Odds, Ends, and Polishing Visualizations

Polishing & Hacking Your Visualizations

Packages

Code

library(RColorBrewer)
library(knitr)
library(kableExtra)
library(plyr)
library(broom)
library(modelr)
library(lme4)
library(broom.mixed)
library(tidyverse)
library(ggdist)
library(patchwork)
library(cowplot)
library(DiagrammeR)
library(wordcloud)
library(tidytext)
library(ggExtra)
library(distributional)
library(gganimate)

Custom Theme:

Code

my_theme <- function(){
  theme_classic() + 
  theme(
    legend.position = "bottom"
    , legend.title = element_text(face = "bold", size = rel(1))
    , legend.text = element_text(face = "italic", size = rel(1))
    , axis.text = element_text(face = "bold", size = rel(1.1), color = "black")
    , axis.title = element_text(face = "bold", size = rel(1.2))
    , plot.title = element_text(face = "bold", size = rel(1.2), hjust = .5)
    , plot.subtitle = element_text(face = "italic", size = rel(1.2), hjust = .5)
    , strip.text = element_text(face = "bold", size = rel(1.1), color = "white")
    , strip.background = element_rect(fill = "black")
    )
}

Diagrams

• In research, we often need to make diagrams all points in our research, from
  • conceptualizing study flow
  • mapping measures
  • mapping verbal models
  • SEM models
  • and more

DiagrammeR

• DiagrammeR is a unique interface because it brings together multiple ways of building diagrams in R and tries ot unite them with consistent syntax
  
• We could spend a whole course, not just part of one class parsing through the DiagrammeR package, so I’m going to make a strong assumption based on my knowledge of your ongoing interests and research:
  • SEM plots
  • network visualizations
  • combinations of both
• Let’s just jump in!

Code

[strict] (graph | digraph) [ID] '{' stmt_list '}'

1. strict basically determines whether we can multiple nodes going into / out of a node
2. We have to tell Graphviz whether want a directed [digraph] or undirected [graph] graph.
3. [ID] is what you want to name your graph object
4. '{' stmt_list '}' is where you specify the nodes and edges the graph (more on this next)

Code

grViz("
digraph ex1 {

  # a 'graph' statement
  graph [overlap = true, fontsize = 10]

  # several 'node' statements
  node [shape = box,
        fontname = Helvetica]
  A; B; C; D; E; F
}"
)

• digraph says we want the graph to be directed
• graph lets us control elements of the graph in the []
  • overlap = true means nodes can overlap
• node means we’re about to specify some nodes (and their properties in [])

Nodes

We can control lots of properties of nodes (either as groups or individually):

• color
• fillcolor
• fontcolor
• alpha
• shape
• style (like linestyle)
• sides

• peripheries
• fixedsize
• height
• width
• distortion
• penwidth

• x
• y
• tooltip
• fontname
• fontsize
• icon

• See documentation for more info!

Edges

But we also want to add edges

Code

grViz("
digraph ex1 {

  # a 'graph' statement
  graph [overlap = true, fontsize = 10]

  # several 'node' statements
  node [shape = box,
        fontname = Helvetica]
  A; B; C; D; E; F
  
  # several 'edge' statements
  A->B B->C C->D D->E E->F
}"
)

• -> indicates directed edges
• -- indicates undirected edges
• A->{B,C} is the same as A->B A->C

Edge properties can be defined like node properties:

• arrowsize
• arrowhead
• arrowtail
• dir
• color
• alpha
• headport

• tailport
• fontname
• fontsize
• fontcolor
• penwidth
• menlin
• tooltip

• See documentation for more information on these!

Example: Big Five

• Let’s do the Big Five because why not?
• But they aren’t orthogonal, so we need to let the factors correlate.

Code

grViz("
digraph b5 {

  # a 'graph' statement
  graph [overlap = true, fontsize = 10]

  # def latent Big Five
  node [shape = circle]
  E; A; C; N; O
  
  # def observed indicators
  node [shape = square]
  e1; e2; e3
  a1; a2; a3
  c1; c2; c3
  n1; n2; n3
  o1; o2; o3
  
  # several 'edge' statements
  E->{e1,e2,e3}
  A->{a1,a2,a3}
  C->{c1,c2,c3}
  N->{n1,n2,n3}
  O->{o1,o2,o3}
}"
)

• But they aren’t orthogonal, so we need to let the factors correlate.
• Mess

Code

grViz("
digraph b5 {

  # a 'graph' statement
  graph [overlap = true, fontsize = 10]

  # def latent Big Five
  node [shape = circle]
  E; A; C; N; O
  
  # def observed indicators
  node [shape = square]
  e1; e2; e3
  a1; a2; a3
  c1; c2; c3
  n1; n2; n3
  o1; o2; o3
  
  # several 'edge' statements
  E->{e1,e2,e3}
  A->{a1,a2,a3}
  C->{c1,c2,c3}
  N->{n1,n2,n3}
  O->{o1,o2,o3}
  
  E->{A,C,N,O} [dir = both]
  A->{C,N,O} [dir = both]
  C->{N,O} [dir = both]
  N->{O} [dir = both]
}"
)

Let’s change the layout to neato because that’s kind of a mess!

