Produce summaries of various types of variables. Calculate descriptive statistics for x and use Word as reporting tool for the numeric results and for descriptive plots. The appropriate statistics are chosen depending on the class of x. The general intention is to simplify the description process for lazy typers and return a quick, but rich summary.
A 2-dimensional table will be described with it's relative frequencies, a short summary containing the total cases, the dimensions of the table, chi-square tests and some association measures as phi-coefficient, contingency coefficient and Cramer's V. Tables with higher dimensions will simply be printed as flat table, with marginal sums for the first and for the last dimension.
Desc(x, ..., main = NULL, plotit = NULL, wrd = NULL)# S3 method for default
Desc(x, main = NULL, maxrows = NULL, ord = NULL,
conf.level = 0.95, verbose = 2, rfrq = "111", margins = c(1,2),
dprobs = NULL, mprobs = NULL, plotit = NULL, sep = NULL, digits = NULL, ...)
# S3 method for data.frame
Desc(x, main = NULL, plotit = NULL, enum = TRUE, sep = NULL, ...)
# S3 method for list
Desc(x, main = NULL, plotit = NULL, enum = TRUE, sep = NULL, ...)
# S3 method for numeric
Desc(x, main = NULL, maxrows = NULL, plotit = NULL,
sep = NULL, digits = NULL, ...)
# S3 method for integer
Desc(x, main = NULL, maxrows = NULL, plotit = NULL,
sep = NULL, digits = NULL, ...)
# S3 method for factor
Desc(x, main = NULL, maxrows = NULL, ord = NULL, plotit = NULL,
sep = NULL, digits = NULL, ...)
# S3 method for ordered
Desc(x, main = NULL, maxrows = NULL, ord = NULL, plotit = NULL,
sep = NULL, digits = NULL, ...)
# S3 method for character
Desc(x, main = NULL, maxrows = NULL, ord = NULL, plotit = NULL,
sep = NULL, digits = NULL, ...)
# S3 method for logical
Desc(x, main = NULL, ord = NULL, conf.level = 0.95, plotit = NULL,
sep = NULL, digits = NULL, ...)
# S3 method for Date
Desc(x, main = NULL, dprobs = NULL, mprobs = NULL, plotit = NULL,
sep = NULL, digits = NULL, ...)
# S3 method for table
Desc(x, main = NULL, conf.level = 0.95, verbose = 2,
rfrq = "111", margins = c(1,2), plotit = NULL, sep = NULL, digits = NULL, ...)
# S3 method for formula
Desc(formula, data = parent.frame(), subset, main = NULL,
plotit = NULL, digits = NULL, ...)
# S3 method for Desc
print(x, digits = NULL, plotit = NULL, nolabel = FALSE,
sep = NULL, nomain = FALSE, …)
# S3 method for Desc
plot(x, main = NULL, …)
the object to be described. This can be a data.frame, a list, a table or a vector of the classes: numeric, integer, factor, ordered factor, logical.
a character vector, containing the main title(s).If this is left to NULL, the title will be composed as: variablename (class(es)),
resp. number - variablename (class(es)) if the enum option is set to TRUE.
Use NA if no caption should be printed at all.
the pointer to a running MS Word instance, as created by GetNewWrd() (for a new one)
or by GetCurrWrd() for an existing one.
All output will then be redirected there. Default is NULL, which will report all results to the console.
integer. With how many digits shoud the relative frequencies be formatted? Default can be set by DescToolsOptions(digits=x).
numeric; defines the maximum number of rows in a frequency table to be reported. For factors with many levels it is often not interesting to see
all of them. Default is set to 12 most frequent ones (resp. the first ones if ord is set to "levels" or "names").
For a numeric argument x maxrows is the minimum number of unique values needed for a numeric variable to be treated as continous. If left to its default NULL, x will be regarded as continuos if it has more than 12 single values. In this case the list of extreme values will be displayed and the frequency table else.
