ggplot2 Geom Exercises: 16 Practice Problems with Solutions
Sixteen graded exercises that drill the core ggplot2 geoms used in real R work: points, bars, histograms, boxplots, violins, lines, smooths, tiles, and annotation layers. Every problem ships with runnable starter code and a hidden step-by-step solution.
Run the setup block once before working through any exercise. Variables and packages persist across blocks, so you can use them anywhere on the page.
Section 1. Points and scatter geoms (3 problems)
Exercise 1.1: Plot mpg vs wt with a basic scatter geom
Task: A junior analyst onboarding to the team wants to confirm the well-known inverse relationship between vehicle weight and fuel economy. Using the built-in mtcars dataset, draw a scatter plot of mpg on the y-axis versus wt on the x-axis with geom_point() and save the plot object to ex_1_1.
Expected result:
#> ggplot scatter object
#> mapping: aes(x = wt, y = mpg)
#> layer: geom_point()
#> visible pattern: clear negative slope, 32 points, mpg falls from ~33 at wt=1.5 to ~10 at wt=5.4Difficulty: Beginner
Think about which two visual channels should carry the numeric columns and which layer draws one mark per row.
Map wt and mpg inside aes(), then add a geom_point() layer.
Click to reveal solution
Explanation: aes() maps data columns to visual channels, and geom_point() then draws a layer of points using those mappings. Because both wt and mpg are numeric, ggplot picks a continuous scale for each axis automatically. Writing the aesthetics inside geom_point(aes(...)) instead of at the top level gives the same picture, but the global form makes the mapping reusable when you add more layers like geom_smooth().
Exercise 1.2: Reduce overplotting on a discrete axis with geom_jitter
Task: A fuel-economy analyst plotting cty against the discrete class column of mpg notices that points stack into vertical strips and obscure density inside each class. Switch from geom_point() to geom_jitter() with a modest horizontal width and zero vertical noise. Save the chart to ex_1_2.
Expected result:
#> Jittered scatter: x=class (7 levels: 2seater, compact, midsize, minivan, pickup, subcompact, suv), y=cty
#> Points spread horizontally inside each class strip
#> compact/midsize cluster ~ cty 18-25; pickup/SUV cluster ~ cty 10-17Difficulty: Intermediate
The points stack because the x-axis is categorical, so you need to spread them sideways without disturbing the real y-values.
Swap the point layer for geom_jitter(width = 0.2, height = 0).
Click to reveal solution
Explanation: geom_jitter() adds random noise to point positions so overlapping marks become visible. Setting height = 0 keeps cty values exact and only the x positions wiggle; jittering on a numeric axis would distort the data you came to read. The width parameter is in units of the discrete axis (so 0.2 spans 20% of one category). alpha gives extra transparency when many points still land on top of each other.
Exercise 1.3: Bubble chart by mapping size and color aesthetics on diamonds
Task: A retailer auditing the diamonds table wants a quick visual of how carat, price, and cut interact together. Sample 500 rows for legibility, then plot price against carat mapping size to depth and color to cut. Save the result to ex_1_3.
Expected result:
#> Bubble scatter: x=carat (0.2-3+), y=price (300-19k)
#> point size encodes depth (~55-70), color encodes cut (5 levels)
#> upward fanning pattern from origin, larger fanning at higher caratDifficulty: Intermediate
A bubble chart encodes two extra variables by sizing and coloring the markers, and the rows should be thinned first so they stay legible.
Draw a reproducible sample with set.seed() and slice_sample(n = 500), then map size = depth and color = cut in aes().
Click to reveal solution
Explanation: Mapping numeric columns (depth) to size and categorical columns (cut) to color is the standard "bubble chart" recipe. alpha = 0.6 is critical with 500 overlapping markers because without it the underlying density structure is unreadable. slice_sample() is the modern dplyr replacement for sample_n(); setting a seed with set.seed(1) makes the random draw reproducible across runs and across collaborators.
