Case Study 2: Gestalt at Work — How The New York Times Designs Charts

Why the Times?

The New York Times graphics desk has been the gold standard for data journalism visualization for over two decades. Under leaders like Amanda Cox, Archie Tse, and Matthew Ericson, and more recently with data journalists like Mira Rojanasakul and Stuart Thompson, the team has produced charts that are simultaneously precise, accessible, and visually elegant. Their work reaches millions of readers, most of whom have no training in statistics or chart reading.

What makes their charts effective is not software sophistication or flashy interactivity. It is a deep — often intuitive, sometimes deliberate — application of the perceptual principles we covered in this chapter. The Times graphics team designs for the human visual system. Their charts work because they align with how the eye groups, compares, and interprets visual information.

This case study analyzes representative NYT chart designs through the lens of Gestalt principles, pre-attentive processing, and the Cleveland-McGill encoding hierarchy. We will not reproduce the specific charts (rights belong to the Times), but we will describe their structure in enough detail for you to understand — and learn from — the design choices.

Example 1: The Connected Scatter Plot — Proximity and Connection at Work

One signature NYT format is the connected scatter plot, used memorably in pieces about the relationship between two variables over time. The canonical example: a chart showing the relationship between unemployment rate (x-axis) and inflation rate (y-axis) over several decades, with data points connected by a line in temporal order.

Encoding analysis. Both variables use position encoding — the most accurate channel. The temporal ordering is encoded through connection (a line joining sequential points). The direction of time is sometimes reinforced with subtle arrowheads or by labeling key years directly on the line.

Gestalt analysis. The line invokes connection, grouping all the points into a single narrative sequence. Without the line, the same points would appear as an unstructured scatter. With the line, the viewer perceives a path — a story moving through two-dimensional space. Continuity reinforces this: the eye follows the smooth trajectory of the line, detecting loops, reversals, and inflection points automatically.

Proximity plays a role in the labeling. The Times typically places year labels directly next to their corresponding points (close proximity between label and data), eliminating the need for a color-coded legend or separate annotation layer. This keeps the label and the data in the same perceptual group.

What makes it work. The chart uses the two highest-accuracy channels (position on x, position on y) for the two quantitative variables, and the strongest grouping principle (connection) for the temporal sequence. The design is perceptually efficient: the viewer decodes two quantitative values and a temporal ordering, all through high-accuracy channels and strong Gestalt grouping.

Example 2: Small Multiples for Election Results — Enclosure and Similarity

NYT election coverage frequently uses small-multiple panels — grids of identically-formatted charts, one per state or one per county, showing the same variables in every panel. A typical layout: a 5x10 grid of tiny charts, each showing the vote margin in one state over the last six presidential elections.

Encoding analysis. Within each panel, the primary variable (vote margin) uses position or length encoding. The temporal variable uses position on a shared x-axis. The categorical variable (state) is encoded by the spatial position of the panel within the grid. Party affiliation uses color hue (blue and red — the standard American political color encoding).

Gestalt analysis. Enclosure is the dominant organizing principle. Each state's panel has a light border or background that groups its data points together. The viewer instantly perceives each panel as a unit — "this box is about Ohio." Without the enclosure, the 50 sets of data points would merge into a single confusing scatter.

Similarity operates through color. All blue marks across all 50 panels are perceived as belonging to the same political category, and all red marks to the other. This cross-panel similarity allows the viewer to detect national patterns: "the blue is growing in the Sun Belt" or "the red is fading in the Upper Midwest." The Gestalt grouping by similarity (color) operates across the Gestalt grouping by enclosure (panel borders), creating a two-level perceptual hierarchy: marks within a state, and patterns across states.

Proximity operates at the grid level. States in the same row or column are perceived as somewhat related. The Times sometimes exploits this by arranging states geographically (Northeast in the upper right, West in the left) rather than alphabetically, so that spatial proximity in the chart grid mirrors spatial proximity on a map.

