• It's not funny if you have to explain it
  • Should be appropriate for your audience
    
  • Should be appropriate for your data
• It boils down to communication–make sure your graph conveys the relevant information in a digestable way

• Created by Hadley Wickham in 2005 for data visualization in R
• Alternative to base R graphics (more personalizable & versatile)
• Utilizes a graphical grammar of layers, where each component of the plot needs to be separately specified
• Extension projects include
  • ggplot for Python (ggpy: https://github.com/yhat/ggpy)
  • ggplot for Matlab (gramm: https://github.com/piermorel/gramm)
  • plotly, a web-based, collaborative data visualization program (https://plot.ly/)

• ggplot works in layers
• Each layer is added sequentially to produce the plot
• Each layer has five components –> makes ggplot call
  • A dataset
    
  • The aesthetic mapping (aes())
    
  • A statistical transformation (stat= )–there are some defaults so we don't always have to specify this
    
  • A geometric object (geom_)
    
  • A position adjustment (position = )

• First, import ggplot2 and your dataset ( + other relevant packages)

library(ggplot2)
library(dplyr)

df = read.csv("data/allANT.csv")
df_run = read.csv("data/ANT_runsheet.csv")

• We can use dplyr to read dataset into ggplot function!

• If we plan to use the same dataset for the whole plot, we only need to specify it once, rather than at each layer

• Note, the package is called "ggplot2" and the function is called "ggplot"

df %>% ggplot(...)

#To assign a plot to a variable...
plot1 = df %>% ggplot(...)

• What does our plot look like?

plot1

-There are no layers yet so the plot can't be displayed yet

plot1 = df_run %>% ggplot(aes(age)) +
  layer(
    mapping = NULL, #already provided above
    data = NULL, #already provided above
    stat = "bin",
    geom = "bar"
  )

Not quite there yet

Now we finally see our plot!

plot1

## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

• Yes! It is very time-consuming to type out all five components of each layer
• Certain components of a layer have inherent/default settings
• For example, each stat has a default geom, and vice versa
• Often, we can just specify the geom and get what we want

df_run %>% ggplot(aes(age)) + 
  geom_histogram() #Should be identical to previous plot

## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

• We can customize the layer by adding information to the geom_ function
• For a histogram, we can modify the size of the bars by increasing/decreasing the number of 'bins'
• The number of bins tells ggplot how many groups to split data into and count the frequencies in that group
• A large number of bins = fewer bars // small number of bins = more bars

df_run %>% ggplot(aes(age)) + 
  geom_histogram(binwidth = 1)

df_run %>% ggplot(aes(age)) + 
  geom_histogram(binwidth = 50) 

df %>% ggplot(aes(correctRT)) + 
  geom_histogram(binwidth = 5)

## Warning: Removed 569 rows containing non-finite values (stat_bin).

• We get a message that ggplot has removed the rows with NA values
• If we want to avoid this error, we can remove NA's

df %>% ggplot(aes(correctRT)) + 
  geom_histogram(binwidth = 5,
                 na.rm = TRUE) #Voila!

• To plot the frequency of a single, continuous variable we have lots of options
• Density plots are very common and function similarly to a histogram
• Rather than "stat = bin", density plots use "stat = density" and draw a line rather than bars

df %>% ggplot(aes(correctRT)) + 
  geom_density(na.rm = T)

That being said, we can tell geom_density to use the bin stat

df %>% ggplot(aes(correctRT)) + 
  geom_density(stat = "bin",
               binwidth = 10, 
               na.rm = T)

• We can add geoms on top of each other
• Remember, ggplot works in layers so each independent object can live under/on top of another

df %>% ggplot(aes(correctRT)) + 
  geom_histogram(binwidth = 10,
                 na.rm = T) +
  geom_density(stat = "bin",
               binwidth = 10,
               na.rm = T)  +
  geom_rug()

• The gray on gray makes this plot difficult to see
• We can adjust the colors and the transparencies to make this plot more readable

df %>% ggplot(aes(correctRT)) + 
  geom_histogram(binwidth = 10, na.rm = T, fill = "gray", color = "black", alpha = 0.5) +
  geom_density(stat = "bin", binwidth = 10, na.rm = T, color = "red") +
  geom_rug()

