Test whether table headers actually work

Write a test question first: How many hours were reported in February? The document owner needs to confirm the correct answer. The test should show whether a reader can find it and understand its context without visually counting rows and columns.

Use a table where February contains 15 cases and 105 hours. Ask: “How many hours relate to February?” Start outside the table and find it using assistive-technology navigation. Establish that the table is identifiable and its subject understandable.

Move to the February row and Hours cell. Readers need to connect 105 with both February and hours. Announcement behaviour can depend on software and settings. Record whether the required context is available, rather than expecting one phrase to be spoken automatically.

Move sideways to 15 and check that the meaning is now cases. Then move to January’s corresponding hours cell. These movements test relationships during exploration, beyond a linear reading of the first row.

If the answer requires visually counting cells or memorising a long opening row, investigate the usability barrier. Compare the tag structure and test settings before deciding its cause.

• The first data cell after the header row.
• A cell in the middle of the table.
• A cell with both row and column headers.
• A cell beside a merged area.
• The first cell after a page break.
• A total and a cell containing a footnote reference.

Compare the visible table, the tagged cell structure and the assistive technology’s presentation. TH must identify a real header, and its relationship to data cells must be correct. Check that an empty cell does not make the next value seem to belong to the wrong column.

Sources: [1]

Merged headings may give one cell several header levels, such as “2026”, “Actual” and “thousands”. Test a cell under each group and one near a group boundary. Ensure a previous group’s heading is not incorrectly carried over.

At page breaks, examine whether the table continues as one logical table and how repeated headings are handled. Readers entering the continuation still need to interpret values. A repeated visual heading must not inadvertently become an ordinary data row.

Empty cells, dashes and zeros do not necessarily mean the same thing. Ask the owner whether a blank means unavailable, inapplicable or no activity. Preserve that distinction. Filling every blank with zero to produce a regular grid changes the content.

Where a table demands excessive reconstruction, an editorial solution may help. Split a broad comparison into two clearly named tables or simplify multi-level headings. Check that the original comparison remains possible.

Start with the failing cell, then test neighbouring cells. If the table structure was rebuilt, review the entire table rather than only the original failing row.