Statistics and averages in reports

Data Center This page documents Time in status, Timer, Stopwatch and Calendar fields for Jira Data Center and Server. On Jira Cloud? See Time in Status (Forge) for Jira Cloud.

“Your average is wrong” is almost always “your average answers a different question than I assumed”. This page states which question each aggregate answers, so a number in a report can be defended in a meeting.


The aggregates

A column in a table report shows a field plus a function that reduces the group to one number. For Time in status, Timer and Stopwatch fields these are:

Function What it returns
SUM Total time across the group.
COUNT Number of records in the group.
MAX Largest value in the group.
MIN Smallest value in the group.
AVG Average time - see How the average is computed.
AVG_PAUSE Average paused time. Paused time is a quantity the app keeps, not a gap it forgets - see Stopwatch events.
percentiles The value at a given percentile of the group, median included.

How the average is computed

AVG = sum ÷ the number of records with a non-zero value.

Records whose value is zero are excluded from the average and from the percentiles. They are not counted in the divisor and they do not sit at the bottom of the distribution dragging the median down.

Why it is defined that way

In a group selected by JQL, a zero is almost never a real measurement. It is an issue that never entered the tracked status, a stopwatch whose start event never fired, or a timer for a workflow path that does not apply. Counting those as “zero minutes spent” would answer “what is the average over every issue in the filter, including the ones this never happened to” - which is a number nobody wants and everybody misreads.

Excluding them means the average answers the question people actually ask: “when this happened, how long did it take?”

This is the number one source of "the report disagrees with my spreadsheet". A hand-computed average that divides by every row in the group is answering the other question, and the two only agree when no row in the group is zero.

What to do if you want zeros counted

Then the zeros are meaningful in your case - which means the group is the wrong shape, not the average. Narrow the report JQL to the issues the measurement genuinely applies to, and the zeros disappear from the group instead of being averaged into it. That also makes COUNT a number you can quote: “how many issues this happened to”.


Percentiles and the median

Percentiles use the same rule: zeros are excluded, and the percentile is taken over the records that have a value.

Reach for them when the average is being pulled around by a long tail. On time-in-status data the tail is usually real - a handful of issues that sat in a queue over a holiday - and the median or the 90th percentile describes normal service far better than the mean does.


The report format applies from the next generation

A report is generated at a point in time and stores that result until it is generated again. The display format is part of the report’s own settings, and the same rule applies to it:

  • Changing the format setting does not restyle a report that has already been generated.
  • The new format takes effect at the next generation - the next scheduled run, or a manual Generate.

There is no retroactive re-render. If you changed the format and the report still shows the old one, generate it; do not go looking for a second format setting somewhere else.

The same "generated at a point in time" rule is why the report's CSV export returns the last generated snapshot rather than recomputing on read. See Field value format and export for BI.

Formats themselves - and their minimum steps, which decide whether a small value renders at all - are listed under Time formats.


The Jira statistics gadget is a different mechanism

A dashboard statistics gadget built on one of these fields does not use the report aggregates. It groups issues into duration buckets and counts how many issues fall in each one - [< 1min], [1min - 10min], [10min - 1h], [1h <].

It buckets issues by the indexed value, while the report computes the value when it is generated. For a running counter those are two different numbers - see Live value vs. indexed value.

  Table report Statistics gadget
Produces One aggregated number per group A count of issues per duration bucket
Groups by A field or a set of JQL conditions Duration ranges
Zeros Excluded from AVG and percentiles Fall in the lowest bucket like any other value
Refreshed On generation - schedule or manual With the dashboard

The two answer different questions and will not match each other, by design. Comparing the gadget’s distribution against a report’s AVG and concluding one of them is broken is a category error.

The bucket boundaries and the value the gadget counts are set on the field configuration - Time Ranges and Value for Gadget Statistics under Export.


Reading a report defensively

Before quoting a number from a report:

  1. Check what the divisor is. AVG over a group full of zeros is an average over the non-zero subset - state that when you quote it.
  2. Check the generation time. The report shows the last generation, not this minute. Reports lists when each one was generated.
  3. Check the format. A format with a coarse minimum step renders small values as empty - the value is there, the format cannot show it.
  4. Check for restarts. A stopwatch with restartCount > 0 measured several cycles on that issue; see Stopwatch events.

JiBrok time in status Time in status | SLA | Timer | Stopwatch
Migration
Release notes
Tutorials
Settings
Troubleshooting
Work calendars
Time in status
Timers and Stopwatches
Stopwatch
Timer
Additional functions
Issue panels
Reports
Field Helpers
Other fields
API
Reference information