1

While I think most of us are familiar with/have encountered discrepancies between GA and Google Ads or other platforms, this one has me rather stumped.

A new client's reporting has always seemed a little bit unusual to me. Until recently, I haven't been able to put my finger on why, but I always felt like the numbers being reported seemed inaccurate. Today a coworker and I discovered that the identical report (in this case, a custom dashboard where data is run through a simple logic filter for source and page url) produced wildly different results. In fact, we noticed at least three different results within a 10 minute period!

We confirmed that our date ranges, views, and segments were identical, but at one point the reports even changed with a simple refresh of the page. We couldn't seem to cause this to happen, but we did notice it on newish data (previous month), and old data (9 months ago). That tells me that there isn't an issue with conversions rolling in last minute.

Thoughts on what could be causing this? One of my first tasks in my new role is to migrate reporting from manual pull to GDS, but I'm nervous about the validity of my data if these results can change so much.

1 Answer 1

1

Depending on how wild your "wildly different" is, this sounds like data sampling.

Standard reports do not get sampled, but once you add a segment (or most secondary dimensions) your data is extrapolated from a random sample of 500K sessions (on the free plan; 100M if you're on 360). This can have a profound effect on ecommerce data; lower-frequency actions are proportionally more affected by the specific collection of sessions that make it into the sample.

If sampling is in effect, the badge next to the report title will be yellow instead of green, and when you hover over it you'll see "This report is based on X% of sessions" in a drop-down panel. (You'll have to go to a report for this; dashboards do not have the badge.)

Sampling happens at the time the report is displayed, which explains the data discrepancies between you and your coworker, and even between page loads.

I recently wrote another answer explaining sampling and ways to mitigate it that might be useful if this is in fact the issue.

5
  • Sampling does seem like the most likely culprit here. To be clear, when I say "wildly different", I am referring to differences in small numbers; at times it showed $1400, $700, $400, or even a very precise $0.00. The zero amount is what troubled me the most, as it seemed that if sampling was the issue, it would surely include SOME of the desired info - right? Additionally, I should point out that this is actually a GA Dashboard, not a Custom Report, so I'm not seeing the badge you refer to. Could this be a factor?
    – Gesundheit
    Dec 6, 2018 at 16:34
  • 1
    @DanielParscale - if only a few sessions included nonzero value for some dimension or combination of dimensions, the sample could potentially miss them entirely and show an erroneous 0. And yes, there's no badge on dashboards - I actually discovered recently that sampling affects dashboards on a per-widget basis! So widgets with only one dimension (and no filters on other dimensions) can be unsampled even if there are other widgets with a setup that triggers sampling. Segments would trigger it on the entire dashboard, though.
    – Reve
    Dec 7, 2018 at 18:40
  • That's very useful information. Thanks Reve for your help!
    – Gesundheit
    Dec 7, 2018 at 18:42
  • @DanielParscale If you adjust a regular report to match one of the widgets that is giving you inconsistent numbers, and apply segments as appropriate, the badge's indication of sampling or no will be correct for the widget as well.
    – Reve
    Dec 7, 2018 at 18:42
  • And you are very welcome! :-)
    – Reve
    Dec 7, 2018 at 18:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.