PUBG Deep Dive: Analysing Circle RNG in PGL

It’s a criticism that has been encountered by just about every competitive PUBG player and spectator – “PUBG has too much RNG [random number generation] to be truly competitive”. It’s been cited by certain large-following streamers as “the reason this game will never be an eSport”, and that sentiment is predictably repeated in many discussions around professional tournaments and competitive play generally.

 

It is however, fairly well established that eSports games, or at least those we have known up to this point, are at their best when RNG is reduced to a minimum. Widespread randomness in fundamental game mechanics just aren’t contingent with a game that wants to reward skill and ability to its maximum.

 

In fact, there’s a strong argument to be made that randomness is often used in games as a “great equalizer” - as a balancing method to allow less skilled or experienced players to “compete” with those with more experience. But these aren’t the kinds of mechanics that we typically want to see in eSports games, as in competitive play, we want to know that the successes of a player are the product of skill, experience, well-honed decision-making abilities, and, in a game like PUBG, teamplay. Randomness serves to disrupt this to varying degrees, depending on how pervasive it is in the game’s mechanics.

A game like PUBG does make use of RNG extensively, and in fundamental mechanics. Circle movements, loot and vehicle spawning, and flight path are those that initially come to mind. PUBG is unlike just about any video game in this regard, outside of the relatively new Battle Royale genre, that has also borne a competitive community before. Admittedly at first glance, it would seem that PUBG isn’t in an ideal spot for a competitive game with eSports aspirations. However, there are examples of games, in the strictest sense of the word, that have both competition and RNG at their core.

 

Poker is one that has been cited numerous times as a ‘comparable’ game to Battle Royales. In poker the cards you are dealt are random, and the cards that are drawn are random. The skill involved centres around how you play those cards. Yet there is still a fundamentally random nature to the base mechanics. Despite this, there still exists a prosperous competitive and professional scene, and the community has developed its own methods for ranking and assessing player performances. In the same vein as card games, Hearthstone and similar have also spawned competitive, even professional scenes around them, despite having significant RNG in their mechanics.

Image: The new 'Savage' map.

It’s not impossible to imagine, therefore, that an FPS game such as PUBG, could follow suit and develop it’s own competitive ruleset which simultaneously would allow the game to retain its inherent appeal (extremely varied gameplay), but also provide a fair playing field and competitive “integrity”, that allows the best players to rise to the top, and reap the rewards for doing so. In truth, most games we currently consider eSports have been through this process to some degree. It may simply be a matter of bringing attention to the right issues and working through them as a community.

The Current Competitive Ruleset

 

Over the year since the game went into early access, the competitive community has gradually experimented with, and adopted various tweaks to the standard PUBG ruleset, in order to make it more suitable for competitive play. This has included the removal of red zones, the switch to first-person perspective (in the NA/EU regions, at least), the mostly-standard increase in spawn rates for assault rifles and ammunition, and adjustments to the circle movements to allow for more fighting, positional adjustments, and strategy to develop. Amongst others.

 

In leagues and tournaments, scoring systems are constantly adjusted to better reward kills and final squad placement. Although there is no current “standard” that has been adopted universally by tournaments. Another area in which there is no current “standard” is the number of matches played during qualifications or finals. Initially with the Gamescom invitational 2017, this started with three games being played in the squads tournament. Intel Extreme Master’s in Oakland, 2017, upped this to eight games being played, split over two days. Other tournaments such as OGN Survival Series 2018 in South Korea played five squads games, alongside an additional solo tournament.

Intel Extreme Master’s in Katowice 2018 continued with eight games over two days, but the StarSeries i-League 2018 in Kiev bumped this up significantly, to 20 games, played over the course of four days. The reasoning for this, most likely, is that in a game that has a significant RNG factor, the more games that are played, the more the effect of RNG is averaged out over the course of the tournament. This, in theory, should produce a final scoreboard that is an accurate reflection of the teams’ and player’s performances in the tournament. Rather than a more mixed set of results in which some teams may have been particularly favoured, or negatively affected by, RNG in their matches - and therefore might have scored higher or lower than they otherwise would have.

