Worksheets and Marks

## The two roles Tableau gives every field When you connect to a data source, Tableau puts each column into the Data pane and assigns it one of two roles: **dimension** or **measure**. This is the first decision the tool makes on your behalf, and almost everything else about a worksheet follows from it. A **dimension** is a field whose values are used to *divide* the data. Product category, region, customer name, order date, a boolean flag — each distinct value carves the data into a group. Put a dimension on the Rows shelf and Tableau draws one row header per distinct value it finds in the data. A **measure** is a field whose values are *aggregated*. Sales, quantity, profit, discount — numbers you want summed, averaged, counted or ranged. Put a measure on a shelf and Tableau does not show the raw values; it shows an aggregate of them, written on the pill as `SUM(Sales)`, `AVG(Discount)` and so on. ## How the two combine The combination is the whole mental model of a worksheet: **dimensions determine how many marks there are, measures determine how big each one is.** A bar chart of `SUM(Sales)` by `Category` has as many bars as there are categories, and each bar's length is the sum of every underlying row belonging to that category. Add `Region` to Color and you now have one mark per Category-Region pair, each one summing fewer rows. Nothing about `Sales` changed; the granularity it is summed at changed. This is also why a measure alone gives you one number. With no dimension in the view, there is exactly one group — all the data — so `SUM(Sales)` renders as a single mark. ## How Tableau assigns the roles The assignment is heuristic and purely type-driven: numeric columns (integers, decimals) become measures; text, date, date-time and boolean columns become dimensions. Tableau has no knowledge of which numeric columns are keys, codes, years or ratios, so the heuristic misfires in predictable ways. In Tableau Desktop 2020.2 and later the Data pane groups fields by table under the current data model, with dimensions listed above a dividing line and measures below it inside each table. Older versions showed two flat Dimensions/Measures sections. The role itself means the same thing in both. ## Converting between them The role is a property of the field in the workbook, not of the source column, and you change it freely: drag the field across the dividing line, or use its menu to convert to dimension or to measure. Typical conversions: - A numeric **ID** (order ID, store number, postal code) is a label, not a quantity — convert to dimension. - A **year** or **month number** stored as an integer — convert to dimension so it slices instead of summing. - A **rating** or **score** you want to bucket by rather than average — convert to dimension. - A text column holding numbers you genuinely want to sum — change the data type first, then convert to measure. You can also override the role for a single use without changing the field: drag a dimension onto a shelf and pick an aggregation such as `COUNTD`, or right-click-drag a measure onto Rows and choose to place it as a dimension (attribute or dimension), which disaggregates it into distinct values. ## Where this goes wrong in practice The two failure modes an interviewer looks for: 1. **Summing an identifier.** `SUM(Order ID)` renders happily and means nothing. The chart looks plausible until someone checks the number. 2. **Assuming the number is wrong when the granularity changed.** A total that "shrinks" after adding a field to Color did not change; it was split across more marks. The measure is always aggregated at the view's current level of detail. A related trap is the *aggregate of an aggregate*. If your source already holds a pre-computed ratio per row, `AVG(that ratio)` weights every row equally and is usually not the ratio you want — you want the aggregate of the numerator over the aggregate of the denominator. ## Role versus discrete/continuous Candidates frequently merge two separate ideas. Dimension-versus-measure decides whether a field **slices or aggregates**. Discrete-versus-continuous (the blue/green pill colour) decides whether the field draws **headers or an axis**. The defaults line up — dimensions are usually discrete, measures usually continuous — but all four combinations exist: a continuous date dimension draws an axis, and a discrete measure draws a header. Saying "blue means dimension" is the answer that gets probed. ## What an interviewer is listening for A crisp statement that dimensions set granularity and measures are aggregated *at that granularity*, an example of the numeric-ID misclassification, and the awareness that the role is editable and independent of the pill colour.