• We can plot the frequency of discrete variables with a bar graph
• Unlike geom_histogram, geom_bar COUNTS the observations whereas geom_histogram returns the count in each bin
• Geom_bar returns a more granular view of the data than geom_histogram
• Let's compare by looking at age

df_run %>% ggplot(aes(age)) + geom_histogram(binwidth = 1, na.rm = T)

df_run %>% ggplot(aes(age)) + geom_bar(na.rm = T)

• Let's say we want to look at the relationship between age and correct reaction time

df %>% ggplot(aes(age, correctRT)) + #Remember, the order is x, y
  geom_point(na.rm = T)

What if we first want to see the distribution of correct RT within a group?

Can use dplyr.

df.young = df %>% filter(group == "Young")
df.young %>% ggplot(aes(age, correctRT)) + #Remember, the order is x, y
  geom_point(na.rm = T, position = "jitter", alpha = 0.3, size = 1)

df.block %>%  ggplot(aes(block, mean_correctRT)) + 
  geom_line(na.rm = T, aes(color = group)) + geom_point(na.rm = T, aes(color = group)) + 
  geom_errorbar(aes(ymin = mean_correctRT - serr, ymax = mean_correctRT + serr), width = 0.3)

Without the error bars, it seemed like perhaps younger subjects got faster with time, but adding error bars suggests that this is likely not a significant change.

• Perhaps we want to compare young and old adults on correct reaction time
• Start by making a summary table of means & standard errors to be plotted

df.group = df %>% select(subnum, group, correctRT) %>% 
  group_by(subnum, group) %>% 
  summarise(mdv = mean(correctRT, na.rm = T)) %>% 
  ungroup() %>% group_by(group) %>% 
  summarise(N = n(), mean_correctRT = mean(mdv, na.rm = T), 
            sd = sd(mdv), serr = sd/sqrt(N)) 

df.group %>% ggplot(aes(group, mean_correctRT)) + 
  geom_bar(stat = "identity", position = "dodge", alpha = 0.7) + 
  geom_errorbar(aes(ymin = mean_correctRT - serr, ymax = mean_correctRT + serr), 
                width = 0.3, position = "dodge") +
  coord_cartesian(ylim=c(350,450)) + 
  scale_y_continuous(breaks = c(350, 375, 400, 425, 450))

## Be cautious of setting limits in scale_y_continuous because it will 
## remove data outside the range. 

df.group.err = df %>% select(subnum, group, block, error) %>% group_by(group, block) %>% 
  summarise(mean_err = mean(error, na.rm = T)) 

df.group.err %>% ggplot(aes(block, mean_err)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "dodge")

df.group.err %>% ggplot(aes(block, mean_err)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "dodge") + coord_flip()

df.group.err %>% ggplot(aes(block, mean_err)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "stack")

df.flank = df %>% select(subnum, group, flank, correctRT) %>% 
  group_by(subnum, group, flank) %>% 
  summarise(mdv = mean(correctRT, na.rm = T)) %>% 
  ungroup() %>% group_by(group, flank) %>% 
  summarise(N = n(), mean_correctRT = mean(mdv, na.rm = T), 
            sd = sd(mdv, na.rm=T), serr = sd/sqrt(N)) 

df.flank %>% ggplot(aes(flank, mean_correctRT, group = group)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "dodge", width = 0.8) + 
  geom_errorbar(aes(ymin = mean_correctRT - serr, ymax = mean_correctRT + serr), 
                width = 0.8, position = "dodge") +
  coord_cartesian(ylim=c(300,500)) + 
  scale_y_continuous(breaks = c(300, 350,400, 450, 500))

• Perhaps we want to see the mean RT for each flank across blocks
• Facets allow you to create panels to display different information

df.facet = df %>% select(subnum, group, block, flank, correctRT) %>% 
  group_by(subnum, group, block, flank) %>% 
  summarise(mdv = mean(correctRT, na.rm = T)) %>% 
  ungroup() %>% group_by(group, flank, block) %>% 
  summarise(N = n(), mean_correctRT = mean(mdv, na.rm = T), 
            sd = sd(mdv, na.rm = T), serr = sd/sqrt(N)) 

df.facet %>% ggplot(aes(flank, mean_correctRT, group = group)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "dodge", width = 0.8) + 
  geom_errorbar(aes(ymin = mean_correctRT - serr, ymax = mean_correctRT + serr), 
                width = 0.8, position = "dodge") +
  facet_wrap(~block)

df.facet.subj = df %>% select(subnum, group, flank, correctRT) %>% 
  group_by(subnum, group, flank) %>% 
  summarise(N = n(), mean_correctRT = mean(correctRT, na.rm = T), 
            sd = sd(correctRT, na.rm = T), serr = sd/sqrt(N)) 

head(df.facet.subj, 3)

