04 · Chart Types & When to Use Them

Tableau can build dozens of chart types, but most real analysis needs only a handful used well. This module covers bar, line, scatter, and pie charts — building each against Orders (Module 1), and critiquing when each is (and isn't) the right choice.

1. Bar charts — comparing categories

1. From Module 3, you already have Region on Columns and SUM(Sales) on Rows. Bar charts are the right default for comparing a measure across discrete categories — regions, products, categories — because humans judge bar length very accurately.
2. Add Category to Rows (after Region) to get a bar per (Region, Category) pair, rather than stacking as color did in Module 3 — this "small multiples" layout lets you compare exact lengths directly, which stacked segments make hard (comparing a middle segment's length across bars requires subtracting, since stacked segments don't share a common baseline).
3. When not to use a bar chart: more than ~15–20 categories, since bars become too thin to label or compare meaningfully — a filtered top-N view or a table is often clearer at that scale (Level 2 covers filtering patterns like this).

2. Line charts — trends over time

1. Remove Category/Region from the shelves (drag off) and start fresh: drag Order Date onto Columns. Tableau adds it as a discrete date hierarchy by default (Year > Quarter > Month > Day, expandable via the + on the pill) — right-click the Order Date pill on Columns and choose Month under the Continuous section (not Discrete) to get a single continuous timeline rather than one header per hierarchy level.
2. Drag Sales onto Rows. The Marks card mark type auto-switches to Line, plotting SUM(Sales) per month across the Orders data (January through April 2024).
3. Line charts are the right choice for showing a trend or change over a continuous field — almost always time. A bar chart with dates on Columns forces one bar per date, which visually competes with the more natural "trajectory" a line conveys.
4. When not to use a line chart: connecting points that aren't naturally ordered/continuous (e.g. a line across unordered categories like Product names) implies a trend or interpolation that doesn't exist — that's exactly the bar chart's job instead.

3. Scatter plots — relationships between two measures

1. Start a new sheet. Drag Sales onto Columns and Profit onto Rows — with two measures and no dimension yet, Tableau plots a single mark (the grand total point). Drag Product onto Detail on the Marks card to break that single point into one mark per product (8 points, matching the 8 rows of Orders, since each product here appears once).
2. Scatter plots answer "how do two measures relate, per item?" — here, whether higher Sales tends to come with higher Profit. Looking at the worked data: the Bookcase order (Sales 950, Profit -40) breaks that pattern — a loss on a mid-sized sale, visually an outlier below the rest of the cluster.
3. Drag Category onto Color to see whether the outlier belongs to a pattern (e.g. "Furniture tends to run thinner margins") — with only 8 points in this course dataset, that's a hypothesis to state and check against more data, not a firm conclusion, which is itself a good scatter-plot habit to build.
4. When not to use a scatter plot: more than one dimension and one measure without a second measure to plot against — a scatter needs two numeric axes to make sense; a single measure against a dimension is a bar chart's job.

4. Pie charts — proportion of a whole (with a warning)

1. Show Me's pie chart thumbnail requires one dimension and one measure — drag Category onto Color and SUM(Sales) onto Angle (or use Show Me directly) to get three wedges: Furniture, Electronics, Office Supplies.
2. Pie charts answer "what share of the total does each category have?" — and only really work well for a small number of categories (2–5) with clearly different sizes. Here: Electronics ≈ 2650 (450+2200), Furniture ≈ 3050 (1200+800+950+1100), Office Supplies ≈ 180 (60+120) — with only 3 wedges and one much smaller than the others, the pie communicates the story reasonably.
3. The general critique: humans compare angles far less accurately than bar lengths, so a pie with more than ~5 similarly-sized wedges becomes nearly unreadable — that same Category-by-Sales comparison as a bar chart (Section 1) would scale to many more categories without losing clarity. Default to a bar chart unless "part of a whole, few categories" is specifically the story you're telling.

5. Choosing quickly — a decision guide

• Comparing a measure across categories → bar chart.
• A measure changing over time → line chart.
• Relationship between two measures, per item → scatter plot.
• Share of a whole, 2–5 categories → pie chart (bar chart otherwise).
• Geographic patterns → map (Module 9).

How It Actually Works

A "chart type" in Tableau is not a separate rendering mode with its own query logic — every chart type in this module issues the same style of GROUP BY aggregate query; what changes is purely how VizQL maps the returned rows onto marks:

1. Bar vs. line: both a Category-bar chart and an Order-Date line chart generate a SELECT <dimension>, SUM(Sales) FROM Orders GROUP BY <dimension> query — the difference is that a continuous date field (Month, treated as continuous per Module 3's technique) is placed on an interpolated numeric axis and connected point-to-point, while a discrete dimension (Category) produces separate header positions with independent bars. Swap Order Date from continuous to discrete and the "line" chart would fall back to disconnected marks — same query, same result set, different mark geometry.
2. Scatter plot: putting a measure on both Columns and Rows (e.g. SUM(Sales) and SUM(Profit) per order) with a dimension on Detail generates one row per Order ID rather than one per Region, because Detail — unlike Color/Label — still forces its field into the GROUP BY even though nothing renders it visually; it exists purely to keep marks from being aggregated together.
3. Pie chart: internally still a GROUP BY Category aggregate, but the Angle shelf tells VizQL to render each group's proportion of the sum as a wedge angle rather than a bar length — verify by hand: Office Supplies' share of total Sales is (60+120) / 6880 ≈ 2.6%, a wedge so thin it's nearly invisible next to Electronics' (450+2200)/6880 ≈ 38.5% — the exact readability failure mode Section 4 describes, and the reason a pie's accuracy problem is perceptual (angle vs. length judgment), not a difference in the underlying arithmetic.

Cheat sheet

Question being answered	Chart type	Key shelf setup
Compare a measure across categories	Bar	Dimension on Columns, Measure on Rows
Trend over time	Line	Continuous date on Columns, Measure on Rows
Relationship between two measures	Scatter	Measure on Columns, Measure on Rows, dimension on Detail
Share of a whole (few categories)	Pie	Dimension on Color, Measure on Angle

Exercise

Using Orders (Module 1), decide which chart type you'd use to answer each of these three questions, and justify why in one sentence each: (1) "Did total Sales grow month over month?" (2) "Which single order had an unusually bad Profit relative to its Sales?" (3) "What fraction of total company Sales came from Office Supplies?"