- v. - Book website is now people.bath.ac.uk/jjf23/ELM
- p27 - clarification: the log-likelihood is given for the logit.
- p29 - last paragraph: n-s degrees of freedom not n-l
- p30 - first displayed equation: right curly brace missing
- p40 - the brlr package has been superceded by the
brglm package. Also replace the brlr() function
call with brglm()
- p41 - Nagelkerke (not Naglekerke)
- p44 - missing parentheses in second displayed equation should be
(1+(k-1)tau^2)mp(1-p)
- p52 - Q2 dataset is wbca not wbcd
- p55 - Definition of the Poisson distribution - should be exp(-mu)
in the first formula, not exp(mu).
- p57 - third displayed equation. Differentiating wrt beta_j gives...
\sum (y_i - \exp(x_i^T \hat\beta)) x_{ij} = 0 (where x_ij refers to the jth element
of vector x_i)
- p64 - second displayed equation: the second sum should go up to y_{i} - 1 ,
not y_{i-1}
-
- p71 - second para should read: "By examining the coefficients, we see
that wafers without particles occur at a significantly higher
rate than wafers with particles. Similarly, we see that
good-quality wafers occur at a significantly higher
rate than bad-quality wafers."
- p74 - 3 lines from bottom: "log" should be omitted.
- p88 - 2 lines from bottom: "can be extracted"
- p85 - "...age is jointly independent of smoking ..."
- p86 - line 6, should be "... all three two-way interactions".
- p86 - mid-page. The odds-ratio is 1.5333, not the log odds-ratio.
- p94 - Q10 dataset is UCBAdmissions not UCBadmissions
- p102 - bottom "...can be viewed as a GLM-type..."
- p112 - 3 lines from top: Unfortunately, the hazards are not proportional. An analogy
to the second equation on p108 can be made, but then it would be log(1-gamma) that is
proportional which is not the hazard in this case. The name of proportional hazards arises because
the model is defined by analogy with a continuous response proportional hazards as
commonly used in survival analysis, where the proportionality does indeed hold.
- p115 - "...we are able to tackle..."
- p121 - Subscript i's are missing in Table 6.2 for the Gamma and Inverse
Gaussian distribution.
- p124 - Spelling is jackknife not jacknife (and elsewhere)
- p125 - First formula, the square root applies to the remainder of the term.
- p141 - line 10, The SD for lognormal is incorrectly set. It should be
sdlog=sqrt(0.61102)
- p146 - gmod <- glm(d ~ Method+Preheating.. should be gmod <- glm(d ~ Material+Preheating..
- p156 - "...standard errors can be obtained..."
- p156 - End of second paragraph should read: "The idea is to find all independent linear combinations of the response, $k$, such that $k^T X = 0$. Form matrix $K$ with
columns $k$ then $K^T y ...."
- p156 and following: The lme4 package (and the underlying
Matrix package)
has been changing since the preparation
of the book. Some differences may be observed between the current version and
that used in the book. In
particular, recent versions of summary() and anova() of
lmer()-produced models now
do not contain p-values. Please
read about changes to these two chapters
to address this as well as some discussion of other inferential methods
that could be used.
- p158 - "...this is a real concern..."
- p161 - last but one displayed equation should have an alphai
rather than an ai
- p180 - last para, Scheffé not Scheffeé
- p181 - Q4a degrees of freedom are no longer output so
question is not meaningful.
- p186 - To produce a corrected version of Figure 9.2, use:
xyplot(log(income+100) ~ year | sex, psid, type="l", groups=person)
- p190 - second para, the standard deviations for the intercept and slope are 0.531 and 0.049. The residual SD is 0.684 not 0.46. (Numbers given are actually the variances)
- p198 - first line 6.3% not 6.5% (twice)
- p198 - line 4, 0.0179 should be 0.0174 (twice)
- p198 One cannot see that the math scores are less variable than
the english scores using the printed output. This effect is, however, visible
in Figure 9.9.
- p205 second para - "...the estimates for the GEE are about half
the size.."
- p212 (and following) - the variable name duration in the
faithful dataset has changed to eruptions. It still
represents the duration of the eruption.
- p215 - completely. (missing .)
- p223 - should be "They are also orthonormal,.."
- p223 - The wavethresh package has been extensively
updated. Some changes are necessary to achieve the same output
- p224 - The plots are now obtained by:
wds <- wd(exa$y,filter.number=1,bc="interval")
draw(filter.number=1,family="DaubExPhase")
-
p225 - last block of code should now be:
wds <- wd(exa$y,filter.number=2,bc="interval")
draw(filter.number=2,family="DaubExPhase")
plot(wds)
- p231 - line 13, should be "variables"
- p235 summary(ammgcv) is missing (needed to get output seen)
- p248-249 Using > summary(lm(ozone[,1] ~ a$x[,-1])) for the three
regression summaries gives the correct R2. These are more
in line with previous models. The printed version is incorrect because
these are calculated by R assuming there is no intercept when in fact
this is already included in the X-matrix. The newly available R
package, earth, provides a much better interface to mars
- p250 - line 6, use predict not predict.mars
- p254 - line 12, ordered factor (not factors)
- p255 - better to say "from 21,115 to 4114+5478 = 9592" for accuracy
- p259 - Should use plotcp(roze) for consistency with output on p258
- p259 - post(roz,digits=3,filename="") is better since it ensures
that the display uses 3 significant digits for the node means.
- p261 - The entropy definition should not contain the 2.
- p261 - Rijksmuseum van Natuurlijke Historie
- p265 - Last para, should be xerror not rel error
- p282 - First displayed equation. LRT is minus the value
given.
- p287 - the brlr package has been superceded by the
brglm package.
- p293 - Nagelkerke (not Naglekerke)
- p294 - Scheffé not Scheffeé