Skip to content

Lesson 29 ยท Performance and Pitfalls

Common Tableau Mistakes and How to Avoid Them

Avoid the Tableau errors that produce wrong numbers or misleading charts: join fan-out, average of averages, nulls, dual axes and truncated axes, with checks.

Tableau makes it very easy to produce a chart. It does not tell you when the chart is wrong. The numbers on a dashboard can be off by 2x because of a join, or the picture can mislead because of an axis, and the tool will happily draw both. This lesson collects the mistakes that experienced analysts see most often, with a quick test for each. The stand-in examples use SQL and pandas so that you can reproduce each error and its fix on a few rows.

Tableau is a graphical application, so the click-by-click steps are instructions. Always validate a new dashboard by recomputing one or two headline numbers independently (a SQL query, a pivot in Excel or pandas). That habit catches most of what follows.

1. Join fan-out: totals that are too big

When you join a table of orders with a table that has several rows per order (for example order lines or payments), each order row repeats once per matching row. A SUM of an order-level amount is then multiplied. The Tableau symptom: "my total sales doubled after I added the Payments table".

SQL Fan-out after a join

Press Run (or Ctrl + Enter) to see the result here.

Shipping is 30 in the source and 40 after the join, because order 1 matches two payment rows. Fixes: use a relationship (the logical layer in Tableau 2020.2+ keeps tables at their own level of detail and aggregates before combining), aggregate the many-side table first, or use a FIXED expression such as { FIXED [Order ID] : MAX([Shipping]) } before summing. Always compare the record count before and after a join.

2. Average of averages

The mean of regional means is not the overall mean unless every region has the same number of rows. In Tableau this happens when you aggregate on a view that already holds averages (for example, adding a grand total to a view of AVG([Sales]), or using AVG on a calculated field that is itself an average).

Python Average of averages

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The true mean is 30.5, the naive one 60.31. Use Total Using > Average (which recomputes on the underlying rows) and check that totals on a crosstab say what you think. Weighted measures, such as SUM(profit) / SUM(sales), avoid the problem: make ratios from sums, never an average of ratios.

3. Nulls

Nulls are missing values, and they behave surprisingly:

  • AVG and COUNT([Field]) skip nulls, so the average of a column with many blanks covers only the filled rows. COUNT of a field is not the same as the number of rows (COUNT([Number of Records]) or SUM([Number of Records]) is).
  • A comparison with null is null, not false, so IF [Discount] > 0.2 THEN 'High' END yields null (blank) for missing discounts.
  • Tableau shows an "N nulls" indicator at the bottom right of the view. Do not ignore it.
  • Null dates and numbers either vanish from the chart or plot at zero depending on the setting; right-click the indicator and choose Filter Data or Show Data at Default Position.

Fix by using ZN() (zero if null) or IFNULL() and ISNULL() deliberately, and decide whether a missing value means "zero" or "unknown", which are different things.

SQL NULL changes the average

Press Run (or Ctrl + Enter) to see the result here.

Skipping nulls gives 0.2, treating them as zero gives 0.1. Both are legitimate, but only one answers your question.

4. Dates: the wrong level or the wrong type

Dragging a date to a shelf gives it a default level, usually Year. Choose carefully:

  • Discrete (blue) dates such as MONTH(Order Date) make separate, labelled panes or headers and the months of different years overlap. Continuous (green) dates make a continuous axis with a continuous line.
  • A "month" date part ignores the year, so January 2024 and January 2025 collapse into one value. Use a Month/Year (truncated) date if you want a timeline.
  • Partial periods such as the current month look like a drop. Filter to complete periods or label them.
  • Fiscal years, week starts (Sunday or Monday) and locale can change results. Set the Fiscal year start and Week start on the data source.

5. Misleading chart choices

  • Truncated bar axes. Bars encode length, so the axis must start at zero. Tableau's automatic axis sometimes does not (usually after you filter). Right-click the axis > Edit Axis > Include zero. Line charts may use a non-zero axis, since they show change.
  • Dual axes with different scales. Two lines on separate axes can be made to cross anywhere you like. Use them only when the relationship is genuinely a pair (for example sales and profit margin), label both axes clearly, and consider Synchronize axis when the units match.
  • Pie charts with many slices. More than three or four slices cannot be compared; use a sorted bar chart.
  • Too many colours. Seven or more categories on Color: highlight one and grey out the rest.
  • 3D and ornaments. Tableau mostly avoids these, so don't add them back with images.
  • Maps for non-geographic data. A map of sales by state mostly shows where people live. Use a normalised measure (sales per capita) or a bar chart.

Check before you share

Reconcile One number checked outside Tableau.
Nulls Look at the "N nulls" note.
Axes Bars start at zero.
Dates Right level, complete periods.
1 / 2

6. Filters that do not do what you think

Ordinary filters, context filters, FIXED calculations and Top N interact in a specific order; see Order of Operations and Filter Pitfalls. Related traps: a "Show Only Relevant Values" quick filter can hide options and confuse users; an "All" option that is not available leaves nothing selected; and filters set on one sheet do not apply to others unless you choose Apply to Worksheets > All Using This Data Source.

7. Data blending and relationships

A blend links a secondary source to the primary one by a field and aggregates the secondary at the linking level only. Totals that mix blended and primary fields may be null or confusing. Prefer a relationship or a join in the data model if you can, and use blends only when the sources cannot be combined (for example separate databases).

8. Dashboards that overload the reader

A dashboard with 12 sheets, 8 filters and 5 colours is a data dump. Follow a simple structure: the key number or message at the top left, supporting charts below, filters in one consistent place, and a tooltip for detail. If the audience cannot say what the dashboard is for in a sentence, simplify it. Slow dashboards are covered in Extracts and Performance Tuning.

A quick self-audit

The script below shows how you might encode a review checklist for a new workbook, with the questions as data so it can be reused across projects.

Python Review checklist as data

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. After joining orders to payments (several rows per order), SUM(Shipping) doubles. What is the cause?

    A one-to-many join repeats the one-side rows, so sums of one-side measures are multiplied.
  2. 2. Which expression avoids the 'average of averages' error for a margin?

    Sum numerator and denominator first, then divide; this weights each row correctly.
  3. 3. Which rule applies to bar charts?

    A truncated bar axis exaggerates differences.
  4. 4. A view shows '4 nulls' in the corner. What should you do?

    Nulls change averages and counts and can hide marks; the right treatment depends on what missing means.

Summary

  • Compare row counts and one total outside Tableau after every join to catch fan-out.
  • Never average averages; make ratios from sums.
  • Treat nulls on purpose with ZN, IFNULL or a filter, and read the null indicator.
  • Pick the right date level and type; handle partial periods.
  • Start bar axes at zero, justify dual axes, limit slices and colours, and keep dashboards focused.

Next: Project: Sales Performance Dashboard