ggplot2: Fix legend for logarithmic color scale

Question

I'm trying to plot fitted model effects in ggplot2 as an alternative to the plots returned by the effects package, and I'm running into issues with mapping a continuous grouping factor to scale_color_gradient. These issues stem from the fact that the grouping factor has a skewed distribution. With the default color mapping, most of the colors are indistinguishable from one another, but a log-transformation messes up the legend. I've looked at a few related SO answers (Is there a built-in way to do a logarithmic color scale in ggplot2?, R colour scale for logarithmic data?, and Logarithmic color scale in ggplot2 squishes certain legend numbers) that don't quite fit the bill.

Here's my data:

myEffs <- structure(list(PrimeShiftIndex = c(-4, -0.2, 4, -4, -0.2, 4,
-4, -0.2, 4, -4, -0.2, 4), PrimeVowelDur = c(0.03, 0.03, 0.03,
0.06, 0.06, 0.06, 0.09, 0.09, 0.09, 1.59, 1.59, 1.59), fit = c(-0.184306629528313,
-0.164313919815862, -0.142216714344205, -0.200749305969527, -0.178039844592615,
-0.152939913597082, -0.210367655099129, -0.186068995874736, -0.159212583047775,
-0.278488972243709, -0.242934925102426, -0.203638346683111),
    se = c(0.0437103286485701, 0.0342751848548937, 0.0446524040373885,
    0.0417352317881704, 0.0340007746839495, 0.042093900962637,
    0.0441609220226782, 0.0341565687974652, 0.0442166991273061,
    0.0995662189943997, 0.041203801253227, 0.0993299532144987
    ), lower = c(-0.269979086288845, -0.231493448847753, -0.229735643449126,
    -0.282550563276701, -0.24468152835471, -0.235444164230796,
    -0.296923277064622, -0.253016036857112, -0.245877528409642,
    -0.473639245768052, -0.323694575976204, -0.398325538129639
    ), upper = c(-0.0986341727677806, -0.0971343907839703, -0.0546977852392849,
    -0.118948048662354, -0.11139816083052, -0.0704356629633676,
    -0.123812033133635, -0.119121954892359, -0.0725476376859078,
    -0.0833386987193667, -0.162175274228647, -0.00895115523658357
    )), class = "data.frame", row.names = c(NA, -12L), .Names = c("PrimeShiftIndex",
"PrimeVowelDur", "fit", "se", "lower", "upper"))

Here, the grouping factor PrimeVowelDur is skewed right, with values at 0.03, 0.06, 0.09, and 1.59. Here are some failed attempts to get distinguishable colors and a readable legend (with obnoxiously wide lines to highlight the color contrast or lack thereof).

p <- ggplot(myEffs, aes(x=PrimeShiftIndex, y=fit, group=PrimeVowelDur, color=PrimeVowelDur)) +
  geom_line(size=6)
##Legend fine but line colors indistinguishable
p
##Missuse's suggestion yields the same issue
p + scale_color_gradientn(colors = colorRampPalette(colors = c("#132B43", "#56B1F7"))(nrow(myEffs)), 
                          values = scales::rescale(log(sort(myEffs$PrimeVowelDur))))


##Colors distinguishable but legend messed up
p + scale_color_gradient(trans="log")
##Using trans="log" with pre-defined breaks as per Gregor doesn't make legend much better
brks <- seq(0, 1.6, length.out=5)
p + scale_color_gradient(trans="log", breaks=brks, labels=brks)
##Nor does S Rivero's suggestion
p + scale_color_gradient(trans="log", breaks=brks, labels=brks, guide="legend")

My intuition is that ggplot should be able to naturally handle the p + scale_color_gradient(trans="log") solution without messing up the legend. Anyway, I've got a proposed solution, but I want to see if there's anything savvier out there that I'm missing.

If you are grouping by a factor which takes discrete values, shouldn't you be using a discrete colour scale? — neilfws, Aug 10 '18 at 01:06
Here, the PrimeVowelDur grouping factor is based on a continuous predictor, but it's set at four values for the purposes of constructing model predictions by the effects package. These values represent the minimum (=1Q), median, 3Q, and maximum for PrimeVowelDur in the original data. — Dan Villarreal, Aug 10 '18 at 03:23

score 2 · Accepted Answer · answered Aug 10 '18 at 00:22

If you pass breaks based on the values of the continuous grouping factor to the breaks and labels arguments of scale_color_gradient, the legend will appear as expected. This works whether you use the default colorbar guide, or the legend guide:

brks2 <- sort(unique(myEffs$PrimeVowelDur))
p + scale_color_gradient(trans="log", breaks=brks2, labels=brks2)
p + scale_color_gradient(trans="log", breaks=brks2, labels=brks2, guide="legend")

ggplot2: Fix legend for logarithmic color scale

1 Answers1