# 2020-04 # Today we are going to go over some steps for how to create # graphics using the ggplot2 package # The package is based on the grammar of graphics, # and attempts to set up attractive and easy defaults # while still allowing flexibility to customize graphics # Like plotting with base R, the more things you want to change # the more code you will need to learn # There is a ton of information on graphing with ggplot2 online # In particular, for a lot of the basics see: # http://www.cookbook-r.com/Graphs/ # For examples and descriptions of every function available # see the ggplot2 help page: # http://ggplot2.tidyverse.org/reference/ # Folks answer questions about ggplot2 pretty much # every day online. See and search in: # http://stackoverflow.com/questions/tagged/ggplot2 # Also see the RStudio community # https://community.rstudio.com/ # Active development of ggplot2 is done on its GitHub repository # This is where to report or check for bugs (under "Issues") # https://github.com/tidyverse/ggplot2 # See ggplot2 extensions in other packages # https://exts.ggplot2.tidyverse.org/gallery/ # In the first part of the workshop, we'll learn some of the basic syntax # by making some simple exploratory plots on a built-in dataset # We won't spend time "prettifying" these plots # In the second part of the workshop, we will switch our focus to creating graphics # meant to display information that could be included # in a presentation or manuscript, # which will allow us to explore a small portion of the many # options available to customize plots # First, load the package ggplot2 ---- # The current version of ggplot2 is 3.3.0 # If you don't have the current version, # you will need to install it prior to loading # Check if your version is up to date using packageVersion() packageVersion("ggplot2") # To install: # Either type install.packages("ggplot2") in your Console pane # Or in RStudio go to the Packages pane, # click Install, type ggplot2 and press enter # Installation code should not be included as part of # your script, as you don't need to install every time # you use a package # Once ggplot2 has been installed, load it via library() library(ggplot2) # Part 1 ---- # Basic graphics # In the first part of the workshop we will be creating graphics # with a dataset already loaded in R, called "mtcars" # The help page tells us a bit about the dataset ?mtcars head(mtcars) str(mtcars) # We will be working with the continuous variables mpg, wt, and disp, # and categorical variables cyl and am # R doesn't recognize these latter variables as categorical # since they have numeric categories # Getting start with ggplot() ---- # Let's start by diving right in to the ggplot() function # This function is the first one you'll use whenever # you are building a plot with ggplot2 # Defining the dataset we are using for the plot # is an important part of ggplot2 # This allows us to refer to variables directly # We don't need (and shouldn't use) dollar sign notation # The first argument in ggplot() is the dataset # Just defining the dataset gives a completely blank graph # Note the default grey background ggplot(data = mtcars) # A standard way use ggplot() is to define # the dataset and the axis variables # The x and y position are defined within aes() # "aes" stands for "aesthetics" # which we will be talking a lot more about today # We now get a blank plot with axes based # on the axis variables with the default # grey background and white gridlines ggplot(data = mtcars, aes(x = wt, y = mpg) ) # Scatterplot # We add the desired geometric objects or *geoms* # as "layers" using the plus sign # to make different kinds of plots, # which looks like: ggplot(data = mtcars, aes(x = wt, y = mpg) ) + geom_point() # You can put this code all on one line # but it's standard to organize ggplot code by putting layers # on new lines with indentations for readability # Boxplot # Now let's make a boxplot of mpg for each cylinder type # The x axis for boxplots is categorical, # and we can't forget to define the "cyl" variable # as categorical using factor() # The boxplot geom function is geom_boxplot() ggplot(data = mtcars, aes(x = factor(cyl), y = mpg) ) + geom_boxplot() # Mapping aesthetics ---- # Let's start making things a little more complex by adding # more variables to the graphic as *aesthetics* # "aesthetics" are not only the x and y position, # but also things like color, fill, and shapes or other # attributes that affect the way observations are displayed # While ggplot2 doesn't support 3D, # using additional variables as aesthetics # is one way to display more than two dimensions # Assigning a variable to an aesthetic is often # referred to as "aesthetic mapping" # Aesthetic mapping *always* happens inside aes() # Let's color the points in our scatterplot # by the number of cylinders # I'm going to stop writing out the data argument in ggplot() # because it is always the first argument in the ggplot() function