If maxrows is < 1 it will be interpreted as percentage. In this case just as many rows, as the maxrows% most frequent levels will be shown. Say, if maxrows is set to 0.8, then the number of rows is fixed so, that the highest cumulative relative frequency is the first one going beyond 0.8.
Setting maxrows to Inf will unconditionally report all values and also produce a plot with type "h" instead of a histogram.
character out of "name" (alphabetical order), "level", "asc" (by frequencies ascending), "desc" (by freqencies descending) defining the order for a frequency table as used for factors, numerics with few unique values and logicals.
Factors (and character vectors) are by default orderd by their descending frequencies, ordered factors by their natural order.
a string with 3 characters, each of them being 1 or 0, defining which percentages should be reported. The first position is interpreted as total
percentages, the second as row percentages and the third as column percentages.
"011" hence produces a table output with row and column percentages. If set to NULL rfrq is defined in
dependency of verbose (verbose = 1 sets rfrq to "000" and else to "111", latter meaning all percentages will be reported.) Applies only to tables and is ignored else.
a vector, consisting out of 1 and/or 2. Defines the margin sums to be included.
Row margins are reported if margins is set to 1. Set it to 2 for column margins and c(1,2) for both.
Default is NULL (none). Applies only to tables and is ignored else.
integer out of c(2, 1, 3) defining the verbosity of the reported results. 2 (default) means medium, 1 less and 3 extensive results. Applies only to tables and is ignored else.
confidence level of the interval. If set to NA no confidence interval will be calculated. Default is 0.95.
a vector with the probabilities for the Chi-Square test for days, resp. months, when describing a Date variable.
If this is left to NULL (default) then a uniform distribution
will be used for days and a monthdays distribution in a non leap year (p = c(31/365, 28/365, 31/365, ...)) for the months. Applies only to Dates and is ignored else.
logical, determining if in data.frames and lists a sequential number should be included in the main title. Default is TRUE. The reason for this option is, that if a Word report with enumerated headings is created, the numbers may be redundant or inconsistent.
boolean. Should a plot be created? The plot type will be chosen according to the classes of variables (roughly following a
numeric-numeric, numeric-categorical, categorical-categorical logic). Default can be defined by option plotit,
if it does not exist then it's set to FALSE.
character. The separator for the title. By default a line of "-" for the current width of the screen
(options("width")) will be used.
logical, defining if labels (defined as attribute with the name label, as done by Label) should be plotted.
a formula of the form lhs ~ rhs where lhs gives the data values and rhs the corresponding groups.
an optional matrix or data frame containing the variables in the formula formula.
By default the variables are taken from environment(formula).
an optional vector specifying a subset of observations to be used.
logical, determines if the main title of the output is printed or not, default is TRUE.
further arguments to be passed to or from other methods. For the internal default method these can include:
a vector of probabilities of the same length of x. An error is given if any entry of p is negative. This argument will be passed on to chisq.test. Default is rep(1/length(x), length(x)).
add_nilogical. Indicates if the group length should be displayed in the boxplot.
smoothcharacter, either "loess" or "smooth.spline" defining the type of smoother to be used in num ~ num plots. Default is loess for n < 500 and smooth.spline else.
A list containing the following components:
the length of the vector (n + NAs).
the valid entries (NAs are excluded)
number of NAs
number of unique values.
number of zeros
arithmetic mean
standard error of the mean, as calculated by MeanSE.
a table of quantiles, as calculated by
quantile(x, probs = c(.05,.10,.25,.5,.75,.9,.95), na.rm = TRUE).
standard deviation
coefficient of variation: mean(x) / sd(x)
median absolute deviation (mad)
interquartile range
skewness, as calculated by Skew.
kurtosis, as calculated by Kurt.
the lowest and the highest values, reported with their frequencies in brackets, if > 1.
a data.frame of absolute and relative frequencies given by Freq if maxlevels > unique values in the vector.
