The non-binary state of affairs
However Zakas' claim isn't substantiated by the data he provides. He has made two fundamental errors:
The first is acknowledged in the post's comments, but not expanded on in any meaningful detail. And the second fundamental error is not mentioned at all.
This isn't just about network outages.
The resource must be retrievable at the specified URL. This is within the hands of the web developer, so they are able to get that part right.
It doesn't take a genius to figure out that the above scenarios are all additive, and as the Web continues to further embrace diversity and flexibility, not only in the choice and configuration of user-agents, but also the networking mechanisms being used; progressive enhancement is very much alive as a cornerstone of web development best practices.
Sure, the iPhone has a particularly capable and very functional mobile version of Safari, but the experience of a web site on the device also depends on the stability and speed of the network connecting the browser to the web server.
The numbers Zakas provides are skewed in several ways:
- C-grade browsers (as per Yahoo's Graded Browser support) were excluded; already this strongly suggests the actual numbers are under-reported
- Mobile browsers are redirected away; again another source of under-reported numbers, this time a fairly substantial one
- No numbers from Germany, China, Hong Kong, Australia (data only mentions 4 out of 40-plus localised Yahoo homepages), grey area as to where the Nordics and Dutch visitor data is collated.
- No explicit mention of how they protected the results from 'errors through omission', like browser set-ups that filter out making requests for tracking pixels.
- Major question mark as to how the data reflects third-world countries and people using assistive technologies
The omission of the above scenarios means that the end-totals are likely to be under-reported, and from the detail Zakas has provided there is no practical way of gauging whether his numbers are off by a few percentage points, or something more substantial.
Despite the appearance of a disciplined scientific approach to collating the data, the post fails to follow the typical scientific pattern of clearly documenting the methodology and the assumptions made during the approach.
This makes it difficult to both clearly understand the actual bias inherent in the collected data, and, more importantly, deprives the web community of a way of independently verifying the results, or producing comparable data to isolate discrepancies.
Asking the right question
But if any of the scenarios mentioned above apply, then the modern web development best practice of progressive enhancement is definitely a more applicable approach.
Progressive enhancement: it's the safety net when things outside your control interfere with your website.