Code

grViz("
digraph b5 {

  # a 'graph' statement
  graph [overlap = true, fontsize = 10, layout = neato]

  # def latent Big Five
  node [shape = circle]
  E; A; C; N; O
  
  # def observed indicators
  node [shape = square,
        fixedsize = true,
        width = 0.25]
  e1; e2; e3
  a1; a2; a3
  c1; c2; c3
  n1; n2; n3
  o1; o2; o3
  
  # several 'edge' statements
  E->{e1,e2,e3}
  A->{a1,a2,a3}
  C->{c1,c2,c3}
  N->{n1,n2,n3}
  O->{o1,o2,o3}
  
  E->{A,C,N,O} [dir = both]
  A->{C,N,O} [dir = both]
  C->{N,O} [dir = both]
  N->{O} [dir = both]
}"
)

• That was all very lavaan, wasn’t it?
• Well, sometimes we want to create diagrams using code or pipelines, which isn’t easy or intuitive using the syntax we’ve been using
• So instead, we can create the same visualizations using create_graph() and accompanying functions
• Unfortunately, we don’t have time for that today, but there’s a great tutorial online

Basic Text Visualization

• In some ways, the hardest part of text visualization is getting the text into R.
• Once text is in R, there are lots of great tools for tokenizing, basic sentiment analysis, and more
• We’ll be relying on Tidy Text Analysis in R
• Today, we’ll use some data from an ongoing project of mine that applies NLP to Letters from Jenny (Anonymous, 1942), which were published in the Journal of Abnormal and Social Psychology
  
• The PDF’s have been converted to a .txt file

Code

text_df <- read.table("https://github.com/emoriebeck/psc290-data-viz-2022/raw/main/08-week8-polishing/01-data/part2_pymupdf.txt", sep = "\n") %>%
  setNames("text") %>%
  mutate(line = 1:n()) %>%
  as_tibble() %>%
  mutate(text = str_remove_all(text, "[0-9]"))
text_df$text[1:10]

 [1] "CASE REPORTS"                                 
 [2] "LETTERS FROM JENNY (continued)"               
 [3] "ANONYMOUS"                                    
 [4] " (continued)"                                 
 [5] "N.Y.C. Sunday /"                              
 [6] "My dearest Boy and Girl:"                     
 [7] "This is not a regular letter, but even if it" 
 [8] "were I could never begin to express my"       
 [9] "gratitude to you. I believe that when two"    
[10] "persons really love each other in the highest"

Tokens

• The first step with text data is to clean and tokenize it.
• Cleaning basically means makoing sure that everything parsed correctly
  
• Tokenizing means that we break the text down into tokens that we can then analyze
• We tokenize for lots of reasons. It let’s us:
  • Remove filler words
  • Group words in different forms, tenses
  • Get rid of punctuation, etc.
  • And more

A token is a meaningful unit of text, most often a word, that we are interested in using for further analysis, and tokenization is the process of splitting text into tokens (Silge & Robinson, Tidy Text Mining in R)

Code

tidy_text <- text_df %>%
  unnest_tokens(word, text)
tidy_text

Now, let’s remove stop words (articles, etc.) that we don’t want to analyze:

Code

data(stop_words)

tidy_text <- tidy_text %>%
  anti_join(stop_words)

Let’s count the frequency of words:

Code

tidy_text %>%
  count(word, sort = T)

Let’s plot the frequencies of the top 20 words:

Code

tidy_text %>%
  count(word, sort = T) %>%
  top_n(20) %>%
  mutate(word = reorder(word, n)) %>%
  ggplot(aes(n, word)) +
  geom_col() +
  labs(y = NULL) + 
  my_theme()

Sentiments

We can also do some basic sentiment analysis to see how positive or negative word usage was.

For example, we can ask: How negative is Jenny?

Code

tidy_text %>%
  inner_join(get_sentiments("bing")) %>%
  count(sentiment, sort = T)

We can also create an “index” variable to chunk the text. In this case, since the texts are across time, we can get a sense of changes in word usage over time.

Does her negativity change over time?

Code

tidy_text %>%
  inner_join(get_sentiments("bing")) %>%
  count(sentiment, index = line%/%100, sort = T)

We can also plot that:

Code

p <- tidy_text %>%
  inner_join(get_sentiments("bing")) %>%
  count(sentiment, index = line%/%100, sort = T) %>%
  ggplot(aes(x = index, y = n, color = sentiment)) + 
    geom_line() + 
    geom_point() + 
    my_theme()
p

We see a bifurcation later on that may correspond to the death of her son.