ggplot(mtcars, aes(x = wt, y = mpg, color = factor(cyl) ) ) + geom_point() # One convenience in ggplot2 is that we get legends automatically # when we map variables to aesthetics # Today we'll mostly work with aesthetics with categorical variables, # but you can certainly use some aesthetics with continuous variables # (this is what you do to make heat maps or bubble plots) # For example, we can map the "size" aesthetic to the # continuous variable engine displacement ("disp") # to get a bubble scatterplot ggplot(mtcars, aes(x = wt, y = mpg, size = disp) ) + geom_point() # Adding more layers ---- # We can put more layers on a graphic by adding more geoms # Let's add points on top of the boxplots, # mapping colors to change by cylinder categories ggplot(mtcars, aes(x = factor(cyl), y = mpg, color = factor(cyl) ) ) + geom_boxplot() + geom_point() # Both the boxplots and the points get colored by "cyl" # because we mapped color the "cyl" *globally* # by putting it in the ggplot() layer # *Global* settings effect all layers # The "color" aesthetic on boxplots and histograms and such # will color only the outlines # The "fill" aesthetic is for changing the color of the inside # of the boxplots/bars etc. # Setting aesthetics to constants ---- # We can add arguments to individual geoms, not just inside ggplot() # Let's modify the last plot by coloring the points purple # We will still map number of cylinders to boxplot color # Adding the color argument to geom_point() overrides # the original aesthetic mapping for that layer ggplot(mtcars, aes(x = factor(cyl), y = mpg, color = factor(cyl) ) ) + geom_boxplot() + geom_point(color = "purple") # Notice the color argument in `geom_point` is *not* inside aes(), # and the points do not appear in the legend # This is because we set the point color to a constant rather # than mapping the point color to a variable # Understanding when to include an aesthetic in aes() # and when not to can be confusing when you first start with ggplot2 # Generally, if you are *mapping* an aesthetic to a variable # do it within aes() # If you are *setting* an aesthetic to a constant, do it outside of aes() # You only get automatic legends when mapping inside aes() # Mapping multiple aesthetics in geoms ---- # We can map aesthetics to variables separately for each geom # rather than setting them globally in ggplot() # For example, maybe we want to map the fill color of the boxplots # as number of cylinders # but color the points by the transmission type # Note the use of aes() within the layers as well as ggplot() # x and y are mapped globally for all layers # but fill is mapped only for the boxplots and # color only for the points ggplot(mtcars, aes(x = factor(cyl), y = mpg) ) + geom_boxplot( aes(fill = factor(cyl) ) ) + geom_point( aes(color = factor(am) ) ) # When we map more aesthetics to variables, we add more legends # Edit the dataset to change the graphic ---- # This is a good time to discuss how the dataset and the graphic # go hand-in-hand in the ggplot2 framework # Notice the labels for "am" in the legend are not useful and # we've been writing out factor() every time we use # these categorical variables # because they are quantitative in the dataset # We can control these things by changing the dataset # If we make these changes in the dataset from the start # we also avoid later work making changes to the legends and such # There are many, many times that the question of # "How do I change xxx in ggplot2?" # has the answer "Preprocess your dataset." # Let's make these changes now mtcars$numcyl = factor(mtcars$cyl) mtcars$am = factor(mtcars$am, labels = c("Auto", "Manual") ) str(mtcars) # Now we can recreate the graphic using our new variables ggplot(mtcars, aes(x = numcyl, y = mpg) ) + geom_boxplot( aes(fill = numcyl) ) + geom_point( aes(color = am) ) # Changing the colors for the fill and color in graphics ---- # We won't be working with colors in the second part of the workshop # so I wanted to take a minute to show how to change fill/color values # This involves using the appropriate "scale" function, # in this case for fill or color, # and changing the values for the colors # See http://www.cookbook-r.com/Graphs/Colors_(ggplot2)/ and # http://colorspace.r-forge.r-project.org/articles/hcl_palettes.html # for some good information about colors # as well as some useful color palettes # When we change the colors using scale_() functions, # both the graphic and legend will change # We will define new color or fill using the "values" argument # New values are assigned in the order of the levels of the mapping variable # So in this example the cyl order is 4, 6, 8 # and the am order is Auto, Manual # See the "limits" argument for more info on # changing the order of the factor levels # and how the legend is displayed # Setting names to the colors can control which # color goes to which factor level # You can assign new colors by name or by hexadecimal values ggplot(mtcars, aes(x = numcyl, y = mpg) ) + geom_boxplot( aes(fill = numcyl) ) + geom_point( aes(color = am) ) + scale_fill_manual(values = c("light green", "sky blue", "pink") ) + scale_color_manual(values = c(Manual = "#009E73", Auto = "#E69F00") ) # Note that if you only map a variable to fill and you use # scale_color_manual(), nothing will happen # The lesson here is to use the correct scale for each aesthetic # Dot plot ---- # Instead of using the points geom for a categorical x variabale # we could use a dotplot geom # Dot plots are more or less # a kind of histogram made out of dots # This allows us to see all the points at once # if they have the same/close-to-the-same value, # as points with the same value are plotted next to each other # instead of on top of each other # Here is a dot plot along the y axis, # which is like making a histogram turned on its side # We will set the dot fill color to purple # so we set the aesthetic to a constant outside of aes() ggplot(mtcars, aes(x = numcyl, y = mpg) ) + geom_boxplot( aes(color = numcyl) ) + geom_dotplot(fill = "purple", binaxis = "y") # Histograms and density plots ---- # And speaking of histograms, # we haven't made a histogram or density plot yet # These are common exploratory graphics # With histogram/density plots you # only set a continuous variable for the x axis # as the y axis is counts or densities that ggplot2 # calculates from the dataset for you ggplot(mtcars, aes(x = mpg) ) + geom_histogram() ggplot(mtcars, aes(x = mpg) ) + geom_density() # How about a histogram with a density overlay? ggplot(mtcars, aes(x = mpg) ) + geom_histogram() + geom_density() # That doesn't look right because the histogram is on the count scale # and the density is on the density scale # We can change this by setting the y axis # to one of the computed variables # these geoms create, called count and density, # which we can use via the after_stat() function # Here is the density on the count scale # We are mapping after_stat(count) to the y axis, # so we do this within aes() for the density geom ggplot(mtcars, aes(x = mpg) ) + geom_histogram() + geom_density( aes(y = after_stat(count) ) ) # Map fill color to transmission type globally, # which fills both the histogram and the density overlay ggplot(mtcars, aes(x = mpg, fill = am) ) + geom_histogram() + geom_density( aes(y = after_stat(count) ) ) # Notice that layer order can matter ggplot(mtcars, aes(x = mpg, fill = am) ) + geom_density( aes(y = after_stat(count) ) ) + geom_histogram() # This graphic might be more useful if we make the # fill color of the density curves more transparent # by setting "alpha" to less than 1 (can be between 0-transparent to 1-opaque) # alpha is an aesthetic; here we set alpha to a constant, so it is done outside of aes() ggplot(mtcars, aes(x = mpg, fill = am) ) + geom_histogram() + geom_density( aes(y = after_stat(count) ), alpha = .5) # Bar graphs --- # Bar graphs are closely related to histograms, but involve # counts of categorical variables so will # always have a discrete x axis instead of a continuous x axis ggplot(mtcars, aes(x = numcyl) ) + geom_bar() # If you want to make a bar graph using a y variable # from a dataset instead of counts see geom_col() # There are a just a few more basics to cover before # moving away from basic exploratory graphics # into creating higher-quality graphics # Facets ---- # Now we'll explore what is called "faceting" in ggplot2, # where instead of mapping aesthetics to variables # we graph different levels of a categorical variable # in separate plots/panels # We'll make separate scatterplots for each # level of number of cylinders # Here I'm showing you faceting with facet_wrap() # There is another function called facet_grid() # that is another option for faceting # Both of these functions have many options that we will # barely touch on today ggplot(mtcars, aes(x = wt, y = mpg) ) + geom_point() + facet_wrap(~numcyl) # Notice that each facet has the same x and y limits by default # We can allow either (or both) of these to differ for each facet # by using the "scales" argument with "free", "free_y" or "free_x ggplot(mtcars, aes(x = wt, y = mpg) ) + geom_point() + facet_wrap(~numcyl, scales = "free_x") # Adding layers of summary statistics ---- # ggplot2 also has some built-in functions that allow # us to add basic summary stats to a graphic # For example, we can add the mean weight as red squares # on top of the boxplot/dotplot graphic using stat_summary() # Since x is categorical, we use the "fun" argument # to define the function to summarize with # Things are a little different if x were continuous, # so be aware of that ggplot(mtcars, aes(x = numcyl, y = mpg) ) + geom_boxplot( aes(color = numcyl) ) + geom_dotplot(fill = "purple", binaxis = "y") + stat_summary(fun = mean, geom = "point", size = 5, shape = 22, fill = "red") # We also set three aesthetics to constants, size, shape, and fill # I used shape = 22, which is a square # The cookbook for R lists