Is Among Us competitive?

(topic: Game design; state: incomplete)

Among Us is unique among popular video games as, rather than superior dexterity or strategy, winning depends on deception—the ability to tell and detect lies.

Because it’s so different from games such as CS:GO, LoL, and Starcraft, surely it can’t be competitive. It’s just a party game, right?

To answer that, we should think about what exactly a party game is.

What is a party game?

Party games are designed around one key goal: to make it impossible for stronger players to dominate weaker ones. If this is achieved, the game can be fun for everyone regardless of the skill level of the play group.

This is usually achieved with:

  1. Major outcomes decided by randomness—heavy RNG
  2. Advantages being tenuous and easily overcome—e.g., rubberband mechanics
  3. Losing players being able to decide who amongst the leaders wins—kingmaking

Mario Party is the prototypical example of a party game. While the minigames themselves are competitive and stronger players can dominate those, winning them has little impact on who wins the overall game. Major outcomes are decided by dice rolls, and kingmaking is rampant with coin and star stealing.

Making a game which satisfies this design goal and is also fun is so tremendously difficult that there are very few popular games in the genre.

What is a competitive game?

A competitive game, by contrast, is designed so that the stronger player wins as much as possible. If this is achieved, the game is often not fun if players aren’t equally matched.

This is usually achieved with:

  1. Minor or no outcomes decided by randomness
  2. Advantages accumulating slowly and being hard but possible to overcome
  3. Losing players being eliminated once they can no longer win

These are polar opposites, and more like a spectrum. One can turn a game from being more like a party game into something more like a competitive game by tinkering with these mechanics.

The clearest example of this is Super Smash Bros. If played in free-for-all and with items on, there’s both a lot of kingmaking and and a lot of randomness, and the game is more of a party game. If played with items off and with only 2 players or teams, the game is more competitive.

Where does Among Us fit?

Almost none of the outcomes in Among Us are decided by randomness. Votes always work the same way. Tasks are always done the same way. There is randomness behind which tasks are assigned, but this isn’t a major decider of who wins.

There’s no rubberband mechanics. The advantage crew gets from accurately sussing an imposter early is strong and takes work to overcome. The advantage imposters get from framing crew and building rapport is similarly strong and hard to overcome. If an imposter is left in a 1v7 situation, they don’t get extra help.

There’s no kingmaking, since there are only two teams.

By these criteria, Among Us is more of a competitive game than a party game.

This should not be surprising to most players, since it’s usually clear that stronger players dominate weaker ones.

However, if you run a tournament which doesn’t consider these mechanics, you can make the game less competitive by unintentionally introducing them.

FaZe clan tournament

The FaZe clan tournament ran on 19 Oct 2020 demonstrates many of the potential pitfalls. The format was, briefly:

This format is almost tailor-made to make the game as uncompetitive as possible.

Randomness

Because the crew winrate (>90%) in these games is substantially higher than the imposter winrate, a major outcome is decided purely by randomness. Most players who roll crew and stay alive get 8 points per game, while imposters get 1 or 2.

Not only that, the average imposter win is only worth ~7 points, which is less than an average crew win, despite being substantially harder.

Advantages can be impossible to overcome

As a phase nears its end, more and more players in the lobby have no reason to play properly, because they’re already technically through or eliminated.

In the final lobby anyone not in the top ~6 is effectively eliminated—there’s no difference between placing 5th and placing 10th—and anyone not within a few points of first has almost no chance to win.

Kingmaking

In the final lobby, Ludwig rolled imposter 4 times in a row, guaranteeing he had no chance to win. However, because a dead crew member can’t vote and get the points for a correct vote, he could kingmake by killing a leading player who he didn’t want to win. So he killed 5uppp two games in a row—without regard for if the kills were good—allowing Yeti to win decisively.

Exploring the issues

Impossible advantages and kingmaking

As soon as you run a series of games, it goes from being a team game to a free-for-all, which makes kingmaking a problem.

