MetricRaven

Tracking

What is data sampling in analytics?

Data sampling is when an analytics tool computes a report from a subset of the data and scales the result up to estimate the total. If a report reads 10% of 50 million events, every figure in it is an estimate built from 5 million.

Updated

What data sampling is

Counting every event in a big date range takes time and computing power. To answer fast, some tools read only part of the data, compute the report on that part and multiply up. Totals usually stay close. Small slices suffer most: one campaign, one country or one landing page can be off by a wide margin, because it is built from few sampled rows.

That matters most for revenue questions. A campaign with 12 sales in a sampled report might have had 8 or 16. The smaller and more specific the slice, the less you should trust it.

When Google Analytics 4 samples

In Google Analytics 4, sampling may happen when the events behind a report, exploration or API request go over the property's quota: 10 million events per query for standard properties, and up to 1 billion for Google Analytics 360. It shows most often in Explorations over long date ranges, and Google notes that filtering large datasets by country can trigger it in standard reports too. A sampled result shows a data quality icon with the share of data used. Even unsampled reports estimate distinct counts such as users and sessions with HyperLogLog++, so those are very close but not exact.

Google's advice for avoiding sampling is to shorten the date range or use other analysis methods. Many teams export raw events to BigQuery for that.

Example

An Exploration covers a year with 40 million events, so it reads 10 million: a 25% sample. It shows 2,000 conversions from Paid Social. The true number is probably close. Filter it to one ad set with 40 sampled conversions and the scaled figure of 160 could easily be 120 or 200, which is the difference between keeping and cutting the ad set.

How to check for sampling

  • Look for the tool's sampling indicator (in GA4, the data quality icon) and the percentage of data used.
  • Compare a sampled exploration with the standard report for the same total; a gap means estimation.
  • Shorten the date range until the report is unsampled, then add periods together for counts (not for unique visitors).
  • For money, check totals against your payment provider, which never samples. Small slices also make conversion rate swing.

Common mistakes

  • Trusting small slices of a sampled report. Error grows as the slice shrinks.
  • Ignoring the indicator. A sampled report looks exactly like an unsampled one apart from the icon.
  • Mixing sampled and unsampled numbers in one dashboard. They won't add up.

Does MetricRaven sample data?

Traffic and revenue reports in MetricRaven count every recorded event and payment; there is no query quota that switches them to estimates. A few heavy features have limits instead, and each says so where it applies:

  • Paths to payment: above 1,000,000 pageviews in the period, visitors who didn't pay are estimated from a sample, and the note under the list says how large. People who paid are always all counted.
  • Errors in funnel drop-offs: above 10,000 visitors who left a step, the first 10,000 are checked and the count is marked as a sample.
  • Page speed: at most 50,000 snapshots per website per day are kept.
  • Session replay: you can choose to record every matching visit, half, 1 in 4 or 1 in 10.

See Paths to payment, and the GA4 alternative page for how the two tools compare.

Questions

When does GA4 sample data?
When the events behind a report, exploration or API request exceed the property's quota: 10 million events per query for standard properties. It is most common in Explorations over long date ranges. The data quality icon shows when a result is sampled.
How do I avoid sampling in Google Analytics?
Shorten the date range so the query stays under the quota, or export raw events to BigQuery and query them there. GA4 360 raises the limit.
Is sampled data accurate?
Totals are usually close. Small segments are not: a campaign or page built from a few dozen sampled rows can be far from the real number. Treat small slices of sampled reports as rough.
Full glossary

Revenue reports built from every payment

MetricRaven counts every visit and payment in its traffic and revenue reports, and marks the few features that estimate.

14-day free trial, no card required.