some of the basic shapes for us, # http://www.cookbook-r.com/Graphs/Shapes_and_line_types/ # I don't have a good algorithm for # changing point sizes as above; it takes # trial and error for me most of the time # Regression lines ---- # We can add regression lines and confidence envelopes # to graphics using geom_smooth() or stat_smooth() # The default line type is a smoothed loess line, so we have to set # the method here to lm for linear regression lines # We also get confidence intervals by default, which may # or may not be appropriate; we'll remove with se = FALSE ggplot(mtcars, aes(x = wt, y = mpg) ) + geom_point() + geom_smooth(method = "lm", se = FALSE) # We can fit a regression line separately by group by # adding aesthetic mapping ggplot(mtcars, aes(x = wt, y = mpg, color = numcyl) ) + geom_point() + geom_smooth(method = "lm", se = FALSE) # Part 2 ---- # Polishing graphics # Let's change gears now # and start thinking about polishing our graphics # to make a final product with ggplot2 # A high-quality graphic takes much more time that exploratory graphics, # as you will spend time fine-tuning the appearance of the graphic # for inclusion in, e.g., a manuscript or presentation # For the first (of 2) high quality graphic, # we'll be working with a dataset I saved and # included below eggs = structure(list(id = c(198L, 199L, 200L, 201L, 202L, 203L, 204L, 205L, 206L, 207L, 208L, 209L, 210L, 211L, 212L, 224L, 225L, 226L, 227L, 228L, 229L, 230L, 231L, 232L, 233L, 234L, 235L, 236L, 237L, 238L), length = c(23.1, 23.5, 24.1, 23.4, 23.2, 22.5, 21.9, 21.9, 25, 24.1, 22.2, 21.1, 22.7, 22, 24.1, 19.8, 22.1, 21.5, 20.9, 22, 21, 22.3, 21, 20.3, 20.9, 22, 20, 20.8, 21.2, 21), width = c(16.4, 16.8, 17.1, 16.4, 16.8, 16.6, 16.1, 16.1, 16.9, 15.9, 16.3, 17.2, 16.1, 17, 17.3, 15, 16, 16.2, 15.7, 16.2, 15.5, 16, 15.9, 15.5, 15.9, 16, 15.7, 15.9, 16, 16), species = structure(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L), .Label = c("Pied Wagtail", "Reed-wren"), class = "factor") ), class = "data.frame", row.names = c(NA, -30L) ) head(eggs) # Take a look at the structure of "eggs" in the Environment pane # These data were analyzed using two two-sample t-tests # on length and width between species # Making graphics of the results of two-sample t-tests is rarely # useful, so we'll make a graphic displaying the observed data # We want to make a graphic that shows the # data for egg lengths and egg widths # by species in separate panels # This should make you think about faceting # However, the dataset isn't set up correctly for this, # as length and width are currently in two separate columns; # i.e., there isn't a variable to facet on # How will we make the graphic we want? # This is a question that comes up a lot on the ggplot2 help forums # and the answer is pretty much always the same # We need to reshape the dataset to take # full advantage of ggplot2 # In this case, we have a wide dataset and want a long dataset # I will use gather() from package tidyr # to reshape the dataset # This will create a column of measurement type with two levels, # length and width, # and a column with the quantitative measurements # Load package tidyr first library(tidyr) # I am not going to talk about this code much, as it has more # to do with data manipulation than graphing # The main point is that we want two columns combined into one column, # We define the name of the new grouping column using "key", # and the name of the new quantitative column using "value" # Then we simply type the names of the columns we want to combine # This type of reshaping involves taking data in a *wide* format # and putting it into a *long* format eggs2 = pivot_longer(eggs, cols = c(length, width), names_to = "type", values_to = "measurement", names_ptypes = list(type = factor() ) ) head(eggs2) # Take a look at the structure in the Environment pane # I'm going to change the style of how I present # the creation of these graphics # so you can see what happens to the graphic # as we add each new layer # I will name the initial graphic "g1", which will only have # the x and y axes defined and a boxplot geom # I will then add layers to "g1", # renaming it as I go until we hit the final graphic # You will see this coding style in some of the help documents, # although I use it only for teaching purposes # At the end, we will put all the code together so you can # see what it looks like all together rather than one line at a time # Here are the initial boxplots # The extra pair of parentheses prints the graphic to the plotting window ( g1 = ggplot(eggs2, aes(x = species, y = measurement) ) + geom_boxplot() ) # We want a separate panel for each type of measurement # (width and length), so let's add faceting g1 + facet_wrap(~type) # Not surprisingly, width and length measurements # cover a different range of values # Let's allow the scale of the y axis to be # different for each panel # so the distributions of the two different variables is