## # A tibble: 3 x 7
## # Groups:   subnum, group [1]
##   subnum  group       flank     N mean_correctRT       sd     serr
##    <int> <fctr>      <fctr> <int>          <dbl>    <dbl>    <dbl>
## 1      1  Young   Congruent    96       380.1302 48.80336 4.980972
## 2      1  Young Incongruent    96       456.1841 58.80482 6.001742
## 3      1  Young     Neutral    96       378.9418 64.77877 6.611455

df.facet.subj %>% ggplot(aes(flank, mean_correctRT, group = group)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "dodge", width = 0.8) + 
  geom_errorbar(aes(ymin = mean_correctRT - serr, ymax = mean_correctRT + serr), 
                width = 0.8, position = "dodge") +
  facet_wrap(~subnum)

df.facet %>% ggplot(aes(flank, mean_correctRT, group = group)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "dodge", width = 0.8) + 
  geom_errorbar(aes(ymin = mean_correctRT - serr, ymax = mean_correctRT + serr), 
                width = 0.8, position = "dodge") +
  facet_wrap(~block) + 
  scale_y_continuous(name = "Mean Correct RT") + scale_x_discrete(name = "Flank Position")

df.facet %>% ggplot(aes(flank, mean_correctRT, group = group)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "dodge", width = 0.8) + 
  geom_errorbar(aes(ymin = mean_correctRT - serr, ymax = mean_correctRT + serr), 
                width = 0.8, position = "dodge") +
  facet_wrap(~block) + 
  ylab("Mean Correct RT") + xlab("Flank Position")

library(wesanderson)

df.facet %>% ggplot(aes(flank, mean_correctRT, group = group)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "dodge", width = 0.8) + 
  geom_errorbar(aes(ymin = mean_correctRT - serr, ymax = mean_correctRT + serr), 
                width = 0.8, position = "dodge") +
  facet_wrap(~block, labeller = labeller(block = block.names)) + 
  scale_y_continuous(name = "Mean Correct RT") +
  ggtitle("Correct RT for Young and Old Adults") +
  theme_bw() + theme(plot.title = element_text(hjust = 0.5, size = 20),
                     strip.text.x = element_text(size =14)) +
  scale_fill_manual(name = "Age Group", values=wes_palette(n=2, name="GrandBudapest"))

final_plot = df.facet %>% ggplot(aes(flank, mean_correctRT, group = group)) + 
  geom_bar(aes(fill = group), stat = "identity", position = "dodge", width = 0.8) + 
  geom_errorbar(aes(ymin = mean_correctRT - serr, ymax = mean_correctRT + serr), 
                width = 0.8, position = "dodge") +
  facet_wrap(~block, labeller = labeller(block = block.names)) + 
  scale_y_continuous(name = "Mean Correct RT") +
  ggtitle("Correct RT for Young and Old Adults") +
  theme_bw() + theme(plot.title = element_text(hjust = 0.5, size = 20),
                     strip.text.x = element_text(size =14)) +
  scale_fill_manual(values=wes_palette(n=2, name="GrandBudapest"))

ggsave("output/final_plot.png", final_plot)

## Saving 8 x 4.5 in image

• http://www.storybench.org/getting-started-data-visualization-r-using-ggplot2/
• https://rpubs.com/hadley/ggplot2-layers
• https://rstudio-pubs-static.s3.amazonaws.com/85508_2affd4b9027d4c97b601d338bbf419d7.html#sec:data
• https://www.rstudio.com/wp-content/uploads/2015/03/ggplot2-cheatsheet.pdf
• http://www.sthda.com/english/wiki/ggplot2-colors-how-to-change-colors-automatically-and-manually
• http://sape.inf.usi.ch/quick-reference/ggplot2/colour