Member Avatar for Dani
Dani 5,664 The Queen of DaniWeb

I was impressed when I logged into Google Analytics and noticed that instead of sampled data I was presented with a drop down menu where I could choose if I cared more about accuracy or speed. I was then thoroughly less impressed when I selected to focus on accuracy and still got a sampling that was only based on 10% of my audience. Am I the only one frustrated by this?

  • 1 Contributor
  • 0 Replies
  • 226 Views

Dani AI

Generated

— this is a very common frustration and it’s not a UI bug: the “Greater precision / Faster response” control only changes how large a sample the report will try to use, it does not override the product’s processing quotas or make a sampled report unsampled. . ()

Why that 10% shows up: Analytics will use the largest sample it’s allowed for that property and query, then scale the results up — if the dataset (sessions or events) is much larger than the maximum sample, the report can still be based on a small fraction of the whole. GA4/standard properties use event-based quotas (for example, default sample/query limits are in the millions), and 360 customers get much higher limits or unsampled options. [How Analytics stores and displays data / About data sampling].() ()

Practical ways to get better (or unsampled) results:

  • Shorten the date range or split the date range into smaller slices and stitch totals together offline (this often drops the query below sampling thresholds). ()
  • Use standard (pre-aggregated) reports where possible — those are served from unsampled tables. ()
  • If pulling via the API, request higher precision (samplingLevel / HIGHER_PRECISION or the equivalent) to increase the sample used for that query. (developers.google.com)
  • For truly unsampled, authoritative data: export raw hits/events to BigQuery (or use Analytics 360 unsampled reports / unsampled explorations) and run deduplication/SQL there. BigQuery export or a 360 unsampled request is the only practical way to avoid sampling for large properties. [BigQuery export / Unsampled explorations].() ()

One caution: some “user-level” counts in the UI are generated with approximations (HLL++ sketches) and for exact unique-user or session-de-duplication the BigQuery raw data approach is the reliable path. [About data sampling / HLL++].() ()

TL;DR — not alone, and the slider did exactly what it can: increase the sample up to the system limit. For exact figures on big datasets, split queries or use API/BigQuery (or upgrade to 360).

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.