1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
|
\name{dotchartpl}
\alias{dotchartpl}
\title{Enhanced Version of dotchart Function for plotly}
\description{
This function produces a \code{plotly} interactive graphic and accepts
a different format of data input than the other \code{dotchart}
functions. It was written to handle a hierarchical data structure
including strata that further subdivide the main classes. Strata,
indicated by the \code{mult} variable, are shown on the same
horizontal line, and if the variable \code{big} is \code{FALSE} will
appear slightly below the main line, using smaller symbols, and having
some transparency. This is intended to handle output such as that
from the \code{summaryP} function when there is a superpositioning
variable \code{group} and a stratification variable \code{mult},
especially when the data have been run through the \code{addMarginal}
function to create \code{mult} categories labelled \code{"All"} for
which the user will specify \code{big=TRUE} to indicate non-stratified
estimates (stratified only on \code{group}) to emphasize.
When viewing graphics that used \code{mult} and \code{big}, the user
can click on the legends for the small points for \code{group}s to
vanish the finely stratified estimates.
When \code{group} is used by \code{mult} and \code{big} are not, and
when the \code{group} variable has exactly two distinct values, you
can specify \code{refgroup} to get the difference between two
proportions in addition to the individual proportions. The individual
proportions are plotted, but confidence intervals for the difference
are shown in hover text and half-width confidence intervals for the
difference, centered at the midpoint of the proportions, are shown.
These have the property of intersecting the two proportions if and
only if there is no significant difference at the \code{1 - conf.int}
level.
Specify \code{fun=exp} and \code{ifun=log} if estimates and confidence
limits are on the log scale. Make sure that zeros were prevented in
the original calculations. For exponential hazard rates this can be
accomplished by replacing event counts of 0 with 0.5.
}
\usage{
dotchartpl(x, major=NULL, minor=NULL, group=NULL, mult=NULL,
big=NULL, htext=NULL, num=NULL, denom=NULL,
numlabel='', denomlabel='',
fun=function(x) x, ifun=function(x) x, op='-',
lower=NULL, upper=NULL,
refgroup=NULL, sortdiff=TRUE, conf.int=0.95,
minkeep=NULL, xlim=NULL, xlab='Proportion',
tracename=NULL, limitstracename='Limits',
nonbigtracename='Stratified Estimates',
dec=3, width=800, height=NULL,
col=colorspace::rainbow_hcl)
}
\arguments{
\item{x}{a numeric vector used for values on the \code{x}-axis}
\item{major}{major vertical category, e.g., variable labels}
\item{minor}{minor vertical category, e.g. category levels within
variables}
\item{group}{superpositioning variable such as treatment}
\item{mult}{strata names for further subdivisions without
\code{group}s}
\item{big}{omit if all levels of \code{mult} are equally important or
if \code{mult} is omitted. Otherwise denotes major (larger points,
right on horizontal lines) vs. minor (smaller, transparent points
slightly below the line).}
\item{htext}{additional hover text per point}
\item{num}{if \code{x} represents proportions, optionally specifies
numerators to be used in fractions added to hover text. When
\code{num} is given, \code{x} is automatically added to hover text,
rounded to 3 digits after the decimal point.}
\item{denom}{like \code{num} but for denominators}
\item{numlabel}{character string to put to the right of the numerator
in hover text}
\item{denomlabel}{character string to put to the right of the
denominator in hover text}
\item{fun}{a transformation to make when printing estimates. For
example, one may specify \code{fun=exp} to anti-log estimates and
confidence limites that were computed on a log basis}
\item{ifun}{inverse transformation of \code{fun}}
\item{op}{set to for example \code{'/'} when \code{fun=exp} and
effects are computed as ratios instead of differences. This is used
in hover text.}
\item{lower}{lower limits for optional error bars}
\item{upper}{upper limits for optional error bars}
\item{refgroup}{if \code{group} is specified and there are exactly two
groups, specify the character string for the reference group in
computing difference in proportions. For example if
\code{refgroup='A'} and the \code{group} levels are \code{'A','B'},
you will get B - A.}
\item{sortdiff}{\code{minor} categories are sorted by descending
values of the difference in proportions when \code{refgroup} is used,
unless you specify \code{sortdiff=FALSE}}
\item{conf.int}{confidence level for computing confidence intervals
for the difference in two proportions. Specify \code{conf.int=FALSE}
to suppress confidence intervals.}
\item{minkeep}{if \code{refgroup} and \code{minkeep} are both given,
observations that are at or above \code{minkeep} for at least one of
the groups are retained. The defaults to to keep all observations.}
\item{xlim}{\code{x}-axis limits}
\item{xlab}{\code{x}-axis label}
\item{tracename}{\code{plotly} trace name if \code{group} is not used}
\item{limitstracename}{\code{plotly} trace name for \code{lower} and
\code{upper} if \code{group} is not used}
\item{nonbigtracename}{\code{plotly} trace name used for non-big
elements, which usually represent stratified versions of the "big"
observations}
\item{col}{a function or vector of colors to assign to \code{group}.
