Important Stats of Each Hero

Last time, I posted about the important 'Team' stats for winning a map (https://owimt.blogspot.com/2021/01/overwatch-league-owl-provides-player.html).

Here I evaluate the importances of the stats, specific to certain heroes.

Hero Time Played per Map. Error bars mean s.e.m.

This is the hero time played (absolute time per map) of the latter half of the 2020 season. We can see that Genji was played the most. It is understandable if we think the Genji's OP updates (https://playoverwatch.com/en-us/news/patch-notes/live/2020/06). Then let's take a look for Genji first.

In the previous post for team stats analysis, I applied normalized values to compare each team stat in standard manner. However, in case of a specific hero, it is hard to apply this normalization method because the hero usage could be asymmetric (e.g. A team used Genji but B team didn't). Therefore, here I applied stat values 'per 10 min' to standardize them over maps and players. 

It is obvious that common stats which can be applied to 'All Heroes', such as 'Final Blows', 'Deaths', 'Objective Kills' are as important as team's average stats analyzed in the previous post. The purpose of this post is extracting important stats of specific hero, so I removed the common stats and analyzed only the unique hero stats of a specific hero (e.g. 'Damage - Swift Strike' is a specific hero stat of Genji, while 'Deaths' is a common stat). I used logistic regression model with targets of map-win and map-lose.

There are 24 hero stats for Genji (including Time Played for calculating 10min average). As a lot of people may agree, 'Dragonblade' is the strongest and important skill for any Genji player. Also the Importance (absolute coefficient value) shows that 'Dragonblade Kills' is the most important stat for Genji to win a map. 'Quick Melee Hits' and 'Damage - Swift Strike Dragonblade' followed. 

As you can see in the distributions below, the winning Genji (orange) has more 'Dragonblade Kills' than losing Genji (blue)

Also, the second important stat, 'Quick Melee Hits' shows slightly different distribution between the winner and the loser,

I believe the reason 'Quick Melee Hits' was shown as an important stat is because the most common but crucial skill combos of Genji, 'shuriken-dash-melee' or 'dash-shuriken-melee' both use the quick melee to finish off their opponent for the final blow.

Then who made the most 'Dragonblade Kills' as Genji?

In aggregate numbers for the latter half of the 2020 season, Yaki (151) made the most 'Dragonblade Kills'. Doha (111) and SP9RK1E (106) followed. As for NYE players, Haksal made 74 kills and WhoRU made 12 kills. 

How about if we look at the stats per 10min?

I removed players who played less than 3600 s in total. Interestingly, SP9RK1E got first place on 'Dragonblade Kills/10min' with 7.91 kills/10min. There is a huge difference with the second place, Rascal (6.34 kills/10min). As for former NYE players, Haksal made 4.72 kills/10min and WhoRU made 2.75 kills/10min.

Now let's take a look at Mercy.

Interestingly, in this model 'Blaster Kills' and 'Damage - Weapon Pistol' was the most important stats for Mercy for winning a map (note that 'Offensive Assists' or 'Defensive Assists' were not included because they are common stats). 

Then who earned the most 'Blaster Kills' per 10 min?

Leejaegon (1.6 kills/10min) and tobi (1.41 kills/10min) got first and second place, respectively. This is plausible if we think that they are well known 'aggressive' Mercy players. A former NYE player, Anamo made 0.45 kills/10min.

Please note that common stats were not considered here, and I only considered hero's specific stats. Thus, it might be impossible to say that these specific stats are the most 'critical' among all stats, but we could say that these specific stats were important among the specific hero's stats.