8.2 Part-to-Whole & Correlation Charts: Treemaps, Scatter Plots & Bubbles

8.2 Part-to-Whole & Correlation Charts: Treemaps, Scatter Plots & Bubbles

Visual analytics excels when revealing structural proportions within a whole and identifying correlations across continuous variables. In business analytics, understanding how revenue breaks down across hundreds of products, or whether marketing spend correlates with customer lifetime value, requires specialized chart architectures. For the Salesforce Certified Tableau Data Analyst, mastering Treemaps, Pie/Donut Charts, Scatter Plots, and Packed Bubble Charts involves understanding not just how to construct them, but also their mathematical underpinnings, perceptual constraints, and handling of edge-case data values.

---

Treemaps: Hierarchical Rectangular Tiling

A Treemap displays quantitative data as a set of nested rectangles. Developed by Ben Shneiderman to visualize hierarchical file directory sizes on disk storage, treemaps have become an indispensable tool in modern business intelligence for part-to-whole evaluations.

1. The Tiling Algorithm and Space Optimization

Tableau utilizes an optimized squarified treemap algorithm. The algorithm tiles rectangles within the canvas while striving to maintain an aspect ratio as close to 1:1 (a square) as possible. Square-like rectangles are substantially easier for the human eye to perceive, compare, and label than elongated, needle-thin slivers.

• Ordering Convention: Tableau orders the tiled rectangles automatically from the top-left corner (largest value) to the bottom-right corner (smallest value).
• Space Efficiency: Unlike bar charts that require linear axis space, or pie charts that waste peripheral canvas area, treemaps occupy 100% of the allocated rectangular worksheet space.

+-----------------------------------------------------------------------+
| TECHNOLOGY                              | FURNITURE                   |
| (Outer Parent Box)                      | (Outer Parent Box)          |
| +-----------------------+-------------+ | +-------------+-----------+ |
| | Phones                | Machines    | | | Chairs      | Tables    | |
| | $330k                 | $188k       | | | $328k       | $206k     | |
| +-----------------------+-------------+ | +-------------+-----------+ |
| | Accessories           | Copiers     | | | Bookcases   | Furnish.  | |
| | $167k                 | $160k       | | | $114k       | $91k      | |
| +-----------------------+-------------+ | +-------------+-----------+ |
+-----------------------------------------+-----------------------------+
| OFFICE SUPPLIES                                                       |
| +-------------------+-------------------+---------------------------+ |
| | Storage $223k     | Binders $203k     | Paper $78k | Envelopes $16k | |
| +-------------------+-------------------+---------------------------+ |
+-----------------------------------------------------------------------+

2. Multi-Level Dimensional Nesting

Treemaps support hierarchical breakdowns by layering dimensions on the Marks card:

• Single Dimension: Placing Sub-Category on Detail and SUM(Sales) on Size generates a flat treemap of 17 individual rectangles.
• Multiple dimensions: A higher-level category on Color and a lower-level member on Label or Detail can encode hierarchy, but color and labels—not a guaranteed set of heavy parent bounding boxes—must make the levels understandable. Inspect the rendered layout because small marks may lose labels.

3. Dual-Metric Encoding (Size and Color)

A powerful capability of treemaps is encoding two independent continuous metrics simultaneously:

• Size: Encodes volume or magnitude (e.g., SUM(Sales) or Record Count).
• Color: Encodes an efficiency ratio, growth rate, or profitability margin (e.g., SUM(Profit) or [Profit Ratio]) using a diverging palette.

Stakeholders can instantly identify products that generate high volume (large rectangle) but negative profitability (saturated orange/red color), directing strategic pricing reviews.

4. Critical Data Limitation: Handling Negative Values on Size

[!CAUTION] Sizing with non-positive values Area cannot communicate a negative magnitude. Before placing a field such as Profit on Size, inspect how Tableau handles zero and negative values in the current mark type and version. A common, clearer design is to size marks with a nonnegative volume measure such as Sales and place signed Profit on a diverging Color encoding.

---

Pie Charts and Donut Charts: Rules, Governance & Workarounds

While popular in business presentations, Pie Charts are frequently misused. Human visual perception is fundamentally inefficient at evaluating 2D angles, arc lengths, and radial wedge areas compared to linear length along a baseline.

1. Design Checks for Pie Charts

Pie charts work best for a small set of nonnegative parts of one meaningful whole. Treat the following as design checks rather than universal certification rules:

1. Limit categories deliberately: A pie chart with many similarly sized slices forces difficult angle comparisons, crowded labels, and excessive colors. For twelve categories, prefer a sorted bar chart; use a treemap when compact part-to-whole display is more important than precise comparison. There is no universal five-slice cutoff.

