Kill-Death Influence on Objective Games in Halo 5

Josh Menke once commented that in the global Halo 5 data, a team that kills the most will usually win.

I wanted to test this against my own data. I’ve been pulling and archiving statistics through the API for a few months, and now feel like I have enough data points to perform some relevant analysis.

You can play with the raw data in this Google Spreadsheet. Use the Data -> Filter Views -> Objective Games with No Quitters menu to filter.

I found that the team that outslays by even 1 kill has an 78% chance of winning the match. That percent only goes up the bigger the kill differential. Now this obviously doesn’t mean “just focus on slaying and you will win”, but rather, “if you told me Blue outslayed Red, I could guess that Blue won and be right 78% of the time.”

If you can outslay the other team by more than 10 kills, you have a 88% chance of winning (though of course you still have to play the objective).

For games where KD Spread did not correctly predict the outcome (i.e. the outslayed team still managed to win) the average kill differential was only 6.9 kills, suggesting that the matches were still fairly close.

Limitations

This data is pulled from 455 games in the HCS playlist. After filtering out Slayer and games with quitters, there were only 251 games to analyze.

These games are all from the silver and gold skill tiers. Numbers could be different in higher level tiers: there might be even less chance for the outslayed team to still pull off a win because players generally know how to play the objective.

Methodology

I’ve been pulling data from the API and storing it in a normalized SQLite database. I ran the following query against my schema to generate the raw table in the Google Spreadsheet.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
SELECT
	*,
	ABS(RedKills - BlueKills) AS KillDifference,
	CASE WHEN (RedKills > BlueKills AND Winner = 'red') OR (BlueKills > RedKills AND Winner = 'blue') THEN 1 ELSE 0 END AS KDPredicted
FROM (
	SELECT
		AMR.Id,
		GBV.Name AS GameType,
		(SELECT SUM(TotalKills) FROM PlayerStats WHERE MatchId = AMR.Id AND TeamId = 0) AS RedKills,
		(SELECT SUM(TotalKills) FROM PlayerStats WHERE MatchId = AMR.Id AND TeamId = 1) AS BlueKills,
		(SELECT CASE TeamId WHEN 0 THEN 'red' ELSE 'blue' END FROM TeamStats WHERE MatchId = AMR.Id AND Rank = 1) AS Winner,
		(SELECT SUM(DNF) FROM PlayerStats WHERE MatchId = AMR.Id) AS Quits -- DNF is 1 if quit (did not finish), 0 if stayed in the match
	FROM
		ArenaMatchResults AMR
		JOIN GameBaseVariants GBV ON GBV.Id = AMR.GameBaseVariantId
	WHERE
		AMR.PlaylistId = '5c9808b003a34577b9667217f9dda52d' -- HCS
	GROUP BY
		AMR.Id
)

This spits out a table with the following columns:

  • MatchID (Guid)
  • GameType (string): the type of game mode (CTF, Strongholds, etc)
  • RedKills (int): number of kills the red team got
  • BlueKills (int): number of kills the blue team got
  • Winner (string): “red” if the red team won, “blue” if the blue team won
  • Quits (int): the number of players that quit the match
  • KillDifference (int): the differential in kill counts between the two teams
  • KDPredicted (int): 1 if the team with the most kills won, 0 otherwise

I then performed a couple filters to get rid of some noise.

To get rid of games which had quitters, I removed any game where the Quits count was not zero.

I also removed Slayer games, because the team with the most kills wins by definition, so its kind of pointless to analyize. Interesting, I found one slayer game where the kill count was even (50-50) but we lost by one point because someone had an accidental team kill.

Other interesting tidbits

I found that 26% of my 455 games had at least one quitter. If asked, I think I would have guessed a quit rate of closer to 10% or something. Its apparenlty much more prevalent than I thought.

I found one game where we outslayed by 73 kills over almost 12 minutes. I don’t recall this game, but evidently my random teammates were stat padding, or the losing team was really good at flag stops and we sucked at running the flag. Heh silver tier.

Almost half of games where the winning team was outslayed by more than 10 kills were Strongholds games. In the two biggest upsets, the winning team was outslayed by almost 40 kills yet still won! I can’t imagine this happening in Platinum tier or above. You might think this suggests that Strongholds is a little less susceptible to high kill differentials, i.e. its easer to get the right kills instead of the most kills and still win. But I found that for Strongholds, the predictive power of KD spread was still at about 78% - the same as every other game type.