
Correlation 229
The function pcf is generic, with a method for point patterns. By default, pcf.ppp uses the esti-
mator (7.26), with the Epanechnikov kernel, with bandwidth h selecte d by Stoyan’s rule of thumb
(7.29) with c = 0.15 so that c
′
= 0.067.
In analysing a po int pattern dataset, attention is often fo cused on the behaviour of g (r) for small
r, which reflects th e correlation between close pairs of points. Unfortunately, estimation of g(r)
at small distances is intrinsically difficult, because geometry dictates that there will typ ically be
few observations at small distances. The estimator (7.26) h as poor performance at small distances.