Section 2. Bar and column geoms (3 problems)
Exercise 2.1: Count diamonds by cut using geom_bar
Task: A pricing manager preparing a Monday status update wants a simple frequency chart showing how many diamonds in the diamonds table fall into each cut category. Use geom_bar() (which counts rows automatically) on the cut column with no manual y aesthetic and save the chart to ex_2_1.
Expected result:
#> Bar chart: x=cut (5 levels: Fair, Good, Very Good, Premium, Ideal)
#> y = count of rows per cut
#> Ideal tallest (~21551), Fair shortest (~1610)Difficulty: Beginner
A frequency chart needs only the category on one axis; the bar layer can tally the rows for you.
Map x = cut in aes() and add geom_bar() with no y aesthetic.
Click to reveal solution
Explanation: geom_bar() defaults to stat = "count", which silently runs a frequency tally on the categorical column you map to x. There is no y aesthetic in the original call because ggplot computes it via the stat. The bar order matches the factor levels of cut; since diamonds$cut is an ordered factor (Fair < Good < ... < Ideal), the bars come out in that natural order without any extra work.
Exercise 2.2: Plot pre-summarised values with geom_col
Task: A fleet analyst already has a summary table of mean mpg per cylinder count and wants ggplot to plot those values literally, not to recount rows. Compute mean mpg grouped by cyl from mtcars with dplyr, then plot the result using geom_col() and save to ex_2_2.
Expected result:
#> Bar chart: x=cyl (factor: 4, 6, 8), y=mean mpg
#> bar heights ~ 26.66, 19.74, 15.10Difficulty: Intermediate
When the heights are already computed you want a bar layer that plots values literally instead of recounting rows.
Summarise mean mpg by cyl with group_by() and summarise(), then plot with geom_col() over factor(cyl).
Click to reveal solution
Explanation: geom_col() is the right choice when y-values are already computed; it uses stat = "identity" so heights map straight to your y column. geom_bar(stat = "identity") does exactly the same thing; geom_col() is shorthand for the common case. Wrapping cyl in factor() prevents ggplot from drawing a continuous x-axis with extra integer ticks for what is conceptually a three-category variable.
Exercise 2.3: Compare stacked, dodged, and filled bar positions
Task: A marketing analyst presenting to leadership wants three side-by-side views of how clarity distributes within each cut level of diamonds. Build three bar charts with geom_bar() using position = "stack", "dodge", and "fill" respectively, then combine them into a named list saved to ex_2_3.
Expected result:
#> Named list of 3 ggplot objects
#> $stack: stacked bars, total per cut, colored segments by clarity
#> $dodge: 8 clarity bars side by side within each cut
#> $fill : 100% filled bars, all reaching 1, comparing clarity shareDifficulty: Advanced
The same bar chart can show absolute totals, side-by-side groups, or proportions depending on how overlapping groups are arranged.
Build three geom_bar() charts using position = "stack", "dodge", and "fill", then collect them with list().
Click to reveal solution
Explanation: The position argument tells geom_bar() how to handle overlapping groups. "stack" (the default for filled bars) sums counts into one bar; "dodge" shifts each group side by side so you compare absolute counts; "fill" rescales every bar to the same height so you compare proportions rather than totals. The fill version is the right pick when group totals vary widely and you only care about composition. position_dodge2() handles uneven group counts more gracefully than plain "dodge".
Section 3. Distribution geoms (4 problems)
Exercise 3.1: Histogram of diamond prices with a sensible binwidth
Task: A pricing intern auditing the catalog wants to see the shape of the price distribution in the diamonds table. Plot a histogram of price with geom_histogram() using binwidth = 500 to control granularity and a white outline so individual bins are visible. Save the chart to ex_3_1.
Expected result:
#> Histogram: x=price ($300-$19000), y=count
#> binwidth=500; ~38 bins
#> strongly right-skewed; tallest bar near price ~$700-1200 with count ~9000+Difficulty: Beginner
To see distribution shape you bucket a numeric column into intervals and count what lands in each.