 

The most recent of these tournaments is the PGL PUBG Spring Invitational 2018, Bucharest. In PGL teams played 16 games over the course of four days, with standardized competitive settings.

Image: IEM Oakland PUBG Event. 

Credit: ESL

The scoring system was quite radically different to previous tournaments, in that it awarded 20 points per kill – significantly higher than normal. StarSeries for example offered 8 points per kill, and IEM offered 10 points. Though of course, these values are all to be considered relative to the amount of placement points given for a squad’s placement.

 

Another change PGL introduced, with huge benefits for spectators, is the addition of a map stream and individual team streams, which provided coverage of all matches over the entire tournament. This has given us with the opportunity to look closer than before at how the matches played out, and how teams performed in this tournament ruleset.

The Problem

 

Anyone that has played more than a few hours of PUBG will know that circle spawns can often put your team at a significant disadvantage, when compared to other teams out on the map. Counter to this, it’s also possible for your team to have a lucky streak of circle RNG, which consistently places your team at the centre of the new safe zone, significantly reducing your risk of getting killed compared to others – who may need to drive or run into the new safe, usually requiring them to expose themselves to the enemy in the process.

 

It’s desirable then, in a competitive setting, that the effect of this circle RNG is minimized, across all teams, across the competition. As described in the previous section, the main methods currently used to achieve this are increasing the number of matches played during qualification and finals, and by adjusting circle spread and movement speeds.

However the effects of circle RNG cannot be completely nullified, no matter the ruleset. Teams will always be affected slightly differently, no matter the number of games played, so long as circle spawns are dictated by RNG in some fashion. The difference may be small enough as to be negligible, but it will still exist. What is currently unknown is the extent to which this RNG is influencing tournament standings.

It’s fair to assume that the effect is not enormous. If it were, it’s likely that players and spectators would be quite vocal about that fact. Especially given that these tournaments have significant prize pools, and the difference in winnings between, for example, a third and a fifth position, can be thousands of dollars. So we would expect the current ruleset to be reasonably fair for most teams.

 

At the same time however – in a game such as PUBG which includes 64 players in each match, and such a significant number of games, it can be difficult to fully gauge how teams are affected by RNG in the same tournament. Additionally, professional matches are bringing together many of the best players in the world, that are playing at the furthest edge of the current skill range. Within such competition can exist very fine margins of performance differential, and the final standings can contain extremely small points differences as a result.

Image: A wild blue zone field rat, in it's natural habitat.

Credit: News Ledge

For example, in the PGL, the top two teams were separated by only 15 points, with first place taking a total of 6155 points. That’s a difference of about 0.25% of the final score. With margins that small, it’s still worth looking very closely at how RNG currently effects teams in tournament play.

Methods

 

The goal is to determine whether teams are equally affected by circle RNG over the course of a tournament. Further to that, we may be able to correlate team’s average RNG influence with final scoring. With the PGL map stream, we have a record of all circle shifts and team movements over the full 16 matches. From these matches, we can grab statistics for each team that played in the tournament and analyse the extent of circle RNG that they experienced. We can do this by:

Screenshotting the map the moment a circle spawns. This is done for several reasons. Firstly, we can map where a team was positioned the moment RNG comes into effect. This allows us to snapshot the second that a team stops considering the previous circle as “safe”, and instead works on the information given by the new, safe circle. Secondly, we can map the central circle point (for both safe and blue circles), and therefore measure the relative distances of the teams to these circles.

1. 

2.

Overlay the safe and blue zones with a circle that has a central point. This gives us a fairly accurate circle centre to measure the above distances.

Measure the team’s distance to the previous circle’s centre (red), as well as the team’s distance to the new safe circle centre (green). To be accurate with these measurements, this could be done by either eyeballing distances using the map’s gridlines (and Pythag), or use an interactive map (https://pubgmap.io/en-gb/erangel.html) to measure the distance between two points.

