errorbargraph {cg}

Create an error bar graph based on pairwise multiple comparisons

Description

Creates a graph to see pairwise comparisons amongst groups. The method of Andrews, Sarner, and Snee (1980) is applied to visualizes significant differences via non-overlapping error bars.

Usage

errorbargraph(estimates, centralvar, critpoint, endptscale="log",
 analysisname="", endptname="", alpha=0.05, digits=NULL, approxstamp=FALSE,
 titlestamp=TRUE, offset=NULL, ticklabels=NULL, ...)

Arguments

estimates

A named vector of estimates. Each estimate element is a measure that will be the center of the error bar of the group. The name of each group must be present in the names attribute of the vector.

centralvar

A single variance value to be used for each group's error bar construction. In the canonical case it is the square of the estimated standard error of the mean estimate of the group, where each group also has the same standard error (and sample size). If the standard errors / variances are similar enough across the groups, the visualization may still be effective even though the error bar lengths will be approximations.

critpoint

The single critical value of the theoretical reference distribution. In the canonical case it is the t-distribution quantile for estimates derived from a standard linear model with homoscedastic variance. It could also reflect a multiplicity adjustment, or like the centralvar discussion in the previous item, it may serve as part of a visually useful approximation for other cases.

endptscale

Must be specified as "log" or "original". If the default "log" then the y-axis will be created with a logarithmic spacing. The tick marks will be calculated accordingly and expressed in the original scale of the estimates. The estimates vector must already be in the logarithmic scale.

analysisname

Optional, a character text or math-valid expression that will be set for default use in graph title and table methods. The default value is the empty "".

endptname

Optional, a character text or math-valid expression that will be set for default use as the y-axis label of graph methods, and also used for table methods. The default value is the empty "".

alpha

Significance level, by default set to 0.05, which equates to a 95% confidence level. This is just used for labelling purposes.

digits

Optional, for output display purposes in graphs and table methods, values will be rounded to this numeric value. Only the integers of 0, 1, 2, 3, and 4 are accepted. No rounding is done during any calculations. The default value is NULL, which will examine each individual estimates value and choose the one that has the maximum number of digits after any trailing zeroes are ignored. The max number of digits will be 4.

approxstamp

Add text to the graph that acknowledges that the error bar method is approximate.

titlestamp

Add text to the top margin above the graph area.

offset

Optional, if for example a numeric constant was added to all response values before calculation of the estimate as a mean, this could be used to shift the axis marks appropriately. The default value is NULL.

ticklabels

Optional, before graphing the data, remove any automatically generated tickmarks for the y-axis, and use these tickmarks instead. A vector of tickmarks to be placed on the y-axis. Any numeric representations will be coerced to character.

...

Additional arguments. None are currently used.

Details

The statistical method of Andrews, Sarner, and Snee (1980) is applied to visualizes significant differences via non-overlapping error bars. The method is exact when there are equal standard errors amongst the groups, and approximate otherwise. The method's usefulness declines as the standard errors become more disparate.

When two groups are compared, nonoverlapping error bars indicate a statistically significant pairwise difference. Conversely, if the error bars overlap, there is no such significant difference. In cases of approximation, or borderline overlap that is seen, the actual comparison needs to be consulted to judge significance with a p-value.

The minimum and maximum values across all the bar ends are added inside the plot region in blue, flush against the y-axis. The number of decimal places are determined by the digits value.

Value

errorbargraph returns an invisible NULL. The main purpose is the side effect of graphing to the current device.

Warning

This function was created for internal use in the cg package as its use can be seen in the errorBarGraph methods code. Therefore any direct use of it needs to be done cautiously.

Note

Contact [e-mail] for bug reports, questions, concerns, and comments.

Author(s)

Bill Pikounis [aut, cre, cph], John Oleynick [aut], Eva Ye [ctb]

References

Andrews, H.P., Snee, R.D., Sarner, M.H. (1980). "Graphical Display of Means," The American Statistician, 34, 195-199.

Examples

data(canine)
canine.data <- prepareCGOneFactorData(canine, format="groupcolumns",
                                      analysisname="Canine",
                                      endptname="Prostate Volume",
                                      endptunits=expression(plain(cm)^3),
                                      digits=1, logscale=TRUE, refgrp="CC")
canine.fit <- fit(canine.data)

## Easier way: notice the camel case of the errorBarGraph call
errorBarGraph(canine.fit, model="olsonly")

## Manual way
## Instead of errorBarGraph(canine.fit, model="olsonly")
errorbargraph(estimates=canine.fit@olsfit$coef,
              centralvar=((summary(canine.fit@olsfit)$sigma^2) /
                          unique(sapply(canine, length))),
              critpoint=qt(0.975, df=canine.fit@olsfit$df.residual),
              endptscale="log",
              analysisname="Canine",
              digits=1,
              endptname=expression(paste( plain('Prostate Volume'),
                                      ' (', plain(cm)^3  ,  ')' ))
              )