Use geom_histogram(binwidth = 500) and pass a color for the bin outlines.
Click to reveal solution
Explanation: binwidth is preferred over bins because it carries interpretable units (dollars here, not "number of buckets"). The default of 30 bins triggers a warning ggplot prints because it almost never matches what your data wants. The white color argument outlines each bar so bin boundaries stand out against the fill. Right-skew is typical for prices since there is a floor at $0 but no ceiling on luxury items.
Exercise 3.2: Overlay density curves by cut with alpha blending
Task: A jeweller curious whether the price distribution shape changes across cut quality wants overlapping density curves. Plot geom_density() of price with fill = cut and an alpha of 0.4 so all five curves remain visible. Save the chart to ex_3_2.
Expected result:
#> Overlayed density plot: x=price ($), y=density (smoothed)
#> 5 colored curves keyed by cut; all right-skewed
#> peaks cluster around $700-$1200 with slight variation across cutsDifficulty: Intermediate
Comparing distribution shapes across a category calls for smoothed curves you can see through one another.
Map fill = cut and add geom_density(alpha = 0.4).
Click to reveal solution
Explanation: geom_density() runs a kernel density estimator (default bandwidth via bw.nrd0()) for each level of the fill grouping. The y-axis is a probability density, not a count, so curves are area-comparable regardless of group size. Without alpha, the topmost fill (the highest factor level) hides everything beneath it. For very different group sizes, use aes(y = after_stat(count)) to weight each density by group n so peak heights reflect frequencies, not pure shape.
Exercise 3.3: Reorder factor levels with fct_reorder for cleaner boxplots
Task: A car-magazine reviewer wants a boxplot of mpg::hwy by class ordered from lowest to highest median (not alphabetical, which is the ggplot default). Use forcats::fct_reorder() to reorder class by median hwy, pipe into ggplot(), and add geom_boxplot(). Save the chart to ex_3_3.
Expected result:
#> Boxplot: x=class (reordered low->high by median hwy), y=hwy
#> 7 boxes
#> leftmost: pickup (median ~17); rightmost: compact (median ~28)Difficulty: Intermediate
The default category order is alphabetical, so re-rank it by a summary of the numeric variable to make the chart read in order.
Inside mutate(), apply fct_reorder(class, hwy, .fun = median) before adding geom_boxplot().
Click to reveal solution
Explanation: Alphabetical ordering is the default but rarely what readers need; a sorted boxplot tells the story at a glance. fct_reorder() re-levels the factor by some summary of another variable (median by default; pass .fun = mean for the mean). The reorder happens inside the data frame, not on the plot, so any subsequent layer using class inherits the new ordering. For descending order, use fct_reorder(class, hwy, .desc = TRUE).
Exercise 3.4: Layer violin, narrow boxplot, and mean marker for richer distribution view
Task: A health-economics analyst wants a richer view of iris::Sepal.Length by Species showing both the full distribution shape and the median/IQR summary side by side. Layer geom_violin() with geom_boxplot(width = 0.1) and add a red mean point using stat_summary(). Save the chart to ex_3_4.
Expected result:
#> Combined violin + narrow boxplot per Species, with red mean dot
#> setosa narrow at Sepal.Length ~5.0
#> versicolor wider at ~5.9
#> virginica widest at ~6.6Difficulty: Advanced
A richer distribution view layers the full shape, the IQR summary, and a central-tendency marker on the same axes.
Layer geom_violin(), then geom_boxplot(width = 0.1), then stat_summary(fun = mean, geom = "point").
Click to reveal solution
Explanation: Layering geoms top to bottom is the ggplot way to combine views: violin shows the full distribution, boxplot narrows in on the IQR, and stat_summary() lets you mark any summary statistic without precomputing it. The width = 0.1 shrinks the boxplot inside the violin so both stay legible. If you want a mean line rather than a point, swap geom = "point" for geom = "crossbar" and the same stat draws a horizontal segment at the mean.
