Local smoothing of scatterplots

Lowess can be used to smooth time series but was originally developed as a general way to draw a smooth curve on any type of scatterplot. Again, the smoothed value for any observation is obtained by fitting a least squares line to the observations with adjacent values for the explanatory variable. The fitted values that are obtained in this way are joined with lines.

(Most computer software implements a version of lowess that is a actually bit more complex than has been described here but our simpler version gives a good flavour of the method.)

Rose powdery mildew

The following scatterplot shows the results of an experiment that was conducted to determine how the latent period (time in days from initial infection to sporation) of rose powdery mildew is affected by temperature.

As in the previous page, drag over the crosses to see the 'adjacent' values that are used for the least squares line that provides each 'smoothed' value.

Use the slider to adjust the 'window width'. Windows of about 10 adjacent points give a fairly smooth curve.