Leave, Sickness and the Denominator

An averaged limit divides total hours by a number of weeks. The hours are the obvious part. The weeks are where the errors are, because some weeks are not ordinary weeks and regimes handle them differently.

The calculation in “Leave, Sickness and the Denominator” depends on complete, consistently coded time records rather than a polished total at the end of the month. When considering how to monitor employees without being intrusive for how to monitor employees without being intrusive, teams should verify exports, missing-time corrections and period boundaries against their own averaging method before relying on a dashboard.

The error is nearly always in the same direction: a configuration that treats absence as zero hours, pulling the average down, making a heavy pattern look moderate. It is the one mistake in this subject that actively conceals the thing being monitored.

For an independent reference relevant to “Leave, Sickness and the Denominator”, consult the CIPD working-time guidance. Use it to test working-time definitions, recordkeeping, access, retention and exception handling against the organisation’s real process rather than treating one software report as conclusive.

The three possible treatments

Counted as zero: the week is in the denominator with no hours in it. This lowers the average.

Excluded, with the period extended: the week is removed and an earlier week is pulled in to keep the span at its proper length. This preserves the average's meaning.

Excluded, with the period shortened: the week is removed and nothing replaces it. This leaves the average unchanged in level but computed over fewer weeks.

Which applies to which kind of absence is set by the regime, and it is commonly different for statutory leave, sick leave and unpaid absence.

Why counted-as-zero is so damaging

Consider somebody who works four heavy weeks and takes a fortnight of leave. Counting the leave as zero divides four weeks of heavy work across six weeks, and the resulting figure says the pattern is comfortable.

It is not comfortable. The person worked those hours, and the limit exists to constrain the hours worked rather than to describe an abstract mean. A system that smooths real load against holiday is producing a number that is reassuring and wrong, and it is reassuring in exactly the cases where the reassurance matters most.

The categories to check separately

Annual leave. Statutory sick leave. Long-term absence. Maternity, paternity and related leave. Unpaid leave and authorised absence. Lay-off and short-time working, which in some arrangements produce weeks with hours that genuinely were zero.

Six categories, each potentially treated differently. The configuration screen in most systems offers one setting, which means somebody chose one treatment for all six, probably in the first week of the implementation, probably without reference to anything.

Finding out what yours does

Take one person with a recent period of leave and recompute their average by hand under each of the three treatments. Compare with what the system says.

That identifies the behaviour in an hour. It also tends to produce the first serious conversation anybody has had about this question, because the three numbers are usually far enough apart that the choice is visibly consequential.

Part-time and variable hours

The same question in a different form. Somebody contracted for sixteen hours who works forty in a particular week is at risk of a limit in the same way as anybody else, and their average sits well below it most of the time.

The thing to avoid is excluding part-time staff from the monitoring on the grounds that they cannot possibly be near a limit. They can, particularly where they hold more than one post, and the exclusion is usually done at the report level by somebody filtering on contract type.

Where the long-term absence case bites

A person returning after six months is in a reference period that may be almost entirely empty. Under one treatment their average is close to zero and they have enormous headroom; under another the period extends back into last year and their old pattern still counts.

Return to work is exactly when somebody is likely to be rostered heavily to clear a backlog, which makes this the case most worth getting right. It is also the one least likely to have been thought about when the system was configured.

What to write down

For each absence category: which treatment applies, what the source is, and what the system actually does. Three columns, six rows.

Where the system does not match the source, that is a defect with a date, and it affects every figure the system has ever produced. Finding it is uncomfortable and considerably better than the alternative, which is that the figures have been wrong in a flattering direction for as long as anybody has been relying on them.

Part weeks, which fit no rule cleanly

A person who worked two days and then went on leave has a week that is neither a full working week nor an absence. How the calculation treats it is frequently undefined, and different parts of the same system can treat it differently.

Pick a treatment, write it down, and apply it consistently. The choice matters less than the consistency: a figure computed one way every time can be explained and audited, while one that varies with how the absence was coded cannot be reconciled by anybody, including the person who built it.