2. Use an intentional order: Sorting by magnitude can make large-to-small comparison easier. A consistent start angle can help, but Tableau does not impose a universal 12-o'clock certification rule.

3. Validate the whole: The Angle measure should represent nonnegative, additive parts of the same whole. If signed values are present, use a chart that shows direction explicitly rather than implying that losses are ordinary shares.

4. Make identification easy: Direct labels can reduce legend lookup when they fit. For a small or constrained view, a clear legend and accessible tooltip may be preferable to overlapping labels.

Pie Chart Marks Card Setup:  [Mark Type: Pie  ]  [Color: Category]  [Angle: SUM(Sales)]  [Label: SUM(Sales) %]

2. Donut Charts: Construction Mechanics in Tableau

Tableau Desktop does not feature a native 'Donut' mark type in the Marks card or Show Me menu. Instead, a Donut Chart is engineered using a Dual-Axis Chart that overlays a smaller, solid-colored circle on top of a larger pie chart.

+-----------------------------------------------------------------------+
|                      DONUT CHART CONSTRUCTION                         |
|                                                                       |
|   Layer 1: Outer Pie Chart              Layer 2: Inner Circle Mask    |
|   (Mark: Pie, Size: Large)              (Mark: Circle, Size: Small)   |
|         . - ~ ~ ~ - .                         . - ~ ~ ~ - .           |
|     . '  Furniture  ' .                   . '               ' .       |
|    /   (Color & Angle) \                 /     White Fill    \      |
|   |                     |     +         |      Blank Mask     |       |
|   | Technology   Office |               |   (or KPI Label)    |       |
|    \                   /                 \                   /        |
|     ' .             . '                   ' .               . '       |
|         ' - ~ ~ ~ - '                         ' - ~ ~ ~ - '           |
|                                                                       |
|                      COMBINED (DUAL AXIS):                            |
|                              . - ~ ~ ~ - .                            |
|                          . '  Furniture  ' .                          |
|                         /   +-----------+   \                         |
|                        |    | Total KPI |    |                        |
|                        |    |  $2.3M    |    |                        |
|                         \   +-----------+   /                         |
|                          ' .  Technology. '                           |
|                              ' - ~ ~ ~ - '                            |
+-----------------------------------------------------------------------+

Step-by-Step Construction Guide:

1. Generate Dummy Axis Placeholders: Double-click the Rows shelf and type MIN(0). Repeat this by typing a second MIN(0) directly beside it on Rows. Tableau generates two identical vertically stacked chart panes.
2. Configure Layer 1 (Outer Pie):
   • In the Marks card, select the top MIN(0) pane.
   • Change mark type from Automatic to Pie.
   • Drag the categorical dimension (e.g., Segment) to Color.
   • Drag the continuous measure (e.g., SUM(Sales)) to Angle.
   • Increase the Size slider to make the pie adequately large.
3. Configure Layer 2 (Inner Circle Hole):
   • Select the second MIN(0) Marks card.
   • Change mark type to Circle.
   • Remove any fields from Color or Angle.
   • Click Color and set the swatch to pure White (or match the dashboard canvas background color).
   • Adjust the Size slider so the circle is approximately 60% of the pie's diameter.
   • Optional: Drag SUM(Sales) to Label to render total enterprise sales inside the center cutout.
4. Merge and Clean Axes:
   • Right-click the second MIN(0) pill on the Rows shelf and select Dual Axis.
   • Right-click either vertical axis header and uncheck Show Header.
   • Navigate to Format > Lines and remove zero lines and grid lines to yield a pristine donut visual.

---

Scatter Plots: Correlation, Distribution, and Regression

A Scatter Plot visualizes the mathematical relationship, distribution, and correlation between two continuous quantitative measures plotted along Cartesian coordinates (X and Y axes).

1. The Single-Mark Aggregation Phenomenon

A foundational concept tested on the exam is Tableau's default aggregation behavior:

• When an analyst drags SUM(Sales) to Columns and SUM(Profit) to Rows, Tableau evaluates the entire underlying data table and plots exactly one single mark.
• Why? By default, Tableau issues aggregate SQL queries (SELECT SUM(Sales), SUM(Profit) FROM Data). Because no categorical dimensions exist on Rows, Columns, or Marks shelves, the entire dataset represents a single partition.

