What’s up guys and gals, CaptainPlanet here to not present the Overwatch Hero Tier List and Meta Report. I was out of town last weekend, and Season 4 just dropped with a plethora of hero balance changes so the meta of the Alienware Monthly Melee and Carbon Series no longer applies! Instead, it’s looking like we’re headed for a second Omnic Crisis: Bastion escaped the PTR without any tweaks to his proposed rework and players like Seagull are already wreaking havoc on the ladder.

While the meta sorts itself out, I have something else to share with you all. After recording professional tournament match results for Seasons 2 and 3, I now have enough data to do some actual statistical analysis. What follows is a study that investigating a phenomenon most players from professional to casual agree on: there appears to be a correlation between capturing the point first in a King of the Hill match and winning the sub-map.

First, let’s define the data, the limitations, and the assumptions of the study.


Things this data is:

I have collected King of the Hill match results from professional tournaments dating back to the beginning of Season 2 through to the end of Season 3. In addition to overall team winrates, hero usage, player names, team names, etc, I also recorded which team captured the point first, and which team won each sub-map. I then calculated how often the team that won the first point ultimately won the sub-map.


Limitations:

I did not record every single tournament match that occurred during this time period, each week of data should average roughly one tournament’s worth. Not every tournament had the same amount of KotH matches, nor did every tournament have the same teams playing in them. Some maps are more represented than others. Meta is not taken into account. Team strength isn’t taken into account either, mostly because I’m not sure how you’d quantify that anyway.

There’s also a personal limitation: I haven’t done any “real” stats (you know, with math and stuff) since college so there’s a decent chance I don’t know what I’m talking about. The analysis I perfomed seemed pretty straightforward to me, but I’m sure if I made a mistake a horde of actual statisticians will be sure to correct me. -_-


Null Hypothesis:

There is no correlation (winrate after first point capture = 50%) between capturing the point first in King of the Hill sub-maps in Overwatch and winning (or losing) the sub-map.


Alternative Hypothesis:

There is a correlation (winrate after first point capture > 50%, or winrate after first point capture < 50%) between capturing the point first in King of the Hill sub-maps in Overwatch and winning (or losing) the sub-map.


Assumptions:

I am going to assume that this data…assumes…the form of a binomial distribution, and thus a binomial proportion confidence interval can be calculated. From Wikipedia (relevant portion highlighted):

Wiki_1.png

To marry this concept to my dataset, each King of the Hill sub-map is its own trial of an experiment, where each trial of the experiment has two possible outcomes: the team that captures the point first wins the sub-map, or they don’t. Assuming that the outcome of a prior sub-map has no impact on the next*, we can also say that the trials are statistically independent. The observed binomial proportion for this study will thus be the winrate after capturing the first point – the fraction of sub-maps where the team that captured the point first ultimately won.

* You could argue that there is a psychological impact to being up or down 2-0, or that after the first map teams know each other’s rough team composition. These influencing factors are hard to quantify, however.


Results/Methodology:

Calculating the winrate of teams that capture the point first gives us a value of 65.44% – noticeably greater than our null hypothesis of 50%. However, how much can we trust this statistic? Perhaps if we ran this experiment many times, the average winrate of teams capturing the point first would be a range – a range that includes the null hypothesis. To determine whether we can reject the null hypothesis, we must now calculate the binomial proportion confidence interval using a normal approximation. Once again, to Wikipedia:

wiki_2.png

Since Wikipedia conveniently provides the Z value for a 95% confidence interval (1.96), let’s start there. My dataset has 616 total KotH sub-maps played (n), 403 of which resulted in wins for the team who captured the point first (n_s). This leaves us with 213 sub-maps that resulted in losses for the team that captured the point first (n_f), and now we have all of our variables to plug in:

CI_1.png

For fun, let’s also calculate a 99% confidence interval (z = 2.58):

CI_2.png

Now, on to discussion!


