Sampling in Google Analytics: How to avoid it

Google Analytics Data Sampling: Everything You Need to Know

You can use Google Analytics to find and understand your target audience and web pages. In Google Analytics, data sampling has become quite popular and it has had direct impacts on a lot of data. Data sampling in Google Analytics involves comparing a single dataset with other data sets. Google Analyzes a portion of your site data to help you quickly generate an actionable report based upon conversions. How can Google Analytics be used in the future? Continue to read!

Advanced Google Analytics users will have experienced sampling before. Especially when it comes to working with custom dimensions.

What is data sampling?

Let’s say in your company you have 250 employees and you want to know how many below the age of 30. Instead of spending hours counting them one by one, you can randomly take some out of the bag and check how many of them fall under your criteria . Based on that sample, you can easily estimate the number of people aged 30 or less.

That’s a rough description of sampling.

In statistical analysis, data sampling means taking a small slice of the whole dataset and analyzing it for trends or for verifying hypotheses.

Since Google Analytics is the first used web analytics tool around the world, it has to process and handle huge volumes of data relatively quickly. For the purpose of speed and accuracy, Google randomly samples a portion of your traffic data.

What is data sampling,

The biggest advantages of sampling are, of course, time-saving and cost-saving. Google can deal with a much smaller and manageable sample yet still produce similar results.

But if you’ll be getting the same result, why waste the time with tangled and huge unsampled data instead of sampled?
Here is why?

Why sampling can misguide decisions making from your data?

Remember the personnel example?

In some cases, your calculations can go wrong: The sample size is too small.

For example, if you only count 5 of 250 people. The younger employees are unevenly distributed.

For example, you accidentally included more younger employees in the sample because they clustered together. In both cases, the pattern does not represent the entire picture. That’s the problem with sampling – it creates uncertainty and mistrust in your reports. While smaller datasets are easier to work with, it doesn’t give you statistical significance. Your sample may or may not reflect the true nature of the data.

Suppose you run two campaigns – A and B. Campaign A has a conversion rate of 10.5%; Campaign B has a conversion rate of 8.3%. The results seem clear, with Activity A the clear winner.

In reality, however, the sample you analyzed may not be representative of the entire population, and there is no discernible difference between the two activities.

This ambiguity is the opposite of how we expect analytics to work. The only reason we decided to use Google Analytics in the first place was to get accurate numbers about our traffic and users.

In the following we will guide you through multiple ways on how to avoid Google Analytics data sampling in GA4. Continue reading!

What are the ways to avoid Google Analytics Sampling In Looker Studios?

Sampling is one of the biggest obstructions that most data analysts and marketers have to deal with. Google Analytics, although a great tool for marketers, unreliable data is something that all marketers should avoid at any cost.

In this post, we will discuss how sampling affects your data. Besides, you will also see how Windsor.ai can ease your sampling pain in the Looker Studio.

Sampling with Google Analytics

Defining samples in data investigation is simply the practice of analyzing a subset of available data to highlight the most meaningful information present in the projected data set. GA4 applies sampling to reports, Explorations, and API queries once the number of events used exceeds your property’s quota. Especially; when you are attracting a large number of visitors each month.

Default reports

GA4’s standard reports are less prone to sampling than custom Explorations, since they’re generally pre-aggregated. Filtering a large dataset by country can still trigger sampling in standard reports, though, even under the quota, since filtering activates a different data-processing method.

Custom reports

When you start modifying your default reports or build custom ones (Explorations), sampling might affect your data. Hence; you should be more careful when you are

• Applying table filters
• Generating custom reports
• Applying custom segments
• Adding secondary dimensions

In the mentioned cases above, sampling might affect your data set.

Sampling Thresholds In GA4

GA4’s quotas are based on events, not sessions. Standard GA4 properties can query up to 10 million events before sampling kicks in. GA4 360 properties get an initial default of 100 million events per query, expandable up to 1 billion events by selecting “more detailed results” in Explore.

In general, reducing your date range is still the most reliable lever if you’re hitting the quota, since a shorter range means fewer events to process.

Native data connector for Google Analytics

A beautiful data visualization can be built in a Looker Studio. The native data connector for Google Analytics works impeccably; when there is no need to worry about sampling. However, when you compare the Google Analytics reporting environment with API, the same data sampling challenges apply to Looker Studio. In brief, analyzing google analytics data in a data studio is not an answer to your sampling challenges.

Sampling effects

For a better understanding of how sampling affects your data, you may set up a test to compare sampled vs unsampled data. A small sample and low values on specific metrics may lead to bigger inaccuracies in your data.

It Is recommended to use Windsor.ai to find out how the different data sets are affected by sampling.

Case study: e-commerce site

Ecommerce, lead generation, and also services websites have to deal with sampling. Some companies often suffer selecting multiple years of data. But it won’t be a great threat as others can analyze data sets for seven days or even less. It is a bigger issue when we want to achieve trend analysis for a longer period.

