Construct Mean-Variance plots for Multivariate Abundance Data

Construct mean-variance plots, separately for each column of the input data, and separately for each level of any input factor that is given (via a formula). This function was specially written for high dimensional data where there are many correlated variables exhibiting a mean-variance structure, in particular, multivariate abundance data in ecology.

Usage

meanvar.plot(x, ...)

# S3 method for mvabund
meanvar.plot(
  x, n.vars=NULL, var.subset=NULL, subset=NULL, table=FALSE, ...)

# S3 method for mvformula
meanvar.plot(
  x, n.vars = NULL, var.subset=NULL, subset=NULL, table=FALSE, 
  overall.main=NULL, overlay=TRUE, ...)

Arguments

x

an mvabund objects or a Model Formula (can be a formula or a mvformula) to be used.

n.vars

the number of variables to include in the plot.

var.subset

vector of indices indicating the variables to be included on the plot, (default: the n.vars most abundant variables).

subset

an optional vector specifying a subset of observations to be used.

table

logical, whether a table of the Means and Variances should be returned

overall.main

an overall title for the window.

overlay

logical, whether overall means/variances for all variables are calculated and drawn on a single plot or calculated and plotted separately for different variables.

...

arguments to be passed to or from other methods.

Details

meanvar.plot calculates a mean-variance plot for a dataset with many variables (e.g., Warton D. I., Wright S., and Wang, Y. (2012)).

The mean values and variances are calculated across all observations, unless a formula is given as the first argument which specifies a factor as the dependent variable. In this latter case the means and variances are calculated separately within the groups defined by these factors.

By default the means and variances of all variables (and all factor levels) are displayed on the same plot. If a formula is given and the explanatory variables contain factor variables, the mean values and variances for each factor level can be calculated and displayed either for all variables together or for each variable separately.

For the latter, set overlay to FALSE. The mean-variances corresponding to the different factors will be drawn in different colors, that can be chosen by specifying col. col can then either be a single color value (see par) with the number of values being at least the maximum number of levels of the factors. The same applies to pch.

If mfrow is NULL and mfcol is NULL, par("mfrow") is used. If all.labels = FALSE, only the x-axis labels at the bottom plot and the y-axis labels of plots on the right side of the window are printed if furthermore main=NULL only the graphics on the top contain the full title, the other ones an abreviated one.

Note, that if a log-transformation is used for displaying the data, a specific mean-variance relation will not be visible in the plot, if either the calculated mean is zero and log!="x" or log!="xy" or if the calculated variance is zero and log!="y" or log!="xy".

By default the y/x ratio of the axis, specified by asp, will be set to 1 if log!="xy". If the mean-variance relation is not displayed on a log scale and overlay is FALSE, it is most often not advisable to specify asp, as there might not be one choice of asp that is sensible for each of the plots.

Value

If table is TRUE a table of the Means and Variances is returned. Otherwise, only the plot(s) is/are drawn.

References

Warton D. I., Wright S., and Wang, Y. (2012). Distance-based multivariate analyses confound location and dispersion effects. Methods in Ecology and Evolution, 3(1), 89-101.

Warton D.I. (2008). Raw data graphing: an informative but under-utilized tool for the analysis of multivariate abundances. Austral Ecology 33(3), 290-300.

Author

Ulrike Naumann, Stephen Wright and David Warton <David.Warton@unsw.edu.au>.

See also

plot.mvabund plot.mvformula.

Examples

require(graphics)

## Load the tikus dataset:
data(tikus)
tikusdat <- mvabund(tikus$abund)
year <- tikus$x[,1]

## Plot mean-variance plot for a mvabund object with a log scale (default):
meanvar.plot(tikusdat)   
#> 5 mean values were <=0 and could not be included in the log-plot
#> 5 variance values were <=0 and could not be included in the log-plot


## Again but without log-transformation of axes:
meanvar.plot(tikusdat,log="")   


## A mean-variance plot, data organised by year, 
## for 1981 and 1983 only, as in Figure 7a of Warton (2008):
is81or83 <- year==81 | year==83
meanvar.plot(tikusdat~year, subset=is81or83, col=c(1,10))
#> Error in eval(parse(text = variables[1])): object 'year' not found