How does a sampled number work?

The formula is: estimated total = count in the sample / share of the data read. Google Analytics defines sampling as analyzing a subset of data to find meaningful information in a larger data set. When it samples, it scales the result up, and the data quality icon shows the share of data used.

Sources: Google Analytics Help: About data sampling in GA4.

What does an example look like?

Example numbers: a report reads 20% of the events and finds 1,000 level_complete events. The estimate is 1,000 / 0.20 = 5,000. For one small level with 12 fails in the sample, the estimate is 60, but a few more or fewer fails in the sample move it by 5 each time.

Sources: worked example.

Why does sampling matter for indie PC and web games?

Indie decisions often rest on small groups: one build, one creator link or one level. Small groups are where a scaled estimate is least stable. Google suggests a shorter date range as one way to reduce sampling.

Sources: Google Analytics Help: About data sampling in GA4.

How do I get exact counts in Manalens?

In SQL, count events with uniqExact(message_id), because a client retry can repeat a row. ClickHouse says uniqExact calculates the exact number of different values, and it uses more memory than uniq. Ask AI and chart SQL count player and session IDs with uniqExact as well.

Sources: Manalens SQL explorer guide and Ask AI documentation; ClickHouse documentation: uniqExact.

Frequently asked questions

When does Google Analytics 4 sample data?

When the events that a report or an exploration needs are more than the quota of the property. Google gives 10 million events as the quota for event level queries on a standard property. A shorter date range is one way to reduce sampling.

How can I tell if a number is sampled?

In Google Analytics 4, the data quality icon shows the share of data that a sampled report used. In SQL that you write yourself, use exact functions. In Manalens SQL, count events with uniqExact(message_id) for an exact count, because a client retry can repeat a row in the raw events table.

Why is a sampled number risky for a small group?

A sampled report scales each event in the sample up to the whole. If a report reads 20% of the events, each sampled event stands for 5. For one small level with 12 fails in the sample, the estimate is 60, and one more or one fewer fail in the sample moves it by 5.

Sources and implementation