3.

With these measurements taken, we now want to factor out any distance between the team and centre circle that could be considered “planned”, or accounted for by the team’s choice to take a certain position. We can do this by subtracting the team’s distance to the new centre of the safe circle from the team’s distance to the previous safe circle;

4.

RNG Circle Deviation = Team Distance to New Safe Centre – Team Distance to Previous Safe Centre

 

e.g. as shown in the screenshot in step 3.

 

RNG Circle Deviation = 164m – 635m

 

RNG Circle Deviation = -471m

 

Showing the new safe circle RNG shifting in favour of this team (TSM), by reducing their distance to the safe centre.

Because we take this screenshot the moment the new safe circle pops, all player-determined distance to a central safe spot should be accounted for in the team’s distance to the previous circle centre. This leaves us with what I would call ‘Unexpected RNG Deviation’. That being, a measure of how much more or less distance a team would need to travel in order to reach the centre of the new safe circle, as determined by circle spawn RNG.

Repeat this process for all circles in all matches during the tournament.

5.

A few additional points on the measuring process:

  • Measuring from the team member closest to the relevant circle seems to be the most sensible approach. Rather than trying to work out an “average” position of a team based on multiple members. It’s probably fair to assume that, on average, the player closest to the centre of the circle would be the one most likely to make it to that centre safely.

  • In cases of split teams, or where different members of a team are closer to the blue/safe centres, we measure from whoever is closest to the particular circle we are measuring. E.g:

  • In the case of the first circle, there is no clearly defined “central” location that we would expect to be the most likely spot for first circle spawn. In this instance, it seemed best to approximate a land mass “centre” and use this as the anchor point for first circle spawns. My best guess was the road at the compound South-West of School, dotted in blue below. This spot was used throughout all games as the centre for the map (or land mass centre).

  • There is a small margin of error to be considered with all these measurements. Factors such as image pixilation preventing good vision of a player’s location, nameplates preventing good vision, inaccuracies with my circle overlays, human error with the measuring tool, etc, all contribute a small amount of inaccuracy. I’d estimate that each measurement could be off by up to ± 2% of the circle’s diameter. This should average out in the full data set, however.

After collecting all this data, we should be able to make good comparisons between teams in terms of how strongly positively or negatively they were affected by RNG shift. I did, however, discover very quickly that this process is extremely time intensive. Ghost’s first game alone took about three hours to get through. This improved once I started using the interactive map and got into a routine with things.

 

However, a single game for a single team still took about 45 minutes to get through. To collect data on all circles, in all games, for all teams in the tournament, would take about 200 hours’ worth of manual work. Unfortunately, I don’t have the time to give that much, so I decided on a middle ground; collect all circles for all games, for four teams.

 

This should give us enough data to compare how two teams are affected by RNG side-by-side, and perhaps enough to speculate on general trends that might be in effect for all teams. These four teams would be Ghost in first place, Grubie second place (since they were separated by such fine margins), eighth TSM (middle of the pack), and last place in placement points, LGD.

 

In total, this means the total sample set is 128 circle spawns, and around 1024 distances and RNG deviations being measured. All circles and measurements are included as images and a spreadsheet in a link at the bottom of this article.

Results & Analysis

 

Before diving into the full data set, let’s look at a couple of example circle spawns to get an idea of what these values mean. This is from game seven, circle six. The team we’re considering is TSM. In this case, Viss was 10m away from the centre of the previous safe zone. The new circle spawns, and they’re now 94m away from the safe zone.

 

This gives an RNG deviation of 84m – quite unfavourable considering the size of the circles at this stage in the game, and the fact they were almost perfectly central in the previous zone. In actuality, out of the four teams analysed, this was one of the most unfavourable circle shifts in the whole tournament.

