It's reasonable to assume that Dvorak and Colemak are about equally optimal – within the uncertainty of the scanty evidence we have. Therefore, it's reasonable for most newcomers to pick Colemak because it means moving and relearning about half as many keys! Some claim that it doesn't matter how many keys you move, but I guess most people will think as I do.

Beyond that, it's a matter of how well implemented the two contestants are. Dvorak still has an edge there, which will matter to some.

I think there must be Colemak users that aren't forum users (and vice versa as we know).

AZERTY is just a mangled QWERTY and shouldn't be counted separately.

You could say that the transitional layouts are just 'mangled QWERTY' too then.  I happen to think them all individual layouts.  How would you define a distinct layout?

I know you have a lot of love for Tarmak, but have you really that much faith?  I still think that new typists would do well with Dvorak.  Qwerty touch typists may have less work going the Tarmak/Colemak route rather than choosing something more radical such as Dvorak.

Anyway it's kind of by the by, tablet sales are beginning to  outstrip laptops/desktops, and as such I think the standard keyboard is likely to fall by the wayside.

I didn't read that post as aggressive.  It was really a concise summary of what I was trying to put forward.

Qwerty was conceived as an optimised layout.  It was designed to be better than an alphebetical layout (well that's my take on it anyway).  So the term optimised layout doesn't really mean much.

Azerty's history seems a little less clear, but it is almost as old as Qwerty, it would be interesting to know a little more about it, but regional variations are themselves micro-optimisations/dialects.

I guess you could make a distinction between modern and typewriter inspired layouts.  Perhaps optimised for the computer keyboard (what a mouthful), or neo-optimised layout would be better.  The latter is probably not worth using as it could soon be lost to the annals of history.

Well, the easiest way of thinking about it is that, given two indistinguishable optimized layouts, the one with the easiest transitioning method would eventually prevail.  Of course, Colemak and Dvorak aren't quite indistinguishable (most importantly in this context, Dvorak has a lot of inertia), so things are a bit more complicated.  Some potentially useful clarifications about my prediction:

1. Long term, here, means really long term:

   1. I don't see Tarmak penetrating the public consciousness in less than ten years (short of very fortuitous circumstances).

   2. Even then, it'll probably take a lot longer for the new Colemakers to actually outnumber the Dvorakers, depending on rates of adoption and the actual (as yet unknown) populations.

2. That said, Tarmak's utility, once it does take hold, is extremely high.  Reasonable predictions include:

   1. Opening Colemak to more people - including many who could not otherwise afford to transition at all.

   2. Capturing most or all of the QWERTY touchtypists who would've otherwise transitioned to Dvorak

   3. After several generations of this, endowing Colemak with a positive reputation for ease of transition (which may even sway non-Tarmak-users due to the bandwagon effect).

How would you define a distinct layout?  Are national variants of AZERTY to be considered distinct? Dvorak and Programmer Dvorak?

While one could use the technically correct (best kind of correct!) definition of "distinct layout", that kind of rigidity just isn't very useful in most analysis.  Instead, "distinct layouts" should be defined to maximize your signal-to-noise ratio.  In this case, listing tons of minor local variations of QWERTY as 2nd, 3rd, 4th, etc tells you little more than that a lot of people live in certain regions, while missing almost everything interesting about OP's question. 

"Mangled" seems to imply that something was screwed up, or at least rearranged in a way that's not an improvement.  I find it hard to apply that term to layouts such as Tarmak 1 and Minimak 1. AZERTY, on the other hand, I'm fairly certain was mangled in most senses of the word.  Even in French, Q still only appears 1.4% of the time, while A (wait for it) 7.6%.  Yes, the same A that AZERTY moves to arguably the hardest position on the keyboard.  I believe you've placed way too much confidence in historical well-intentions.

Hold back!  It's fine to say that Azerty is worse that Qwerty or that you don't like either layout.  But the way I read your comment was that Azerty doesn't warrant it's own classification as a distinct layout.  Or rather it was dumped in the Qwerty camp.  Anyway it's not worth spitting hairs over.

It's hot and balmy, but I have yet to see the storm.

It's made me think about layout taxonomy (did I just say that out loud?)...

The "large Chinese keyboard" I guess you'd be hard-pressed to call QWERTY.  With the others, it still looks as if you're typing QWERTY characters.

You do have a point (in that certain non-Romantic layouts cannot be considered the same as QWERTY), but I think you're also missing my point, which was to maximize the signal-to-noise ratio.  Sometimes, substantively similar data can be combined (as in the case of "mangled" QWERTY) to reduce that noise.  Other times, if data is uninteresting and unsalvageable, you may be forced back into Shai's approach of discarding it entirely. 

I can see how a survey of incomparable-layouts-that-don't-even-use-the-same-characters might serve as an interesting factoid.  From an optimization perspective, however, it's just noise.

Fortunately, most languages' use of extra symbols fall in the low percentages. Therefore, a system like mine where some keys (brackets plus <LSGT> plus some AltGr mappings) are used for these signs will work well enough for many short of creating an optimized layout for that particular language (group). My language, Norwegian, is an example of this: The æøå letters rest comfortably on the aforementioned keys and it's no hassle for typing my language. The special symbols are in my experience less of a problem than some bigrams which are more common in my language ('kj' for instance). If I wanted a layout fully optimized for Norwegian I'd have to go about things differently, but I don't because I don't want to use more than one layout and I type more English than Norwegian anyway.

Other scripts really should be kept out of this I think. The Chinese "QWERTY" layout has a QWERTY setup for the latin letters so it's a QWERTY layout, but that's mainly for ease of recognition. If a Chinese Colemak user wanted this kind of layout it'd be easy to make a Chinese Colemak layout the same way. It looks to me as if the words are placed based on physical key location and not phonetics (but I may be wrong); in the first case I'd just move the latin letters and in the latter I'd move the keys with all their mappings.

The Greek, Kyrillic and Hebrew Colemak phonetic layouts I and others have made are for people who type Colemak and want the other script(s) to follow what they already know. They aren't significantly optimized for typing the other scripts comfortably in a different language, and that's not the point with them.

If we're discussing which is the most-used keyboard layout today and tomorrow, I'd prefer to limit the discussion to latin script and lump together locale variants for the above reasons. For usage statistics it would be fair to include the phonetic layouts for other scripts, but they shouldn't contribute much at any rate. Thus, the **ERT* layouts may safely be counted together in my opinion as they're all part of the old school before optimization set in, various Dvorak flavors and various Colemak flavors including locale ones are best counted together with their "mothership" layouts in my opinion etc. The larger picture is the more interesting here I believe.

From the point of view of optimisation, it's a question of who you're optimizing for and once several languages enter the mix this becomes very individual. In that case, counting flavors becomes fairly irrelevant as it's a different question.

You're entitled to that opinion. But the distinctions I have made haven't been about the result but the methodology. I think it's fairly easy to group layouts into a QWERTY family, a Dvorak family, a Colemak family etc, for counting. You don't have to believe in any of those layouts to do that.