Discussion

There are two….ish possible outcomes of this study: accepting the null hypothesis, rejecting the null hypothesis and showing a positive correlation, and rejecting the null hypothesis and showing a negative correlation. Examining our confidence intervals calculated in the Results section, we can state that the 50% null hypothesis value falls outside the 95% confidence interval for our parameter (0.6167 - 0.6918), and also for the 99% confidence interval (0.6048 - 0.7037). Therefore, it is not plausible that the observed winrates of teams that capture the point first conform to the null hypothesis (50%), and we can reject it. In this individual study, the calculated winrate was 65.44%, however this does not necessarily mean the true winrate of teams who capture the point first is 65.44%. Using data from this study, however, we can say with 99% confidence that true winrate falls between 60.48% and 70.37%.

** Current meaning no changes to map design in this case – I couldn’t fit that in here


Further Analysis

Here, I plotted the % First Point Capture Rate (how often a specific team captures the first point) on the X axis and the % First Point Conversion Rate (how often a specific team converts first point capture into a win) on the Y axis.

tea.png

Direct Link

You can also see a reference line for 50% conversion rate, as well as the overall average conversion rate (65.44%). Notice that the average conversion across all teams is noticeably greater than what the rate would be if there was no relationship between first point capture and winning the sub-map – which we can expect given the confidence interval calculations above. In many cases, we can also see that even teams that have a lower chance to take the first point in the first place still have a very high chance to win the sub-map when they manage to: NRG, LDLC, and Luminosity for example. Some teams were more likely to take the first point often, and then go on to win the sub map: teams like Rogue, EnVyUs, Luxurywatch Red, and NiP stand out. These teams would be prime examples of skill skewing the results somewhat, perhaps something to consider for future studies.

I’ve also included charts with the same X and Y axes, but split out by sub-map and by tournament.

Sub Map Split

a.png

Direct Link

Tournament Split

n.png

Direct Link

These are more for those who are curious, but we can see that outside of a couple of individual tournament days, the majority of tournaments and every sub-map available had a positive correlation to first point capture subsequent wins.


Why is this happening?

Ultimate charge economy. It’s really as simple as that: whichever team wins the first engagement (we’ll call them Team A) on a King of the Hill map tends to take the point first and generates far more ultimate percentage than their opponents in the process. Team A’s ultimate economy advantage then spills over into the next fight, where they will have a positional and economic advantage over the attacking Team B. This second engagement won, and point percentage having climbed above 60-70%, Team A will likely lose control of the point as Team B finally charges their own ultimates to full. However, Team A must only bide their time and re-charge once more to take the point a final time and can play conservatively. Team B must not only win the next two or so engagements, but must also stave off the usually multiple-ultimate combo final push to have any hope of winning the match. The snowball effect is real on King of the Hill.


What’s next?

Pros and casual players alike have already noticed this phenomenon, but now there’s some statistical weight behind it. The pros have already figured out how best to abuse the ultimate economy to ensure they win the point first as often as possible: pick heroes that charge their ultimates quickly and have the potential to even charge and use an ultimate during the first fight. Heroes like Tracer, Genji, and Winston fit the bill and often see higher usage on king of the hill maps for this reason. Sombra has also been played lately because her EMP can charge very quickly, provided her team uses hacked health packs at a decent rate. Even more ambitious teams will use Mei as well: her ultimate charges rapidly if you can hit your icicles, her Ice Wall is extremely effective on Nepal Village, and Blizzard is a perfect ultimate for King of the Hill maps. Due to the close-quarters nature of most sub-maps, strong Zarya players can also build enough ultimate charge to get Graviton Surge during the second fight, shutting down any hope for the attacking team stealing the point. Heroes with slower ultimate charge rates (Reaper), ultimates that charge quickly and don’t fit the map type (Symmetra, Torbjorn, Widowmaker) or ultimates take too much setup (McCree) are generally avoided.

So what can you, the probably non-professional player do to improve your king of the hill winrates? While I cannot say for certain that these results hold true farther down the ladder, I still try to emulate the pros whenever I can. I would try to stick to heroes that charge their ultimates as quickly as possible and make sure to advise your teammates to do so as well. If you have a regular group of players that you play with, try to come up with strategies, lineups, and combos that maximize the potential of winning that first fight – then make sure to use your ultimates in unison on the second round of fighting as well. When the other team inevitably wins an engagement thanks to many fights’ worth of banked up ultimate charge, be patient! Re-charge your own ultimates and use them all at once to take back control for the final point percentage you need to claim the sub-map. Or, just play the new Bastion.


 

Until next time,

 

CaptainPlanet