Desc is a generic function. It dispatches to one of the methods above depending on the class of its first argument. Typing ?Desc
+ TAB at the prompt should present a choice of links: the help pages for each of these Desc methods (at least if you're using RStudio, which anyway is recommended).
You don't need to use the full name of the method although you may if you wish; i.e.,
Desc(x) is idiomatic R but you can bypass method dispatch by going direct if you wish:
Desc.numeric(x).
This function produces a rich description of a factor, containing length, number of NAs, number of levels and
detailed frequencies of all levels.
The order of the frequency table can be chosen between descending/ascending frequency, labels or levels.
For ordered factors the order default is "level".
Character vectors are treated as unordered factors
Desc.char converts x to a factor an processes x as factor.
Desc.ordered does nothing more than changing the standard order for the frequencies to it's intrinsic order, which means order "level"
instead of "desc" in the factor case.
Description interface for dates. We do here what seems reasonable for describing dates. We start with a short summary about length, number of NAs and extreme values, before we describe the frequencies of the weekdays and months, rounded up by a chi-square test.
A 2-dimensional table will be described with it's relative frequencies, a short summary containing the total cases, the dimensions of the table, chi-square tests and some association measures as phi-coefficient, contingency coefficient and Cramer's V. Tables with higher dimensions will simply be printed as flat table, with marginal sums for the first and for the last dimension.
Note that NAs cannot be handled by this interface, as tables in general come in "as.is", say basically as a matrix without any further information about potentially previously cleared NAs.
Description of a dichotomous variable. This can either be a boolean vector, a factor with two levels or a numeric variable
with only two unique values.
The confidence levels for the relative frequencies are calculated by BinomCI, method "Wilson"
on a confidence level defined by conf.level.
Dichotomous variables can easily be condensed in one graphical representation. Desc for a set of flags (=dichotomous variables) calculates the frequencies, a binomial confidence intervall and produces a kind of dotplot with error bars.
Motivation for this function is, that dichotomous variable in general do not contain intense information. Therefore it makes sense to condense the description of sets of dichotomous variables.
The formula interface accepts the formula operators +, :, *, I(), 1 and evaluates any function.
The left hand side and right hand side of the formula are evaluated the same way.
The variable pairs are processed in dependency of their classes.
Word This function is not thought of being directly run by the enduser. It will normally be called automatically, when
a pointer to a Word instance is passed to the function Desc.
However DescWrd takes some more specific arguments concerning the Word output (like font or fontsize), which can make it necessary to call the function directly.
# NOT RUN {
opt <- DescToolsOptions()
# implemented classes:
Desc(d.pizza$wrongpizza) # logical
Desc(d.pizza$driver) # factor
Desc(d.pizza$quality) # ordered factor
Desc(as.character(d.pizza$driver)) # character
Desc(d.pizza$week) # integer
Desc(d.pizza$delivery_min) # numeric
Desc(d.pizza$date) # Date
Desc(d.pizza)
Desc(d.pizza$wrongpizza, main="The wrong pizza delivered", digits=5)
Desc(table(d.pizza$area)) # 1-dim table
Desc(table(d.pizza$area, d.pizza$operator)) # 2-dim table
Desc(table(d.pizza$area, d.pizza$operator, d.pizza$driver)) # n-dim table
# expressions
Desc(log(d.pizza$temperature))
Desc(d.pizza$temperature > 45)
# supported labels
Label(d.pizza$temperature) <- "This is the temperature in degrees Celsius
measured at the time when the pizza is delivered to the client."
