Definition
In GA4, a data threshold is a privacy rule applied when a report is viewed. If the reporting identity includes Google signals or modelled data, rows that could identify individuals are withheld, which in practice means small segments over short date ranges. The report shows a warning icon to say that thresholding has been applied.
Three features of the mechanism matter for anyone reading reports. It operates at view time, so the stored data is not changed. It is triggered by the reporting identity and by the dimensions in the report, especially demographic and interest dimensions that Google signals adds. And it is not described in the report's numbers: the only clue is the icon, which is why a figure that looks suspiciously low with the icon showing should be treated as withheld rather than as low.
Thresholds, cardinality and sampling compared
The guide to GA4 reports and dimensions groups three mechanisms that hide data in standard reports. They are easy to confuse because all three make a report less complete than expected.
| Mechanism | What triggers it | What you see | What helps |
|---|---|---|---|
| Data thresholds | Reporting identity that includes signals or modelled data, with small segments | Rows withheld; a warning icon | A wider date range, fewer demographic or interest dimensions, or a device-based reporting identity |
| Cardinality limit | Too many distinct values of a dimension in a day | Values grouped into an (other) row | Fewer distinct values, for example by cleaning identifiers out of page paths |
| Sampling | Explorations that cover more events than the documented quota | Estimated figures, with a sampling icon | Checking the indicator, and the standard report, which is not sampled |
The causes are unrelated, so the fixes are too. Widening a date range helps thresholds and can make sampling worse; cleaning up page paths helps cardinality and does nothing for thresholds.
In practice: reporting identity and mitigations
The decision that most affects thresholds is made at set-up. The GA4 set-up tutorial asks the team to decide on Google signals deliberately: signals add demographic and interest reporting, and the cost is that reports containing those dimensions become subject to thresholds. Because the way signals feed the reporting identity has changed over time, the tutorial asks for the current help page to be read before deciding.
When a report is thresholded, the documented mitigations are to widen the date range, to remove demographic or interest dimensions, or to switch the reporting identity to device-based. The last is a view-time setting, so it can be changed to read a report and changed back without altering the data. The consent mode guide adds a related point: behavioural modelling needs a reporting identity that includes modelled data, and that identity is also one under which thresholds can apply, so the choice is a trade-off rather than a free switch.
Thresholds also explain some differences between tools. The comparison of Google Search Console and GA4 notes that Search Console anonymises rare queries while GA4 withholds small rows under thresholds, so neither tool shows every row and the two lose different ones.
A wedding venue group finds its missing rows
Deerlarn Wedding Venues runs 4 venues and reviews enquiries by age band each week, using a GA4 report with Google signals switched on. For its smallest venue, which takes about 15 enquiries a week, the 25 to 34 and 35 to 44 age rows kept disappearing, and the weekly total for the venue showed 9 enquiries when the booking inbox held 14. The analyst noticed the warning icon at the top of the report. Switching the same report to a monthly range and removing the age dimension restored the rows: 58 enquiries for the month, matching the inbox within 2. The team moved age analysis to a quarterly view for all 4 venues and kept the weekly report to venue and channel only, with no demographic dimension.
Related terms
Thresholds are a feature of GA4Google's event-based analytics platform. reporting and depend on its reporting identity settings. CardinalityThe number of distinct values a dimension or field holds; high-cardinality dimensions, such as page paths with IDs, can exceed report row limits. limits hide data by grouping too many distinct values into an (other) row, a separate cause with a separate fix. Data SamplingWhen an analytics tool calculates a report from a subset of the data and scales the result up, rather than processing every event. estimates figures from part of the data in explorations, again a different mechanism. Consent ModeA Google tag feature that adjusts what tags collect according to the visitor's consent choices. matters because modelled data depends on it, and a reporting identity that includes modelled data is one of the conditions under which thresholds can be applied.
Common confusions
- Thresholded rows were never collected. The data exists; the report withholds it from that view.
- Thresholds and the (other) row are the same problem. The (other) row comes from cardinality limits and is fixed by reducing distinct values, not by changing the date range.
- Standard reports are sampled, so missing rows mean sampling. Standard reports are not sampled; sampling applies to some explorations, with its own icon.
- Thresholds can simply be switched off. The documented route is indirect: the reporting identity, the date range and the dimensions in the report decide whether they apply.
- The data threshold for modelling eligibility is the same thing. Consent mode's behavioural modelling has its own documented data volume requirements; that is an eligibility condition, not the privacy mechanism that withholds rows.