clearer ( g1 = g1 + facet_wrap(~type, scales = "free_y") ) # I really want to give a good picture of what these data look like, # which means I want to show the actual data points in the plot # I'm going to add the raw data values as a dot plot g1 + geom_dotplot(binaxis = "y", stackdir = "center") # I think the black color for the dots is too dark # so we'll change the dots to some shade of grey # Color choice usually takes a certain amount of trial and error # I don't show you all the different # colors I tried before finding one I liked # but I mostly worked with a series of greys (grey24-grey74) ( g1 = g1 + geom_dotplot(binaxis = "y", stackdir = "center", color = "grey64", fill = "grey64") ) # Now I'll add the mean measurement of each group as a diamond, # but in a darker color grey ( g1 = g1 + stat_summary(fun = mean, geom = "point", shape = 18, size = 5, color = "grey24") ) # So far this is all a review things we've already done today # Let's start working on the appearance of the graph a little bit more # When we change the overall appearance, # such as the color of the panel background # or the gridlines, we are changing what ggplot2 calls "themes" # First, for printing we would not generally want a grey background, # although it is nice for presentations that have colors # We can change the background manually, # but there are a few pre-built themes in ggplot2 # that I often start with # The one I use the most is called theme black and white (theme_bw) # You might be interested in theme_minimal() or theme_classic(), as well g1 + theme_bw() # There are many themes out there, including a whole # package called ggthemes # The default size of the text in a ggplot graphic tends to # be too small for presentations or publications # I like to make this larger right off that bat within # the theme_*() function I use by changing # the "base_size" to something larger than # the default of 11 ( g1 = g1 + theme_bw(base_size = 16) ) # I think the gridlines on the y axis are visually useful for boxplots, # but the gridlines along the x axis seem like overkill # Let's remove the x axis gridlines # This is the kind of thing we can change within theme() # The help page for theme() lists all the things we can change ?theme # The gridlines we want to remove are the "major" gridlines # To remove items in theme() involves setting items to element_blank() g1 + theme(panel.grid.major.x = element_blank() ) # Changing the appearance of the facet strips # is also done within theme(), # including changing the background color and # changing the appearance and position of the strip text # Here we remove the background color of the strips all together # and move the text to the far left of the strip and # change the text size through element_text() # See the help page for element_text() for more info ( g1 = g1 + theme(panel.grid.major.x = element_blank(), strip.background = element_blank(), strip.text = element_text(hjust = 0, face = "bold", size = 14) ) ) # At this point I realized I had made an error # While I can change a lot of things about the appearance of # the facet labels (such as size, font, etc.), # simple changes to the label names require # a certain amount of effort using the "labeller" argument # I should have changed the names in the original dataset # One of the reason I didn't need to change the tick labels # on the x axis is because I had edited # the species names in the original dataset to make them look nicer # Changing the names in the dataset is easy levels(eggs2$type) # We can change the names by assigning new levels levels(eggs2$type) = c("Egg length", "Egg width") levels(eggs2$type) # So do we have to go back and restart the whole graphic? # Sure, we could, and it wouldn't actually take that much effort, # especially outside of a workshop # where we aren't going through this one layer at a time # However, we can take advantage of the function %+% for this, # which updates the data.frame used to make the graphic # This is especially handy if you want to make a series of graphics # that all look the same but are created with a different datasets # Here we'll recreate the dataset so far using the newly-edited "eggs2" ( g1 = g1 %+% eggs2 ) # The label on the x axis seems redundant, so I'm going to remove it # by setting it to NULL # Let's capitalize and add units to the y axis, as well ( g1 = g1 + labs(x = NULL, y = "Measurement (mm)") ) # This could have been where I stopped, but to make a really complete final graphic # I decided I wanted to add some summary stats directly to the graphic # I will put the mean and standard deviation for each group # as text in each pane # I like graphics that combine text information with graphics # and I wanted to give an example of how to think through this # as well as the kind of effort this will take in terms of coding # The first step I need to do is to calculate the summary stats by group # I do this with group_by() and summarise() from the dplyr package, # naming