What makes it work. Small multiples are one of the most perceptually efficient designs in data visualization because they exploit enclosure for separation, similarity for cross-panel patterns, and position encoding (the highest-accuracy channel) for the quantitative data in every panel. The Times adds geographic arrangement to create an additional proximity signal. The result is a chart that shows 50 individual stories and one national story simultaneously.

Example 3: Annotated Line Charts — Connection, Continuity, and Direct Labeling

For economic and public health reporting, the Times frequently uses annotated line charts — simple line charts with contextual annotations placed directly on the chart rather than in captions or footnotes. A representative example: a line chart showing the national COVID-19 case count over time, with shaded bands marking lockdown periods, text annotations marking key events ("vaccines authorized," "Omicron wave begins"), and multiple series distinguished by color.

Encoding analysis. The time series uses position encoding for both time (x) and case count (y). Multiple series (e.g., cases, hospitalizations, deaths) use color hue for categorical distinction. The shaded bands for lockdown periods use a combination of color intensity (light gray or light blue background) and enclosure.

Gestalt analysis. Connection is the primary grouping mechanism: each line connects sequential data points, and the viewer perceives each line as a single evolving story. Continuity supports trend perception: the eye follows each line's smooth trajectory, detecting surges and declines automatically.

Enclosure (shaded bands) groups time periods. When the Times shades the background behind a recession or a lockdown period, every data point within that shaded region is perceived as belonging to that period. The viewer sees "this is what happened during the lockdown" without needing to read the exact dates on the x-axis.

Proximity governs the annotations. The Times places event labels immediately adjacent to the point on the line where the event occurred. "Vaccines authorized" appears right next to the corresponding date on the curve, not in a separate legend or margin note. This proximity ensures that the label and the data are in the same perceptual group, eliminating the serial lookup that a remote legend would require.

Similarity distinguishes the series. When cases are shown in red and deaths in gray, the viewer groups all red marks together (the case trajectory) and all gray marks together (the death trajectory). The two stories can be perceived simultaneously because color similarity creates two distinct perceptual groups.

What makes it work. The annotated line chart combines the highest-accuracy encoding (position), the strongest temporal grouping (connection and continuity), contextual grouping (enclosure for time periods), and direct labeling (proximity between annotation and data). The viewer does not need to decode a legend, identify a time period, or cross-reference a footnote. Everything is where the eye expects it.

Example 4: The "You Draw It" Interactive — Fighting Inattentional Blindness

In 2015, the Times published an interactive feature where readers were asked to draw a trend line before seeing the actual data. For example: "You draw the line showing the share of Americans who identify as having no religion, then see the actual data." The reader draws with their mouse, and the real line is then revealed.

This design is a direct response to inattentional blindness and confirmation bias. When readers simply look at a chart, they often fail to notice trends that contradict their expectations — inattentional blindness in action. By asking readers to commit to a prediction first, the Times forces attention to the gap between expectation and reality.

Gestalt analysis. The design uses connection (the drawn line and the actual line) and similarity (the drawn line and the actual line are typically in different colors) to create a visual contrast. The viewer perceives two lines — their prediction and reality — grouped by connection within each line and separated by color similarity into two distinct series. The gap between the two lines is perceived through the Gestalt principle of proximity (where the lines are close, the prediction was good; where they diverge, it was poor).

What makes it work. The design does not just show data — it makes the viewer's own perception a part of the visualization. It is a powerful demonstration that seeing data accurately requires not just good encoding (the Times uses position throughout) but also active engagement with one's own cognitive biases.

Common Design Patterns Across NYT Graphics

Analyzing NYT graphics reveals several recurring patterns that align with the perceptual principles in this chapter:

Direct labeling over legends. The Times almost always labels data series directly on the chart (a text label placed near the line or bar it describes) rather than using a separate color legend. This exploits proximity (label and data are perceptually grouped) and eliminates the serial lookup required by a remote legend.

