Beyond Bar Graphs
M. Toomey
2022-10-10
For a detailed discussion of the issues in this lesson please read:
Comparing the values of measures among treatment groups is a common approach to hypothesis testing in biology. However, the reliability of these comparisons depends on the sample size and distribution of the data. One of the most common ways to plot such comparisons is bar plot of the group means with standard error:
require(tidyverse)## Loading required package: tidyverse## ── Attaching packages ─────────────────────────────────────── tidyverse 1.3.2 ──
## ✔ ggplot2 3.3.6 ✔ purrr 0.3.4
## ✔ tibble 3.1.8 ✔ dplyr 1.0.10
## ✔ tidyr 1.2.1 ✔ stringr 1.4.1
## ✔ readr 2.1.2 ✔ forcats 0.5.2
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()# order factor levels in the full dataset
mpg_fact<-mpg %>%
mutate(class = factor(class, levels = c("2seater","subcompact","compact","midsize","minivan","pickup","suv")))
# plot a bar plot of the means with standard error
p1<-ggplot(mpg_fact, aes(class, hwy)) +
stat_summary(fun = mean,
geom = "col",
width = 0.5,
color = "red", fill = "white") +
stat_summary(geom = "errorbar",
width = 0.3) + # by default the standard error will be calculated from the "errorbar" geom
ylim(0,45) +
coord_flip()
p1## No summary function supplied, defaulting to `mean_se()`
Plots of these types are essentially “visual tables” of summary statisitics and do not inform the reader about the distribution of the data or sample size.
Here we will explore a variety of alternatives that offer the reader better opportunities to assess the data and consider the conclusion drawn from it.
Boxplots
As we have discussed previously in the Intro to ggplot, boxplots offer information about the distribution of the data.
p2<-ggplot(mpg_fact, aes(class, hwy)) +
geom_boxplot(color = "red") +
ylim(0,45) +
coord_flip()
p2
However, they do not provide information about sample size and the interpretation of distribution of data is not intuitive. For example:
from: https://nightingaledvs.com/ive-stopped-using-box-plots-should-you/Violin plots
Violin plots present symmetrical smoothed density profiles that reveal more of the variation in the frequency of values.
p3<-ggplot(mpg_fact, aes(class, hwy)) +
geom_violin(draw_quantiles = c(0.25, 0.5, 0.75))+ #added the draw_quantiles option to show the 25th, 50th, and 75th quantile of the data
stat_summary(fun = mean,
geom = "crossbar",
width = 0.4, color = "red") + #calculated the mean for each group and plotted a crossbar.
ylim(0,45) +
coord_flip()
p3
These violin plots give more information about the the distribution of the data, but sample size still remains obscure.
Univariate scatter plots
Why hide behind summary statistics when you can simply show the complete data. Dot plots allow for the easy plotting of continuous data among categorical groupings.
p4<-ggplot(mpg_fact, aes(class, hwy)) +
geom_jitter(width = 0.3,
alpha = 0.4, size = 1.6)+ # adding transparency with the alpha option makes overlaps easier to discern
stat_summary(fun = mean, geom = "crossbar",
width = 0.5, color = "red") + #plot a crossbar with the mean values
stat_summary(geom = "errorbar",
width = 0.3,
color = "red") + #plot standard error of the means
ylim(0,45) +
coord_flip()
p4## No summary function supplied, defaulting to `mean_se()`
This allows us to visualize all of the data and the mean in the same plot.
Let’s compare all four approaches:
require(patchwork)## Loading required package: patchwork(p1 + p2) / (p3 + p4) ## No summary function supplied, defaulting to `mean_se()`## No summary function supplied, defaulting to `mean_se()`
Which do you think conveys the most useful information?
Combining multiple plot elements
These various elements communicate different aspects of the distribution of the data. Therefore, it can sometimes be effective to combine elements in a single plot.
p5<-ggplot(mpg_fact, aes(class, hwy)) +
geom_violin(draw_quantiles = c(0.25, 0.5, 0.75),
position = position_nudge (x = 0.5, y = 0),
width = 0.4)+
geom_jitter(width = 0.2,
alpha = 0.4, size = 1.6)+
stat_summary(fun = mean, geom = "crossbar",
width = 0.4, color = "red") +
ylim(0,45) +
coord_flip()
p5
Paired/Repeated Observations
When the observations across the levels of a factor are linked (e.g. repeated measures of the same individual over time) it is essential to show the data points at each level. Summary statics like group means will not capture the patterns of change at the individual level.
# Let's generate some paired observations
ID<-c("A","B","C","D","E","F","G","A","B","C","D","E","F","G")
Obs<-c("before","before","before","before","before","before","before","after","after","after","after","after","after","after")
Measure<-c(runif(1:7),runif(1:7)*4)
RepEx<-tibble(ID,Obs,Measure) #make these observations into a tibble
RepEx<-RepEx %>%
mutate(Obs = factor(Obs, levels = c("before","after"))) #order the levels of observation as a factor
# build a point and line plot
p6<-ggplot(RepEx, aes(Obs, Measure))+
geom_point(alpha = 0.4,
size = 4,
aes(color = ID))+
geom_line(size = 2,
alpha = 0.4,
aes(color = ID, group = ID)) #group the line aesthetic to connect before and after observations by ID
p6
Compare this to a simple plot of the means
p7<-ggplot(RepEx, aes(Obs, Measure))+
stat_summary(fun = mean,
geom = "col",
width = 0.5,
fill = "white") +
stat_summary(geom = "errorbar",
width = 0.3)
p7## No summary function supplied, defaulting to `mean_se()`