the data.frame "sumdat" library(dplyr) # Don't forget to round to a reasonable number of digits ( sumdat = eggs2 %>% group_by(type, species) %>% summarise(Mean = round(mean(measurement), 1), SD = round(sd(measurement), 1) ) ) # Let's create the labels to add to the graph # I wanted the mean and sd on separate lines (one on top of another) # "\n" indicates a line break in R, so I add that in the paste0() function # along with everything else I want in the labels (values and units) # The paste0() function pastes values together with no separator # This is done using mutate(), also from dplyr sumdat = mutate(sumdat, label = paste0("Mean = ", Mean, " mm", "\n", "SD = ", SD," mm") ) # The resulting column looks like this sumdat$label # As you can see in the final plot, # I decided to add text under each boxplot # This means the x position of the text will still # be defined by "species" and the facets by "type", # but I need a reasonable value to define the position # of the text along the y-axis # I decided on the minimum value of each facet minus .5, # which took some trial and error on my part # The main advice I have when trying to do this # is to think about the range of the axis as your starting point # and don't be afraid of trying a few things - # you can always change the plot back # Here I calculate the y position for each facet by "type" # using dplyr functions again ( loc = eggs2 %>% group_by(type) %>% summarise(yloc = min(measurement) - .5) ) # To get the y position variable "yloc" into the summary dataset # we can simply use inner_join() from dplyr ( sumdat = inner_join(sumdat, loc, by = "type") ) # This is the first time we will be working with a different dataset # for a geom layer (in this case, geom_text() ) # We just define the new dataset # using the "data" argument within the geom # Unlike in the ggplot() function, we need to write out "data =" # in a geom when defining the dataset # Adding a dataset to a geom will override # the overall dataset for this layer only - # it does not affect any other previous or subsequent layers # We'll need to define the y position for the text in aes() # because it's different than the rest of the graphic # geom_text() requires us to set the labels argument within aes(), # as well. The help pages always indicate if # there is a required aesthetic g1 + geom_text(data = sumdat, aes(label = label, y = yloc) ) # That doesn't work very well to start, # but after adjusting the lineheight, # vertical justification, and text size # things are looking a little better in the current plotting window ( g1 = g1 + geom_text(data = sumdat, aes(label = label, y = yloc), lineheight = .9, vjust = .3, size = 4) ) # This is what I used as my final graphic, # and this is what the code looks like all together # Notice we can still name the graphic object ( g1 = ggplot(eggs2, aes(x = species, y = measurement) ) + geom_boxplot() + facet_wrap(~type, scales = "free_y") + geom_dotplot(binaxis = "y", stackdir = "center", color = "grey64", fill = "grey64") + stat_summary(fun = mean, geom = "point", shape = 18, size = 5, color = "grey24") + theme_bw(base_size = 16) + theme(panel.grid.major.x = element_blank(), strip.background = element_blank(), strip.text = element_text(hjust = 0, face = "bold", size = 14) ) + labs(x = NULL, y = "Measurement (mm)") + geom_text(data = sumdat, aes(label = label, y = yloc), lineheight = .9, vjust = .3, size = 4) ) # I can use ggsave to save this graphic # By default, the last graphic plotted is saved # I can also save a graphic by name, which I'll also show you # Finding the ideal size of the graphic # takes some trial and error for me, # although some manuscripts require very specific sizes, # which would make this easier # The default plot from ggsave will look # like what is in the plotting window in RStudio # so if you save things to a different size # you might find you don't like how things look # You cannot set the size of the plotting window in # RStudio at this time # One option is to use the RStudio Export drop down menu # and preview different sizes # Otherwise you can save a plot and take a look at it # Here are some examples of ggsave() with some different arguments ggsave("final plot 1.pdf", width = 7, height = 7) # setting width and height (default inches) ggsave("final plot 1.png", plot = g1) # using default size # At this point I decided I didn't like the size of the dots # and I thought the tick labels on the x-axis should be bigger # I edited the plot code with a new dotsize for that geom ( g1 = ggplot(eggs2, aes(x = species, y = measurement) ) + geom_boxplot() + facet_wrap(~type, scales = "free_y") + geom_dotplot(binaxis = "y", stackdir = "center", color = "grey64", fill = "grey64", dotsize = .5) + stat_summary(fun = mean, geom = "point", shape = 18, size = 5, color = "grey24") + theme_bw(base_size = 16) + theme(panel.grid.major.x = element_blank(), strip.background = element_blank(), strip.text = element_text(hjust = 