Heroes of the Storm, and Statistics

I’ve had a bit of writer’s block recently, so I thought I’d try writing on a lighter topic: Blizzard’s Heroes of the Storm (HotS).

I’ve been playing this game for several months now (just under 1000 games played so far), and I enjoy it a lot. Especially since Blizzard instituted some penalties for trolls, my ingame experience has improved dramatically.

But one thing about the game bugs me: the obsession many players have with “the meta”.

For those not in the know: Heroes of the Storm is a 5-a-side team game, where players attempt to destroy the home base of the other team. There’s some defensive structures in the way, so ultimately it’s a game about planning, timing, reflexes, and quick-thinking.

Heroes come in 4 semi-distinct flavours: warriors (who block and control the enemy), assassins (who have high damage output), support (who heal and/or shield their teammates), and specialists (who are not so easily categorized, but tend to be really, *really* good at damaging structures, more than enemy heroes). A good team should be some combination of these things.

The Meta

If you ask 10 different players what “the meta” is, you’ll likely get several different answers. There are two views that I see most commonly:

  1. There is a ‘right’ combination of hero-types, generally 1 warrior, 1-2 support, 2-3 assassins, and *maybe* 1 specialist (pick any 5). As many heroes cross into the other categories to some degree, people are a little flexible on this point.
  2. There are heroes who are objectively good, and heroes who are objectively bad.

While I dislike the first view, I’m going to focus on the second view in this post: how do we know which heroes are good or bad? Players who insist on the objectivity of their view will point to the statistics that we have access to: Hots Logs.

Surprisingly for people who insist on the importance of these statistics, these folk don’t seem to understand how statistics work. Statistics are a general overview of a population. There can, of course, be an issue with how the statistics are generated: in this case, players can submit their game logs to the site, and from that log the scores of *every* player in that game will be uploaded. So long as a sizeable chunk of the playerbase of HotS submit their stats, I think the stats here are pretty reliable.

But what these players want to do is to draw inferences from the win rate of a particular hero (across the population) to the worth of that hero in a particular game, being controlled by a particular player. Which is something that you just can’t do.


Let’s use two examples: Lt. Morales has a win rate, according to Hots Logs, of 42.8% (for the week of Oct 11 to 17, 2015), and this is one of the lowest win-rates of any hero. This translates, to players ignorant of statistics, to Lt. Morales being a ‘bad hero’, and thus should avoided especially in competitive play. However, knowing the statistics for Lt. Morales tells us precisely *nothing* about how well a particular player will do with that character.

The second example: in that same time period, Azmodan (my favourite character) has a win rate of 49.8%, so while not as bad as Lt. Morales is not ‘good’, but merely ‘middling’. However, if you look at my stats for Quick Match over the last 30 days, my win-rate is a significantly higher 63.6%. [see end note] So do I have a 49% chance of winning my next game with Azmodan? Or 63%?

The answer is, of course, not something you can figure out from looking at those stats. To illustrate the point: if I and a group of random folk end up playing against Cloud9, my chance of winning with Azmodan is roughly 0%. If, on the other hand, I manage to play *with* Cloud9 against some random folk, my chance of winning is somewhere close to 100%. Notice how neither of these results depend on the population-based win-rate of Azmodan.

Basically, these stats are merely descriptive, and to attempt to use them prescriptively is to declare to the world that you don’t know how statistics work: the statistics have been generated from across the entire player-base, from people who play well, and people who play poorly. It might be reasonable to infer from these statistics that some characters are more difficult to succeed with, perhaps, but you have no way of knowing 1) how well the player in your team has done with that hero in the past, and 2) how well ye will mesh as a team. And the second point there is vastly more important that player-base-wide stats of heroes: a team that works well together will do better with the 5 lowest win-rate heroes than a team that works poorly together with the 5 highest.

So play well, be constructive, and don’t whine when a player picks “the worst character ever“.

End Note

So yes, I’m comparing stats from a 1 week period across all characters with ‘the last 30 days’ of my own playtime so the timeframe is different, and for Hero League across all characters vs Quick Match on my own. My defense of the first is that these are the only stats available. I don’t have much defense of the second except to say that I don’t play Hero League much, so I don’t have the stats to compare. Moreover, if these stats are “Objective”, then it shouldn’t matter a whit what game mode they are from…

Follow Brian on Twitter!



Leave a Reply

Your email address will not be published. Required fields are marked *