This is unavoidable, since you can’t run one game and call it a day, so it needs to be addressed directly by the tournament format. It can’t be ignored.

The best game to take notes from here is Poker, as it’s the only major competitive game which uses a free-for-all format.

In most Poker tournaments, every difference in placement matters. Aside from prize money having a long and flat distribution, there’s often an intricate system of qualification points for making it into the next tournament. By doing this, the player in 9th place is still interested in fighting for 8th place, even if they can’t place 1st.

On top of this, collusion of any kind is strictly forbidden.

However, unlike Poker, there’s no way to reduce the table to only 2 players. Even the biggest disadvantage can be overcome in Poker as long as you’ve got one chip left. In Among Us, it’s impossible for every player in the final game to have a chance to win right up until the end. This is by far the biggest issue facing any attempt at running an Among Us tournament.

Randomness

Awarding crew wins around the same number of points as imposter wins causes problems when the crew wins more often, because it means rolling into imposter too often makes it impossible to win.

This can be addressed by giving imposters more points for a win.

Aside from the randomness behind how often you roll imposter, there’s also randomness behind who you roll imposter with. If imposter wins are worth a lot of points, and most players only roll 1–2 imposters in a series, then getting a strong or weak imposter teammate is a significant outcome decided by pure chance.

This can be addressed by the crew winrate not being so high so that imposter wins aren’t so important.

Point system

Generally speaking, if points are awarded for anything but match wins, it warps gameplay, and usually not in the way you want it to. Tournament organisers are usually not game designers.

Assigning points for correct and incorrect votes seems fine in theory (“We want to distinguish the effective crew from the ineffective crew.”), but in practice creates bad incentives—it makes staying alive optimal gameplay.

Almost no competitive tournament format awards points for subgoals in game. LoL teams don’t get extra points for pentas. CS:GO teams don’t get extra points for knife kills. Federer doesn’t get a bye because he scored the most aces. At most, tournaments hand out MVP awards to players for such things. The actual tournament is never decided by anything but match wins, with subgoals sometimes used only for tiebreakers.

There are a few ways we can assign points based only on match wins.

Win ratio

Add up the players’ win ratio on each role. For example, if someone goes 7-1 as crew (87.5%) and 1-1 as imposter (50%), their total score is 137.5.

This has many issues:

All of these issues suggest that using ratios is a mistake.

Dynamic scoring

Award points for wins based on the actual win rate for that role. For example, if a lobby plays a series with 1 imposter win and 6 crew wins, the imposter win is worth 6 points and the crew wins worth 1 point each.

This has a few weird effects:

The first can be addressed by giving a minimum of 1 point for a win.

The second could be considered a feature in that it could help address kingmaking by making it less clear who is and isn’t eliminated.

Fixed scoring

Award points for wins based on the expected win rate for that role. For example, if we expect the crew win rate to be 80%, we give 20 points for a crew win and 80 points for an imposter win.

This works well if we can accurately guess at the crew win rate. But if this guess is wrong, such as the crew rate actually being >90% when our scoring system suggests it’s 60%, then this causes a big problem.

We can help avoid ties in this system by making the points for each win not share any common divisors. This is most relevant for a crew win rate of 50% where all wins are worth exactly 50 points (but is also relevant for 60%, 75%, and 80%). In that situation if one player goes 4-2 on crew and 0-1 on imposter, and another player goes 3-3 on crew and 1-0 on imposter, they both end up with 200 points. But if we instead say crew win rate is 51%, the tie would be resolved with the player who got the imposter win having 2 more points than the other player.

Another feature of this system is we can make the total number of points per game constant with a crew win rate of 80% (for a 10 player lobby). If crew wins, 8 players get 2 points. If imposters win, 2 players get 8 points. So every game gives out exactly 16 points. This could be useful if our tournament format wants to compare points across lobbies or carry points across phases.