Now, I will discuss a short story on how I used Windsor.ai to deal with sampling for one of my big clients. The configuration was done even before the new connector was introduced, which we will explain later.

Background

E-commerce companies function internationally and have millions of visitors. Although it was a popular and profit-making online sale platform, it didn’t want to convert to a higher-tier analytics package. The company wanted to create a data studio dashboard for an easy track of the e-commerce performance. Overall, it is not difficult to get unsegmented data in the data studio. All that’s needed is just connecting the native Google Analytics connector to receive all the metrics and dimensions required.

Besides, they also desired to get an in-depth view of the goings-on of their e-commerce business. It is the point where we get into sampling challenges. A few segments are like

• Visitors who show specific interest in buying a product
• Visitors who are showing specific interest in repairing a product
• Visitors who are navigating to the store locator page (it is a sign that they are more interested in offline buying)

As you may have already guessed, these reporting needs and sections have led to data sampling challenges. Hence, we decided to extract a basic set of metrics daily on the channel level.

Solution

We together discussed the different options to tackle the sampling challenges of the company. Also, I used various functionalities of Windsor.ai as well as Google Sheets to solve the challenge.

Try Windsor.ai, free forever plan, no credit card

Google explains sampling in GA4 this way

Data sampling is the data-analysis practice of analyzing a subset of data in order to uncover meaningful information from a larger data set. The practice enables you to retrieve data more quickly with minimal impact on data quality.

Google’s current stated limits for GA4 are:

The quota limit for event-level queries is 10 million events for standard Google Analytics properties and up to 1 billion events for Google Analytics 360 properties. Google Analytics 360 properties have an initial default of 100 million events per query, to provide faster and directionally accurate results. When increased accuracy is required, the data quality icon lets you access the higher sampling limit in Explore by selecting “more detailed results.”

Now let’s have a look at what issues it causes and why relying on sampled data is such a problem when doing data analysis.

How to identify it and why is it a problem

In GA4, reports and Explorations show a data quality icon next to the results. Hovering over it tells you the percentage of data used to produce the report, 100% means no sampling was applied, anything lower means you’re looking at an estimate.

Now why exactly is this a problem?

A report built on a small percentage of your data will not tell you the true story of what is happening and will almost certainly be too small for even looking at trends.

If you are looking at a sample rate of 50%> it may help you to analyze demographics of your audience or similar high-level insights, but definitely will not help you if you want to do any kind of comparative analysis. If you make decisions based on sampled data you basically work with inaccurate data. These decisions can lead to:

1. A loss of trust in the data and risk of the reputation in the data/marketing team
2. A financial loss for the company as you make budgeting decisions based on incomplete data

So let’s explore the options you have to avoid sampled data.

Sampling: How to avoid it when working with data

Option 1: Work with standard Google Analytics reports

GA4’s standard reports are pre-aggregated and less prone to sampling than custom Explorations. If you look at top level metrics this will be the way to go. Chances are however, that this will not suffice for you. Especially when you made it as far that you have hit the quota in an Exploration, I doubt that looking at these top level reports will bring you one step further ;-).

Option 2: Use short date ranges

Another way to avoid sampling is to use a short date range. If you reduce monthly to weekly or even daily, the sampling will at one point disappear. This approach might work to look at very short date ranges but makes analysis of longer date ranges hard as you would need to export the reports into Google Sheet documents or CSV files and then somehow patch it together (which is a time waster you should probably avoid).

Option 3: Export to BigQuery

GA4 includes a free, native daily export (or streaming, on GA4 360) of your raw, event-level data into BigQuery, unsampled. Once it’s there, you can query it directly with SQL, or connect it to Looker Studio, Power BI, or a tool like Windsor.ai without the query-level quota limits that apply to the standard reporting API.

Option 4: Use Windsor.ai

Windsor.ai connects to GA4 the same way most reporting tools do, through the standard GA4 Data API, so very high-volume properties can still hit the same event-quota limits described above through that route. For those cases, connect your GA4 BigQuery export (see Option 3) as a Windsor.ai data source instead, and you get the same unsampled event data blended with the costs from your various sources (Google Ads, Facebook Ads, Bing Ads, LinkedIn Ads, Salesforce, Hubspot, GA4 …) and makes it available for you to work with in raw format via API, Looker Studio, Microsoft PowerBI or our own dashboard.

The steps to get started for free are:

1. Connect your Google Analytics and your costs data here
2. Load data for the date range you need, checking the data quality icon in GA4 if you’re close to the quota
3. Setup your dashboard in the platform of your choice (links above) and analyse data
4. (Optional) Connect your GA4 BigQuery export for full, unsampled event data, or customize the setup to connect your Google Analytics Tableau data with your CRM or e-Commerce data or enable pulling and visualization of custom dimensions from your Google Analytics setup

Conclusion

It depends greatly on your technical abilities and your wallet to what option you choose. The most important takeaway which I’m sure you understood by now is that making decisions based on sampled data leads to many problems.

If you have another way of tackling this problem feel free to share it with us.