This is an example of a neutral shift, where the RNG deviation is near-zero. We’re looking at the second circle, game eight, focus on Grubie. Their distance to the previous zone centre was 884m, and the new distance to centre is 876m. This is an RNG deviation of only -8m in a blue zone 4km in diameter. Notice that this is not a circle that has landed perfectly on the team. The circle has shifted away from the previous centre, but the important fact is that Grubie’s travel distance to the new centre is about the same as was expected anyway. As such, the circle could be considered very neutral.

Finally, here is an example of a favourable, or negative shift. This occurs in a situation where centre circle has shifted to be closer to the player/team than it was before. Grubie’s game 12, circle 2. Their RNG deviation in this spawn was -609m. This situation seems to be somewhat less common than neutral shifts, and much less frequent than positive shifts.

Moving on to the full data set, let’s look at the averages for each circle stage across the tournament, in raw distance form. Note that LGD are missing statistics for Circle 8, as they never reached this circle in any game.

Graphing these out makes it slightly more readable:

One point that we can at least start with here, is that it appears that the first circle is prone to affecting teams very differently. On average, Ghost had to deal with over 1800m of additional rotation distance due to shift RNG. Meanwhile TSM and LGD were less affected, with 243m and 395m of additional shift respectively.

 

Since both TSM and LGD tended to land near the coast (primarily around Lipovka and occasionally North George), we could assume that the first circles here tended to shift to the coasts throughout the tournament.

This would also explain Ghost’s high deviation average since they were primarily landing Pochinki throughout the tournament. Since Pochinki is very central, first circles that push to the coast are likely to give them a high RNG factor.

 

Beyond this though, it’s hard to read anything more as average shift distances tend to approach zero in later circles, due to shrink. If we convert the raw distance into a percentage using the following formula:

This gives us the RNG distance as a percentage of the “potential” maximum shift distance. For example, if we are considering the second circle, which has a radius of 1175m, and sits in the first circle, which has a radius of 2km, we get this formula:

Using the example of TSM’s circle 6 in game 7, which we looked at earlier in this section, we get these values (circle 5 radius is 232.5m, circle 6 radius is 125m):

Note that whilst this was a very unfavourable shift for TSM, it wasn’t too close to hitting 100% unfavorability. That’s because even though it was a hard shift away from them, the safe circle was still on their side of the previous circle’s central point.

 

Had the circle landed on the opposite side of circle 5, north-east of Viss, the RNG shift distance would have been higher, and their RNG unfavorability would be higher as a result. 

 

If we apply this calculation across each circle, and average it for all games based on the circle number, we get these figures:

Now we can generate a graph with the same data as before, but make it more readable in later circles:

Something to consider at this point is that there are very few data points to work with for the eighth circle. There were only nine instances in which one of these teams made it to the eighth circle. This isn’t a reasonable sample set, so we’ll exclude this from consideration for a while.

 

At the same time, the first circles are producing wildly different results for each team. As mentioned earlier, Ghost is averaging far more RNG shift distance than the others, likely due to their very central drop location, and the high number of coastal circles that this tournament seemed to produce. Let’s also exclude this circle, so we’re left with circles 2 to 7.

Now we can generate a graph with the same data as before, but make it more readable in later circles:

Ghost quite obviously has more favourable RNG shift when we consider these circles combined. Grubie has some quite mixed results, but they’re pulled down quite low in the combined average by their highly favourable second circle shifts. LGD and TSM both suffer heavily in circle 2 and continue to have generally unfavourable circles from that point onwards.

At this point, we could potentially conclude that the answer to the question “are teams equally affected by circle RNG over the course of a (16 game) tournament?” is no. There is variation of up to 34% additional RNG shift distance in these teams in circle two, and 43% if we consider the first circle. Variation does seem to decrease in later circles, however based on the data we have here, it still seems to remain around 15% through circles three to seven.

