“What proportion of people is that?” an inquisitive mind asked.
The answer is:
Yes, we do support them.
How did we calculate these numbers?
Well, we tried this, but in short, it wasn’t accurate enough. There was enough variance in the data as a result of local and corporate caching, bots, analytics blockers, timing, latency in the disparate logging etc to worry us about the accuracy of the data coming back. Particularly when we’re dealing with relatively small proportions.
What was the solution?
So @tombaromba hacked some code in the GOV.UK homepage (similar to an approach inspired by an experiment Yahoo! Conducted in 2010). We chose this page because of its high volume of traffic and low likelihood of any bias towards a particular user group or demographic.
This code included three images, of which browsers should request two.
First, an image that virtually all browsers would request (the ‘base image’).
We deployed this code and then collected the log data from over half a million visits. I expected that number of ‘base image’ requests would closely equal the combined ‘script image’ and ‘noscript image’ requests.
I was wrong.
509,314 visits requested the ‘base image’.
503,872 visits requested the ‘script image’.
1,113 visits requested the ‘noscript image’.
Which meant that 4,329 visits weren’t requesting either the ‘script image’ or ‘noscript image’. Significantly higher than the 1,113 visits requesting the ‘noscript image’.
Why is there such a big difference?
- page being left between requesting the base image and the script/noscript image
- browsers that pre-load pages they incorrectly predict you will visit
- network errors, especially on mobile devices
- any undoubtedly many more I haven’t even thought about…
So while these are interesting reasons, ultimately the reason why someone doesn’t receive the enhancements is largely irrelevant. What’s important is understanding how many people this is, and now we know.
Is there a trend?
We will look to repeat this analysis on a more regular basis and will share anything interesting we find.
Pete Herlihy is a Product Manager, GDS
Follow Pete on Twitter: @yahoo_pete