BookmarkSubscribeSubscribe to RSS Feed

Smoother graph with confidence levels


Community Trekker


Sep 28, 2011

I have attached a couple of images of graphs I have created.

The raw data is a series of temperature probe measurements showing cooling curves by pallet. The image 'data trends' shows each trend, the image 'smoother' shows a summary curve. I want to be able to represent the distribution on this summary curve. A suggestion has been to have a shaded area around the line that indicates 95% confidence, but we have no idea how to construct this graph. Any ideas?



Staff (Retired)


Sep 6, 2011

Hello Phill,

If you are asking how to generate a shaded area, you could use the BAND statement in PROC SGPLOT which is part of ODS graphics.

Here is an example:

ods graphics on;

data toplot;

    set sashelp.air;

    L95 = air - 0.30*air;

    U95 = air + 0.30*air;


proc sgplot data=toplot;

    band    x=date lower=L95 upper=U95 / fillattrs=GraphConfidence legendlabel="Confidence" name="band";

    series  x=date y=air / legendlabel="Summary Curve" name="series";

    keylegend "series" "band"/ location=outside position=bottom;


Note that the order of the BAND and the SERIES statements is important. If reverted, the line produced by the series is not seen.






Jun 23, 2011

I'm not aware of a good definition of a confidence interval for a spline smoother. (There is no real parameter being estimated.) Your data may suggest a definition -- it looks like it's made up of multiple series. For instance, you might take the 95% confidence interval of the mean for each X. With some effort, you can plot such an interval in GB using the Area element in Range mode. You have to make a plot with 3 Ys (data, upper and lower) and use the right-click > element > Y submenu to turn off the data Y for the Area element and turn off the upper/lower Ys for the Smoother element.


Another technique is to make lots of copies of the variable using random samples, and plot smoothers for the copies transparently. Here's an attempt at that with 30 resamples:



Community Trekker


Sep 28, 2011

Thanks for the ideas. We'll have a look at whether we can apply this to what we're doing. Phill