Let’s format that a bit more:

Code

p + 
  scale_color_manual(values = c("grey40", "goldenrod")) + 
  scale_x_continuous(limits = c(0,18), breaks = seq(0,15,5)) + 
  annotate("label"
           , label = "negative"
           , y = 32
           , x = 15.5
           , hjust = 0
           , fill = "grey40"
           , color = "white") + 
  annotate("label"
           , label = "positive"
           , y = 13
           , x = 15.5
           , hjust = 0
           , fill = "goldenrod")  +
  labs(x = "Chunk", y = "Count") + 
  theme(legend.position = "none")

We can also look at the most common negative and positive words:

Code

tidy_text %>%
  inner_join(get_sentiments("bing")) %>%
  count(sentiment, word, sort = T) %>%
  group_by(sentiment) %>%
  top_n(10)

and plot those:

Code

p <- tidy_text %>%
  inner_join(get_sentiments("bing")) %>%
  count(sentiment, word, sort = T) %>%
  group_by(sentiment) %>%
  top_n(10) %>%
  ungroup() %>%
  mutate(word = reorder(word, n)) %>%
  ggplot(aes(x = n, y = word, fill = sentiment)) +
  geom_col() +
  labs(y = NULL) + 
  facet_wrap(~sentiment, scales = "free_y") +
  my_theme()
p

Some small aesthetic touches:

Code

p + 
  scale_fill_manual(values = c("grey40", "goldenrod")) + 
  theme(legend.position = "none")

Word Clouds

Word clouds are another way to depict word usage / frequency. Rather than having an axis like our bar graph, it uses relative text size to communicate the same information.

Code

tidy_text %>%
  count(word) %>%
  with(wordcloud(
    word
    , n
    , max.words = 100)
    )

We can also use custom color palettes:

Code

pal <- brewer.pal(6,"Dark2")
tidy_text %>%
  count(word) %>%
  with(wordcloud(
    word
    , n
    , max.words = 100
    , colors = pal)
    )

And split by positive v negative words:

Code

par(mar = c(0, 0, 0, 0), mfrow = c(1,2))
tidy_text %>%
  inner_join(get_sentiments("bing")) %>%
  count(sentiment, word, sort = T) %>%
  filter(sentiment == "negative") %>%
  with(wordcloud(
    word
    , n
    , max.words = 100
    , colors = "grey40")
    )
title("Negative", line = -2)

tidy_text %>%
  inner_join(get_sentiments("bing")) %>%
  count(sentiment, word, sort = T) %>%
  filter(sentiment == "positive") %>%
  with(wordcloud(
    word
    , n
    , max.words = 100
    , colors = "goldenrod")
    )
title("Positive", line = -2)

ggplot2 hacks

Data

• Data cleaning is often the hardest, most time consuming part of our research flow
• Whether we are cleaning raw data, or cleaning data that come out of a model object, we have to be able to wrangle it to the shape we need for whatever program we’re using
  
• Other than lots of tools in your toolbox for reshaping (see Week 1), the biggest data cleaning hack I have has nothing to do with cleaning, per se

Two Key Rules of Data Cleaning:

• Specifically, data cleaning requires two things:
  • You have to know what the output you want is (in our case, plots)
  • You have know how what the data need to look like to produce that

Example: Corrlelograms and Heat Maps

• Let’s consider an example, going back to when we wanted to make correlelograms / heat maps.
• Here’s the plot we wanted to create:

Code

load(url("https://github.com/emoriebeck/psc290-data-viz-2022/blob/main/04-week4-associations/04-data/week4-data.RData?raw=true"))

r_data <- pred_data %>%
  select(study, p_value, age, gender, SRhealth, smokes, exercise, BMI, education, parEdu, mortality = o_value) %>%
  mutate_if(is.factor, ~as.numeric(as.character(.))) %>%
  group_by(study) %>%
  nest() %>%
  ungroup() %>%
  mutate(r = map(data, ~cor(., use = "pairwise")))

r_reshape_fun <- function(r){
  coln <- colnames(r)
  # remove lower tri and diagonal
  r[lower.tri(r, diag = T)] <- NA
  r %>% data.frame() %>%
    rownames_to_column("V1") %>%
    pivot_longer(
      cols = -V1
      , values_to = "r"
      , names_to = "V2"
    ) %>%
    mutate_at(vars(V1, V2), ~factor(., coln))
}

r_data <- r_data %>%
  mutate(r_long = map(r, r_reshape_fun))

hmp <- r_data$r_long[[1]] %>%
  ggplot(aes(x = V1, y = V2, fill = r)) + 
  geom_raster() + 
  geom_text(aes(label = round(r, 2))) + 
  scale_fill_gradient2(limits = c(-1,1)
    , breaks = c(-1, -.5, 0, .5, 1)
    , low = "blue", high = "red"
    , mid = "white", na.value = "white") + 
  labs(
    x = NULL
    , y = NULL
    , fill = "Zero-Order Correlation"
    , title = "Zero-Order Correlations Among Variables"
    , subtitle = "Sample 1"
    ) + 
  theme_classic() + 
  theme(
    legend.position = "bottom"
    , axis.text = element_text(face = "bold")
    , axis.text.x = element_text(angle = 45, hjust = 1)
    , plot.title = element_text(face = "bold", hjust = .5)
    , plot.subtitle = element_text(face = "italic", hjust = .5)
    , panel.background = element_rect(color = "black", size = 1)
  )

