Chart Types & When to Use Each

The Visualization Fundamentals gave you the science. This topic gives you the vocabulary: every major chart type, when each one is the right answer, and — just as importantly — when each one is the wrong answer.

There are over 100 named chart types in the wild, but you need to master roughly 17 to cover 95% of business analysis work. For each, you will learn: what question it answers, what it looks like, when to use it, when to avoid it, and a real-world example. By the end, mapping a business question to the right chart should be automatic.

Complete Visualization Fundamentals — you need to understand pre-attentive attributes and the data-ink ratio before choosing charts.

Before looking at individual types, internalize this decision framework. Every business question falls into one of five categories:

This table is your 80/20: memorize it and you will never be stuck choosing a chart.

Line charts are the gold standard for time-series data. The position of each point along a continuous axis encodes the value — the most accurate pre-attentive attribute.

• Use when: showing change over continuous time (daily revenue, monthly active users, yearly temperature)
• Avoid when: categories are not time-based (use bars instead), or when you have fewer than 4 time points (not enough data for a line)
• Multiple lines: up to 5–7 is readable; beyond that, use a highlight (one accent color, rest gray) or small multiples

Area charts are line charts with the area below filled. They emphasize volume over time, and stacked area charts show composition over time. But stacked areas distort the comparison of middle series — use stacked bars for accurate category comparison.

Slope charts are a special case: two time points connected by a line. They are the most efficient way to show before/after change across categories.

Bar charts (horizontal) and column charts (vertical) encode value by length — the second most accurate pre-attentive attribute after position.

• Use when: comparing discrete categories (regions, products, departments)
• Sort bars by value — never alphabetically unless the category has an inherent order
• Horizontal bars when labels are long (better readability)
• Vertical columns when time is the x-axis (column chart) or when there are few short labels
• Grouped bars for comparing two sub-groups side by side
• Stacked bars for showing composition within each category

Always start the value axis at zero for bar and column charts. A truncated axis on a bar chart distorts the length encoding and is the most common way charts mislead.

Pie charts encode value by angle — the third most accurate encoding. They have strict rules:

• Maximum 5 slices — beyond that, the eye cannot distinguish angles
• Never use 3D — it distorts the perspective
• Sort slices by value, starting at 12 o'clock
• Direct labels — no legends; put the percentage and label on or next to each slice

Donut charts are a pie chart with the center removed. They are aesthetically cleaner and the center can display a KPI number. Same rules apply: ≤5 slices, direct labels.

100% stacked bar/column charts are almost always better than pie charts for composition. They compare proportions across categories using length (more accurate than angle) and support more than 5 categories.

When pies fail: comparing proportions across groups (use stacked bars), showing more than 5 categories (use a treemap or bar chart), or when the exact values matter (use a table or bar chart with direct labels).

Scatter plots encode two numeric variables by position on X and Y — the most accurate encoding for relationships.

• Use when: exploring the relationship between two continuous variables (ad spend vs revenue, age vs income)
• Add a trend line (linear or polynomial) to make the pattern visible
• Watch for: non-linear relationships, clusters, outliers, and overplotting (too many points)

Bubble charts add a third variable as the size of each point. They are useful but less accurate than scatter — area is harder to judge than position. Use bubble charts when the size variable is a secondary emphasis, not the primary comparison.

Heatmaps encode a third variable as color intensity on a 2D grid. They are excellent for:

• Correlation matrices (relationship between many numeric variables at once)
• Activity patterns (hour-of-day × day-of-week, or geographic density)
• Large datasets where individual points would overplot

Waterfall charts show how an initial value is affected by a series of positive and negative changes. They answer: "How did we get from A to B?"

• Use for: profit/loss breakdowns, revenue bridges, budget variances, before/after with intermediate steps
• Rule: each bar starts where the previous one ended; positive bars go up, negative bars go down, the total bar starts from zero

Funnel charts show a process with sequential stages where the count decreases at each step. They answer: "Where do we lose people?"

• Use for: sales pipelines (lead → qualified → proposal → closed), user journeys (visit → signup → activate → retain)
• Rule: stages must be sequential and monotonically decreasing; label each stage with the count and percentage

Sankey diagrams show flow from one set of categories to another. They answer: "Where does X go?"

• Use for: user journey flows (which pages lead to purchase), budget allocation (where does revenue go), energy flow (source → use → loss)
• Rule: width encodes volume; keep to ≤10 source categories or the diagram becomes unreadable

Box plots (box-and-whisker) show the distribution of a numeric variable: median, quartiles, range, and outliers in one compact shape.

• Use for: comparing distributions across groups (salary by department, response time by region)
• Read: the box = middle 50% (IQR), the line = median, the whiskers = 1.5× IQR or min/max, dots = outliers
• Avoid when: the audience does not know how to read them — for general audiences, a bar chart of medians with error bars is simpler

Violin plots combine a box plot with a kernel density estimate (KDE) — they show the full shape of the distribution, not just summary statistics.

• Use when: you need to see if a distribution is bimodal, skewed, or has heavy tails
• Pair with: a strip/swarm plot showing individual data points for full transparency

Ridge plots (joy plots) stack multiple density plots vertically, each slightly overlapping. They are excellent for showing how a distribution shifts across ordered categories (e.g., temperature by month).

Treemaps encode hierarchical composition as nested rectangles. Area = value. They answer: "How is the whole distributed across categories and subcategories?"

• Use for: budget allocation, market share with sub-segments, file system sizes, product catalog revenue
• Rule: larger rectangles = larger values; color can encode a second metric (e.g., growth rate); label the most important rectangles directly

Radar charts (spider charts) plot multiple variables on axes radiating from a center. They answer: "How does this entity score across multiple dimensions?"

• Use for: competitor comparison (5–8 metrics), skill assessments, product feature scoring
• Limitation: the area of the shape is misleading (perimeter depends on the number of axes). Use only for profile comparison, never for precise value reading

Bullet charts combine a bar with a performance marker (target line) and qualitative ranges (background shading for poor/good/excellent). They answer: "How are we performing against the target?"

• Use for: KPI tracking (revenue vs target, with poor/good/excellent zones)
• Origin: Stephen Few designed them as a compact alternative to gauge charts

Build your personal chart-type reference card:

1. For each of the 17 chart types above, build one example in Python (matplotlib/seaborn) or your preferred tool. Use the same dataset — the Superstore or retail sales dataset works well.
2. Write a one-line rule for each: when to use it, when not to.
3. The misfit test: try to force every dataset into every chart type. Notice which ones look wrong — this builds your intuition for the limits.
4. The grayscale test: convert all 17 charts to grayscale. Which survive? Which need color labels or shape encoding to remain readable?

In professional settings, the ability to choose the right chart quickly and defend the choice is what separates a junior from a senior analyst. Interviewers will ask: "You have sales data by region and month — what chart do you build first?" The answer is not "a bar chart" — it is a systematic question-to-chart mapping. This topic's 17-chart vocabulary, combined with the pre-attentive hierarchy, gives you the framework to answer any such question with confidence and evidence.

Business problem: A marketing director wants a quarterly review deck with 5 key visuals. They have data on campaign spend, conversions, revenue by channel, user retention, and regional performance.

Dataset: Use a public marketing dataset (Google Analytics sample, Facebook ads data, or a Kaggle marketing dataset) with at least 5 metrics across multiple dimensions.

Objective: Build 5 different chart types, each answering a different question, each following the question-to-chart map.