Restrained color palette. Most NYT charts use 2-3 colors, not 8-10. This respects the working memory limit of 3-5 visual items and ensures that color similarity creates clean, unambiguous groupings.

Generous whitespace. NYT charts use wide margins, generous gutters between panels, and ample spacing between elements. This strengthens proximity grouping (elements that belong together are close; elements that do not are far apart) and reduces visual clutter.

Gray as default, color as emphasis. In many NYT charts, the majority of data is drawn in gray, with a single highlighted series or data point drawn in a saturated color. This is pre-attentive pop-out in action: the colored element is immediately detected against the gray field. It also keeps the total number of color categories low (often just "highlighted" vs. "context"), well within working memory limits.

Annotations as part of the chart, not separate. Titles, subtitles, source attributions, and contextual notes are placed within the chart area, near the data they describe. This exploits proximity and enclosure to keep all relevant information in a single visual group that the eye can process without jumping between chart and caption.

Lessons for Practitioners

You do not need to work at the New York Times to apply these principles. The core lessons are:

1. Use position encoding for your most important variables. The Times does this in virtually every chart.

2. Use connection (lines) for temporal sequences. Do not scatter time-series points without connecting them.

3. Use enclosure (panels, shading) to create groups. Small multiples work because enclosure is a powerful grouping mechanism.

4. Use proximity for annotation. Label directly on the chart, not in a remote legend.

5. Use color sparingly and for emphasis. Gray for context, color for the story.

6. Align Gestalt principles — do not let them fight. Proximity, similarity, connection, and enclosure should all point the viewer to the same groupings.

The Times graphics team may not explicitly invoke Gestalt terminology in their design meetings. But the principles are embedded in their practice. Every effective chart exploits the same perceptual machinery. The vocabulary gives you the ability to analyze why a design works, diagnose why it does not, and make principled improvements.

What Distinguishes NYT Charts from the Average Dashboard

The analysis above may give the impression that NYT graphics are special because they use "advanced" techniques. They do not. The techniques — position encoding, color hue for categories, direct labeling, small multiples, whitespace — are available in every charting tool, including matplotlib, seaborn, and Plotly. What distinguishes NYT graphics is the discipline with which these techniques are applied.

The average corporate dashboard or Jupyter notebook chart violates Gestalt principles routinely: legends placed far from the data they describe (breaking proximity), 10-color categorical palettes (exceeding working memory limits), multiple chart types crammed into one panel (creating conflicting enclosure and proximity signals), and heavy gridlines competing with the data for visual attention.

The NYT avoids these problems not through exotic methods but through restraint and perceptual awareness. Each chart asks one question and answers it using the minimum number of visual elements required. Color is used for meaning, not decoration. Whitespace is used for grouping, not just for padding. Labels are placed where the eye needs them, not where the software's default puts them.

This discipline is learnable. It requires understanding the perceptual principles (which you now have) and the willingness to revise a first-draft chart until the Gestalt signals align, the encoding channels match the data types, and the pre-attentive attributes highlight what matters. Every chart you build in Parts III through VII of this book is an opportunity to practice this discipline.

Discussion Questions

1. The NYT uses direct labeling instead of legends. Under what circumstances might a legend be more practical? (Hint: think about the number of series and the density of the plot.) How would you decide when to switch from direct labeling to a legend?

2. The "gray as default, color as emphasis" pattern assumes that the designer knows which data is most important. What happens when the viewer's question is different from the designer's intended emphasis? How might interactive visualization address this tension?

3. Small multiples can show dozens of panels, but at some point, the individual panels become too small to read. What Gestalt principles still work at very small panel sizes, and which require a minimum amount of visual space to be effective?

4. The "You Draw It" format requires interaction — the reader must actively engage. Most data visualization is passive (the reader looks at a finished chart). What are the perceptual advantages and disadvantages of passive vs. interactive visualization?

5. Find a recent NYT chart (nytimes.com/section/upshot is a good source) and analyze it using the Gestalt principles from this chapter. Which principles are at work? Are any in conflict?