• This seems like it should be straightforward because we’re taking a correlation matrix and… visualizing it as a matrix
  
• But ggplot2 doesn’t communicate with correlation matrices because they are in wide format
• So we need to figure out how to make the correlation matrix long format in ways that gives us:
  • Variables on the x-axis
  • Variables on the y-axis
  • Correlations for fill
  • Correlations (rounded) for text
  • no double dipping on values

Code

hmp

• If you remember nothing else from this course, please remember this:
  • AESTHETIC MAPPINGS CORRESPOND TO COLUMNS IN THE DATA FRAME YOU ARE PLOTTING
• So if want all of the above we need the following columns:
  • V1 (x)
  • V2 (y)
  • r (fill, text)
• But what do we currently have?
  • A p*p correlation matrix
  • ggplot2 wants a data frame
• Where are the variable labels (our eventual V1 [x] and V2 [y])?
  • Column names (colnames()) and row names (rownames())
• Where are our correlations?
  • In wide format (unindexed by explicit columns)

Code

r_data$r[[1]]

               p_value          age       gender    SRhealth       smokes
p_value    1.000000000 -0.005224085  0.053627861  0.15917525 -0.069013463
age       -0.005224085  1.000000000 -0.057243245 -0.22438335 -0.078788619
gender     0.053627861 -0.057243245  1.000000000 -0.03182278  0.022275557
SRhealth   0.159175251 -0.224383351 -0.031822781  1.00000000 -0.129241536
smokes    -0.069013463 -0.078788619  0.022275557 -0.12924154  1.000000000
exercise   0.048576025 -0.361768736  0.061659017  0.34546038 -0.155018841
BMI       -0.019741798  0.036151816  0.012217132 -0.09340105 -0.037713371
education  0.001465775 -0.173399716 -0.001603648  0.11008540 -0.096936630
parEdu     0.019871078 -0.374733606  0.055468171  0.08273023  0.005215303
mortality -0.089637524  0.627069166 -0.092109448 -0.31142292  0.035759332
             exercise         BMI    education       parEdu   mortality
p_value    0.04857602 -0.01974180  0.001465775  0.019871078 -0.08963752
age       -0.36176874  0.03615182 -0.173399716 -0.374733606  0.62706917
gender     0.06165902  0.01221713 -0.001603648  0.055468171 -0.09210945
SRhealth   0.34546038 -0.09340105  0.110085399  0.082730234 -0.31142292
smokes    -0.15501884 -0.03771337 -0.096936630  0.005215303  0.03575933
exercise   1.00000000 -0.06217297  0.210204022  0.176766791 -0.32138385
BMI       -0.06217297  1.00000000 -0.048914825 -0.075000576  0.01643219
education  0.21020402 -0.04891483  1.000000000  0.232321970 -0.17215791
parEdu     0.17676679 -0.07500058  0.232321970  1.000000000 -0.18796244
mortality -0.32138385  0.01643219 -0.172157913 -0.187962436  1.00000000

• As a reminder, here’s our criteria for what we want our data to look like to plot:

  • V1 (x)
  • V2 (y)
  • r (fill, text)
  • no double dipping on values
  • Must be a data frame

• But these aren’t in the right order

• It should be these steps:

  • no double dipping on values
  • Must be a data frame
  • V1 (x)
  • V2 (y); r (fill, text)

• Last but, BUT we have also been learning lots about ggplot2 default behavior, and one of those things is that it will treat columns of class() character as something that should be ordered alphabetically via scale_[map]_discrete()

  • If we don’t want it to, we need to make it a factor with levels and/or labels we provide
  • For a heat map / correlelogram, it is imperative that this order is the same order you gave cor() with the raw data.

• You can see that order by looking at the row and column names:

Code

r_data$r[[1]]

               p_value          age       gender    SRhealth       smokes
p_value    1.000000000 -0.005224085  0.053627861  0.15917525 -0.069013463
age       -0.005224085  1.000000000 -0.057243245 -0.22438335 -0.078788619
gender     0.053627861 -0.057243245  1.000000000 -0.03182278  0.022275557
SRhealth   0.159175251 -0.224383351 -0.031822781  1.00000000 -0.129241536
smokes    -0.069013463 -0.078788619  0.022275557 -0.12924154  1.000000000
exercise   0.048576025 -0.361768736  0.061659017  0.34546038 -0.155018841
BMI       -0.019741798  0.036151816  0.012217132 -0.09340105 -0.037713371
education  0.001465775 -0.173399716 -0.001603648  0.11008540 -0.096936630
parEdu     0.019871078 -0.374733606  0.055468171  0.08273023  0.005215303
mortality -0.089637524  0.627069166 -0.092109448 -0.31142292  0.035759332
             exercise         BMI    education       parEdu   mortality
p_value    0.04857602 -0.01974180  0.001465775  0.019871078 -0.08963752
age       -0.36176874  0.03615182 -0.173399716 -0.374733606  0.62706917
gender     0.06165902  0.01221713 -0.001603648  0.055468171 -0.09210945
SRhealth   0.34546038 -0.09340105  0.110085399  0.082730234 -0.31142292
smokes    -0.15501884 -0.03771337 -0.096936630  0.005215303  0.03575933
exercise   1.00000000 -0.06217297  0.210204022  0.176766791 -0.32138385
BMI       -0.06217297  1.00000000 -0.048914825 -0.075000576  0.01643219
education  0.21020402 -0.04891483  1.000000000  0.232321970 -0.17215791
parEdu     0.17676679 -0.07500058  0.232321970  1.000000000 -0.18796244
mortality -0.32138385  0.01643219 -0.172157913 -0.187962436  1.00000000