Desc(d.pizza$temperature)
# try as well: Desc(d.pizza$temperature, wrd=GetNewWrd())
z <- Desc(d.pizza$temperature)
print(z, digits=1, plotit=FALSE)
# plot (additional arguments are passed on to the underlying plot function)
plot(z, main="The pizza's temperature in Celsius", args.hist=list(breaks=50))
# formula interface for single variables
Desc(~ uptake + Type, data = CO2, plotit = FALSE)
# bivariate
Desc(price ~ operator, data=d.pizza) # numeric ~ factor
Desc(driver ~ operator, data=d.pizza) # factor ~ factor
Desc(driver ~ area + operator, data=d.pizza) # factor ~ several factors
Desc(driver + area ~ operator, data=d.pizza) # several factors ~ factor
Desc(driver ~ week, data=d.pizza) # factor ~ integer
Desc(driver ~ operator, data=d.pizza, rfrq="111") # alle rel. frequencies
Desc(driver ~ operator, data=d.pizza, rfrq="000",
verbose=3) # no rel. frequencies
Desc(price ~ delivery_min, data=d.pizza) # numeric ~ numeric
Desc(price + delivery_min ~ operator + driver + wrongpizza,
data=d.pizza, digits=c(2,2,2,2,0,3,0,0) )
Desc(week ~ driver, data=d.pizza, digits=c(2,2,2,2,0,3,0,0)) # define digits
Desc(delivery_min + weekday ~ driver, data=d.pizza)
# without defining data-parameter
Desc(d.pizza$delivery_min ~ d.pizza$driver)
# with functions and interactions
Desc(sqrt(price) ~ operator : factor(wrongpizza), data=d.pizza)
Desc(log(price+1) ~ cut(delivery_min, breaks=seq(10,90,10)),
data=d.pizza, digits=c(2,2,2,2,0,3,0,0))
# response versus all the rest
Desc(driver ~ ., data=d.pizza[, c("temperature","wine_delivered","area","driver")])
# all the rest versus response
Desc(. ~ driver, data=d.pizza[, c("temperature","wine_delivered","area","driver")])
# pairwise Descriptions
p <- CombPairs(c("area","count","operator","driver","temperature","wrongpizza","quality"), )
for(i in 1:nrow(p))
print(Desc(formula(gettextf("%s ~ %s", p$X1[i], p$X2[i])), data=d.pizza))
# get more flexibility, create the table first
tab <- as.table(apply(HairEyeColor, c(1,2), sum))
tab <- tab[,c("Brown","Hazel","Green","Blue")]
# display only absolute values, row and columnwise percentages
Desc(tab, row.vars=c(3, 1), rfrq="011", plotit=FALSE)
# do the plot by hand, while setting the colours for the mosaics
cols1 <- SetAlpha(c("sienna4", "burlywood", "chartreuse3", "slategray1"), 0.6)
cols2 <- SetAlpha(c("moccasin", "salmon1", "wheat3", "gray32"), 0.8)
plot(Desc(tab), col1=cols1, col2=cols2)
# use global format options for presentation
Fmt(abs=as.fmt(digits=0, big.mark=""))
Fmt(per=as.fmt(digits=2, fmt="%"))
Desc(area ~ driver, d.pizza, plotit=FALSE)
Fmt(abs=as.fmt(digits=0, big.mark="'"))
Fmt(per=as.fmt(digits=3, ldigits=0))
Desc(area ~ driver, d.pizza, plotit=FALSE)
# plot arguments can be fixed in detail
z <- Desc(BoxCox(d.pizza$temperature, lambda = 1.5))
plot(z, mar=c(0, 2.1, 4.1, 2.1), args.rug=TRUE, args.hist=list(breaks=50),
args.dens=list(from=0))
# The default description for count variables can be inappropriate,
# the density curve does not represent the variable well.
set.seed(1972)
x <- rpois(n = 500, lambda = 5)
Desc(x)
# but setting maxrows to Inf gives a better plot
Desc(x, maxrows = Inf)
# Output into word document (Windows-specific example) -----------------------
# by simply setting wrd=GetNewWrd()
# }
# NOT RUN {
# create a new word instance and insert title and contents
wrd <- GetNewWrd(header=TRUE)
# let's have a subset
d.sub <- d.pizza[,c("driver", "date", "operator", "price", "wrongpizza")]
# do just the univariate analysis
Desc(d.sub, wrd=wrd)
# }
# NOT RUN {
DescToolsOptions(opt)
# }
Run the code above in your browser using DataLab