Effective Data Communication

From Visualization to Communication

The previous two lessons focused on what to visualize — how to represent distributions and relationships using the right chart type for the data at hand. This lesson turns to the harder question: how to communicate that visualization effectively. A chart that is technically correct but poorly designed fails its audience; one that is beautifully clear succeeds even if it is simpler.

Effective data communication is a skill that combines statistics, design, and psychology. The statistician provides the numbers; the designer shapes their presentation; and an understanding of human perception determines what actually lands in the viewer’s mind. This lesson covers the principles that unite all three: choosing the right chart, applying Tufte’s guidelines, avoiding common traps, selecting colors wisely, and building a narrative that carries the reader from data to insight.

Choosing the Right Chart Type

The first decision is always chart type selection. The right chart depends on three factors: the data type (continuous, categorical, ordinal, time-series), the number of variables (one, two, or many), and the message you want to convey (comparison, distribution, relationship, composition, or trend).

A rough decision tree:

Comparison across categories: Bar chart (vertical for named categories, horizontal for long labels). Avoid pie charts whenever a bar chart will work — the eye compares lengths much more accurately than areas or angles.

Distribution of a single variable: Histogram (continuous data) or box plot (comparing distributions across groups). Density plots when you want a smooth curve rather than discrete bins.

Relationship between two continuous variables: Scatter plot, possibly with a smoothed trend line. Line plot when one variable is time.

Composition: Stacked bar chart for showing how parts add up to a whole at several points. A single pie chart is acceptable only when there are two or three slices and the message is truly about proportions, not precise magnitudes.

Many variables simultaneously: Heatmap for a matrix of correlations or a time × category grid. Pair plot for exploratory analysis with up to ~10 variables.

Tufte’s Principles of Good Visualization

Edward Tufte, the pioneering information designer, articulated a set of principles in his 1983 book The Visual Display of Quantitative Information that remain the gold standard of data visualization practice. His core insight is elegantly simple: show the data, and nothing but the data.

Tufte’s key principles:

Maximize the data-ink ratio. Every drop of ink in a visualization should encode data. Ink that does not represent data — decorative borders, background gradients, redundant grid lines — should be eliminated or minimized. The data-ink ratio is the fraction of a graphic’s ink devoted to the non-redundant display of data.

Erase non-data ink. Remove chart borders, unnecessary tick marks, redundant labels, and background shading. Each element removed makes the data stand out more clearly.

Erase redundant data ink. If two channels encode the same information — say, both the height of a bar and its color encode the same category — remove the redundant one. Redundancy clutters without adding clarity.

Revise and edit. Good data graphics are the product of multiple iterations. The first draft is rarely the clearest one. Each revision should ask: does this element earn its space?

Common Mistake: Misleading Axes

The y-axis is one of the most powerful tools for distortion in data visualization — and one of the most commonly abused. The core issue is axis truncation: starting the y-axis at a value other than zero to make a small change appear large.

Consider a bar chart of monthly sales that vary between $980,000 and $1,020,000. If the y-axis starts at $970,000, the bars look dramatically different in height — a bar at $1,020,000 appears five times as tall as one at $980,000, even though the actual difference is only 4%. If the y-axis starts at zero, the bars all look nearly identical, which is the honest picture.

The rule: bar charts must start at zero because bar length is the visual encoding, and length is always measured from the baseline. Line charts have more flexibility — because the message is the shape of the change rather than absolute magnitude, starting a line chart’s y-axis above zero is often reasonable, provided you label the axis clearly.

Related tricks to watch for: dual axes (two y-axes with different scales can make any two lines appear to track each other), cherry-picking time ranges (choosing a start date that makes a trend look better than it really is), and reversing the y-axis (plotting downward what everyone expects to go upward).

