Home / Tableau Tutorial / Day 10
Day 10 — Building a view that gets read
The calculations are right. Now make something a manager will actually look at.
Choose the chart from the question
| Question | Chart |
|---|---|
| How is this changing over time? | Line |
| Which category is biggest? | Bar, sorted by value |
| What is the headline number? | Big number (text mark) |
| Two measures against each other | Scatter |
| Numbers by two dimensions | Text table or heat map |
| Parts of a whole | Stacked bar - rarely a pie |
The chart is decided by the question, not by what looks impressive
People cannot compare angles accurately, so a pie with more than three or four slices is decoration. Packed bubbles and word clouds have the same problem. A sorted bar chart answers most questions faster than anything else, which is why good dashboards are full of them.
Sorting is part of the message
Every calculation on this page runs against the Orders data source below - 48 rows, dimensions in blue, measures in green, exactly as Tableau colours them.
Read down the Sales column and you have to hunt for the biggest. Sorted descending, the ranking is the finding - and the eye gets it instantly. Always sort a bar chart by its measure unless the categories have a natural order.
Colour carries meaning, or it should carry nothing
- Use one colour when bars are just a measure. Rainbow bars imply a difference that is not there.
- Reserve red and green for bad and good. Never use them as ordinary category colours.
- Diverging palettes need a meaningful midpoint, usually zero.
- Colour-blind safe: about 8% of men cannot separate red from green. Tableau ships a colour-blind palette - use it.
Text tables are underrated
Finance teams often want the number, not a picture of the number. A clean text table with a few well-chosen columns beats a chart nobody trusts.
Dashboard layout
+-----------------------------------------------+
| Sales Profit Ratio Growth | <- headline numbers
+-----------------------------------------------+
| Sales over time | Filters |
+----------------------------------+ |
| By region | By category | |
+-----------------------------------------------+
| Detail table |
+-----------------------------------------------+
One dashboard, one question
A dashboard answering "how did we perform this month?" is useful. One with nineteen worksheets answering nothing in particular is a data dump. If it needs scrolling, it is probably two dashboards.
Performance habits
- Use an extract rather than a live connection unless you genuinely need live data
- Filter at the data source, not in every view
- Hide unused fields - they still get queried
- Prefer a calculated field to a long chain of table calculations
- Fewer worksheets per dashboard; each one is a separate query
Before you publish
- Do the parts add to the total? Check one region against the grand total by hand.
- Does every calculation still work with no filters, and with each filter applied?
- Are there Nulls that should be zeros, or zeros that should be blank?
- Does any total row show something meaningless, like an averaged percentage?
- Would someone who has never seen it understand it without you in the room?
Try these yourself
- Choose the right chart for "sales by month over two years".
- Choose the right chart for "profit against sales, one point per product".
- Give two reasons to avoid a pie chart with eight slices.
- Explain why a bar chart should usually be sorted by its measure.
- List three checks you would run before publishing a dashboard.