Get variable order from correlation matrix

Code

r <- r_data$r[[1]]
coln <- colnames(r)
coln

 [1] "p_value"   "age"       "gender"    "SRhealth"  "smokes"    "exercise" 
 [7] "BMI"       "education" "parEdu"    "mortality"

No double dipping on values

Code

r <- r_data$r[[1]]
coln <- colnames(r)
r[lower.tri(r, diag = T)] <- NA
r

          p_value          age      gender    SRhealth      smokes    exercise
p_value        NA -0.005224085  0.05362786  0.15917525 -0.06901346  0.04857602
age            NA           NA -0.05724324 -0.22438335 -0.07878862 -0.36176874
gender         NA           NA          NA -0.03182278  0.02227556  0.06165902
SRhealth       NA           NA          NA          NA -0.12924154  0.34546038
smokes         NA           NA          NA          NA          NA -0.15501884
exercise       NA           NA          NA          NA          NA          NA
BMI            NA           NA          NA          NA          NA          NA
education      NA           NA          NA          NA          NA          NA
parEdu         NA           NA          NA          NA          NA          NA
mortality      NA           NA          NA          NA          NA          NA
                  BMI    education       parEdu   mortality
p_value   -0.01974180  0.001465775  0.019871078 -0.08963752
age        0.03615182 -0.173399716 -0.374733606  0.62706917
gender     0.01221713 -0.001603648  0.055468171 -0.09210945
SRhealth  -0.09340105  0.110085399  0.082730234 -0.31142292
smokes    -0.03771337 -0.096936630  0.005215303  0.03575933
exercise  -0.06217297  0.210204022  0.176766791 -0.32138385
BMI                NA -0.048914825 -0.075000576  0.01643219
education          NA           NA  0.232321970 -0.17215791
parEdu             NA           NA           NA -0.18796244
mortality          NA           NA           NA          NA

Must be a data frame

Code

r <- r_data$r[[1]]
coln <- colnames(r)
r[lower.tri(r, diag = T)] <- NA
r %>% data.frame()

V1 (x)

Code

r <- r_data$r[[1]]
coln <- colnames(r)
r[lower.tri(r, diag = T)] <- NA
r %>% data.frame() %>%
  rownames_to_column("V1")

V2 (y); r (fill, text)

Code

r <- r_data$r[[1]]
coln <- colnames(r)
r[lower.tri(r, diag = T)] <- NA
r %>% data.frame() %>%
  rownames_to_column("V1") %>%
  pivot_longer(
    cols = -V1
    , values_to = "r"
    , names_to = "V2"
  )

Preserve variable order through factors

Code

r <- r_data$r[[1]]
coln <- colnames(r)
r[lower.tri(r, diag = T)] <- NA
r %>% data.frame() %>%
  rownames_to_column("V1") %>%
  pivot_longer(
    cols = -V1
    , values_to = "r"
    , names_to = "V2"
  ) %>%
  mutate(V1 = factor(V1, levels = rev(coln))
         , V2 = factor(V2, levels = coln))

Final Words

• Data cleaning is anxiety-provoking for lots of really valid reasons
  
• You probably outline your writing, so why not outline your data cleaning? It’s writing, too
• Start by figuring out three things:
  • What do you data look like now
  • What’s your final product (table, visualization, etc.)
  • What do your data need to look like to be able to feed into that final product?
• Then, start filling out the middle:
  • How you do get to that end point?
• Don’t be afraid to use cheat sheets!
  • tidyr
  • dplyr
  • plyr
  • purrr
• And also don’t be afraid to ask questions!

Axes

Axes: Bar Charts

• Remember when we talked about bar charts?
• When we measure things, we are careful about scales, wording, etc.
• But when we plot our measures, we sometimes fail to give it the same thoughtfulness
• Our axes should be representative of our measures!

