In Tableau, why do bars stack when you drop a dimension on Color?

## Why the bar splits at all Dropping a dimension on Color does two things at once, and people only notice one of them. It assigns a colour per member, and it **adds that dimension to the view's level of detail**. The bar that was one mark for `Furniture` becomes four marks — one per region within Furniture. Colour is the visible half; the granularity change is the half that alters the picture. ## What stacking is Once a bar position holds several marks, Tableau must decide where to draw them. The `Stack Marks` setting under the **Analysis** menu has three states: Automatic, On and Off. Automatic is the default, and for **bar** and **area** marks it stacks: each segment starts where the previous one ended, so the bar's total length equals the sum of its segments — the category total. Line, circle and square marks do not stack; several marks simply sit at their own positions. This default is defensible. When you split a total by a categorical field, the reader usually still wants the total to be legible, and stacking preserves it. ## Where stacked bars mislead **Only one segment has a baseline.** Human position judgment is accurate against a common baseline and poor otherwise. In a stacked bar, the bottom segment is easy to compare across categories; every segment above it starts at a different offset, so comparing the third segment across five bars is close to guessing. If the question is "which region grew", a stacked bar is the wrong chart. **Negatives are ambiguous.** With a mix of positive and negative values, segments extend in both directions and the visible bar length no longer reads as the total. Profit by category split by sub-category is the classic case where the chart quietly stops meaning what people assume. **Ordering is invisible.** Segment order comes from the colour field's sort, and changing that sort changes which segment gets the baseline — so two views of the same data can support different visual conclusions. **Turning stacking off is not free.** With `Stack Marks` set to Off, all segments start at zero and overlap, so a small segment disappears behind a large one. Readers report this as "the data is missing". Off is useful for deliberately overlapping comparisons — such as actual drawn over target — usually combined with transparency or a different mark type, not as a general fix. ## The alternatives - **Side-by-side bars.** Move the dimension from Color to Rows or Columns (keeping it on Color if you still want the hue). Now every segment shares the baseline and cross-category comparison is accurate. You lose the immediate read of the total. - **Small multiples.** One small chart per member, all on the same axis scale. Excellent for comparing shapes across many members; needs space. - **100% stacked.** Convert values to a percent of total so every bar fills the same length and the chart is about composition, not magnitude. This is done with a percent-of-total table calculation, which belongs to Tableau's calculation surface rather than the mark card. - **Just show the total, and let interaction do the rest.** A plain bar of the total with the breakdown on tooltip or a linked view is often the honest choice. ## Related mark card behaviour worth knowing Stacking interacts badly with **dual axis** charts: two axes each with their own Marks card, both stacking, produce a chart very few readers decode correctly. If you build a dual axis with bars, check the stacking state on both marks cards explicitly rather than trusting Automatic. Also note that stacking is a *rendering* decision, not an aggregation one. The measure values are identical whether stacking is on or off; only the drawn positions change. If the numbers themselves changed, the cause was the granularity change from the dimension, not the stacking. ## The interview answer Say that a dimension on Color both colours and refines the grain; that Automatic stacking for bar and area marks preserves the total at the cost of a shared baseline; that switching stacking off overlaps rather than separates; and that side-by-side bars or small multiples are the right move when comparison beats total. Naming the negative-values failure earns the rest of the point.