`Modern Applied Statistics with S-Plus' by W.N. Venables & B.D. Ripley
======================================================================

Errata for the second printing of the second edition.

p. ii	We supplied PostScript not LaTeX files.
	Dr Sheather's address is the Australian Graduate School of
	>Management<.

p. 8	S-mode has been replaced by ESS, from

	http://franz.stat.wisc.edu/pub/ESS/

p. 43	There is a new function regexpr() in 4.0 for Windows that
	behaves like grep() on Unix.  It is not used in objects(), though.

p. 49 l.-15  at least >1<.  Arrays can be of dimension 1 (although this
	is not very useful).

p. 67	Splus /BATCH under S-PLUS 4.0 can have a errfile specified
	following the outfile, in which case the errors are written
	to errfile.

p.119	l.-2  sweep(A, 2, apply(A,2,mean), "-")

p.151, 3 The incorrect uses of Integrate with a scalar return value
	give an error in S-PLUS 4.0 and later.

p.168 l.-3 estimate of the >skewness<.

p.179 (5.4) 1/b should be 1/(nb)

p.183	l.3  summation is over i, not j.

p.215	The calculation of Vars fails in S-PLUS 4.0, since
	var reports an error (rather than NA) for less than 2 cases.

p.277   first display.  theta/log_2

p.292	To work around a bug in deriv() introduced in S-PLUS 4.0,
	mix.gr needs to be altered to

mix.gr <- function(p, x) 
{
   u1 <- p[2]; s1 <- p[3]; u2 <- p[4]; s2 <- p[5]; p <- p[1]
   e <- lmix2a(x, p, u1, s1, u2, s2)
   rep(1, length(x)) %*% attr(e, "gradient")
}

p.313	The correlation here is in fact between observations, as lme
	has taken the ranks of the times. Set

	serial.covariate.transformation = "none"

	to achieve what we intended (but note this fails to fit correctly).

p.327   display.  ... - \alpha . (This is necessary as the smoothers
	used by gam() do not include the constant.)

p.340 l.7,8  x is the independent variables, y the dependent ones.

p.346	(12.3) The S code follows Fleming and Harrington in breaking
	ties, so if there were 3 deaths from 10 at risk the summand
	would be 1/10 + 1/9 + 1/8.

p.347 third displayed equation. - is needed before H(t).

p.354	Figure 12.3 is of the >leuk< dataset

p.362 l.17  The scaling used in the Schoenfeld residuals is actually to
	give I^{-1}schoenfeld where I is the (partial) information
	matrix from the Cox model.

p.370	We should have had age=c(50,50) - 48, but in fact survfit
	would in error here even with the correct age.

p.397 l.18 The first term of (13.3) is the >squared< Mahalanobis ...

p.411 (13.9) The scond Psi should be inverted too.

p.416 Caption: The numbers m/n denote the proportion of training
	cases reaching that node that are classified incorrectly by
	the label in the node

p.431	rts(scan(n=48)), not 47.

p.441 l.2  sqrt(m) + 0.11 + 0.12/sqrt(m).

p.457	The coh component is the _squared_ coherency, so these plots
	need some adjustment (and people need to use more helpful names).

p.478	The 3D spherical covariance should have r/d rather than d/r, twice.

p.489 l.-10 For >classification trees< we can use

p.542	offset.  Add page 229 as the primary reference.


Typos
-----

p. 91 l.-3 The axes >are< omitted
p.154 l.6 >A< small example
p.167 first line after code.  In S-PLUS 3.2 this >has< been modified..
p.169 in nclass.Scott, nclass.FD, ^(-1/3) would be better.
p.179 ll.-19,-18 "rectangular", "triangular" and "cosine" should be in
      typewriter font.
p.183 l.11 using the distances >between< all pairs of points
p.238 l.-7 For large models >it< is often helpful to 
p.253 l.-8 with one value >that< appears to be out by
pp.359, 529 Nagelkerke
p.454 l. -1 missing X after \Delta \Delta_{12} 
p.455 l. 3 missing X after \Delta \Delta_{12}
p.456 l.-5 missing ( after I_ii
p.490 l. 2 >logistic< regression
p.522 Box & Cox >An< analysis ...