Don't know if you've seen this. Very interesting, certainly worth considering.

Strength hero for every Doob?

http://www.pcgamer.com/2012/10/18/dota-2-data-yields-ideal-team-composition/?ns_campaign=article-feed&ns_mchannel=ref&ns_source=steam&ns_linkname=0&ns_fee=0

Ranged heroes tend to dominate lanes harder, and at least this is my opinion. In the new metagame if you can dominate your lanes especially hardcore then you can win the game if you keep applying pressure be it a split push or what have you.

Statistics of games are kinda useless for determining what's good for multiple reasons.
First of all, the vast majority of all games are played at a relatively low MMR, i.e. the normal skill bracket. Thus, what wins games, statistically, is not necessarily what's good or powerful, but rather what is easy to execute. For example, Lich's win rate is ~50% simply because he is easy to understand and has a powerful ultimate even if underfarmed. Anti-Mage or Morphling, despite previously being extremely powerful, only won around 45% of the time due to them being difficult to play.
That being said, disablers and initiators are always good in any skill bracket. So to sum up, if you want to win, pick heroes that you know how to play, and that are easy to execute well for you.

I'd have to agree with you and disagree with you gambit.

Statistics are useful for getting general ideas, they are by no means set in stone.
Things like oh dagon 5 on phantom lancer has a 90% winrate, does that mean its viable? probably not. Those statistics are sort of trollish or are ones that you'd never use.

But statistics like who's being played the most or being banned the most are good to figure out trends in picks.

Like before 6.75 there was the holy trinity of bans. Based on statistics and just all around game knowledge, in competative play, a rediculous majority of dire teams banned naga lycan and darkseer? is it because they didnt want to play against them, were they too strong? or were they just generally accepted heroes that people didnt want to play against. It practically became the norm before the international, but once the international came you didnt see that banning combination, you saw something else.. so those statistics were thrown out the window and hence useless.

Another example of statistics being beneficial:

Looking at certain heroes win rates against another hero. like specific hero to hero win/loss %'s. This isnt the be all end all of oh he hard counters someone else. but its the building blocks for theory craft. saying why is that the case?

Now it is true that a fraction of 1% is the top tier competitive play, but that trickles into the pub scene, everyone has their favorite teams and players, and thus their favorite heroes because they see their favorite pro's playing a certain hero. So in a sense that top 1% of play influences the rest of the 99% of play in a way.

I do definately agree with you in saying that theres hero by hero cases of heroes being hard to play, theres a reason morph and meepo have the lowest win rates in the entire game. They have a HUGE learning curve for the hero, its why like you said lich, lion, veno, all those heroes like that have a higher win rate, they have a more shallow learning curve and thus are easier to play super effectively!