10-16-15 EDIT: Looks like Safari now supports the HTML5 method necessary for Google Analytics to track page load times! You should now see your 0 data filled in with numbers. 

In Google Analytics, page load time is automatically retrieved by the page tracking snippet within the JavaScript tracking code on every page of your website. But Google has been making some subtle changes in the past year to the process that happens after it receives this data.

To put it simply, they are using a method only available in HTML5 and only supporting certain browsers with that method. This grid lists which browsers are supported. It means Google is only covering 73.57% of the browsers. Unfortunately, the unsupported browsers include a large and growing audience: Safari and mobile iOS Safari. Which means your traffic from Safari users on desktop computers, iPad Safari users, and iPhone Safari users are not being tracked in terms of these metrics: Avg. Page Load Time, Avg. Domain Lookup Time, Avg. Redirection Time, Avg. Server Connection Time, and Avg. Server Response Time.

For some, or maybe most of us, the percentage of your website’s Safari users may be more than 26.43%. The excluded browsers are not “evenly” distributed in any way, so your traffic may include a lot of the other excluded browsers, as well. You think you can just exclude these users and think you have accurate data? Well, no. Not only would you be excluding a large portion of your customers, but those 0’s are ruining the averages for page timing. When you look at Avg. Page Load Time, Google Analytics is creating an average from all sessions, including your 0’s, which is not accurate.

Numbers are not what they seem…

To make matters worse, they actually removed past data in page timing, too. Let’s take a deeper look…

Click for full version

Click for full version

The report we’re looking at here is using all the metrics mentioned, plus displays the Dimensions for Browsers and then Regions for a particular client. Looking at Browsers first, you can clearly see what Safari looks like now in comparison, big fat zeros. Worse yet, if you drill down to versions to some other excluded browser versions, that data is now also a big fat zero.

Click for full version

Click for full version

I’m comparing the same time period, year over year, in this report using the Regional Dimension. We can see a few states have data in one row and not the next, going both forward or backward in time. Strange, right? Especially backwards! Well, because of those 0’s both going forward, and the old data being replaced with 0’s going backward, we can see better how they are changing the averages. Even where numbers are available, there are huge differences in timings for a site that hasn’t changed in a year. If you drill down into one of these states with 0’s and look at the browsers in that region, you can see that there is data. But the data available has more of those excluded users than not, making the averages ZERO. Going back to the overview of regions, the average is using those zeros. This means any singular report on page timing you may be looking at has an inaccurate count in those average timing metrics; aggregated data using a lot of zeros tends to throw things off.

Those (not set) folks in the Region report, by the way, 70% of those are Safari users. All zeros…

Page load speed is so important though…

What to do? You can drive yourself crazy by digging into the data yourself thinking you can get a better average, but that’s almost impossible, because it isn’t an accurate average of your customers. The numbers you use will mainly be Chrome and Firefox users over time, that’s about it, but I know that isn’t a good sample of your customers. I did it for a couple hours the other night, and redoing the numbers I found the percentages in change over time to be wildly different anyway. Larger numbers will of course produce better results – but that only means you are excluding a larger audience, as well.

So, take it with a grain of salt until Google stops removing your data and changing its algorithms when it comes to page load time. In the meantime, I will be finding an alternative tracking method for this measurement. If you have any suggestions, please email me!

Last updated by .

Leave a Reply