Visualisation Intermediate ggplot2 R data visualisation reproducibility R ggplot2 ggplot2 is easiest to learn as a system for constructing graphics rather than a collection of commands for making particular charts. Once the underlying grammar is clear, the same ideas transfer from a simple scatter plot to faceted distributions, model summaries and publication ready figures. The examples use datasets included with ggplot2. Select Run R above any code block to execute it in the browser. The first ggplot2 example may take a little longer because R and the package need to load. The basic grammar Most ggplot2 figures contain at least three components: 1. Data: the data frame containing the observations. 2. Aesthetic mappings: the variables assigned to visual properties such as the x position, y position, colour or shape. 3. A geometric layer: the marks used to display the observations, such as points, lines or bars. The components are joined with + . r library ggplot2 p < ggplot data = mpg, mapping = aes x = displ, y = hwy + geom point print p Here, each row of mpg describes a car. Engine displacement is mapped to the horizontal position and highway fuel economy to the vertical position. geom point represents each row with a point. The shorter form below is equivalent: r library ggplot2 p < ggplot mpg, aes displ, hwy + geom point print p The first form is useful while learning because it makes the role of each argument explicit. The shorter form is convenient once the structure is familiar. Mapping a variable versus setting a value Code placed inside aes creates a mapping from a variable to a visual property. Code placed outside aes sets one appearance for the whole layer. r library ggplot2 Colour is mapped to the class variable p mapped < ggplot mpg, aes displ, hwy, colour = class + geom point size = 2, alpha = 0.75 + labs title = "Colour mapped to vehicle class" Colour is set to one value p set < ggplot mpg, aes displ, hwy + geom point colour = " 214c57", size = 2, alpha = 0.75 + labs title = "One colour set for every point" print p mapped print p set The mapped version needs a legend because colour carries information. The set version does not. Accidentally writing aes colour = "blue" asks ggplot2 to treat the word “blue” as a category; it does not simply colour the points blue. Common aesthetics include: x and y ; colour for lines and point borders; fill for the inside of bars, areas and many shapes; shape , size and alpha ; group for observations that belong to the same line or polygon. Use colour, shape and size deliberately. Mapping too many variables at once usually makes a figure harder to read. Choose a geometry for the question The geometry should reflect both the type of variables and the comparison being made. Question Useful starting geometry How are two quantitative variables related? geom point How does a value change over an ordered variable or time? geom line What is the distribution of one quantitative variable? geom histogram or geom density How does a quantitative variable differ between groups? geom boxplot or geom violin How many observations belong to each category? geom bar How do already calculated values compare? geom col The distinction between geom bar and geom col matters. geom bar counts rows. geom col uses supplied y values. r library ggplot2 p count < ggplot mpg, aes class + geom bar fill = " 214c57" + labs title = "Number of cars in each class", x = NULL, y = "Count" mean hwy < aggregate hwy ~ class, data = mpg, FUN = mean p value < ggplot mean hwy, aes class, hwy + geom col fill = " 6e2538" + labs title = "Mean highway fuel economy", x = NULL, y = "Mean MPG" print p count print p value Build figures in layers Each geometry is a layer. Layers may use the plot's main aesthetic mappings or provide their own. r library ggplot2 p < ggplot mpg, aes displ, hwy + geom point colour = " 214c57", alpha = 0.45 + geom smooth method = "lm", formula = y ~ x, se = TRUE, colour = " 6e2538", fill = " efe2e5" + labs title = "Engine displacement and highway fuel economy", subtitle = "Linear fit with a 95% confidence band", x = "Engine displacement litres ", y = "Highway fuel economy MPG " print p The points show the observations. The line shows the fitted mean relationship, and the shaded band represents uncertainty in that estimated mean. It is not a prediction interval for individual cars. Layer order matters. Later layers are drawn over earlier layers. Put broad intervals, reference regions and large marks first; put the observations or annotations that must remain visible later. Display distributions A histogram divides a quantitative variable into bins and counts observations within each bin. The apparent shape can change when the bin width changes, so choose and report it rather than relying automatically on the default. r library ggplot2 p < ggplot diamonds, aes carat + geom histogram binwidth = 0.1, boundary = 0, fill = " 214c57", colour = "white" + coord cartesian xlim = c 