0, face = "bold", size = 14) ) + labs(x = NULL, y = "Measurement (mm)") + geom_text(data = sumdat, aes(label = label, y = yloc), lineheight = .9, vjust = .3, size = 4) ) # Notice you can change the dpi for png, jpeg, etc. # The final size is 9 by 8 inches ggsave("final plot 1.pdf", plot = g1, width = 9, height = 8) ggsave("final plot 1.png", dpi = 600, width = 9, height = 8) # We have one final graphic to make ---- # Plotting results from comparisons of all groups with the control # For this graphic, I saved the results of the # estimated differences in means and CI around those differences # for 5 groups vs the control group # from a two-factor linear model # Let's read those results in res = structure(list(Diffmeans = c(-0.27, 0.11, -0.15, -1.27, -1.18 ), Lower.CI = c(-0.63, -0.25, -0.51, -1.63, -1.54), Upper.CI = c(0.09, 0.47, 0.21, -0.91, -0.82), plantdate = structure(c(2L, 3L, 3L, 1L, 1L), .Label = c("Feb25", "Jan2", "Jan28"), class = "factor"), stocktype = structure(c(2L, 2L, 1L, 2L, 1L), .Label = c("bare","cont"), class = "factor") ), class = "data.frame", row.names = c(NA, -5L) ) res # Take a look at the structure # I added two columns to this dataset before I saved this so I knew # which comparison (combination of two factors vs control) each row represented # The control had a plantdate of Jan2 with bare root stock # The default factor levels in R are alphanumeric, # so Feb25 is listed first # We'll want plantdate in order by date in my graphic, # so let's relevel the factor # We'll make the labels look a little nicer while we're at it res$plantdate = factor(res$plantdate, levels = c("Jan2", "Jan28", "Feb25"), labels = c("January 2", "January 28", "February 25") ) # We're working towards an plot with horizontal error bars # This is a fairly complex example because # there is only one stocktype for Jan2 but two for the other dates # Much of the time you will be in a simpler situation, # which means you will have fewer tweaks to do # Let's start by plotting the estimated differences in means as points ggplot(res, aes(x = Diffmeans, y = plantdate) ) + geom_point() # Now we can add error bars to represent # the confidence intervals using geom_errorbar() # We are required to provide the "ends" of the error bars # in geom_errorbar() by providing "xmin" and "xmax" positions in aes() # to make horizontal error bars # We'd use ymin and ymax if making vertical error bars # We use the upper and lower limits of the CI in the dataset ggplot(res, aes(x = Diffmeans, y = plantdate)) + geom_point() + geom_errorbar(aes(xmin = Lower.CI, xmax = Upper.CI)) # That's not exactly what we want because # the two stock types at one date are on top of each other # We'll want to "dodge" whenever a geom overlaps in x space # using the position argument with position_dodge() # To do this, we have to define stocktype as the overlapping "group" # and decide how much to dodge the points # (this may take some trial and error, although .75 or .9 are common) # We can only dodge horizontally, and we can set the width of the dodge # (how much space between groups) ggplot(res, aes(x = Diffmeans, y = plantdate, group = stocktype) ) + geom_point(position = position_dodge(width = .75) ) + geom_errorbar( aes(xmin = Lower.CI, xmax = Upper.CI), position = position_dodge(width = .75) ) # That's better, but still not very pretty # There is a problem with the width of the error bars # This is where the single result on Jan 2 complicates things, # There is only 1 group level instead of 2, # which causes the width of the error bar # to be twice as wide as the others # I can set widths in geom_errorbar() outside the aes() # by defining widths for each comparison, # with the first set as half the width of others # In a simpler situation I would set a single width for every error bar ggplot(res, aes(x = Diffmeans, y = plantdate, group = stocktype) ) + geom_point(position = position_dodge(width = .75) ) + geom_errorbar( aes(xmin = Lower.CI, xmax = Upper.CI), position = position_dodge(width = .75), width = c(.2, .4, .4, .4, .4) ) # That looks much better # However, we can't tell which line is for which stocktype # We can get a legend by mapping "linetype" to stocktype ( g2 = ggplot(res, aes(x = Diffmeans, y = plantdate, group = stocktype) ) + geom_point(position = position_dodge(width = .75) ) + geom_errorbar( aes(xmin = Lower.CI, xmax = Upper.CI, linetype = stocktype), position = position_dodge(width = .75), width = c(.2, .4, .4, .4, .4) ) ) # Since we're ready to start changing the plot appearance, # I'll name the plot and start adding to it # one layer at a time # First, let's change the theme to theme_bw() # with a larger base_size # and make the axis labels look nicer ( g2 = g2 + theme_bw(base_size = 16) + labs(x = "Difference in Growth (cm)", y = "Planting Date") ) # It can be informative to show the area that indicates # a practically unimportant difference # Any difference in mean growth over/under +/-.25 cm # when compared to the control