Code

load(url("https://github.com/emoriebeck/psc290-data-viz-2022/raw/main/05-week5-time-series/01-data/ipcs_data.RData"))

First, let’s wrangle the data to long form:

Code

ipcs_long <- ipcs_data %>%
  filter(SID == "02") %>%
  select(SID:purposeful) %>%
  pivot_longer(
    cols = c(-SID, -Full_Date)
    , values_to = "value"
    , names_to = "var"
    , values_drop_na = T
  ) %>%
  mutate(valence = ifelse(var %in% c("afraid", "angry", "guilty"), "Negative", "Positive"))
ipcs_long

Now let’s get the means and SD’s and plot them:

Code

ipcs_long %>%
  group_by(var, valence) %>%
  summarize_at(vars(value), lst(mean, sd)) %>%
  ungroup() %>%
  ggplot(aes(x = var, y = mean, fill = valence)) + 
    geom_bar(
      stat = "identity"
      , position = "dodge"
      ) + 
    geom_errorbar(
      aes(ymin = mean - sd, ymax = mean + sd)
      , width = .1
      ) +
    facet_grid(~valence, scales = "free_x", space = "free_x") + 
    my_theme()

• But our scale is 1-5, so it doesn’t make much sense to have 0 as the bottom of our y-axis
  
• But ggplot2 won’t just let us change the scale minumum, so we have to hack it to allow us to to be able to show the first point scale
• To do this, we simply have to subtract 1 from the means, which will effectively make the scale 0-4
• Then, we can “undo” this by changign the y-axis labels

Code

ipcs_long %>%
  group_by(var, valence) %>%
  summarize_at(vars(value), lst(mean, sd)) %>%
  ungroup() %>%
  ggplot(aes(x = var, y = mean - 1, fill = valence)) + 
    geom_bar(
      stat = "identity"
      , position = "dodge"
      ) + 
    geom_errorbar(
      aes(ymin = mean - 1 - sd, ymax = mean - 1 + sd)
      , width = .1
      ) +
    scale_y_continuous(limits = c(0,4), breaks = seq(0,4,1), labels = 1:5) + 
    facet_grid(~valence, scales = "free_x", space = "free_x") + 
    my_theme()

Let’s add the raw data in, too!

Code

p <- ipcs_long %>%
  group_by(var, valence) %>%
  summarize_at(vars(value), lst(mean, sd)) %>%
  ungroup() %>%
  ggplot(aes(x = var, y = mean - 1, fill = valence)) + 
    geom_bar(
      stat = "identity"
      , position = "dodge"
      ) + 
    geom_jitter(
      data = ipcs_long
      , aes(y = value - 1, fill = valence)
      , color = "black"
      , shape = 21
      , alpha = .5
      , width = .2
      , height = .1
    ) + 
    geom_errorbar(
      aes(ymin = mean - 1 - sd, ymax = mean - 1 + sd)
      , width = .1
      ) +
    scale_y_continuous(limits = c(-.1,4), breaks = seq(0,4,1), labels = 1:5) +
    facet_grid(~valence, scales = "free_x", space = "free_x") + 
    my_theme()
p

And do soem small aesthetic touches.

Code

p + 
  labs(
    x = NULL
    , y = "Mean Rating (1-5) + SD"
  ) + 
  theme(
    legend.position = "none"
    , axis.text.x = element_text(angle = 45, hjust = 1)
    )

Axes: Another Example

• Here’s a plot I was making for a grant last week, demonstrating different mean-level patterns of a behavior across situations from 1 to n. 
• Note the … in the axis, which is normal notation to indicate some unknown quantity.
• How would we create this?

Here’s the data:

Code

tibble(
  p = as.character(rep(1, 4))
  , x = paste0("S", c(1,2,3,"p"))
  , y = c(1, 2, 4, 3)
  ) 

Let’s add the core ggplot code:

Code

tibble(
  p = as.character(rep(1, 4))
  , x = paste0("S", c(1,2,3,"p"))
  , y = c(1, 2, 4, 3)
  ) %>%
  ggplot(aes(x = x, y = y, group = p))

And our geoms, labs, and theme:

Code

tibble(
  p = as.character(rep(1, 4))
  , x = paste0("S", c(1,2,3,"p"))
  , y = c(1, 2, 4, 3)
  ) %>%
  ggplot(aes(x = x, y = y, group = p)) + 
  geom_line(size = 1, color = "#8cdbbe") + 
    geom_point(size = 2.5, color = "black", shape = "square") + 
    labs(x = "Situation", y = "Mean Response", title = "Intraindividual Variability", subtitle = "Person 1") + 
    my_theme()

But how do we add the …?

Let’s switch to a continuous scale, then we can use labels to add it!

Code

tibble(
  p = as.character(rep(1, 4))
  , x = paste0("S", c(1,2,3,"p"))
  , x2 = 1:4
  , y = c(1, 2, 4, 3)
  ) %>%
  ggplot(aes(x = x2, y = y, group = p)) + 
  geom_line(size = 1, color = "#8cdbbe") + 
    geom_point(size = 2.5, color = "black", shape = "square") + 
    labs(x = "Situation", y = "Mean Response", title = "Intraindividual Variability", subtitle = "Person 1") + 
    my_theme()

Now that our scale is continuous, we can use scale_x_continuous() to set breaks and labels where we want them and saying what we want:

