Re: 48-Hour Consensus Call: InstateLongdesc CP Update

I know that in the past people have questioned Ian Hickson's study of a large corpus because it was not possible to independently reproduce the results.

 - Maciej

On Sep 19, 2012, at 10:00 AM, Leif H Silli <xn--mlform-iua@xn--mlform-iua.no> wrote:

> Maciej, e.g. Janina did not reject Steve's findings. She only questioned their relevance. It would be more interesting - now - to conclude about: how to interpret the misuse and why it is negative or does not matter. Leif 
> 
> ------- Opprinnelig melding -------
>> Fra: Maciej Stachowiak <mjs@apple.com>
>> Til: bhawkeslewis@googlemail.com
>> Cc: joshue.oconnor@cfit.ie, xn--mlform-iua@xn--mlform-iua.no, silviapfeiffer1@gmail.com, faulkner.steve@gmail.com, john@foliot.ca, rubys@intertwingly.net, public-html-a11y@w3.org
>> Sendt: 19/9/'12,  18:43
>> 
>> 
>> On Sep 19, 2012, at 3:11 AM, Benjamin Hawkes-Lewis <bhawkeslewis@googlemail.com> wrote:
>> 
>>> On Wed, Sep 19, 2012 at 11:07 AM, Joshue O Connor
>>> <joshue.oconnor@cfit.ie> wrote:
>>>>> I did not have time too look through it, but those I looked at either
>>>>> contained only a "#" or they contained (another) image file. With
>>>>> regard to the first (#) then I agree "misinformed" about the potential
>>>>> negative effect. With regard to image URLs inside @longdesc, then there
>>>>> are image light box solutions - libraries - that  more or less
>>>>> consciously makes incorrect use of longdesc. (Today they would perhaps
>>>>> picked at @data-foo attribute instead - but that was not 'valid' then.)
>>>>> Of the few I scanned, no one contained text.
>>>> 
>>>> 
>>>> Yikes, maybe it is the former Silvia. Thanks for doing that Leif. It does
>>>> therefore sound like an inappropriate sample population or at least
>>>> partially so.
>>> 
>>> How does the reason why longdesc was misused make it in an
>>> inappropriate sample population for client software developers trying
>>> to make a decision about whether to expose longdesc via UI to their
>>> users?
>>> 
>>> (My problem with these approaches to sampling is that randomly
>>> sampling the web corpus doesn't match the pattern of usage by typical
>>> users, it just tells you about long tail effects, so the relationship
>>> with user impact is unclear.)
>> 
>> Some browser vendors (including Apple) have the ability to gather data on real-world usage as actually observed by users. Generally for privacy considerations we cannot log individual URLs. But we could log data such as:
>> 
>> - What proportion of images have a longdesc attribute
>> - What proportion of those images have obviously wrong longdesc URLs (empty, #, appears to be an image, top-level URL of a domain, url of the same page that contains the image, etc)
>> 
>> Would folks see such data as more credible? It would be significant effort and we could not reveal the raw numbers. I suspect many would reject such data as not publicly reproducible.
>> 
>> Regards,
>> Maciej
>> 
>> 
> 
> 

Received on Wednesday, 19 September 2012 17:06:11 UTC