# Measure zeros and missing values

Measure the actual values before deciding whether a column's representation needs attention. A zero value, an SQL NULL and an absent row have different meanings and should be counted separately.

```sql
CREATE TABLE sampled_measurements(value INTEGER);
INSERT INTO sampled_measurements VALUES (0), (0), (5), (NULL);

SELECT count(*) AS total_rows,
       count(*) FILTER (WHERE value = 0) AS zero_rows,
       count(*) FILTER (WHERE value IS NULL) AS null_rows,
       1.0 * count(*) FILTER (WHERE value = 0) / nullif(count(*), 0)
           AS zero_fraction
FROM sampled_measurements
FORMAT csv;
```

The denominator includes every row. If the reader task concerns only present measurements, use `count(value)` as the denominator and explain that choice. `nullif` protects the calculation for empty input rather than turning an undefined ratio into a misleading zero.

This distribution describes the selected rows, not on-disk compression savings. Coordinator DB does not document a sparse-serialization tuning switch tied to a zero ratio. Choose a suitable declared type, preserve the meaning of missing values and measure representative query behavior before changing the schema.

See [aggregate functions](https://coordinatordb.com/docs/reference/functions/aggregate-functions), [NULL handling](https://coordinatordb.com/docs/reference/data-types/nullable) and [type selection](https://coordinatordb.com/docs/concepts/best-practices/select-data-type).