Code

tibble(
  p = as.character(rep(1, 4))
  , x = paste0("S", c(1,2,3,"p"))
  , x2 = 1:4
  , y = c(1, 2, 4, 3)
  ) %>%
  ggplot(aes(x = x2, y = y, group = p)) + 
  geom_line(size = 1, color = "#8cdbbe") + 
    geom_point(size = 2.5, color = "black", shape = "square") + 
    scale_x_continuous(
      limits = c(.9, 4.1)
      , breaks = c(1,2,3,3.5,4)
      , labels = c("S1", "S2", "S3", "...", "S4")
      ) + 
    labs(x = "Situation", y = "Mean Response", title = "Intraindividual Variability", subtitle = "Person 1") + 
    my_theme()

Almost there, but we don’t want the tick mark at “…”

We can actually supply a vector of length breaks to axis.ticks.x specifying the size of the ticks!

Code

tibble(
  p = as.character(rep(1, 4))
  , x = paste0("S", c(1,2,3,"p"))
  , x2 = 1:4
  , y = c(1, 2, 4, 3)
  ) %>%
  ggplot(aes(x = x2, y = y, group = p)) + 
  geom_line(size = 1, color = "#8cdbbe") + 
    geom_point(size = 2.5, color = "black", shape = "square") + 
    scale_x_continuous(
      limits = c(.9, 4.1)
      , breaks = c(1,2,3,3.5,4)
      , labels = c("S1", "S2", "S3", "...", "Sn")
      ) + 
    labs(x = "Situation", y = "Mean Response", title = "Intraindividual Variability", subtitle = "Person 1") + 
    my_theme() + 
    theme(axis.ticks.x = element_line(color = c(rep(.5, 3), 0, .5)))

Scales

• coord_cartesian(): the default and what you’ll use most of the time
• coord_polar(): remember Trig and Calculus?
• coord_quickmap(): sets you up to plot maps
• coord_trans(): apply transformations to coordinate plane
• coord_flip(): flip x and y

coord_polar()

Here’s some data we’ll use

Code

ipcs_m <- ipcs_data %>% 
  filter(SID %in% c(216, 211, 174)) %>%
  select(SID, Full_Date, afraid:purposeful, Adversity:Sociability)
ipcs_m

• Let’s:
  • Grab the variable names (vars)
  • Make the data long (pivot_longer())
  • get means and sd’s for the participant

Code

vars <- colnames(ipcs_m)[c(-1, -2)]
ipcs_m <- ipcs_m %>%
  pivot_longer(
    cols = c(-SID, -Full_Date)
    , values_to = "value"
    , names_to = "var"
    , values_drop_na = T
  ) %>%
  group_by(SID, var) %>%
  summarize(m = mean(value)
         , sd = sd(value)) %>%
  ungroup()
ipcs_m

Let’s use the vars vector to create a data frame that also gives each variable:

• A category label
• A integer value

Then, we can use the integer value as the x aesthetic mapping, which will let us “hack” that axis later:

Code

vars <- tibble(
  var = vars
  , cat = c(rep("Emotion", 10), rep("Situation", 8))
  , num = 1:length(vars)
)

ipcs_m <- ipcs_m %>%
  left_join(vars %>% rename(var2 = num)) 

p <- ipcs_m %>%
  ggplot(aes(x = var2, y = m, fill = cat)) + 
    geom_bar(stat = "identity", position = "dodge") +
    my_theme() +
    facet_wrap(~SID)
p

Let’s change the fill colors:

Code

p <- p + 
  scale_fill_brewer(palette = "Set2")
p

Change the scale to polar:

Code

p <- p + 
  coord_polar()
p

• Now, let’s:
  • set the angle we want the text labels on
  • add the text labels
    
  • change the y scale to add some white space in the middle (kind of like a donut chart)

Code

angle <- 90 - 360 * (ipcs_m$var2-0.5) / nrow(vars)

p <- p + 
  geom_text(
    aes(label = var, y = m + .5)
    , angle = angle
    , hjust = 0
    , size = 3
    , alpha = .6
    ) + 
  scale_y_continuous(limits = c(-2, 6.5))
p

And do some aesthetic stuff:

Code

p <- p + 
  labs(
    fill = "Feature Category"
    , title = "Relative Differences in Intraindividual Means"
    , subtitle = "Across Emotions and Situation Perceptions"
    ) + 
  theme(
    axis.line = element_blank()
    , axis.text = element_blank()
    , axis.ticks = element_blank()
    , axis.title = element_blank()
    , panel.background = element_rect(color = "black", fill = NA, size = 1)
  ) 
p

Points

• You can make points any text character. Here, we’ll change points representing men to “M” and women to “W”

Code

pred_data %>%
  filter(study == "Study1") %>%
  ggplot(aes(x = p_value, y = SRhealth)) + 
    geom_point(aes(shape = gender, color = gender), size = 3, alpha = .75) + 
    scale_shape_manual(values = c("M", "W")) + 
    scale_color_manual(values = c("blue", "red")) + 
    my_theme()

Annotations

• Annotations are a great way to hack because they don’t require data frame input
  • "text"
  • "label"
  • "rect" and here
  • lines
    
  • "segment"
  • "arrows"
    
  • … and more!