2. Disaggregating Marks: Two Distinct Approaches

To expand the single mark into a scatter plot of individual data points, the analyst has two architectural options:

Default State:               [Columns: SUM(Sales)] [Rows: SUM(Profit)] -> 1 mark (Total Sum)
With Dimension on Detail:    Adds [Customer ID] to Detail              -> 793 marks (1 per customer)
With Aggregate Measures OFF: Analysis > uncheck Aggregate Measures    -> 10,000 marks (1 per raw record)

3. Advanced Statistical Analytics on Scatter Plots

Once marks are plotted, the Analytics pane unlocks advanced statistical modeling tools:

Trend Lines

Dragging Trend Line from the Analytics pane onto the scatter plot fits a mathematical regression model across the marks. Tableau supports five trend line models:

• Linear: Fits $Y = b_0 + b_1 X$. Ideal for constant proportional relationships.
• Logarithmic: Fits $Y = b_0 + b_1 \ln(X)$. Useful when rate of change flattens rapidly as X increases (e.g., diminishing marginal returns).
• Exponential: Fits $\ln(Y) = b_0 + b_1 X$. Used when values double or multiply over intervals.
• Polynomial: Fits higher-order curves (degree 2 to 8). Degree 2 models parabolic arcs (e.g., peak profit curves).
• Power: Fits $\ln(Y) = b_0 + b_1 \ln(X)$. Models physical scaling laws.

Trend-line details can report the model equation, $R^2$, and p-values. A larger $R^2$ means the fitted model explains more observed variation in that sample. Compare a p-value with a chosen significance level, such as 0.05, as evidence about the tested coefficient; it does not prove causation or guarantee predictive value.

Clustering (k-Means)

Dragging Cluster from the Analytics pane initiates Tableau's built-in k-means clustering algorithm:

• Tableau evaluates the selected measures (and any optional dimensions) and groups marks into $k$ distinct statistical clusters.
• Automatic Variable Standardization: Tableau automatically normalizes measure scales (subtracting the mean and dividing by standard deviation) so that large-scale metrics (e.g., Sales in millions) do not mathematically drown out small-scale metrics (e.g., Profit Ratio from -0.2 to +0.4).
• Saving clusters as a field: Authors can drag the generated Clusters pill from the Marks card into the Data pane to save a cluster group for reuse in other worksheets. Treat it as a saved analytical field, not an immutable business classification; review it when the data or clustering inputs change.

---

Packed Bubble Charts: Perceptual Reality vs. Aesthetic Appeal

A Packed Bubble Chart displays relational data as a cloud of clustered circular marks:

• Configuration: Dimension on Detail (or Color), continuous measure on Size, and Mark type set to Circle.
• Layout Engine: Tableau executes a non-spatial circle-packing layout algorithm, clustering circles tightly within the canvas borders.

The Perceptual Flaw of Packed Bubbles

While visually eye-catching in infographics, packed bubble charts have severe analytical limitations:

1. Area vs. Length Distortion: According to Stevens' Power Law of psychophysics, humans systematically underestimate differences in circular areas compared to differences in straight linear lengths. Differentiating whether Circle A ($r=14px$) is 20% larger than Circle B ($r=12px$) is virtually impossible without tooltips.
2. Absence of Cartesian Axes: Packed bubbles lack X and Y coordinates. Proximity between two bubbles carries zero statistical meaning; bubbles sit next to each other solely because the packing algorithm found an open geometric gap.
3. When Packed Bubbles Are Acceptable: Exploratory overviews where the user only needs to spot massive order-of-magnitude anomalies (e.g., one colossal bubble dominating dozens of tiny specks).

---

Part-to-Whole & Correlation Comparison Matrix

---

Exam Traps & Practical Scenarios

Critical Pitfalls

[!WARNING] Exam Trap: Why Does My Scatter Plot Only Show One Mark? A classic exam scenario asks why a newly created scatter plot displays a single dot in the center of the canvas. The answer is measure aggregation. Tableau aggregates the entire table into SUM(Columns) and SUM(Rows). To fix it, the analyst must either drop a dimension on the Detail shelf or deselect Analysis > Aggregate Measures.

[!CAUTION] Pitfall: Reusing K-Means Clusters A generated cluster pill initially belongs to the current view. Drag it from the Marks card into the Data pane to save a cluster group for reuse, then validate whether the saved grouping remains appropriate after data or model inputs change.

[!NOTE] Pitfall: Signed values on Size Treemap area cannot show sign. Inspect warnings and mark behavior when a Size field contains zeros or negatives, and prefer a nonnegative size measure with the signed measure on a diverging Color encoding when both magnitude and direction matter.