 

For arguments sake, one could argue that we should consider the average RNG shift distance for all circles combined, in order to get a fair overview of circle RNG in the entire tournament. In this case, there would be approximately 7.8% variation between the worst-affected, and the least-affected teams in circles two to seven (Ghost – 3.16%, LGD – 10.99%).

 

Considering all circles, bearing in mind our incomplete/anomalous data set when including circle one and eight - we have a variation of around 4%. However, I would personally argue that we should consider each circle stage individually, and aim to balance RNG in each of these, rather than averaging the random shift over all circles. This is because the inherent ‘risk’ in rotating during each circle phase is not the same across all stages of a match.

 

Rotating during the first circle phase is not likely to carry as much risk as rotating during the seventh circle phase, due to the relative distances between teams etc. Therefore, a team having to move a distance of 20% the width of the previous circle in stage one, is not going to be as risky as having to move the equivalent distance in circle seven. Moving an additional 1000m to get into circle one, is not as dangerous as moving 50m to get into circle seven.

 

This is more of a discussion for players and tournament organisers to have, however. It’s up to the community to consider what a “fair” amount of RNG variation and shift distance would be in each given circle, and whether this amount of variation is likely to affect team scores. For example, the community may decide that one team having to move an extra 500m than another to get to the first circle, on average, across a tournament, would be considered balanced. They may also decide that 50m of variation in circle seven is unfair, that it should be reduced to be approximately 20m of variation between teams. This kind of discussion and evaluation would be extremely useful in refining the competitive ruleset in the future.

One factor that needs to be considered before we finish here, however, is centralisation. Up to this point, our methodology has assumed that circle spawns are entirely random but with a few ground rules – the new circle must spawn within the previous circle, and that the new circle must be entirely contained within the previous circle. The assumption was that so long as these rules were adhered to, the circle spawn location was completely random. This is not strictly true.

 

There seems to be a minor bias towards central circle spawns from stage two onwards. Not to say extreme shifts don’t happen in later circles, but there is a slight favourability towards centre. PUBG Corp’s notes on circle settings, as well as custom circle settings, seem to confirm this.

 

As a result, our previous data is likely to be skewed slightly to favour teams that centralise more than others.

Fortunately, we can use our existing data to gauge how central teams tend to be.  We can take the teams distances to the last safe centre, in each circle, and average these out across the games to give us a fairly accurate depiction of how centralised that team tended to be:

We can then convert these to percentages based on the following formula, which compares the team’s average distance to the previous safe centre to the radius of the previous circle:

This gives us a value that tells us how close the team was to the previous circles centre, as a percentage of the previous circle’s radius. For example, for Ghost’s second circle centralisation:

Using this method across the full set of teams and circles:

Based on the totals for each team, it’s obvious that Ghost are centralising far more than any other team, in both the all-circles average, and circles two through seven. TSM, Grubie and LGD have quite similar centralisation on the whole. Graphing the averages out would look like this:

An interesting side-note on this is how teams tend to centralise less and less as the circles progress. This is perhaps obvious if you’re a competitive player, but I find it interesting shown like this anyway. Presumably the reason for that is that as the circles become smaller, centralising becomes riskier and riskier. On top of that, it’s more likely you have lethal range on your opponents as the game progresses, so there’s less motivation to move unless absolutely necessary.

Now that we have a good idea of how much teams are centralising, we can also have a go at correlating this with the RNG deviations. From this, we should be able to make an estimation as to how favourable or unfavourable a given team’s RNG was based on their centralisation.

 

Unfortunately though, after attempting to make correlations for each circle stage, there are wildly varying correlation strengths. As can be shown in circles two and three, where the trend would suggest that being further away from the centre of the circle would actually be beneficial in terms of minimizing RNG deviation. Which is most likely not true:

This is most likely due to the small number of teams sampled, along with a fairly small number of circle samples per phase. Sixteen circles of each stage for four teams isn’t likely to be a big enough data set to get a good understanding of how centralisation is influencing a team’s RNG susceptibility.

