Replay Analysis Part 2: Battle Duration

Welcome to Part 2 of my Replay Analysis. In this series I am analysing over 66,000 replays sent to me by you. Today we will look at the battle duration.

In Part 1 we looked at the Matchmaking Spread – the spread being your tier vs. the tiers of vehicles which are in the battle: -2 would mean you were bottom tier by 2, a Tier VI in a Tier VIII battle for example, with -1, -1/+1, =, +1 and +2 being self explanatory. There were some very surprising statistics, such as: only 7/3,215 Tier VII battles were of equal spread (Tier VII vehicles only in the battle) and, although it doesn’t feel like it at times, an equal spread is actually the most common spread for every map.

Battle duration is probably the easiest data set to understand and to make the results even easier to read I have taken the battle duration, which is presented in seconds, and rounded it to the nearest minute. (e.g: 470/60 ≈ 8). The results are quite fascinating when comparing them against other factors such as the Tier of the vehicle the player was using in the replay, the Map the battle took place on and of course the server and region the player originates from.

An important factor: The Assault game mode has a ten minute battle timer, whereas Encounter and Standard battles have a fifteen minute timer. This report will focus primarily on standard battles, but we’ll touch on Assault and Encounter after. It is not possible to know if Assault and Encounter were disabled by the player when they entered the recorded battle, so the proportion of Standard battles to Assault or Encounter battles is not an accurate representation of every day matchmaking.

Due to the new way I am filtering the data (by Standard, Random battles) I will revisit Part 1 and update the data at a later date. The figures won’t change significantly, but it will better filter out rouge data such as team training, ranked battles etc. from the data, to provide a much stronger overview of Standard, Random battles.


Of the 66,849 replays, 35,030 had the battle duration data and were Standard Random Battles. 18,159 of these were played on the EU servers and 16,871 on the NA servers. The most common cause for a replay to not have the duration data is if the player left the battle before it ended (i.e. they were eliminated from the battle and returned to the garage without watching to the end). The average battle duration was 7.21 minutes (3 s.f.). Please note that all figures relating to duration in minutes are decimals, and not minutes:seconds – 7.21 minutes is 7 point 21, not 7 minutes and 21 seconds, unless I specifically list both values.

Reading the data...

The most common battle duration was six minutes (23% of all battles) and the least common was three minutes (0.4% of all battles). There were no one or two minute battles. The second and third most common battle duration was seven minutes (19%) and five minutes (18%) respectively. These are of course either side of six minutes. It is common to suggest a five minute simulator, however the results suggest battles run on average a minute or two longer.

Although five, six and seven minute battles were the most common from both EU and NA regions, on average the NA battles were longer (7.45 minutes to 3 s.f.) than those from the EU region (6.98 minutes to 3 s.f.). We can see this in the above diagram with multiple pie charts. Note: The majority of battles on the NA servers were played in SPG's. This may influence the recorded battle duration, as the duration is only recorded in a complete replay. In a later post I will attempt to analyse, or rather exclude, SPG data.


Reading the data...

Of the 35,030 battles:

  • 2,805 (8%) were -2,
  • 5,163 (15%) were -1,
  • 2,604 (7%) were -1 / +1,

  • 13,161 (38%) were =,
  • 6,035 (17%) were +1, and
  • 5,262 (15%) were +2.

The average battle duration in minutes for each spread (to 3 s.f.) was:

  • 7.10 for -2,
  • 7.13 for -1,
  • 7.11 for -1 / +1,

  • 7.19 for =,
  • 7.24 for +1, and
  • 7.40 for +2.

Looking at these averages and if we transpose the data and rank each row (smallest to largest), we can see minor variation in the spread no matter the length of the battle, suggesting the Matchmaking Spread has very little effect on the duration of the battle:


 

Reading the data...

42.96% of battles were played in a Tier 10 vehicle, 14.49% in a Tier 9, and 22.1% in a Tier 8, whereas only 7.92% of all battles were in Tier 1 to 5. (Tier 1 to 5 have been grouped together in the pie chart above to better visualise Tier 6 to 10).

The average duration in minutes for each Tier (to 3 s.f.) was:

  • 6.80 for Tier 1,
  • 6.32 for Tier 2,
  • 5.45 for Tier 3,
  • 6.73 for Tier 4,
  • 6.71 for Tier 5,

  • 6.76 for Tier 6,
  • 7.03 for Tier 7,
  • 7.04 for Tier 8,
  • 7.34 for Tier 9, and
  • 7.47 for Tier 10.