Common Mistake: Cherry-Picking and Selective Display

Cherry-picking means selecting only the data points, time periods, or subgroups that support your preferred conclusion while ignoring contradictory evidence. It is perhaps the most pervasive form of statistical dishonesty because it looks like straightforward data presentation. The selection is invisible to the viewer.

Common forms of cherry-picking in visualization:

Selective time windows: Starting a trend chart at a market low to show maximum growth, or ending it before a recent decline.

Selective subgroup comparison: Showing only the age group, region, or demographic where your product wins while omitting the others.

Omitting error bars: Showing point estimates without their uncertainty ranges makes small differences look definitive when they may be within the noise.

The antidote is transparency: show the full time series, all relevant subgroups, and always include uncertainty bounds where they exist. When you do restrict the view (which is sometimes legitimate to focus attention), make the restriction explicit in the title or caption.

Color Usage: Sequential, Diverging, and Categorical Palettes

Color is one of the most powerful encoding channels in data visualization — and one of the easiest to misuse. The key is matching the color palette to the data type.

Sequential palettes encode a single ordered quantity from low to high. They use a single hue that varies in lightness or saturation: light yellow → dark blue, or white → dark green. Use sequential palettes when your data has a natural order and a meaningful zero (temperature, count, probability).

Diverging palettes encode quantities that have a meaningful midpoint, with one color for values above the midpoint and another for values below. A classic diverging palette goes from deep red (negative) through white (zero) to deep blue (positive). Use diverging palettes for correlations (−1 to +1), temperature anomalies, or any measure with a meaningful baseline.

Categorical palettes assign a distinct color to each category. The colors should be perceptually distinct but not ordered — there is no implied hierarchy. Use categorical palettes for nominal data: product lines, countries, animal species. Limit categorical palettes to 8–10 colors; beyond that, the eye cannot reliably distinguish the hues.

Accessibility: Colorblind-Safe Palettes

Approximately 8% of men and 0.5% of women have some form of color vision deficiency, the most common being red-green color blindness (deuteranopia and protanopia). A chart that relies on red versus green to distinguish two series is unreadable to a significant fraction of your audience.

Best practices for accessible color use:

Do not rely on color alone. Use both color and shape to distinguish data series in line charts (filled circles vs. triangles). Use both color and pattern in bar charts (solid vs. hatched). A legend that says “blue series” and “orange series” is meaningless to a colorblind viewer if the colors are the wrong ones.

Use colorblind-safe palettes. The Okabe-Ito palette (originally designed for scientific figures) is an excellent default: it distinguishes eight categories reliably for viewers with the most common forms of color vision deficiency. The viridis family of sequential palettes was designed with colorblind safety as a primary goal.

Test your charts. Tools like the Coblis color blindness simulator and the colorblindr R package let you preview how your chart appears to viewers with different types of color vision deficiency. This takes under a minute and is one of the highest-impact accessibility checks you can perform.

Storytelling with Data

The most technically correct visualization can fail if it does not have a clear narrative purpose. Data storytelling — the practice of constructing a coherent argument from data visualizations, context, and explanation — is what separates an analyst who produces charts from one who drives decisions.

A data story has three components:

Context: What is the situation? What question are we trying to answer? Without context, the viewer does not know what the data represents or why it matters. A good visualization always includes a title that states the conclusion (“Q3 Revenue Declined 12% vs. Prior Year”) rather than just the variable (“Revenue by Quarter”).

Data: The visualization itself. What does the data show? Good storytelling charts have a single clear message per chart rather than trying to show everything at once. If you need to make three points, use three charts.

Insight: What does the data mean? The “so what?” The visualization is a means to an end — that end being a decision, a recommendation, or a change in understanding. State the insight explicitly, either in the title, caption, or annotation directly on the chart.

Practical techniques for data storytelling include: annotations that call out key events or anomalies directly on the chart; progressive disclosure in presentations that starts with a simple view and adds complexity; and consistent framing across a series of charts so the viewer builds a mental model rather than starting from scratch each time.