Section 4. Lines, areas, and smooths (3 problems)
Exercise 4.1: Time-series line plot of US unemployment rate
Task: An economist preparing a one-pager wants to plot the US unemployment rate over time. Using the economics dataset, compute a rate = unemploy / pop column with dplyr, then pipe into ggplot() and draw geom_line() with date on the x-axis. Save the chart to ex_4_1.
Expected result:
#> Line chart: x=date (1967-2015), y=unemployment rate (0.012-0.045)
#> single continuous dark red line
#> multiple peaks during recession years (~1975, 1982, 1992, 2009)Difficulty: Intermediate
A time series is a connected path, so derive the metric you want and order it along the date axis.
Add a rate column with mutate(), then map x = date and draw geom_line().
Click to reveal solution
Explanation: geom_line() connects points in x-order, which is exactly what you want for time series. linewidth is the modern (ggplot2 3.4+) replacement for size on line geoms; size still works but emits a deprecation warning. If date were a character column instead of a Date class, the line would order alphabetically rather than chronologically (and the chart would be wrong); always cast time-like columns to Date or POSIXct before plotting.
Exercise 4.2: Linear fit with confidence band via geom_smooth
Task: A car reviewer plotting mpg vs wt from mtcars wants to overlay a linear trend line and its 95% confidence band so readers can judge the fit at a glance. Add geom_smooth(method = "lm") on top of geom_point() with se = TRUE (the default) and save the chart to ex_4_2.
Expected result:
#> Scatter + linear fit line with gray confidence ribbon
#> negative slope (~-5.3 mpg per ton of weight)
#> ribbon narrow at center of wt, wider at extremesDifficulty: Intermediate
Overlaying a trend on a scatter means adding a second layer that fits a model and shades its uncertainty.
Add geom_smooth(method = "lm", se = TRUE) on top of geom_point().
Click to reveal solution
Explanation: geom_smooth() fits a model and draws both the predicted line and its uncertainty ribbon in one layer. method = "lm" runs ordinary least squares; the default is "loess" for under 1000 rows and "gam" otherwise. The se = TRUE (default) ribbon is the confidence interval for the mean prediction, not a prediction interval; readers often misread this, so consider adding a caption clarifying which one you are showing.
Exercise 4.3: Build a custom confidence ribbon from a fitted lm
Task: A statistics consultant wants a publication-quality plot of mpg ~ wt with explicit control over the 95% confidence ribbon (rather than the one geom_smooth builds internally). Fit lm(), predict over a grid with interval = "confidence", then layer geom_ribbon() and geom_line() on the scatter. Save to ex_4_3.
Expected result:
#> Lightblue ribbon between lwr/upr + dark blue fit line + scatter points
#> ribbon narrows at the data centroid, widens at the wt extremes
#> ribbon, line, and points all on the same plotDifficulty: Advanced
To control the confidence band yourself, fit the model, predict across a grid of x-values, and draw the band as its own layer.
Use lm() and predict(..., interval = "confidence"), then add geom_ribbon(aes(ymin = lwr, ymax = upr)) and geom_line() with inherit.aes = FALSE.
Click to reveal solution
Explanation: Building the ribbon manually with predict(..., interval = "confidence") gives you control that geom_smooth() hides: you can swap to interval = "prediction" (wider, accounts for residual variance) or transform the predictions before plotting. inherit.aes = FALSE is important because it prevents the ribbon and line layers from inheriting y = mpg from the top-level call; without it ggplot would error since band has no mpg column.
Section 5. 2D density and heatmap geoms (2 problems)
Exercise 5.1: Correlation heatmap with geom_tile
Task: A feature engineer wants to spot collinearity in mtcars quickly before fitting a regression. Compute the correlation matrix of the 11 numeric columns, pivot it to long format with pivot_longer(), then draw a geom_tile() heatmap and use scale_fill_gradient2() so positive and negative correlations are visually distinct. Save to ex_5_1.