Looking at these averages and if we transpose the data and rank each row (smallest to largest), we can see a slightly wider variation, suggesting the battle will last longer in a higher Tier. (On Average, a Tier 10 battle lasts 0.67 minutes - 40.2 seconds - longer than a Tier 1 battle, and 0.43 minutes - 25.8 seconds - longer than a Tier 8 battle).


Reading the data...

The Range has been tricky for me to define. In fact, I was going to focus on it in this post, but decided in order to do so I needed to redefine it. For the purpose of this post, we are only looking at one aspect (Battle Duration), which is why I have included it. The Range is the number of surviving team members minus the number of surviving enemies. A 15-5 win would have a range of 10, whereas a 7-10 defeat would have a range of -3. It is possible to have a negative range in a win when your team captures the base to win the battle and has less surviving vehicles in the field than the enemy does. The issue I have here is it only approaches it at a surface level: a 15-14 win has a range of 1, but so does a 3-2 win. This is why I need to approach the Range in a different way in the future and is why we are only looking at the table of values for now.

The most common Ranges were +10, +9, -9 and -10 respectively. These were all six minute battles and the most common values surrounded them in terms of Range and Duration, forming some beautiful butterfly visualisation in the statistics. The longer a battle lasted, the less likely it would end in a wipe of the allied or enemy team and more likely it would end with a range of 0. The closer the Range was to 0, the longer the battle lasted.


Reading the data...

The map with the shortest average battle duration was Ensk (4.93 minutes) and the map with the longest average battle duration was Prohorovka (8.98 minutes). Most battles were fought on Mannerheim Line (1,211 battles) and the least frequent map (excluding Safe Haven, as it is a new map, and Province and Widepark, as they are exclusive to lower tiers) was Berlin, with 458 battles.

Note: It is not possible to know which maps each player had on their ban list for each battle. This can skew this data significantly, for example if one person submits 15,000 replays and had the same maps on their ban list for every battle, in those 15,000 battles it is guaranteed (providing they played solo or as the commander in a platoon) not to see those maps. Considering the average frequency of each map (excluding Safe Haven, Province and Widepark for the same reasons as above) was 907, I would say the above scenario is true for Berlin, Empires Border and Pearl River. Looking closer at the statistics on a per-player basis strengthens this theory: Player 23 had 31 battles in Berlin, 68 on Empires border and 34 on Pearl River (out of 11,982) but it is still not possible to see exactly when a map was banned - we can only theorize that the player had them banned (or the commander of their platoon had them banned) for a significant number of their battles.


Reading the data...

Of the 66,849 battles, 3,889 (5.82%) were Random Battle, Encounter. The same butterfly effect can be seen in the heatmap for the range, focused around 5, 6, and 7 minute battles, with Himmelsorf and Ensk being the most common five minute battle.

Of the 66,849 battles, 2,383 (3.56%) were Random Battle, Assault. The same butterfly effect can be seen in the range, but this time it is not primarily surrounding 5, 6, and 7 minute battles. There was more variation in battle duration with Karelia having the most 10 minute battles, but with 6 and 7 minute battles still a strong factor.


In conclusion, we can see that the average Standard Random Battle length was 7.21 minutes (3 s.f.), but was longer for the NA region (7.45 minutes to 3 s.f.) than the EU region (6.98 minutes to 3 s.f.). Matchmaking Spread does not appear to effect the battle duration, however battles lasted on average 40.2 seconds longer at Tier 10 than they did at Tier 1, or 25.8 seconds longer at Tier 10 than they did at Tier 8. The most common ranges (+10, +9, -9 and -10 respectively) all occurred in six minute battles, further strengthening the five minute battle or fifteen to five simulator curse so many people complain about.

Thank you for taking the time to read through this report. I've had to focus on coursework so far this year, which is why it took so long to publish part 2 - I'm hoping to have part 3 published by the end of March / beginning of April. I would like to look at the Range, but still need to figure out a way to better represent that data, so may look at a different aspect first to give me more time to come up with that plan.

If you would like to send me your replays to be included in the next analysis, you can find more information [HERE].

Thank you!

Leave a Reply

Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors