Email content

Heya John,

Different data sources will often show different site speed results. Why is that, and what can you do to interpret the data?

Why are Google’s real user metrics better than lab test results?

This is a typical test result from PageSpeed Insights.

The data from real Chrome users says the LCP is 2.3 seconds, while the lab data says it takes 5.7 seconds to render the page.

Lab tests are run in a controlled test environment, with a specific network speed and CPU speed.

For mobile devices, Google runs lab tests using a bandwidth of 1.6 Megabits per second and a network latency of 150 milliseconds.

This test configuration is a lot slower than the average network connection on a mobile device. In the US, the average bandwidth available is 72 Mbps and the network latency is just 31 milliseconds.

Google’s real user data represents the slowest 25% of user experiences. But the lab data is typically closer to what the 5 to 10% of users with the slowest networks experience.

If you want to test something closer to the average user experience, DebugBear lets you select the “Mobile Fast” test device when setting up a page. Or you can create a custom device configuration based on network data for your country.

Why is DebugBear lab data different from PageSpeed Insights?

If DebugBear and PageSpeed insights use the same network and CPU settings, why are the final test results still different?

A big part of this is that DebugBear data is more accurate. PageSpeed Insights relies on a simulation of a slower network, while DebugBear actually slows down the network before collecting data from Chrome.

PageSpeed Insights collects data on a fast connection and then simulates a slower device. Our Site Speed Chrome extension can surface the original data in addition to the simulated values.

In this example, the Largest Contentful Paint occurred after 1.0 seconds, but Chrome simulated an LCP of 5.7 seconds.

Other differences between site speed tools

However, the way the network is throttled isn’t the only factor that causes differences between tools. Other factors include:

Different test server locations

Different CPU speeds on the test server

Different Chrome or Lighthouse versions