pwpp: Pairwise P-value plot

Description Usage Arguments Details Note Examples

View source: R/pwpp.R


Constructs a plot of P values associated with pairwise comparisons of estimated marginal means.


pwpp(emm, method = "pairwise", by, sort = TRUE, values = TRUE,
  rows = ".", xlab, ylab, xsub = "", = 0, ...)



An emmGrid object


Character or list. Passed to contrast, and defines the contrasts to be displayed. Any contrast method may be used, provided that each contrast includes one coefficient of 1, one coefficient of -1, and the rest 0. That is, calling contrast(object, method) produces a set of comparisons, each with one estimate minus another estimate.


Character vector of variable(s) in the grid to condition on. These will create different panels, one for each level or level-combination. Grid factors not in by are the primary factors: whose levels or level combinations are compared pairwise.


Logical value. If TRUE, levels of the factor combinations are ordered by their marginal means. If FALSE, they appear in order based on the existing ordering of the factor levels involved. Note that the levels are ordered the same way in all panels, and in many cases this implies that the means in any particular panel will not be ordered even when sort = TRUE.


Logical value. If TRUE, the values of the EMMs are included in the plot. When there are several side-by-side panels due to by variable(s), the labels showing values start stealing a lot of space from the plotting area; in those cases, it may be desirable to specify FALSE or use rows so that some panels are vertically stacked.


Character vector of which by variable(s) are used to define rows of the panel layout. Those variables in by not included in rows define columns in the array of panels. A "." indicates that only one row is used, so all panels are stacked side-by-side.


Character label to use in place of the default for the P-value axis.


Character label to use in place of the default for the primary-factor axis.


Character label used as caption at the lower right of the plot.

Numeric value to adjust amount of space used for value labels. Positioning of value labels is tricky, and depends on how many panels and the physical size of the plotting region. This parameter allows the user to adjust the position. Changing it by one unit should shift the position by about one character width (right if positive, left if negative).


Additional arguments passed to contrast and summary.emmGrid


Factor levels (or combinations thereof) are plotted on the vertical scale, and P values are plotted on the horizontal scale. Each P value is plotted twice – at vertical positions corresponding to the levels being compared – and connected by a line segment. Thus, it is easy to visualize which P values are small and large, and which levels are compared. In addition, factor levels are color-coded, and the points and half-line segments appear in the color of the other level. The P-value scale is nonlinear, so as to stretch-out smaller P values and compress larger ones. P values smaller than 0.0004 are altered and plotted in a way that makes them more distinguishable from one another.

If xlab, ylab, and xsub are not provided, reasonable labels are created. xsub is used to note special features; e.g., equivalence thresholds or one-sided tests.


The ggplot2 package must be installed in order for pwpp to work.


pigs.lm <- lm(log(conc) ~ source * factor(percent), data = pigs)
emm = emmeans(pigs.lm, ~ percent | source)
pwpp(emm, method = "trt.vs.ctrl1", type = "response", side = ">")

Example output

emmeans documentation built on Dec. 1, 2019, 1:15 a.m.