Game Pacing and You

Fearless and Riot Boom Bear here! We wanted to talk a bit about how we’ve been thinking about game pacing and where we’re at right now.

Warning: This is going to get a bit long. We’ll have something better… formatted next year, but we wanted to talk about this earlier than later.

A major goal of our preseason changes was to make the game have clear and reasonable expectations on teams as to what they need to do to win. For a long time, some of the actions that were most effective at winning (starvation tactics) weren’t very intuitive or engaging for either team. Thus began our efforts to cut down on starvation and replace it with more interactive paths to victory.

Starvation actually has a few different permutations depending on the level of play, but in each case (by definition) it leads to long stalls in game progress. Both teams essentially stop taking meaningful action. Winning teams are happy to build a farm lead, gobble up safe objectives, and elude any risky team fights. Losing teams, on the other hand, have no choice but to slowly starve out, unsure of how to take action or attempt a comeback. We’ve been wanting to balance out these situations for a while now, but up until recently we didn’t have confidence that we’d be able to make the right changes while understanding their impact.

Enter Insights!

Over the past couple years, a group of researchers, analysts, and data scientists have been working toward getting a better understanding of game pacing, or the general flow and feel of the game. Game pacing, by the way, is a little different from game length, which is purely about the number of minutes a game takes to complete. When we say game pacing, we’re talking about the progression of a game from laning phase to mid game to late game.

We consider game pacing from a variety of perspectives and multiple methods, ranging from game server data, all the way to survey answers that let us understand how players feel. Both sides are important to consider, as a game that feels well-paced but is always getting stomped out in ~15 minutes is just as concerning as a game that has meaningful back-and-forths, but all players didn’t consider it close at all.

On the in-game data side, we’ve started to hone in on a way of thinking about pacing that we’re pretty happy with, so let’s take a look at it: think about your games on Summoner’s Rift. At any given time, one team has an advantage of some sort -- whether it’s in gold, experience, or CS. Given the size of this lead and how far you are through the game, there can be pretty large differences in how confident you are that one team will win. A team that’s up 5.5 k gold 10 minutes into a 20 minute game is almost certain to close it out, while a team that’s up a mere 500 gold 45 minutes into a 50 minute game is likely in a head-to-head sprint to the finish.

We’ve approached pacing by categorizing games into different classes based on: how early we’re (fairly) certain we know the winner. On one end, games that are decided early are “snowbally” or “stompy,” while games that aren’t decided until the very end have an “overtime” feel. Games that fall in the middle, where the outcome of the game becomes consistently more predictable as it progresses, are considered “standard.” The proportion of games that fall into each pacing bucket tells us quite a bit, and gives us a good sense of what we want to do if too many games are falling into any one category. It’s worth noting that having only standard games isn’t a desirable or intended game state. Teams that get super ahead should be able to close games out, while overtime nailbiters - where one fight seals the deal - are also awesome to have in moderation.

As we started using these tools to understand the current state of game pacing in League (and our ability to impact it), we set out to cut down on starvation tactics. More specifically, we wanted to ensure that the burden of winning a game stays on the winning team. This led us to changes like the biased minions waves that very slightly push into the losing team, doing two things. The winning teams get more chances to siege and meaningfully progress the game toward their victory. The other outcome is that the losing team has more access to minions, meaning that games slowly move toward equilibrium when winning teams get complacent. We layered this groundwork with some complementary tuning to gold income, tower durability, and yes, death timers, and we had a strong first attempt at improving game pacing.

When the first Preseason 2016 patch dropped in 5.22, actively monitoring game pacing (and length) was high on our list. In terms of length, the average game was a bit shorter in 5.22 -- about 1.5 minutes on average. This, in itself, isn’t necessarily a bad thing. When we looked at pacing, however, we noticed things were a little out of whack. Compared to 5.21, there were more “stompy” games. In short, games were being decided earlier than they had been, leading to more game time where the outcome was all but determined. A deeper dive showed us that both XP and gold (to a lesser extent) snowballing were the culprit. Our framework for game pacing allowed us to identify what was mucking up and dive deeper to find the cause. We adjusted the XP rewards from kills and shipped the minion changes, expecting the minions to either end stomps faster (thus removing the long painful wait for them to actually resolve) while helping losing teams access gold on the map.

When the data started to come back for 5.23, we saw some pretty encouraging things. Stompy games were down across the board while standard and overtime games picked up the slack. Although we’re still looking into what exactly is the “best” distribution of game types, this was definitely much closer. In 5.23, there were less stomps, closer games, and a larger proportion of competitive game time. Not bad.

This work isn’t done, but we wanted to let you all know where we were at, and how we got here. Thanks for reading!

Riot Boom Bear + Fearless