 

In this instance, we have to use an average of all circles for each team, to get a sensible trend to read from. This is the result if we take the centralisation of teams in circles two to seven, and correlate with the RNG deviation for those teams and circles:

This trend is fairly well correlated with the team results, and also makes logical sense. Ghost have the most favourable average RNG shift, but they are also the more centralised team, on average. Grubie and TSM have less favourable RNG, but they are also centralising significantly less than Ghost. LGD have by far the least favourable RNG, but they are centralising the least out of the four teams.

 

This aligns quite nicely with how most of us have anecdotally come to understand the PUBG competitive meta. As such, I think it’s fair to call this an accurate correlation. From this, we can finally get some accurate figures for RNG variation.

 

If we take each team’s centralisation, and match it to the trendline to get a “predicted” RNG deviation, we can see how much the actual tournament RNG compares to a “normal” amount:

This means that across the tournament and in all circles between two and seven, there was approximately 5.18% variation in RNG between the most-favoured team, and the least-favoured team. In real terms, that is 5.18% of the radius of the blue circle, the moment a new safe circle spawns.

 

For arguments sake, if we apply that logic to the spawn of circle three, it would mean that the worst-affected team would have to move, on average, 118m further than the least-affected team, because of RNG. The full width of circle two is 2250m, for reference, so that’s 118m further in a blue circle 2250m across. If this were applied to circle six, the worst-affected team would have to move an extra 26.7m in a blue circle 465m across. The team would also encounter an equivalent amount in every circle, in every game of the tournament.

 

These figures are not exact, because as I mentioned earlier, it’s possible that each circle stage has a slightly different RNG potential and centralisation bias. Which is why, ideally, we would measure this variation on a circle stage basis, rather than averaging the RNG deviation across all circles (or in this case, circles two to seven).

 

However, it should give us a rough estimate of the kind of variation in RNG teams can encounter across a 16-game tournament. To offer a personal opinion, that degree of variation is probably manageable. It suggests that a team’s circle luck is unlikely to affect their placements in a positive or negative sense, more than 5% of whatever they happen to score. And that’s before you start to consider a team’s preference for playing on the edges of circles, their ability to manage RNG, etc. All of that would minimize the real-terms effect of that RNG variance.

 

I’d perhaps like to see the variation reduced to 2% or even 1%, just as a matter of security. But the implementation of this would depend on how it impacts the gameplay (it may require reducing the potential areas a circle may spawn, for example), spectator value, and most importantly the competitive community’s viewpoint on what constitutes a “negligible” amount of move distance in each individual circle.

 

Flaws and Future Analysis

Much of this analysis could be considered a rough estimation of RNG. There are inherent flaws with the methodology used here. Whilst it should be reasonably accurate, there are a number of issues present that could be improved upon in future, that I haven’t mentioned yet:

  • Water in a circle is a constant problem. When working out the potential shift RNG in any given circle, we must assume that the whole area of the blue circle is a valid spawn location for a new safe circle. With water present, this isn’t true. The amount of water in a given circle reduces the potential area that a safe circle can spawn. As such, this is likely to interfere with our RNG deviation values somewhat.   Fortunately, the potential RNG shift is the same for all teams, in each circle. So, our values are at least balanced across the teams. But it does mean that our scaling for unfavorability-vs-favourability is not perfect in each circle. It may be possible to solve this with an API-accessing script that can determine the amount of water in each circle, though this is likely to be challenging to do still.

  • Our measurements for both distances to circles, and RNG deviation, are both entirely two-dimensional. In other words, it doesn’t factor in the additional distance involved when moving over uneven terrain. Moving across a 1km square on the map that has a mountain in the middle will take longer to do than if the terrain were flat. I suspect the variation this would introduce, if considered, isn’t huge. But if we’re aiming for perfect accuracy, it does need to be considered. An API script would perhaps be able to account for this more accurately.

