proc rangebar


Version 2.33 Jun'06

Scripts


Manual page for proc_rangebar(PL)
proc rangebar draws one boxplot (also known as box-and-whisker plot) using the current data set, in the current plotting area. Rangebars express the range and distribution of one data variable. proc rangebar examines the chosen data field and computes some statistics (usually the median, inter-quartile range, and the 5th and 95th percentiles, and then renders the rangebar. Alternatively, rangebars can be used to show means and standard deviations.

Rangebars may be vertical or horizontal. N will be automatically displayed; there are also options for reporting of outliers and missing observations. See also the gallery rangebar examples.


Attributes

The datafield attribute is required. Default behavior is for statistics to be calculated from the contents of datafield to produce a vertical rangebar at X location 1.0 and for an N= label to be placed just above the X axis.    

datafield     dfield

Specifies the data field on which to compute descriptive statistics. Example: datafield: 2

axis     x | y

Determines which axis to plot against. x results in horizontal rangebars, while y results in vertical ones. Default is y. Example: axis: x

barloc     plotvalue

Location where the rangebar is to be rendered. For vertical boxplots this is a plottable value in X; for horizontal boxplots this is a plottable value in Y.
Example: barloc: 3


Bar appearance details

barwidth     n

The width of the box portion of the rangebar in absolute units. Default is 0.2 inches.
Example: barwidth: 0.1

color     color

Specifies the color of the box area. Example: color: yellow

mediansym     symboldetails | dot | line

Specifies the symbol that will be displayed to show the median. May be a symbol specification (to get dots, etc.) or line which is the default. dot or yes gives a small black dot. This also controls the mean symbol with meanmode: yes. Example: mediansym: shape=diamond

outline     yes | no

If yes, box is outlined with a line.

taildetails     linedetails

Controls color, width, etc. of tail lines.
Example: taildetails: color=blue width=1.8

tailmode     5/95 | minmax | 1.5iqr

Specifies whether the rangebar tails are to extend to the 5th and 95th percentile, to the min and max, or using 1.5 x IQR. Only relevant when statistics are computed internally. Default is 5/95.
Note: 1.5irq mode causes the tails to be rendered using the IQR, which is the interquartile range (the difference between 75th and 25th percentiles). This follows the formal Tukey specification, in that the lower tail will extend from the bottom of the box downward to the nearest data point on or above (25th - (1.5 x IQR)). Similar for upper tail.
Example: tailmode: minmax

95tics     yes | no

If tailmode is minmax, allows display of 5th and 95th percentile by adding tics. Example: 95tics: yes

outlinedetails     linedetails

Controls color, width, etc. of box outline.

truncate     yes | no

If yes, bars are truncated to plotting area. Default is yes.

ticlen     n

Length, in absolute units , of the tics which appear at the end of the tails. Default is 70% of the width of the bar.

meansym     yes | dot | symboldetails

When a median-based rangebar is being rendered, this attribute may be used to indicate that a second symbol be placed at the mean. The result will be two symbols, a symbol at the median, and a second symbol at the mean. If yes or dot, a default symbol (a small black dot) will be placed at the mean. Other symbols may be rendered by giving other symboldetails specifications. If no, a mean symbol will not be rendered. Default is no.
Example: meansym: shape=circle style=filled fillcolor=black radius=0.04)


Selecting data rows

select select expresion

Allows cases to be selected for inclusion using a selection expression.
Example: select: @2 = B


Mean & standard deviation mode

You can create rangebars that display means and standard deviations if you wish, instead of medians and IQR.

meanmode     yes | no

If yes, mean and standard deviation are computed and drawn as an error bar, instead of a boxplot of median/quartiles, etc. Mean is shown as a point (style can be controlled using mediansym). Number of standard deviations may be controlled using nstddevs. Cannot be used with values; use proc bars instead. Appearance of the lines may be controlled using taildetails.

logmean     yes | no

If yes, mean and standard deviation are computed in log space. Useful with data having log characteristics; when plotted in log space the standard deviations will appear equidistant from the mean. If log+1 scaling is in effect for the plot, then this feature will operate in log+1 space (allowing 0.0 values).

nstddevs     n

Used with meanmode. Specifies the number of standard deviations to use in each direction when drawing the error bar. Default is 1.

skipmed     yes | no

When using meanmode, this attribute may be specified as yes to speed things up a bit by avoiding computation of medians and percentiles. Default is always no.


Displaying outliers

showoutliers     yes | no

If yes, outliers will be displayed and/or reported upon. An outlier is any point that is beyond the end of either of the rangebar's tails. The default way of displaying outliers is circles for the near outliers and asterisks for the far outliers. The default boundary between near and far is 3 times the interquartile range.

outlierprint     yes | no

If yes, a report on each outlier will be printed to standard error.

outliernearsym     symboldetails | none

Specifies the symbol for displaying near outliers. Default is a small circle: shape=circle style=outline radius=0.05. Use none if you are displaying outliers as lines or labels and don't want any geometric symbol.

outlierfarsym     symboldetails | none

Specifies the symbol for displaying far outliers. Default is an asterisk: shape=circle style=spokes radius=0.05 Use none if you are displaying outliers as lines or labels and don't want any geometric symbol.

outliernearfarcutoff     n

The boundary between near outliers and far outliers will be n times the interquartile range. Default is 3.0.

outlierlinelen     n

Display outliers as line segments. If specified, all outliers (near and far) are displayed as short line segments rather than symbols. n will be the length of these segments, in absolute space. The color and width of these line segments may be controlled using outlierlinedetails.
Example: outlierlinelen: 0.1

outlierlabelfield     dfield

If specified, the contents of this field will appear as a text label for each outlier. For vertical rangebars the label will appear a bit to the right of the point. For horizontal rangebars the label will appear in rotated text, a bit above the point. The size, style, and location may be adjusted if desired using the outlierlabeldetails attribute.

outlierlabeldetails     textdetails

Details related to the outlier labels.

outlierlinedetails     linedetails

When outliers are being rendered as line segments, specifies color, line width, etc. of these line segments.


