Significantly Different Levels per Group

Extracts, for every factor level, the set of levels it differs significantly from, based on the pairwise comparisons of a post hoc test.

Usage

signifDiff(x, ...)

# S3 method for class 'PostHocTest'
signifDiff(
  x,
  alpha = NULL,
  direction = TRUE,
  labels = "numbers",
  minlength = 3L,
  sep = ", ",
  ...
)

# S3 method for class 'pairwise.htest'
signifDiff(
  x,
  alpha = 0.05,
  direction = FALSE,
  labels = "numbers",
  minlength = 3L,
  sep = ", ",
  ...
)

# S3 method for class 'signifDiff'
print(x, legend = TRUE, ...)

Arguments

x

an object of class PostHocTest as returned by postHocTest(), or of class pairwise.htest as returned by pairwise.t.test() and friends

...

further arguments, not used so far

alpha

the significance level; defaults to 1 - conf.level of the object, or to 0.05 where no confidence level is stored

direction

logical; if TRUE, a sign is appended to every label, giving the direction of the row level's mean relative to the listed level. Requires the differences to be present in the object, which is not the case for the p-value branch and for pairwise.htest objects.

labels

either a keyword defining how the levels are abbreviated, one of "numbers" (the default), "letters", "LETTERS", "abbreviate" or "names", or a character vector of labels with one entry per level

minlength

the minimum length of the abbreviations, used for labels = "abbreviate" only

sep

the separator between the labels

legend

logical; if TRUE, the meaning of the signs is printed as a footer, separated by a rule

Value

a list with one data frame per term, each holding the columns label (the label of the level itself) and diff (the labels of the levels it differs significantly from), with the levels as row names. The class is "signifDiff", the significance level is kept in the attribute alpha.

Details

The result answers the question a reader of a post hoc table usually has: not which of the pairs are significant, but which levels a given level is distinguishable from, and in which direction.

The labels are numbers by default. Letters are available, but carry the opposite meaning of a compact letter display, where levels sharing a letter are those that could not be distinguished; here a label marks a significant difference.

See also

postHocTest(), pairwise.t.test()

Other test.posthoc: conoverTest(), dscfTest(), dunnTest(), dunnettTest(), gamesHowellTest(), nemenyiTest(), plot.PostHocTest(), postHoc, scheffeTest(), steelTest()

Examples

r.aov <- aov(breaks ~ tension, data = warpbreaks)
res <- postHocTest(r.aov, method = "hsd")

signifDiff(res)
#> 
#>   Posthoc multiple comparisons of means : Tukey HSD 
#>     levels a level differs from, at alpha = 0.05
#> 
#> $tension
#>   label diff  
#> L 1     2+, 3+
#> M 2     1-    
#> H 3     1-    
#> 
#> ---
#> Sign codes:  '+' mean of the row level is higher than the listed one, '-' lower
#> 

# stricter level, without recomputing the test
signifDiff(res, alpha = 0.01)
#> 
#>   Posthoc multiple comparisons of means : Tukey HSD 
#>     levels a level differs from, at alpha = 0.01
#> 
#> $tension
#>   label diff
#> L 1     3+  
#> M 2         
#> H 3     1-  
#> 
#> ---
#> Sign codes:  '+' mean of the row level is higher than the listed one, '-' lower
#> 

# readable labels instead of numbers
signifDiff(res, labels = "abbreviate", minlength = 4)
#> 
#>   Posthoc multiple comparisons of means : Tukey HSD 
#>     levels a level differs from, at alpha = 0.05
#> 
#> $tension
#>   label diff  
#> L L     M+, H+
#> M M     L-    
#> H H     L-    
#> 
#> ---
#> Sign codes:  '+' mean of the row level is higher than the listed one, '-' lower
#> 

# the p-value branch carries no differences, hence no signs
signifDiff(postHocTest(r.aov, method = "hsd", conf.level = NA))
#> 
#>   Posthoc multiple comparisons of means : Tukey HSD 
#>     levels a level differs from, at alpha = 0.05
#> 
#> $tension
#>   label diff
#> L 1     2, 3
#> M 2     1   
#> H 3     1   
#> 

signifDiff(pairwise.t.test(warpbreaks$breaks, warpbreaks$tension))
#> 
#>   Pairwise comparisons : t tests with pooled SD 
#>     levels a level differs from, at alpha = 0.05
#> 
#> $warpbreaks$breaks and warpbreaks$tension
#>   label diff
#> L 1     2, 3
#> M 2     1   
#> H 3     1   
#>