Function to draw a multivari chart

A multivari chart of one quantitative response variable depending on two to four categorical variables can be drawn.

multivari(var, fac1, fac2, fac3 = NULL, fac4 = NULL, data, sort = FALSE,
 fun = mean, pch = list(15, 16, 17), col = list("black", "blue", "red"),
 col0 = "black", cex = 1, fac.cex = 2, xlab.depth = 3, legend = FALSE,
 main = paste("multivari chart of", var), add = FALSE, ...)

Arguments

var

variable name (character string) or column index of response variable, required

fac1

variable name (character string) or column index of first level factor, required; precedes fac2 and fac3 (if present) in the hierarchy (see Details)

fac2

variable name (character string) or column index of second level factor, required; follows fac1 and precedes fac3 (if present) in the hierarchy (see Details)

fac3

variable name (character string) or column index of third level factor, optional; if present, fac3 is the last factor in the hierarchy (see Details)

fac4

variable name (character string) or column index of fourth level factor, optional; can only be specified if there is also a third level factor; if present, this factor is the first in the hierarchy (see Details), and separate multivari charts for the first three factors are drawn for each level of this factor

data

a data frame, required

sort

logical, specifying whether or not levels are sorted, when converting character vectors to factors (a single choice for all factors is needed); default: FALSE

fun

a function to be used in aggregation; default: mean

pch

a list of length 2 or 3, depending on whether or not fac3 is specified; the ith list element can be an individual plotting symbol (like the usual pch entry) or a vector of plot symbols for each level of fac_i

col

a list of length 2 or 3, depending on whether or not fac3 is specified; the ith list element can be an individual color or a vector of colors for each level of fac_i; this color specification is used for the plot symbols of fac_i levels and for the lines connecting the symbols for the next level in the hierarchy

col0

the color for the first line to be drawn

cex

the size of axis annotation text (annotation of the fourth level header is 1.5 times this size)

fac.cex

a multiplier for cex; plot symbol sizes are fac.cex*cex; default: 2

xlab.depth

labels for the horizontal axis are printed down to this level of the hierarchy (default: 3); if the depth is reduced, different plot symbols should be used, and a legend should be drawn

legend

logical determining whether or not a legend should be drawn (default: FALSE); the function determines wether top right or bottom right yields a better position (it is not guaranteed that there is no overlab); if this does not work well, one can manually draw a legend in the outer margin

main

title, as usual; a default is provided

add

logical; add to an existing plot (which of course has to have suitable axis limits)?; default: FALSE; note that horizontal axis labeling will always be printed by function multivari, while vertical axis labeling will be omitted for add=TRUE

...

further arguments to functions plot, lines, points, mtext

Details

The function is inspired by Minitabs behavior for multivari charts (see also Bruno Scibilia's blog which is linked in the references). It does not attempt to visualize individual observations.

A multivari chart mainly serves exploratory purposes. It works particularly well with balanced data, but can also be used for messy data. multivari can visualize the dependence of a single quantitative variable on up to four factors (i.e., interactions of order up to four can be visualized). The display is hierarchical: for factors later in the hierarchy, conditional means given level combinations of factors earlier in the hierarchy are displayed. Therefore, the order of the factors can make a big difference in the display. If there is no natural order, it may be worthwhile to inspect several orders.

For interactions with two factors only, it is often preferrable to use function interaction.plot or raw.means.plot.

Value

a list of (lists of) data frames with summary statistics to be plotted

References

Scibilia, Bruno (2013). Using Multi-Vari Charts to Analyze Families of Variations. https://blog.minitab.com/en/using-variability-charts-to-analyze-call-center-wait-times.

Author

Ulrike Groemping

See also

See also interaction.plot, raw.means.plot

Examples

##---- Should be DIRECTLY executable !! ----
##-- ==>  Define data, use random,
##--  or do  help(data=index)  for the standard data sets.
if (FALSE) { # \dontrun{
require(car)
multivari("cycles", "len", "load", "amp", data=Wool, 
    col=list("black","red",c("grey70","grey45","grey20")),
    pch=list(15,17,8), legend=TRUE, xlab.depth = 2, lwd=2)
multivari("cycles", "load", "len", "amp", data=Wool, 
    col=list("black",c("red","blue","darkgreen"),
    c("grey70","grey45","grey20")),
    pch=list(15,17,8), legend=TRUE, xlab.depth = 2, lwd=2)
    
## create a fake fourth factor
fakedat <- rbind(cbind(newfac="blabla",Wool),cbind(newfac="albalb",Wool))
## make it character for demonstrating the effect of sort option
fakedat$newfac <- as.character(fakedat$newfac)

## default: sort order in the data is respected (order of unique is used)
multivari("cycles", "load", "len", "amp", "newfac", data=fakedat, 
    col=list("black",c("red","blue","darkgreen"),
    c("grey70","grey45","grey20")),
    pch=list(15,17,8), legend=TRUE, xlab.depth = 2, lwd=2, cex=0.8)
    
## sort=TRUE: levels are sorted (order of sort(unique))
multivari("cycles", "load", "len", "amp", "newfac", data=fakedat, 
    col=list("black",c("red","blue","darkgreen"),
    c("grey70","grey45","grey20")),
    pch=list(15,17,8), legend=TRUE, xlab.depth = 2, lwd=2, cex=0.8,
    sort=TRUE)
} # }