If a function it will be evaluated with an argument equal to the
number of distinct groups.}
\item{dec}{number of places to the right of the decimal place for
formatting numeric quantities in hover text}
\item{width}{width of plot in pixels}
\item{height}{height of plot in pixels; computed from number of strata
by default}
}
\value{a \code{plotly} object. An attribute \code{levelsRemoved} is
added if \code{minkeep} is used and any categories were omitted from
the plot as a result. This is a character vector with categories
removed. If \code{major} is present, the strings are of the form
\code{major:minor}}
\author{Frank Harrell}
\seealso{\code{\link{dotchartp}}}
\examples{
\dontrun{
set.seed(1)
d <- expand.grid(major=c('Alabama', 'Alaska', 'Arkansas'),
minor=c('East', 'West'),
group=c('Female', 'Male'),
city=0:2)
n <- nrow(d)
d$num <- round(100*runif(n))
d$denom <- d$num + round(100*runif(n))
d$x <- d$num / d$denom
d$lower <- d$x - runif(n)
d$upper <- d$x + runif(n)
with(d,
dotchartpl(x, major, minor, group, city, lower=lower, upper=upper,
big=city==0, num=num, denom=denom, xlab='x'))
# Show half-width confidence intervals for Female - Male differences
# after subsetting the data to have only one record per
# state/region/group
d <- subset(d, city == 0)
with(d,
dotchartpl(x, major, minor, group, num=num, denom=denom,
lower=lower, upper=upper, refgroup='Male')
)
n <- 500
set.seed(1)
d <- data.frame(
race = sample(c('Asian', 'Black/AA', 'White'), n, TRUE),
sex = sample(c('Female', 'Male'), n, TRUE),
treat = sample(c('A', 'B'), n, TRUE),
smoking = sample(c('Smoker', 'Non-smoker'), n, TRUE),
hypertension = sample(c('Hypertensive', 'Non-Hypertensive'), n, TRUE),
region = sample(c('North America','Europe','South America',
'Europe', 'Asia', 'Central America'), n, TRUE))
d <- upData(d, labels=c(race='Race', sex='Sex'))
dm <- addMarginal(d, region)
s <- summaryP(race + sex + smoking + hypertension ~
region + treat, data=dm)
s$region <- ifelse(s$region == 'All', 'All Regions', as.character(s$region))
with(s,
dotchartpl(freq / denom, major=var, minor=val, group=treat, mult=region,
big=region == 'All Regions', num=freq, denom=denom)
)
s2 <- s[- attr(s, 'rows.to.exclude1'), ]
with(s2,
dotchartpl(freq / denom, major=var, minor=val, group=treat, mult=region,
big=region == 'All Regions', num=freq, denom=denom)
)
# Note these plots can be created by plot.summaryP when options(grType='plotly')
# Plot hazard rates and ratios with confidence limits, on log scale
d <- data.frame(tx=c('a', 'a', 'b', 'b'),
event=c('MI', 'stroke', 'MI', 'stroke'),
count=c(10, 5, 5, 2),
exposure=c(1000, 1000, 900, 900))
# There were no zero event counts in this dataset. In general we
# want to handle that, hence the 0.5 below
d <- upData(d, hazard = pmax(0.5, count) / exposure,
selog = sqrt(1. / pmax(0.5, count)),
lower = log(hazard) - 1.96 * selog,
upper = log(hazard) + 1.96 * selog)
with(d,
dotchartpl(log(hazard), minor=event, group=tx, num=count, denom=exposure,
lower=lower, upper=upper,
fun=exp, ifun=log, op='/',
numlabel='events', denomlabel='years',
refgroup='a', xlab='Events Per Person-Year')
)
}
}
\keyword{hplot}
|