Labels

printn     yes | no

If yes, a label showing N (the number of observations) is produced. Default is yes. Example: printn: no

nlocation     locvalue

Where to position the N label. The label will be aligned with the rangebar. For vertical rangebars location indicates where to place the label in Y; for horizontal, X. Example: nlocation: -4

nword     string

A template that determines the format of the N label. Default is N=@@N. The N value is substituted for the @@N symbol. Example: (N = @@N)

printmissing     yes | no

If yes, a label showing the number of non-plottable (missing) values is produced. Only relevant if statistics are calculated internally. Example: printmissing: yes

mlocation     locvalue

Where to position the missing values label. The label will be aligned with the rangebar. For vertical rangebars location indicates where to place the label in Y; for horizontal, X.

mword     string

A template that determines the format of the label showing number of missing values. Default is M=@M. The N value is substituted for the @M symbol. Example: (@M missing)

ntextdetails     textdetails

Set the size, color, font or fine-tune the position of the N labels. Alternate name for backward compatibility: textdetails.
Example: ntextdetails: size=8 style=I

mtextdetails     textdetails

Set the size, color, font or fine-tune the position of the missing labels.
Example: mtextdetails: size=8 adjust=0,0.05

mwhenexists     yes|no

If yes, the missing label will be displayed only if the number of missing observations is greater than 0. Default is no.


Accessing the generated statistics

statsonly     yes | no

If yes, statistics will be computed and internal @variables set, but no rangebar rendered. I've used this to compute with meanmode: yes to compute the overall mean and standard deviation for a set of data points, then render them using lines (via @RANGEBARMEAN, @RANGEBARMIN, and @RANGEBARMAX).

showstats     yes | no | only

If yes or only, all the computed descriptive statistics will be written to the showstatsfile. If only, the statistics will be printed but no bars will be drawn.

showbriefstats     yes | no | only

If yes, and if N > 0, the most important computed descriptive statistics will be written to the showstatsfile. All fields are on one line, TAB-delimited, for convenient use by other programs. The fields are in this order:
tag, datafield, N, mean, standard deviation, median, min, max, #missing.
The tag contents may be set using the briefstatstag attribute; if this attribute is not set then the first result field will be datafield.

briefstatstag     text

Set the contents of the tag that will be written at the beginning of each 'briefstats' record. This may be useful in identifying cases and groups in the 'brief stats' output.

showstatsfile     filename

If specified, statistics will be written to this file. If not specified, statistics will be written to the diagnostic stream, usually stderr. File will be opened in append mode, so the caller may want to remove previous contents of the file before invoking ploticus.


Supplying pre-computed statistics

If you have already computed the summary stats, you can supply them using these attributes, and ploticus will render a rangebar to suit.

values     5thpercentile 25th median 75th 95th [N]

Specify pre-computed descriptive statistics that should just be plotted. 5 plottable values should be given. If a 6th value is given it is taken to be N, and printn is implied. Min/max may be substituted for the 5th and 95th percentiles if desired.
Example: values: 0.3 3.0 5.3 6.2 9.4 236

plotfields     5thpercentile 25th median 75th 95th [N]
or with meanmode: plotfields mean stddev [N]

Similar to values but rather than literal values, a set of dfield s are given. The data will be accessed from one row, which may be specified using plotrecord. (If plotrecord not specified, data will be taken from the first row by default. Proc processdata action: select may be used to isolate one row of data.) Example:
  plotfields: 1 2 3 4 5
  plotrecord: 1


plotrecord     n

Used with plotfields; indicates which data row to get the values from. First row is 1. Example: see plotfields above. If not specified, first row is assumed.


VARIABLES THAT ARE SET

NVALUES is set to N. If nothing was plotted, this will be equal to 0.

The following variables are set as soon as statistics are computed, whether or not a plot is actually drawn.

RANGEBARMEDIAN is set to hold the median value. This might be useful in drawing a line for comparing medians. Not set when using meanmode.

RANGEBARMEAN is set to hold the mean value. This might be useful in drawing a line for comparing means. Set only when using meanmode.

RANGEBARIQRMIN and RANGEBARIQRMAX are set to hold the values of the bottom and top of the box. Not set when using meanmode.

RANGEBARMIN and RANGEBARMAX are set to hold the values at the ends of the tails. Normally this will be the 5th and 95th percentile; with meanmode: yes it will be +/- one standard deviation.


data display engine  
Copyright Steve Grubb


Ploticus is hosted at http://ploticus.sourceforge.net
SourceForge Logo


Markup created by unroff 1.0,    June 02, 2006.