Choose a data shape for the comparison readers need
Compare long and wide data with the same four observations, then choose a source layout and public display without changing the totals.
Use long data when readers need to filter or group repeated observations. Use a wide display when the task is to compare a small, fixed set of periods side by side. The downloadable source and the on-page presentation can serve different needs, provided they represent the same observations.
Consider this invented count of visits to two exhibition rooms:
| Room | January visits | February visits |
|---|---|---|
| North | 120 | 150 |
| South | 90 | 110 |
That is a wide layout: each month gets a separate column. The same four observations in a long layout are:
| Room | Month | Visits |
|---|---|---|
| North | January | 120 |
| North | February | 150 |
| South | January | 90 |
| South | February | 110 |
The longer table has more rows, not more observations. Its row means “one room in one month.” That definition is the important part. Wickham’s Tidy Data describes a structure with variables in columns, observations in rows and each kind of observational unit in its own table. Original paper
Compare the reader’s task
| Task | Long layout | Wide layout |
|---|---|---|
| Read January beside February for one room | Requires scanning two rows | Direct side-by-side view |
| Add March | Add another observation per room | Add a March column |
| Filter all observations to one month | One month field to filter | Select the corresponding column |
| Explain the meaning of one row | Room-month observation | Room with several period values |
These are consequences of the example’s structure, not measured performance claims about a publishing tool.
Decide what should stay a column
Long does not mean putting every value into an undifferentiated “metric” column. If visits and floor area describe the same observation, they can remain separate variables. Conversely, a sequence of columns named January, February and March often indicates that a period variable has been spread across the page.
Write the row definition first. Then list the fields needed to identify one observation. In this example, room alone is no longer unique in the long table; room plus month identifies a row. A real multi-year dataset also needs a year or complete period identifier.
Publish a useful view without losing the source
For a short annual summary, a wide table may be the most readable article display. Keep the underlying long file available when readers need to combine years or perform their own analysis. Name the relationship: “This display pivots the downloadable room-month data.”
Before publication, reconcile the two versions. Both sample tables contain four observations totaling 470 visits. If a pivot combines duplicate room-month records, investigate those records before treating its output as a faithful reshaping. A prettier table is not evidence that the underlying data survived unchanged.
Record the row definition and the room-period key in the data dictionary. The primary-key guide covers how to document an identifier made from more than one field.