plotrix (version 3.4-5)

barNest: Display a nested breakdown of numeric values

Description

Breaks down the elements of a data frame by one or more categorical elements and displays the breakdown as a bar plot.

Usage

barNest(formula=NULL,data=NULL,FUN=c("mean","sd"),ylim=NULL,
 main="",xlab="",ylab="",shrink=0.1,errbars=FALSE,col=NA,
 labelcex=1,lineht=NA,showall=TRUE,barlabels=NULL,showlabels=TRUE,mar=NULL,
 arrow.cap=NA,trueval=NA)

Arguments

formula

A formula with a numeric element of a data frame on the left and one or more categorical elements on the right.

data

A data frame containing the elements in formula.

FUN

The functions to apply to x.

ylim

Optional y limits for the plot, usually necessary for counts.

main

Title for the plot.

xlab,ylab

Axis labels for the plot. The x axis label is typically blank

shrink

The proportion to shrink the width of the bars at each level.

errbars

Whether to display error bars on the lowest level of breakdown.

col

The colors to use to fill the bars. See Details.

labelcex

Character size for the group labels.

lineht

The height of a line of text in the lower margin of the plot in user units. This will be calculated by the function if a value is not passed.

showall

Whether to display bars for the entire breakdown.

barlabels

Optional group labels that may be useful if the factors used to break down the numeric variable are fairly long strings.

showlabels

Whether to display the labels below the bars.

mar

If not NULL, a four element vector to set the plot margins. If new margins are set, the user must reset the margins after the function exits.

arrow.cap

The width of the "cap" on error bars in user units, calculated on the basis of the number of bars in the final breakdown if NA.

trueval

If this is not NA, the call to brkdnNest will return the proportions of the response variable that are equal to trueval. See Details.

Value

• The summary list produced by brkdnNest.

Details

barNest displays a bar plot illustrating the hierarchic breakdown of the elements of a data frame. The breakdown is performed by brkdnNest and the actual display is performed by drawNestedBars. The heights of the bars will be proportional to the values returned by the first function in FUN. If showall is TRUE, the entire nested breakdown will be displayed. This can be useful in visualizing the relationship between groups and subgroups in a compact format.

If trueval is not NA and brkdnNest is called to calculate the values for the heights of the bars, the proportions of the dichotomous response variable that are equal to trueval will be returned.

A number of functions can be passed in the FUN argument. Two functions, propbrk and valid.n that are provided can be used as bar heights, giving proportions and counts in each group and subgroup respectively. Binomial confidence limits can be added to the proportions returned by propbrk with binciWl and binciWu as in the second last example. If valid.n is the first element of FUN, the "overall" bar and label will be suppressed, as they are not informative.

The colors of the bars are determined by col. If showall is FALSE, the user only need pass a vector of colors, usually the same length as the number of categories in the final (last on the right side) element in the formula. If showall is TRUE and the user wants to color all of the bars, a list with as many elements as there are levels in the breakdown should be passed. Each element should be a vector of colors, again usually the same length as the number of categories. As the categorical variables are likely to be factors, it is important to remember that the colors must be in the correct order for the levels of the factors. When the levels are not in the default alphanumeric order, it is quite easy to get this wrong. As a barNest plot with more than a few factors and levels in each factor is quite dense, easily distinguished colors for each level of the breakdown may be preferable.

References

Lemon, J. & Levy, O. (2011) barNest: Illustrating nested summary measures. Statistical Computing and Graphics Newsletter of the American Statistical Association, 21(2): 5-10.

See Also

brkdnNest, drawNestedBars, superbarplot(UsingR)

Examples

# recreate the Titanic data frame and show the three way breakdown
 titanic<-data.frame(
  class=c(rep("1st",325),rep("2nd",285),rep("3rd",706),rep("Crew",885)),
  age=c(rep("Adult",319),rep("Child",6),rep("Adult",261),rep("Child",24),
  rep("Adult",627),rep("Child",79),rep("Adult",885)),
  sex=c(rep("M",175),rep("F",144),rep("M",5),rep("F",1),
  rep("M",168),rep("F",93),rep("M",11),rep("F",13),
  rep("M",462),rep("F",165),rep("M",48),rep("F",31),
  rep("M",862),rep("F",23)),
  survived=c(rep("Yes",57),rep("No",118),rep("Yes",140),rep("No",4),rep("Yes",6),
  rep("Yes",14),rep("No",154),rep("Yes",80),rep("No",13),rep("Yes",24),
  rep("Yes",75),rep("No",387),rep("Yes",76),rep("No",89),
  rep("Yes",13),rep("No",35),rep("Yes",14),rep("No",17),
  rep("Yes",192),rep("No",670),rep("Yes",20),rep("No",3)))
 require(plotrix)
 titanic.colors<-list("gray90",c("#0000ff","#7700ee","#aa00cc","#dd00aa"),
  c("#ddcc00","#ee9900"),c("pink","lightblue"))
 barNest(survived~class+age+sex,titanic,col=titanic.colors,showall=TRUE,
  main="Titanic survival by class, age and sex",ylab="Proportion surviving",
  FUN="propbrk",shrink=0.15,trueval="Yes")
 # now show the actual numbers of passengers
 barNest(survived~class+age+sex,titanic,col=titanic.colors,showall=TRUE,
  main="Titanic passengers and crew by class, age and sex",ylab="Number",
  FUN="valid.n",shrink=0.15)
 # to see this properly displayed, start a wide plot window
 # x11(width=10)
 test.df<-data.frame(Age=rnorm(100,25,10),
  Sex=sample(c("Male","Female"),100,TRUE),
  Marital=sample(c("Div","Mar","Sing","Wid"),100,TRUE),
  Employ=sample(c("FT","PT","Un"),100,TRUE))
 test.col<-list(Overall="gray",Sex=c("pink","lightblue"),
  Marital=c("mediumpurple","orange","tan","lightgreen"),
  Employ=c("#1affd8","#caeecc","#ff90d0"))
 barNest(formula=Age~Sex+Marital+Employ,data=test.df,ylab="Mean age (years)",
  main="Mean age by subgroups",errbars=TRUE,col=test.col)
 # set up functions for 20th and 80th percentiles
 q20<-function(x,na.rm=TRUE) return(quantile(x,probs=0.2,na.rm=TRUE))
 q80<-function(x,na.rm=TRUE) return(quantile(x,probs=0.8,na.rm=TRUE))
 # show the asymmetric dispersion measures
 barNest(formula=Age~Sex+Marital+Employ,data=test.df,ylab="Mean age (years)",
  main="Use median and quantiles for dispersion",FUN=c("median","q80","q20"),
  errbars=TRUE,col=test.col)
 barNest(formula=Employ~Sex+Marital,data=test.df,ylab="Proportion unemployed",
  main="Show the proportion unemployed",FUN=c("propbrk","binciWu","binciWl"),
 errbars=TRUE,col=test.col,trueval="Un")
 barNest(formula=Age~Sex+Marital+Employ,data=test.df,ylab="Counts",
  main="Show the counts in subgroups (final level only)",FUN="valid.n",
  col=test.col,showall=FALSE,ylim=c(0,10))
 barNest(formula=Age~Sex+Marital+Employ,data=test.df,ylab="Counts",
  main="Show all the counts in subgroups",FUN="valid.n",mar=c(5,5,4,2),
  col=test.col)