  • ‘Patterns’ are preference settings available in custom servers that will prefer circle spawns in certain locations. E.g. prefer to spawn circles on a mountain, or in cities. It’s possible that these settings are in effect to some degree in regular and tournament play. This would mess with our deviation settings somewhat, as some of the distance we are attributing to RNG, could actually be a ‘preference’ shift towards or away from a certain pattern. In theory this could be managed by player decision making, as a player could read a circle and expect it to shift a certain direction based on patterns. However, this is a fairly murky area, and after consulting with players, this doesn’t seem to be common.

  • One influence of RNG that we haven’t covered is the ‘cumulative’ effect that favourable shifts may have on a team’s centralisation. Or put another way, the effects that a favourable earlier circle (or circles) may have in reducing the potential RNG shift distance of future circles. I believe that much of the knock-on effect of a favourable circle would be accounted for in our data, as teams that have repeatedly favourable circles will have an RNG deviation percentage that is near zero (or below), at least for that particular set of circles. But there is still some potential for ‘streaks’, the benefits of which may not be obvious in our analysis here. The effect may be something like repeated high-value wins in roulette – your winnings tend to increase exponentially. But in general players tend to come out even after a decent number of games. The counter could also be true, where a player or team has streaks of poor luck that dramatically decrease their winnings over that particular set of games. This may be worth looking at in future, but generally I think the effect on team scoring shouldn’t be too great - there are simply so many circles and matches being played, and also a ‘hard cap’ on the points that could be scored in individual games. If tournaments had a small number of matches, say less than 8, the potential for cumulative favourability to significantly influence team scoring would be greater.

 

All of this is to say – this analysis is a “best guess” on the effects of RNG, based on publicly available information at the time of writing. Despite my best efforts, there may be significant error. This method is also extremely time intensive, I’m not exactly an accomplished statistician, and since beginning this project – PUBG Corp. have been fiddling with circle settings, so the PGL data may not be relevant for much longer.

 

As such, I would personally recommend that going forward, if the community expresses a desire for it, tools and strategies could be developed that would be much more reliable than individuals in analysing Battle Royale mechanics. For example; PUBG Corp. recently opened up a developer API that allows people so inclined to create web apps and other services that can grab player stats from public matches.

 

If this API allows developers to access custom game stats, it should be possible to create an app that can do all the work set out above and more. Such an app could be implemented in all the online leagues and LAN tournaments, collecting data over a massive amount of games. With such a data set, very accurate breakdowns on the influence of circle shift on teams could be done. This could be used to assess the balance of a certain competitive ruleset, and trials could even be done on different rulesets, with a view to hopefully setting a ‘standard’ that is tried and tested.

 

This may occur naturally as players and tournament organisers assess the game themselves, but I generally argue that analytics can significantly speed up that process. Not necessarily by taking a primary role in determining game balance, but by introducing a body of empirical evidence which can inform arguments around particular topics.

 

In my time in game development, I’ve personally also found that data-driven analysis can often highlight problems that are not always obvious to individuals. For example, an interesting point I uncovered whilst performing this analysis was that Grubie had a higher average placement in the tournament than Ghost (7.06 vs 7.25), yet Grubie had scored less placement points by the end of the tournament. I believe this points to a flaw in the placement points scaling in the PGL scoring system, though I may be wrong.

 

This sort of issue may go unnoticed without analysing the scoring/placements of teams. PUBG Corp. themselves may well have analysis tools that are collecting data on public/custom games, and they are perhaps using these to inform general game balancing as well. However, developer transparency is not always as good as we would like, and when we are trying to craft a game into an eSport, it’s always beneficial for the community to be as informed as possible, and to have confidence that the game they dedicate so much time, and even careers to, is as fair as it can be.

 

Conclusions

Through this analysis we can see that random circle shift almost certainly affects teams differently, even in long-form tournaments such as PGL. Whether we consider circle centralisation bias or not, there exists a variation of 5-8% of average ‘unexpected’ shift distance*, between the most-favoured team, and the least-favoured, out of the four teams we have analysed.