Ties

If points are only awarded for match wins, ties will be more common. This isn’t much of an issue, since we can safely use a subgoal point system for tiebreakers without warping gameplay.

If a series run for about 90 minutes and with a >90% crew winrate, it’s possible there are no imposter wins, which will lead to a lot of ties and match result determined entirely by who rolled imposter less.

This is far from ideal, but isn’t quite as bad as it sounds. After all, crew winning every game wasn’t a foregone conclusion, and if the losers—who played the most imposter games, thus having the most chances—had won even a single imposter game they would have instead won the series.

Winner will typically go 1-0 on imposter

If a series runs for about 90 minutes and with a >90% crew winrate, most players won’t roll imposter more than twice, and some won’t roll imposter at all. It’s likely that only 1 game will have an imposter win. Whichever 2 players get that win are almost guaranteed to tie for first, and whoever of them rolls imposter a second time will likely not win.

Crew winrate

The elephant in the room. Why is the crew winrate so high? This isn’t just an issue of making the game less competitive, but of the game not being fun to watch or play—which is actually a far bigger problem.

One might assume that this is just a fatal flaw of the game, that at a certain skill level crew is overpowered and imposters have no viable strategies. This may be true, but there are other explanations as well.

The first is that, by giving crew a lot of points for correct votes, they’re heavily incentivised to stay alive, because they can’t get these points while dead. The easiest way to stay alive is grouping up, which results in a meta that’s hard for imposters to do anything.

The more subtle explanation is that players can increase their imposter winrate at the cost of their crew winrate, and vice versa. If you act unusually sus as crew, it’s easier to deflect accusations when you happen to be imposter. Whereas if you have strong crew tells, you’ll be easy to trust as crew, but as imposter will be unable to get past any real scrutiny. (Many players complain about this type of metagaming, but it’s an essential and unavoidable aspect of the genre.)

As a result of this tradeoff, the crew winrate is dependent on the players’ incentives and playstyles. If imposter wins were worth substantially more than crew wins, the imposter winrate would improve.

Aside from that, a more direct solution is to change the game settings. You could play with only 9 players (2 imposter 7 crew), or you could play with a lower kill cooldown, higher player speed, longer emergency cooldown, anonymous votes, or more tasks. This should be done with caution, since doing so could make the game less fun.

Simulation

If we play a series of games with points awarded only for wins, how likely are ties? How much is the result decided by a single imposter win? How likely are stronger players to win?

Usage

Skill factor

If “skill factor” is 0, crew wins with a chance equal to crew win rate. If it’s 100, a team with the best players will always beat a team with the worst players; and a skill gap of half that will influence the result by 50%.

For example, with a skill factor of 40 and crew win rate of 70%, a match where the skill gap is—

(This is a very naive simulation of skill and may be better off set to 0.)

Dynamic scoring

If “dynamic scoring” is on, an imposter win is worth points equal to the number of actual crew wins, and vice versa.

If “dynamic scoring” is off, an imposter win is worth points equal to the target crew win rate, and vice versa. This is fixed scoring.

Analysis

Optimal crew win rate

One might assume that the optimal crew win rate is 50%, and it is if you’re trying to minimise the effect of randomness on the result.

However, a crew win rate of 60–70% still removes a lot of the randomness, and is arguably a better target overall.

Reducing crew win rate without causing other problems isn’t easy, and if it’s >90% without any counter-measures, then getting it to 70% is surely more practical than getting it to 50%.

More importantly, if imposters win half the time, an imposter win is not so impressive when actually pulled off. By having the odds slightly against the imposters, it adds a bit more hype to the win when it does happen.

Summary

Among Us tournaments can potentially be competitive, but there are many pitfalls due to the nature of the game. Tournament formats must be designed to minimise the effect of chance on the match result. In particular, rolling imposter shouldn’t be a death sentence. Game settings and point systems also need to be set such that crew winrate isn’t too high. It doesn’t need to be 50%, but it shouldn’t be >90%.