Heatmap Colors: Keep Missing, Zero and Not Applicable Distinct

Decide the data state before choosing the fill

Zero is a value; a missing measurement is not. Keep “not applicable” separate too: it says the measurement is outside the record’s defined scope, rather than merely unavailable. The palette should follow those distinctions instead of erasing them.

This article uses explicitly constructed records. An observed zero belongs in numeric calculations. A not-observed cell and a not-applicable cell have no numeric value here. Other datasets may define statuses differently; obtain that definition before replacing blanks or calculating a summary.

Give each nonnumeric state a visible name, abbreviation with a key, or another clear cue. A pale cell alone cannot tell a reader whether it means a low value, absence or exclusion. The accessible table and exported records should preserve the same distinctions.

Basis: [1]

Read the complete six-cell example

Download the original labelled SVG

On a narrow screen, scroll inside the table. The region can receive keyboard focus.

A/P1 was observed and equals 0. A/P2 should have had a measurement under this example’s design, but it was not observed. A/P3 and B/P3 are not applicable. B/P1 and B/P2 were observed as 8 and 4 fictional units.

Download the complete status-preserving CSV

The blank numeric field in the CSV does not stand alone: a separate status field explains it. Importing just the numeric column would lose information. The text table uses an em dash for no numeric value and writes its reason in the status column.

Calculate the summary without converting absence to zero

There are three observed values: 0, 8 and 4. Their known subtotal is 12. Their observed-value mean is 12/3=4 fictional units. The two not-applicable cells and the one not-observed cell do not belong in that denominator.

On a narrow screen, scroll inside the table. The region can receive keyboard focus.

The subtotal 12 is a subtotal of available observations. It is not proof that every eligible record totals 12, because A/P2 is unknown. If the task requires an eligible-population total, report incomplete coverage rather than presenting the observed subtotal as complete.

If a dataset calls for imputation, document its method and retain the original state. This example performs no imputation. Neither a palette nor a charting library’s default replacement rule supplies a valid missing-data policy.

Keep zero inside the numeric scale

The illustration gives 0 a pale numeric fill, 4 an intermediate fill and 8 a dark fill. The number remains visible in each cell. These are original sample colors and values, not universal thresholds or a measured perceptual scale.

The missing and N/A cells use white with different visible words. Their distinction does not depend on a gray-versus-blue difference. A pattern can reinforce a state, but a legend should still say what it means, and the numeric palette should not assign an invented value to it.

Basis: [1] [4]

Check text against each actual fill. Do not assume black works on every dark value cell; the 8 cell uses white text. A white border or outline may help locate cells, but the text and complete table supply the exact state/value information. This is not a blanket claim that heatmaps are exempt from contrast requirements.

Basis: [2] [3]

Keep the same meaning in imports, filters and exports

Preserve row identity, period, numeric value, status and unit together. Validate combinations: an “observed” row needs its numeric value, including zero; a “not applicable” row should not silently carry a numeric replacement. Reject ambiguous input or ask for the data definition before coloring it.

When filtering to observed records, show that the result contains three of six cells. A summary label such as “Mean of observed values; 3 records” explains the scope better than “Average 4” alone. If you show coverage, distinguish the one eligible missing observation from the two out-of-scope records.

For an exported image, keep the status key and the exact-data table or a linked equivalent. Hover-only explanations disappear in print and screenshots. Review the phone layout so the word “Missing” and the N/A definition remain available without interpreting hue.

Try three observation-state checks

These answers apply to the complete six-record example, without imputation.