0, 3 + labs title = "Distribution of diamond size", x = "Carat", y = "Count" print p To compare a histogram with a density curve, map the histogram's y axis to a computed density using after stat . r library ggplot2 p < ggplot diamonds, aes carat + geom histogram aes y = after stat density , binwidth = 0.1, boundary = 0, fill = " abc1c4", colour = "white" + geom density colour = " 6e2538", linewidth = 1 + coord cartesian xlim = c 0, 3 + labs title = "Histogram and density estimate", x = "Carat", y = "Density" print p Density curves are smooth summaries, not direct observations. Their appearance depends on the bandwidth, just as a histogram depends on its bin width. Compare groups Box plots are compact summaries of the median, quartiles and potential outliers. They are useful for comparison, but can conceal sample size and distribution shape. Adding jittered observations often provides useful context. r library ggplot2 small mpg < subset mpg, class %in% c "compact", "midsize", "suv" p < ggplot small mpg, aes class, hwy, fill = class + geom boxplot width = 0.58, outlier.shape = NA, alpha = 0.65 + geom jitter width = 0.12, alpha = 0.45, size = 1.4 + scale fill manual values = c compact = " 214c57", midsize = " 6e2538", suv = " abc1c4" + labs title = "Fuel economy across vehicle classes", x = NULL, y = "Highway fuel economy MPG " print p If a grouping variable is mapped to fill or colour , ggplot2 usually infers the groups. Explicit group mappings are particularly important for line plots when colour or linetype does not already define the intended series. Use facets for small multiples Faceting creates a consistent panel for each level of a variable. This is often clearer than placing many groups in one plotting area. r library ggplot2 p < ggplot mpg, aes displ, hwy + geom point colour = " 214c57", alpha = 0.55 + geom smooth method = "lm", formula = y ~ x, se = FALSE, colour = " 6e2538" + facet wrap vars drv , nrow = 1 + labs title = "The relationship differs by drivetrain", x = "Engine displacement litres ", y = "Highway fuel economy MPG " print p facet wrap arranges one faceting variable across rows and columns. facet grid rows ~ columns is useful when two variables define a meaningful grid. Keep common scales when direct comparisons between panels matter. Free scales can reveal within panel patterns, but they make magnitudes harder to compare and should be used only for a clear reason. Control scales Scales translate data values into visual properties and control axes and legends. They are where you change breaks, labels, limits, transformations and colour palettes. r library ggplot2 p < ggplot mpg, aes displ, hwy, colour = class + geom point size = 2, alpha = 0.75 + scale x continuous breaks = seq 2, 7, by = 1 , expand = expansion mult = c 0.02, 0.06 + scale colour viridis d end = 0.88 + labs title = "A discrete, perceptually ordered colour scale", x = "Engine displacement litres ", y = "Highway fuel economy MPG ", colour = "Vehicle class" print p Use a continuous scale for quantitative variables and a discrete scale for categories. A sequential palette suits values progressing from low to high. A diverging palette suits meaningful deviations around a centre. Unrelated categories need distinguishable colours without implying an order. Transformations such as scale x log10 change the visual scale and its axis labels while preserving the original data. Log scales are helpful for multiplicative relationships and strongly right skewed positive variables, but they must be labelled and interpreted accordingly. Zoom without discarding data Scale limits and coordinate limits behave differently. Setting limits on a scale removes observations before statistical layers are calculated. coord cartesian zooms the view after the calculations. r library ggplot2 p < ggplot mpg, aes displ, hwy + geom point colour = " 214c57", alpha = 0.5 + geom smooth method = "lm", formula = y ~ x, colour = " 6e2538" + coord cartesian xlim = c 2, 5 , ylim = c 15, 40 + labs title = "Coordinate limits zoom without refitting the model", x = "Engine displacement litres ", y = "Highway fuel economy MPG " print p Use scale limits when excluded observations genuinely fall outside the data definition. Use coordinate limits when you simply want a closer view. Labels, reference lines and annotations A reader should be able to understand the main comparison without consulting the surrounding text. Use labs to state the subject, units and meaning of mapped aesthetics. r library ggplot2 threshold < 30 p < ggplot mpg, aes displ, hwy + geom point colour = " 214c57", alpha = 0.6 + geom hline yintercept = threshold, linetype = "dashed", colour = " 6e2538" + annotate "text", x = 6.8, y = threshold + 1, label = "30 MPG reference", hjust = 1, colour = " 6e2538" + labs title = "Engine size and fuel economy", x = "Engine displacement litres ", y = "Highway fuel economy MPG " print p Annotations should clarify a comparison rather than decorate