I am seeing a weird thing in my GA data, where the referral traffic conversions are in multiples of 3! And in some of the referral sites, the user numbers are exactly the same as sessions. Not sure if its natural or something is amiss! I am also seeing this from email source as well.enter image description hereenter image description here

  • Can you share more details? How long a date range, have you added secondary dimensions or segments, what exactly do you mean by "in referral sites"? – Reve Nov 28 '18 at 14:58
  • I have used segments ( segment for traffic from the USA) and looking at the source/medium...I can see patterns of data like same number of users and sessions. I have sorted the data in descending based on the highest number of conversions first..If i scroll at the very bottom where the data trickles down to single digit users, its a strange pattern of 3 users for a bunch of referral sites, then 6 users for a bunch of sites.I am looking at a year worth of data – chicgirl Nov 28 '18 at 15:27
  • Thank you for editing you question, by the way - the screenshots made it very clear! – Reve Nov 28 '18 at 20:53

This looks like the effects of data sampling. When you apply segments or add secondary dimensions (with some exceptions in the latter case), GA uses a random sample of 500000 sessions (at the free account level) to calculate the metrics. Since it is using a fraction of the real data and multiplying the numbers by a scaling factor, you'll see repeated values. I consider the repeated values a good reminder that I'm not looking at exact metrics!

For example, if your sampling rate was 33% (meaning 500K sessions was 33% of the total sessions for your date range in that View), GA would count the number of sessions out of that 500K for each referrer, and then multiply each value by 3 to estimate the whole - so every value in the report will be a multiple of 3. If the sample included a disproportionately large number of sessions with a certain referrer compared to the whole data set, the sampled number will be larger than the real one, and it can likewise undercount.

You can confirm whether you're seeing the effects of sampling by checking the badge next to the report title. If it is green, there is no sampling. If it is yellow, sampling (or some other warning) will show if you hover over the badge, and give you the percentage of sessions that are being used to estimate the report data.

Here is Google's document About Data Sampling for more details on when sampling is applied and its effects.

Upgrading to Google Analytics 360, if that is a possibility, would most likely fix the problem - it is still subject to sampling, but the sample is 100 million sessions and so won't take effect nearly as often. Also, standard reports with no segments applied will not be sampled, so if you can get at any of your data in that way you can get the real numbers for any date range.

If you have to have segments or secondary dimensions to see the correct numbers, the only workaround I know of is to export data from multiple shorter date ranges - short enough that they are not affected by sampling - and put it back together again in a spreadsheet. Of course you can do this manually with exporting to CSV or Excel format, or using Python, PHP, or Java, you can pull data down via the Core Reporting API.

In between "totally manual" and "primarily automated" there is the GA add-on for Google Sheets. That also uses the API, but you don't need to know a programming language to use it. If you want to try that, you'll need the Google Analytics add-on documentation and the Dimensions & Metrics Explorer to get the API names of the dimensions and metrics.

Of course whether a manual solution is practical, or a programmatic solution is worth the time and effort, will depend on your specific situation, but those are the options I know of!

  • wow i wasnt aware that data sampling could inflate the data ( if i am not wrong) ! Does this mean that GA is replicating the metrics? How accurate will my data be then in this case?@reve As you mentioned, i can see the data sampling in the report. – chicgirl Nov 28 '18 at 16:46
  • So what is the workaround for this? In case of data sampling, how can i look at a bigger date range? – chicgirl Nov 28 '18 at 16:55
  • Right, it can inflate or deflate, depending on whether the sample has a larger or smaller proportion of the data than the whole. I will edit my answer with more information! – Reve Nov 28 '18 at 20:16
  • Thank you so much Reve...for taking the time out and explaining in detail. This just cleared all my questions. I agree with considering smaller date ranges as thats what i had been doing all this while but this is for the first time i ever saw data in this form. Thank you again – chicgirl Nov 30 '18 at 8:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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