Resources · 103
Missing data: distinguish absence, zero and estimation
Prevent silent replacement of absent values from changing the meaning of metrics.
· 2 min
What this guide helps achieve
- Name the states
- Check the denominator
- Compare coverage
- Document estimation
- Expose limitations
Quick check
- Should missing values become zero?
- Can a mean rise without a real increase?
Step-by-step method
- 01
Name the states
Separate not collected, unknown, not applicable and observed zero. Define each state at collection; a blank field does not explain its cause.
Deliverable: state dictionary.
- 02
Check the denominator
Record whether calculation uses all records or only supplied values. Reproduce sum, count and mean on a known small dataset including missing values.
Deliverable: explicit-denominator calculation.
- 03
Compare coverage
Measure missingness by period, channel and segment. Check whether a chart change comes from collection changes rather than new behaviour.
Deliverable: coverage by segment.
- 04
Document estimation
When a purpose justifies imputation, preserve original value, method and estimation flag. Compare decisions with and without the method, especially for lightly observed groups.
Deliverable: sensitivity analysis.
- 05
Expose limitations
Show partial-data status near the result. Assign an owner to complete or qualify gaps; preserve corrected versions instead of silently rewriting history.
Deliverable: visible status and history.
Choose how a value is handled
Check the observed state before choosing next steps.
Observed zero
Preserve its meaning and unit.
Missing observation
Show coverage and the known reason for absence.
Estimated value
Expose method, version and decision sensitivity.
Reusable worksheet
Complete with your authorised observations. These fields are a working template, not observed results.
| Field | Information to record |
|---|---|
| State | Cause and handling rule |
| Calculation | Population, count and units |
| Estimation | Method, version and sensitivity |
Worked example
Illustrative situation
Illustrative case: ten records contain two known amounts and eight missing values. Dividing the sum by ten answers a different question from dividing it by two.
Decision and expected evidence
Publish the denominator, supplied count and rule; do not turn the eight missing values into zero purchases.
Distinguish the mechanisms
| Mechanism | Purpose | Check or limitation |
|---|---|---|
| Zero | Measured null outcome | Do not create it to fill a blank |
| Missing | Unavailable observation | Show coverage |
| Estimated | Method-produced value | Preserve provenance |
Management indicators
| Indicator | What it measures | First action |
|---|---|---|
| Coverage | Usable values over expected records | Inspect missing segments |
| Sensitivity | Difference between documented calculations | Review unstable decisions |
Common pitfalls
- Turn every absence into zero
- Change the denominator without showing it
- Impute without provenance and sensitivity checks
Frequently asked questions
Should missing values become zero?
Only when the business definition establishes that absence actually means zero. Otherwise the result changes meaning.
Can a mean rise without a real increase?
Yes, if the supplied records change. Compare population and coverage before interpreting the trend.
Official references
References consulted: . The method and worksheet propose checks to adapt to your context; they do not constitute certification.
Continue exploring
Related analysis