Expected result:
#> 11x11 tile grid colored from red (-1) through white (0) to blue (1)
#> diagonal uniformly blue (self-correlation = 1)
#> strong red tile at mpg-wt (~-0.87) and cyl-disp pairsDifficulty: Intermediate
A heatmap needs three columns - two keys and a value - so the wide correlation matrix has to be reshaped to long form first.
Reshape with pivot_longer(), draw geom_tile(), and color it with scale_fill_gradient2(midpoint = 0).
Click to reveal solution
Explanation: A correlation matrix is wide by default (each row a variable, each column a variable), but geom_tile() expects long format with three columns: x, y, and fill. pivot_longer() reshapes the matrix into one row per (var1, var2, corr) triple. scale_fill_gradient2() is the right scale for a signed metric: it sets a midpoint (0 here) so positive and negative correlations get visually distinct hues with white as the neutral.
Exercise 5.2: 2D density heatmap with geom_bin2d on heavy data
Task: A data scientist wants to visualize the joint density of price vs carat across all 53,940 rows of diamonds without rendering an unreadable point cloud. Use geom_bin2d() with bins = 40 and apply the viridis palette via scale_fill_viridis_c() for a perceptually uniform fill. Save to ex_5_2.
Expected result:
#> Rectangular bin heatmap: x=carat (0-5), y=price ($0-$19k)
#> viridis fill: yellow=high count, dark purple=low count
#> dense band along an upward-curving spine from originDifficulty: Advanced
With tens of thousands of rows a point cloud is unreadable, so bin the plane into cells and color each by how many observations fall inside.
Use geom_bin2d(bins = 40) and apply scale_fill_viridis_c().
Click to reveal solution
Explanation: For 53,940 points, geom_point() produces an unreadable smear. geom_bin2d() partitions the x-y plane into a grid, counts observations per cell, and maps count to fill. geom_hex() is an alternative using hexagonal bins (smoother visually) but requires the hexbin package. The viridis palette is perceptually uniform and colorblind-safe, while the default ggplot continuous palette has uneven brightness that distorts perceived density.
Section 6. Annotation geoms (1 problem)
Exercise 6.1: Label outlier points with geom_text restricted to a subset
Task: A motorsport analyst plotting mpg vs hp from mtcars wants car names called out only for the four highest-hp vehicles, not all 32. Use slice_max() to filter, then draw geom_point() on the full data and geom_text() on the filtered subset with vjust = -1 so labels float above their markers. Save to ex_6_1.
Expected result:
#> Scatter of mpg ~ hp with 32 points
#> 4 text labels above the top-4 hp points
#> Labels: Maserati Bora, Ford Pantera L, Camaro Z28, Duster 360Difficulty: Intermediate
Labelling only a few points means feeding one layer a filtered subset while the marker layer keeps the full data.
Filter the top four rows with slice_max(hp, n = 4), then add geom_text(data = ..., aes(label = car), vjust = -1).
Click to reveal solution
Explanation: Passing a filtered data frame to a single layer is the clean way to highlight a subset; geom_point() uses the global data (mtcars), while geom_text() overrides with top_hp. vjust = -1 nudges the label one unit above the y position so the text does not overlap the marker. For overlapping labels, switch to ggrepel::geom_text_repel() which spaces them automatically and draws short leader lines from label to point.
What to do next
You have practiced the core geom families in ggplot2. The natural next steps in the visualization track are:
- ggplot2 Aesthetics Exercises: drill the mapping rules between data and visual channels.
- ggplot2 Facets Exercises: break a plot into small multiples by a grouping variable.
- ggplot2 Themes Exercises: customize the non-data ink (axis text, legend, panel grid, fonts).
- ggplot2 Customization Exercises: final polish for publication, including titles, captions, scales, and color.
r-statistics.co · Verifiable credential · Public URL
This document certifies mastery of
ggplot2 Geoms (16 problems) Mastery
Every certificate has a public verification URL that proves the holder passed the assessment. Anyone with the link can confirm the recipient and date.
199 learners have earned this certificate