Data Studio
3.2K members online now
3.2K members online now
Ask questions about Data Studio
Guide Me
star_border
Reply

Sample rate in Data Studio - refresh

Follower ✭ ✭ ✭
# 1
Follower ✭ ✭ ✭

I wrote (most of) the below as a reply to an earlier conversation, but since I feel this is a very important topic, I would like to put it to the forefront of the forum.

 

The problem we appear to have is that Data Studio, when fetching data from a GA property, does not make it possible to control the sampling the way it is possible to do in GA. And unfortunately, the choice made in the design is to use the method which gives the least precision, which in many cases make it all useless.

 

In GA I can choose either "Slower response, greater precision", or "Faster response, less precision". Since it is a premium property, I can also choose to get an unsampled report. In Data Studio, however, the results seems to be similar to "Faster response, less precision". Since I have a lot of data, this does not produce anything like reliable results. For instance, I am looking at a subset of devices recently introduced; in GA, with "Faster..." I get 0,11% sampling, which produces results more than 50% off the real figures (activation numbers). In GA, with "Slower...", I get instead 22% sampling, which, while not ideal, is somewhat acceptable.

 

I really, really need a way to control this setting of "Fast.". and "Slow..." response.

 

I have also tried to do an unsampled report, and use that in Data Studio, This gives me good data, but I then get two other problems:

 

1)  If GA creates a new unsampled report with the same name, the old one continues to be used. In some way, it is connected to the specific file. If I, manually, throws the old one in the bin, and permanently deletes it, then Data Studio still does not use the new file with the same name, but instead forces me to delete the old connection, add the new connection, and then, again manually, link each graph or table one by one to the new connection. This makes Data Studio almost useless for data presentations which shall be updated daily.

 

2)  In Data Studio I can set the time span of a graph to some precise date in the future, say 30 september 2016. This means I can be sure that any new data that comes in will be used and all the figures will be updated. That is a very good thing, but this does not work with unsampled data from GA since I cannot set a repeated, unsampled report to start at the most recent date. While that is not a problem in Data Studio, it makes the use of unsampled data still more cumbersome.

 

IF Data Studio could fetch an unsampled report (as stored in the "Customized" section of my property), with the date set to the most recent date, update the connection to the new file automatically, and update all the charts, that would make a world of a difference.

 

Sample rate in Data Studio - refresh

Visitor ✭ ✭ ✭
# 2
Visitor ✭ ✭ ✭

Kjell, I agree. Data Studio should default to using less sampling, the "greater precision, slower response" setting. Google folks, please take note, this is a big problem.

Sample rate in Data Studio - refresh

Visitor ✭ ✭ ✭
# 3
Visitor ✭ ✭ ✭

Definitely a huge problem which i was not aware of up until today. We need this option to get the absolute same numbers in GA and Datastudio. Although my monthly report in analytics isnt sampled (well at least it tells me that it is not), the data in Data Studio is way off there... Cannot use those reports then at all...

Sample rate in Data Studio - refresh

Visitor ✭ ✭ ✭
# 4
Visitor ✭ ✭ ✭

I'm interested in this topic and agree with much of what has been said already.

 

What I would add to the conversation is that when I have cross referenced data with 100% sampling in GA to the GDS figures; they've always matched completely.

 

This typically was for top line figures of Sessions; Users etc that are pre-processed; but essentially if GA claimed "This report is based on 100% of sessions" then GDS showed the same figure.

 

Obviously as soon as you start applying secondary dimensions and advanced filters then the sampling and accuracy between the two platforms starts to vary but I thought I'd just show my experience of the pre-processed figures.

Sample rate in Data Studio - refresh

Visitor ✭ ✭ ✭
# 5
Visitor ✭ ✭ ✭

I believe I have a "semi" answer to this. I was also investigating this as I also had an issue where my GA was not matching my Data Studio. Here is what I found....

 

I found a Google Support doc that states... The sampling rate for the provided data depends (in part) on the date range: when the range does not...

 

I tested this and NOPE that was not true, HOWEVER I did find that if I went back 5 days that it was greater precision, slower response ...

 

So my conclusion is that Google is intending for it to be greater precision when calculating anything that is not today, however, either through lack of provisioning of servers or something else it is not getting to the one day goal target for more precision.

 

Unfortunately like most people I assume, having just a one day lag on precision would be fine, but 5 days is a no go and I am stuck using GA. 

 

I hope this helps.

 

Sample rate in Data Studio - refresh

Participant ✭ ✭ ☆
# 6
Participant ✭ ✭ ☆

I'd love to get fully unsampled data as well, but I think that it's actually an issue with GA rather than DS. For my purposes, I'm not willing to use sampled data, so my workaround is to set up automated downloads into Sheets for those sampled reports and then upload them. This results in many different datasets feeding into the same report, but I have more confidence that the numbers are actually real.

Sample rate in Data Studio - refresh

Visitor ✭ ✭ ✭
# 7
Visitor ✭ ✭ ✭

May I suggest using scitylana.com. Being official Google Analytics partners, they extract raw hit-level data from a free GA account through the API. Data are delivered as local text files or uploaded to a BigQuery account of your choice.

With hit-level data in BigQuery, you can connect with DataStudio and query all dimensions and metrics without sampling og without any 7-dimension limit