the plot. Direct labels are often preferable to a legend when only a few important series need identification. Combine raw data with summaries Different layers can use different datasets. This is useful when raw observations and a calculated summary need to appear together. r library ggplot2 class means < aggregate hwy ~ class, data = mpg, FUN = mean p < ggplot mpg, aes class, hwy + geom jitter width = 0.16, colour = " 214c57", alpha = 0.32 + geom point data = class means, mapping = aes x = class, y = hwy , inherit.aes = FALSE, colour = " 6e2538", size = 3.2 + labs title = "Observations and group means", subtitle = "Large points show the mean within each class", x = NULL, y = "Highway fuel economy MPG " print p inherit.aes = FALSE tells the summary layer not to inherit mappings from the main plot. This becomes important when the secondary dataset has different variable names or represents a different unit of observation. Write reusable plotting functions When the same type of figure is produced repeatedly, place the common structure in a function. The .data ... pronoun allows column names to be supplied as strings. r library ggplot2 scatter plot < function data, x, y, colour = " 214c57" ggplot data, aes x = .data x , y = .data y + geom point colour = colour, alpha = 0.65 + geom smooth method = "lm", formula = y ~ x, se = FALSE, colour = " 6e2538" + labs x = x, y = y p < scatter plot mpg, "displ", "hwy" + labs title = "A plot produced by a reusable function" print p A useful function should encode genuine repetition while leaving the important analytical choices visible. Avoid a single function with dozens of arguments that obscures what the final figure is doing. Export figures reproducibly Use ggsave rather than exporting manually from the plotting pane. State the dimensions and resolution in the script so the figure can be recreated. r library ggplot2 p < ggplot mpg, aes displ, hwy + geom point colour = " 214c57", alpha = 0.6 + labs title = "Engine size and fuel economy", x = "Engine displacement litres ", y = "Highway fuel economy MPG " + theme minimal ggsave filename = "engine fuel economy.png", plot = p, width = 7, height = 5, units = "in", dpi = 300, bg = "white" print p Use a vector format such as PDF or SVG when the journal or document workflow supports it. Use a high resolution PNG when a raster image is required. Check the destination's size, font and file format requirements before export. A final plotting checklist Before treating a figure as finished, ask: Does the geometry match the question and variable types? Is the unit of observation clear? Are the axes labelled with meaningful names and units? Does each colour, shape or size mapping carry useful information? Could faceting or direct labels reduce legend reading? Are uncertainty intervals identified and interpreted correctly? Have scale transformations, exclusions and missing values been made explicit? Is the figure still readable at its final printed or on screen size? My reusable ggplot theme My usual starting point is theme minimal , with bold axis titles, a centred bold plot title, angled category labels when needed and Ubuntu as the preferred font. Applying a 45 degree angle and removing the legend on every plot is too rigid: continuous x axes generally do not need rotated labels, and some figures need a legend. The function below keeps the visual identity while making those choices adjustable. r library ggplot2 theme bodhi < function base size = 11, base family = "Ubuntu", rotate x = FALSE, legend position = "none", title size = 14 x angle < if rotate x 45 else 0 x hjust < if rotate x 1 else 0.5 x vjust < if rotate x 1 else 0.5 theme minimal base size = base size, base family = base family + theme axis.text.x = element text angle = x angle, vjust = x vjust, hjust = x hjust , axis.title = element text face = "bold" , plot.title = element text face = "bold", size = title size, hjust = 0.5 , plot.subtitle = element text hjust = 0.5 , plot.title.position = "plot", legend.position = legend position, legend.box.background = element rect colour = "black", fill = NA , panel.grid.minor = element blank , text = element text family = base family p < ggplot mpg, aes class, hwy, colour = drv + geom jitter width = 0.14, alpha = 0.65 + labs title = "Fuel economy by vehicle class", subtitle = "A categorical axis and a visible legend", x = "Vehicle class", y = "Highway fuel economy MPG ", colour = "Drivetrain" + theme bodhi rotate x = TRUE, legend position = "bottom" print p Use theme bodhi for a plot that does not need rotated labels or a legend. Use theme bodhi rotate x = TRUE for crowded categorical labels, and supply legend position = "right" or "bottom" when colour, fill, shape or linetype needs explanation. Ubuntu is bundled with this website and loaded before the browser executes these examples. When using the theme on another computer, install Ubuntu locally or pass an available family such as base family = "sans" .

Full post