ggplot2 Color Scales Exercises in R: 20 Real-World Practice Problems

Explanation: A named character vector inside values = is the safest pattern because the names match factor levels directly, so the order of assignment is independent of the data order. Compare with an unnamed vector like c("tomato","steelblue","forestgreen"): that maps positionally and silently breaks if levels are reordered. Use named vectors whenever a level can move.

Explanation: Defining the palette as a named vector before the call separates the design decision from the plot code, which makes it easy to reuse the same palette across multiple charts. guides(fill = "none") hides the redundant legend since the x-axis already labels each cut. If a future level is added to cut, scale_fill_manual will throw a missing-value warning unless that level is also added to brand_palette.

Explanation: breaks controls only what appears in the legend and in what order, not the actual category mapping. To also reorder bars on the x-axis you would need to relevel the factor (e.g., mutate(cut = factor(cut, levels = c("Ideal","Premium",...)))) before the call to ggplot. A common bug is to swap colors by reordering values instead of using breaks: that breaks the name-color contract.

Explanation: na.value controls how NA data points render (otherwise they become invisible because the default is NA color, which is dropped). drop = FALSE keeps every declared factor level in the legend even when a level has zero observations in the current data slice, which is critical for dashboards that filter and re-render: without it, legends shrink and grow as filters change, confusing the reader.

Explanation: scale_color_brewer() is designed for discrete data; pairing it with a numeric aesthetic raises a warning and falls back to grey. Set1 has only nine colors maximum, so if a factor has more levels you must switch to Paired, Set3, or interpolate with colorRampPalette(). For a continuous variable, use scale_color_distiller() instead, which uses the same Brewer palettes but interpolates over a range.

Explanation: Dark2 is qualitative, meaning every color is perceptually distinct and carries no ordinal meaning. That makes it correct for nominal variables like month-as-category. If the months were meant to read as an ordered progression (May to September), a sequential palette like Blues via scale_color_brewer(palette = "Blues") would carry that ordering information visually. Pick the palette family that matches the variable's role, not just what looks pretty.

Explanation: scale_fill_brewer() would error here because ..count.. is continuous; scale_fill_distiller() solves that by interpolating the discrete Brewer palette over a continuous range. Spectral is diverging, which can be a poor fit for count data that has no natural center. A sequential palette such as YlOrRd would more honestly encode "more density = darker." Use diverging palettes only when zero or a midpoint matters semantically.

Explanation: colorRampPalette() returns a function; calling that function with (7) produces seven evenly interpolated hex codes between the anchors. This pattern is essential when a categorical variable has more levels than a Brewer palette supports, or when you need a custom corporate gradient. The two anchors here approximate the endpoints of viridis, so you get a hand-rolled version of viridis_d(n = 7).

Explanation: scale_color_viridis_c() is the continuous variant; _d is discrete. Viridis is perceptually uniform (equal data steps look like equal color steps to the eye) and remains distinguishable in greyscale and to viewers with red-green color vision deficiency. That makes it a safe default for any continuous color encoding, especially in scientific or accessibility-sensitive contexts.

Explanation: option = selects the variant: "viridis" (default), "magma", "plasma", "inferno", "cividis", "mako", "rocket", "turbo". direction = -1 flips the ramp without rewriting hex codes. A common style choice: use the dark end for the value the reader should worry about (low fuel economy here), since dark pulls the eye. Reversing keeps the palette properties (perceptual uniformity, colorblind safety) intact.

Explanation: Setting end = 0.85 trims the lightest yellow off the magma ramp, which often improves contrast against a white background. Without this, the highest-value category nearly disappears. Setting the factor levels explicitly preserves the credit-quality order (AAA best, BBB worst) so the color ramp reads as a quality gradient.

Explanation: Color scales obey the same transformation rules as axis scales: trans = "log10", "sqrt", "reverse", or any scales::trans_new(). The square-root transform pulls long-tailed distributions back toward the middle so visual variance reflects real variance instead of being eaten by one or two outlier cells. Always label the legend with the transform applied: a reader expects raw counts unless told otherwise.

Explanation: scale_color_gradient() is the simplest continuous color scale: two endpoints, linear interpolation. White-to-red is fine on a printed page but loses its lowest band against a white panel; for screen plots a pale grey or pale yellow at the low end (low = "#FFF7BC") usually reads better. Reach for gradient2() or gradientn() when one color stop is not enough.

Explanation: gradient2() is the right choice whenever zero (or any natural anchor) divides "good" from "bad." The midpoint argument anchors the mid color to a value; default is 0. Without gradient2, a normal gradient would color the smallest negative value the same shade as the smallest positive value, hiding the sign-flip that risk teams care about most.

Explanation: scale_color_gradientn() is the most flexible continuous scale: pass any number of color stops and ggplot interpolates evenly between them. Use it when two endpoints (gradient) or three (gradient2) aren't enough, e.g., reproducing a brand palette or matching an existing chart in a brand book. For finer control over where each stop lands on the value axis, also pass a values = vector of breakpoints in [0, 1].

Explanation: Without limits = c(-1, 1), the legend would auto-fit to the observed min and max (e.g., -0.3 to 0.9), and "white at midpoint = 0" would slide off-center visually. Locking limits to the variable's theoretical range (correlation is bounded in [-1, 1]) keeps the white band anchored to zero regardless of which subset is plotted. Reuse this trick for any quantity with a known semantic center: log-fold-change at 0, lift at 1, residuals at 0.

Explanation: A single-hue sequential palette is appropriate here because cut is ordinal (Fair worst, Ideal best): the value of being darker carries meaning. If you were comparing brands or unordered campaigns, switch to a qualitative palette like Set1 or Dark2. Named vectors also let the same palette feed scale_color_manual elsewhere in the deck without redefining the colors.

Explanation: direction = 1 reverses the default Brewer interpolation so high ozone is dark and low is pale, which matches the reader's expectation that "darker = more." YlOrRd is sequential and colorblind-safe for the dominant red-green deficiency, unlike Rainbow or RdYlGn which collapse for those viewers. For exact contrast checks, pass your palette through colorblindcheck::palette_check() (separate package).

Explanation: Cividis was published in 2018 specifically to remain perceptually uniform across both deuteranopia and tritanopia, which is rare. The other viridis options handle red-green issues well but can flatten under tritanopia. For accessibility-critical work (regulatory submissions, scientific publication), cividis is a safer default than even viridis. The tradeoff is a narrower color range (blue to yellow only), so it's less visually striking.

Explanation: Hard-coding the brand primary on the control variant (A) is a common reporting pattern: the eye anchors on the brand color and reads the treatments as variations of it. The secondary shades (#33CCFF, #66FFCC) are tints of cyan and teal, both adjacent on the color wheel to the primary blue, so the panel reads as one family. For long-running brand systems, pin these in a separate file (brand_colors.R) and source() it.