is practically important to growers # We could draw horizontal lines at +/- .25 using geom_hline() # Instead we're going to draw a see-through grey rectangle # over the area between -.25 and .25 # to indicate the "Zone of no difference" # This is done with the rectangle geom "rect" in annotate() # The boundaries of the rectangle are set # with xmax, xmin, ymax, ymin # We'll use Inf/-Inf for the y axis to make the rectangle # all the way across the plot ( g2 = g2 + annotate(geom = "rect", xmin = -.25, xmax = .25, ymin = -Inf, ymax = Inf, fill = "grey54", alpha = .25) ) # If 0.25 is a practical difference, we may want more # "breaks" or tick marks on the y axis so we can see the .25 clearly # We do this with the appropriate scale function, scale_x_continuous() # If we were changing the y axis we would use scale_y_discrete() # because y is a categorical variable in this plot ( g2 = g2 + scale_x_continuous(breaks = seq(-1.5, .5, by = .25) ) ) # The y axis is discrete, and by default # the levels go from bottom to top # I'd like the dates to go from top to bottom, instead, # with January 2 at the top # This can be done with scale_y_discrete() ( g2 = g2 + scale_y_discrete(limits = rev) ) # I don't like the default dotted linetype # and the names of the stock categories should be clearer # We can change how aesthetics appear on the graphic # through the "scale" functions # Essentially every aesthetic has a scale function # (we used fill and color scales earlier) # In this case, we'll use scale_linetype_manual() # to change the linetypes # and linetype names as they appear in the legend # It doesn't seem like we need the legend title # We can suppress the title (or rename it) # in scale_linetype_manual(), as well g2 + scale_linetype_manual(values = c("solid", "twodash"), name = NULL, labels = c("Bare root", "Container") ) # It would also be nice to have the lines in the legend # match the order they appear in the graph (dashed first) # We can do that by reversing the order with the guide argument # and guide_legend() function # There are a ton of things we can control in guide_legend(), much like with theme() ( g2 = g2 + scale_linetype_manual(values = c("solid", "twodash"), name = NULL, labels = c("Bare root", "Container"), guide = guide_legend(reverse = TRUE) ) ) # We have plenty of space to put the legend within the graphic # This is something we can adjust in theme() # The legend position coordinates are essentially # between 0 and 1 in both x and y # within the plotting area # This is yet another time where you might need trial and error to get # a perfect placement # While we're using theme(), let's get rid of the y axis gridlines # and the minor x axis gridlines g2 + theme(legend.position = c(.2, .75), panel.grid.major.y = element_blank(), panel.grid.minor.y = element_blank(), panel.grid.minor.x = element_blank() ) # Then I decided to tweak the space around the legend, # leading me to spend some time figuring out the # legend margins and # legend spacing, https://github.com/tidyverse/ggplot2/issues/3587 ( g2 = g2 + theme(legend.position = c(.2, .75), legend.spacing.y = unit(0, "pt"), legend.margin = margin(t = 4, r = 5, b = 5, l = 5), panel.grid.major.y = element_blank(), panel.grid.minor.y = element_blank(), panel.grid.minor.x = element_blank() ) ) # Finally, let's add a label to indicate that the grey rectangle # is the "zone of no difference" # annotate() is the function to use when adding a single label like this # instead of geom_text() to add many labels from a dataset # Placement is a hard decision here # I decided to put it near the bottom of the rectangle on the left ( g2 = g2 + annotate(geom = "text", x = 0, y = 0.5, label = "Zone of no difference", size = 3) ) # Here's what the ggplot code looks like if we built it all together ( g2 = ggplot(res, aes(x = Diffmeans, y = plantdate, group = stocktype) ) + geom_point(position = position_dodge(width = .75) ) + geom_errorbar( aes(xmin = Lower.CI, xmax = Upper.CI, linetype = stocktype), position = position_dodge(width = .75), width = c(.2, .4, .4, .4, .4) ) + theme_bw(base_size = 16) + labs(x = "Difference in Growth (cm)", y = "Planting Date") + annotate(geom = "rect", xmin = -.25, xmax = .25, ymin = -Inf, ymax = Inf, fill = "grey54", alpha = .25) + scale_x_continuous(breaks = seq(-1.5, .5, by = .25) ) + scale_y_discrete(limits = rev) + scale_linetype_manual(values = c("solid", "twodash"), name = NULL, labels = c("Bare root", "Container"), guide = guide_legend(reverse = TRUE) ) + theme(legend.position = c(.2, .75), legend.spacing.y = unit(0, "pt"), legend.margin = margin(t = 4, r = 5, b = 5, l = 5), panel.grid.major.y = element_blank(), panel.grid.minor.y = element_blank(), panel.grid.minor.x = element_blank() ) + annotate(geom = "text", x = 0, y = 0.5, label = "Zone of no difference", size = 3) ) # This now matches the final graphic, and we could save it as before ggsave("final plot 2.pdf", plot = g2, width = 7, height = 6) # end of workshop ----