Text

• You’ve already seen lots of example of using annotate("text", ...)
• But we can also use annotate("text", label = "mu", parse = T) or annotate("text", label = expression(mu[i]), parse = T) to produce math text in our geoms

Here’s another figure from a grant I’m working on that uses several of the features we’ve been discussing: Specifically, notice the line that has label = "mu", parse = T, which creates the Greek letter on the figure.

Code

set.seed(11)

dist_df = tibble(
  dist = dist_normal(3,0.75),
  dist_name = format(dist)
)

dist_df %>%
  ggplot(aes(y = 1, xdist = dist)) +
  stat_slab(fill = "#8cdbbe") + 
  annotate("point", x = 3, y = 1, size = 3) +
  annotate("text", label = "mu", x = 3, y = .92, parse = T, size = 8) + 
  annotate("text", label = "people", x = 2, y = .95) + 
  annotate("segment", arrow = arrow(type = "closed", length=unit(2, "mm")), size = 1, x = 2.8, xend = 1.2, y = .98, yend = .98) + 
  annotate("text", label = "people", x = 4, y = .95) + 
  annotate("segment", arrow = arrow(type = "closed", length=unit(2, "mm")), size = 1, x = 3.2, xend = 4.8, y = .98, yend = .98) + 
  labs(title = "Between-Person Differences") + 
  theme_void()+ 
  theme(plot.title = element_text(face = "bold", size = rel(1.2), hjust = .5))

But if instead we wanted to emphasize one person’s mean and distribution of their psychological states rather than population-level differences, then we want to show \(\mu_i\), not \(\mu\). If we want it to have a subscript, we can use expression().

Code

dist_df %>%
  ggplot(aes(y = 1, xdist = dist)) +
  stat_dots(quantiles = 300, fill = "#b1c9f2") + 
  stat_slab(fill = NA, color = "cornflowerblue") + 
  annotate("point", x = 3, y = 1, size = 3) +
  annotate("text", label = expression(mu[i]), x = 3, y = .92, parse = T, size = 8) + 
  annotate("segment", arrow = arrow(type = "closed", length=unit(2, "mm")), size = 1, x = 1.2, xend = 2, y = 1.4, yend = 1.16) +
  annotate("text", label = expression(Occasion[i]), x = 1.2, y = 1.425) +
  labs(title = "Within-Person Variability") + 
  theme_void() + 
  theme(plot.title = element_text(face = "bold", size = rel(1.2), hjust = .5)
        , plot.background = element_rect(fill = "white"))

Legends

• There are several ways to control legends:
  • use theme(legend.position = [arg]) to change its position
  • use labs([mappings] = "[titles]") to control legend titles
    
  • use guides() to do about everything else

###theme()

• legend.position takes two kinds of arguments
  • text: "none", "left", "right" (default), "bottom", "top"
  • vector: x and y position (e.g. c(1,1))

Code

hmp + 
  theme(legend.position = "right")

• legend.position takes two kinds of arguments
  • text: "none", "left", "right" (default), "bottom", "top"
  • vector: x and y position (e.g. c(1,1))

Code

hmp + 
  theme(legend.position = c(.8, .35))

labs

• I won’t spend too much time here. We’ve seen this a lot
  
• Say that you set color and fill equal to variable V1
• Unless you specify differently, that will be the axis title
• You can change this using labs(fill = "My Title", color = "My Title)
• But make sure you
  • Set both
  • Make the labels the same or they will not be combined into a single legend

guides()

• theme() lets you control the position of the legend and how it appears
• labs() lets you control its titles
  
• scale_[map]_[type] lets you control limits, breaks, and labels
• guides() lets your control individual legend components

Remember correlelograms? Do we need the size legend?

{r. echo = F} p <- r_data$r_long[[1]] %>% ggplot(aes(x = V1, y = V2, fill = r, size = abs(r))) + geom_point(shape = 21) + scale_fill_gradient2(limits = c(-1,1) , breaks = c(-1, -.5, 0, .5, 1) , low = "blue", high = "red" , mid = "white", na.value = "white") + scale_size_continuous(range = c(3,14)) + labs( x = NULL , y = NULL , fill = "Zero-Order Correlation" , title = "Zero-Order Correlations Among Variables" , subtitle = "Sample 1" ) + theme_classic() + theme( legend.position = "bottom" , axis.text = element_text(face = "bold") , axis.text.x = element_text(angle = 45, hjust = 1) , plot.title = element_text(face = "bold", hjust = .5) , plot.subtitle = element_text(face = "italic", hjust = .5) , panel.background = element_rect(color = "black", size = 1) )

We can use guides() to remove only the size legend:

Code

p + 
  guides(size = "none")

And for fill, we could change its direction and the number of columns

Code

p + 
  guides(
    size = "none"
    , fill = guide_legend(
      direction = "vertical"
      , ncol = 2
      )
    ) + 
  theme(legend.position = c(.7,.3))