 3D effects as techniques I avoid. If you plot a polynomial
or other fitted line to data, and present it as such, that is not a
distortion. Showing all the points in addition to the fitted curve can
help illustrate the variability in the data. Depending on the data and
on the proper use of markers, showing several hundred points on the
chart can improve the information content of the chart. Small markers
(2-pt squares) or unfilled ones like crosses work pretty well. Other
ways to indicate variability include judicious use of error bars, or
drawing confidence bands around the fitted curve.

Showing every Nth point only helps to identify different lines in a
chart, helpful if you only have a B&W printer. I don't think you should
show a marker at every Nth point without having the line connect every
point, to retain a sense of the overall distribution.

Following Edward Tufte (I'm a big fan of his, BTW), you should not use
smoothed curves unless there's an underlying reality to the curve. Oh,
I said that. And also, you should probably not eliminate data if the
data adds to the transfer of information about the data (i.e., its
distribution). Hmm, I said that too, not in so many words.

- Jon
Click to expand...

I won't repeat my post but I think we are in essential agreement. It is just
that an *unsmoothed* graph conveys just the same information, with or
without the points. If you are going to plot a fraction of the points with
the trend line or average or whatever to show variability and say so, again
I think we agree.

Best wishes,

Jim.
 
Back
Top