 

Perfect balance in anything involving RNG is of course impossible, however the variation we have discovered here is not a negligible amount either. This variation could affect a team so that on average, they must move an additional ~120m to reach centre circle in circle three, for example. That equates to 120m of extra distance in every third circle in every game in the tournament – and each of the other circles would also involve the equivalent amount of extra shift distance.

 

Though this is not the likely scenario teams will actually encounter – the shift variation is more likely be randomly distributed across the games and circles - it does serve to show the potential random unfavorability in the current implementation of circle mechanics. Whilst not negligible, this amount of variation should also not be considered severe.

 

The fact that Ghost, whilst having the worst RNG shift averages when considering centralisation (+1.86%) out of our four teams, came out the eventual winners over Grubie, who had the move favourable shift (-3.32%), is what makes me consider this a ‘manageable’ amount of variation as a total circle average. However, I’m not the one to make the final decision on this. Whether this is an acceptable amount is up to the community to decide.

Further to this, I’d be extremely keen to see how this variation comes out when we have a sufficient data set for all eight circles and can factor in centralisation as well. We were not able to get a good reading on individual circle stages with centralisation considered, as our data set is too slim to break it down to that degree. The 5-8% quoted above is an average for all circles combined. It’s still possible that this variation is different per circle, and if that were the case, it’s likely that would have ramifications for team scoring. If for example, a team were to have more favourable shifts in the final circles, but another only has favourable shifts in the first two. This is another reason to attempt further analysis with tools that could collect this information automatically.

Additional interesting points would include the influence of centralisation in reducing potential for a team to be hit hard by RNG shift. A team such as LGD which was typically 20% further away from the central point than the most centralised team, only suffered around an extra 7% random shift distance. This suggests that whilst centralisation as a strategy is certainly beneficial, it only reduces your potential for unfavourable RNG a small amount.

 

On another note, the effect of the extremely coastal first circles that have been seen throughout competition recently, is obvious in our data. Potentially the most central-looting team in the tournament had by far the most extra rotation distance to get into the first circle, out of the four teams studied. Whilst it could be argued that this is inconsequential due to Ghost securing the most placement points in the tournament, as well as the first circle being the least risky to move into, it’s still surprising that the RNG shift didn’t even come close to averaging out over the 16 games.

To round up, hopefully this has been a useful read for anyone interested in competitive PUBG mechanics. One topic I have deliberately missed out is the effect the RNG variation covered may have had on team scoring, or final standings. Initially, I thought there may either be clear-cut differences between the RNG teams had encountered, and their final placement points. Or, there would be no correlation whatsoever, and as such we could conclude that RNG had no influence on standings.

 

As I delved into this analysis more, I realised that the situation was not going to be as clear cut as that. As mentioned the variations seem to be small (<10% variance in RNG between teams), and methods are not accurate to a high enough degree to account for that. It would be unfair of me to attempt to draw conclusions around team scoring without practically perfect data and analysis methods – neither of which is the case here.

 

Instead, hopefully the competitive community can take this initial groundwork and build upon it with more accurate data, tools and analysis methods. Such data-driven analysis should help the community build a strong ruleset that would ensure fairness in tournament play, and strengthen PUBG’s claim in becoming the next major eSport.

 

*Averaged RNG shift distance for circles two to seven (5%) and all circles (8%). See Results & Analysis section for details on what this means.

Please feel free to message me on Twitter or Discord if you have any comments, feedback, suggestions, questions or other. 

Cheers

 

- Andy 'TameR_' Hair

  • Twitter - Black Circle

- TameR_ #3648

 

Thanks to thisKye, hypoc, and BALLOC for proof-reading and feedback.

Data Set: Spreadsheet

Source Circle Images: Drive Download 

  • Facebook Round
  • SoundCloud Black Round
  • Vimeo Round
  • YouTube Round
  • Twitter Round

© Andrew Hair 2018.