Skip to main content

Forums / Games / Halo: The Master Chief Collection (Xbox)

Yes Frank, the game session leader IS the P2P host

OP NNMS MXMS

  1. 1
  2. ...
  3. 2
  4. 3
  5. ...
  6. 5
Please note: I have been requested to add this caveat at the beginning of this post. The relative percentage of P2P vs. dedicated server conclusions below apply only to my case and cannot be generalized to all players. This post in no way "disproves" 343i's claim that most games are on dedicated servers. It only demonstrates that in my case, almost all or all of the games are P2P.

In response to complaints by myself and others concerning dedicated servers, several posters have linked to Frank's post at Neogaf and added "/thread". Alternatively, they have linked to the post and requested proof that we are still playing on P2P connections. Both Frank - and the posters who linked to his comment - are at least partially incorrect . . . and the proof is provided.

The MCC performs the matching / voting process via P2P connections. During this time, the session leader is visible ("X" for roster, "A" when the session is highlighted - not on an individual player but on the top roster line itself, look for session leader and the "Privacy: $closed" notification). Once voting is complete, the game either connects to a dedicated server or remains P2P. If it remains P2P, our claim is that the session leader becomes the host. Frank's claim is that this is a UI glitch. This case is demonstrated here.

If the session leader is, in fact, the game host when the connection is P2P, then we would expect the session leader to demonstrate a host advantage. If a significant number of games were migrated to dedicated servers, we would expect the host advantage to be washed out, as the session leader is not the host for dedicated server games.

With a statistical certainty of 99.99999941% (yes, that is the real number), I can demonstrate both that the session leader is the game host and that few if any of my games are on dedis. So when I complain about crappy P2P connections, I would prefer not to be told that I don't know what I'm talking about and I would prefer not to be pointed to statements from 343i that are - in my case, at least - demonstrably wrong. Proof was requested; proof is provided in the post below.

I'm really disappointed at the paucity of information from 343i about the dedicated servers. Of all the problems plaguing MCC, this is one of the top three - if not number one. Yet all we have a vague generalities and incorrect statements. Why do we not know where the datacenters are? Why do we not know what the coverage area is? Why do we not know how many games are P2P and how many are dedis? What is the point of hiding this information? You can find out coverage areas prior to buying a phone . . . so why not when you buy a game for which dedicated servers for multiplayer is one of the primary selling points?

Anyway, statistical dork stuff is in the next post; my raw data and instructions to post your own are in the post after that. The basic conclusions, however, are these:

1. The game session leader is, indeed, the game host for P2P. UI glitches need not apply.

2. Few or zero of the games I play are on dedicated servers.

3. The mean host advantage turns out to be 1.75 places on the post-game leaderboard and 3.23 additional kills (more if any of the games I played were actually on dedis).

4. The host placed first in the match 10 times out of 33 cases, which is more than double the expected rate (approximately 14% based on an average of 7 players per game).

Additional dedi/P2P analysis is possible. If people want to collect data to determine if the countdown or different privacy notices indicate the game has migrated to a dedicated server, read the data rules below. You can post your data here and I can do the number crunching. I cannot use my own data for this purpose because, as far as I can tell, none of my games are on dedis. To determine the validity of the countdown as an indicator, we need data from people who sometimes get dedis and sometimes get P2P.
TL/DR Summary: The claim that P2P hosts can be determined via the pre-game session leader listed in the roster was termed a "glitch" by Frank O'Connor at Neogaf, which implies that the information is inaccurate. Frank's implied claim is demonstrably incorrect. With a statistical certainty of greater than 99%, the pre-game session leader is the host during P2P matches and will place an estimated 1.75 places higher with an estimated 3.23 more kills than when off-host.

Statistical dork stuff

First, the claims to be examined:

1. The pre-game lobby host listed is a UI glitch, implying that the listed information is not accurate and/or bears no relation to the in-game host. Since everyone has the same UI, any result here applied generally - meaning to all players.

2. Most of the lag that it currently attributed to P2P connections is actually the result of netcode issues, and most of the games are actually played on dedicated servers. Since my ability to connect to a dedicated server may not be representative of all players, any result here applies to my situation alone and cannot be generalized to all players.

Since we know that the host in P2P connections has an advantage compared to the other players, then if the pre-game session leader is actually the game host we should be able to detect the host advantage by accumulating enough post-game performance data to see if the session leader tends to perform better on average. If the pre-game session leader is not the host, then being listed as session leader confers no advantage, and the long-term performance of the session leader should be average. This lets us test both claims at the same time. So our null hypothesis, then, for our statistical test is:

Null Hypothesis: Being listed as the pre-game session leader confers no in-game advantage (Frank O'Connor's implied claim).

In order to reject the null hypothesis, we will require a confidence interval of 95%, which is fairly standard for engineering and scientific usage. In other words, we will not conclude that Frank's claim is false unless we are 95% certain that Frank's claim is, indeed, false.

The data we will use comes from 42 consecutive attempts to play matches. No game attempts have been censored (lest we bias our data to the answer we think is "correct"). These 42 consecutive attempts led to 3 games in which P2P was confirmed via the pre-game session leader quitting and causing a host migration, 3 cases in which the pre-game session leader quit prior to the start of the match, but too late for the replacement session leader to be noted, and 3 sets of failed lobbies that required an MCC or Xbox restart to resolve. The data taken during the 33 remaining sessions was pre-game session leader GT, post-game GT score (in kills), post-game GT placement, GT own team kills, and opposing team kills.

The placement (rank) data was processed for analysis as follows:

1. The data was centered such that the median placement was zero, better-than-median placement was positive, and lower-than-median placement was negative. In other words, for a 3v3 game, first place would be "3" and last place would be "-3". This was done to allow combining data from games with a differing number of participants.

2. An additional data set was created using #1, but scaling the data such that first place was "1" and last place was "-1". This data set sacrifices sensitivity in favor of ensuring that a single outlier cannot unduly affect the outcome of the analysis.

The kill data was processed for analysis as follows:

3. The data was centered such that the median number of kills was zero, better-than-median kills were positive, and lower-than-median kills were negative.

4. An additional data set was created using #3, but scaling the data by dividing by the maximum number of kills achieved by any player in the match. This gives the player with the most match kills a "1". This was done for the same reason as #2 above.

The following analyses were performed:

A. T-test: This tests whether the mean of the data is greater than zero with a confidence interval. Since our data is just a sample, the confidence intervals are necessary to avoid us accidentally reaching the wrong conclusion due to purely random factors. This was performed for all 4 data sets.

B. Wilcoxon Signed Rank test: Since the data sets are not pulled from a Gaussian population (ask if you want an explanation), the T-test may yield erroneous results. To guard against this, the non-parametric Wilcoxon test was also performed. This test has the advantage that its results are valid for all distributions. It has the disadvantage that it is less sensitive than the T-test.

The results:

Data set 1:

The T-test yielded a 99.99999914% chance that the pre-game session leader has a real in-game advantage in terms of placement on the post-game leaderboard. The mean benefit is 1.75 places higher.

The Wilcoxon test yielded a > 99% chance that the pre-game session leader has a real in-game advantage in terms of placement on the post-game leaderboard. The median benefit is also 1.75 places higher.

Data set 2:

The T-test yielded a 99.9999999306% chance that the pre-game session leader has a real in-game advantage in terms of placement on the post-game leaderboard. The mean benefit is (0.4876 * # players per team) places higher.

The Wilcoxon test yielded a > 99% chance that the pre-game session leader has a real in-game advantage in terms of placement on the post-game leaderboard. The median benefit is (0.5000 * # players per team) places higher.

Data set 3:

The T-test yielded a 99.9% chance that the pre-game session leader has a real in-game advantage in terms of kills achieved in the game. The mean benefit is an additional 3.23 kills per game.

The Wilcoxon test yielded a > 99% chance that the pre-game session leader has a real in-game advantage in terms of kills achieved in the game. The median benefit is an additional 2.88 kills per game.

Data set 4:

The T-test yielded a 99.99% chance that the pre-game session leader has a real in-game advantage in terms of kills achieved in the game. The mean is an additional (0.3527 * highest # of kills) kills per game.

The Wilcoxon test yielded a > 99% chance that the pre-game session leader has a real in-game advantage in terms of kills achieved in the game. The median is an additional (0.3500 * highest # of kills) kills per game.

Full statistics on Scribd.

It is absolutely false that the in-game session leader has no relationship to the host in a P2P game. This is a universal statement, and it applies to anyone who plays MCC. It is also absolutely false that the lag I experience while gaming is due to poor netcode on dedicated servers. Rather, it is a near certainty that the lag I experience is P2P lag. This is a specific statement, that applies only to myself. It does demonstrate, however, that 343i’s statements about dedicated server availability do not necessarily apply to even players in North America who are located next to large population centers (I live about 1:30 from Chicago and 2:00 from Detroit along the I-94 corridor).
Additional Investigation to be Performed:

If we wish to see if UI indications exist that can distinguish between P2P and dedicated servers, more data is necessary. This data must come from people who can both connect to dedicated servers AND get P2P matches. I cannot provide this data, as all of my connections (or almost all) are likely P2P. In addition, I have not experienced any of the UI indications (countdown, etc.) mentioned in previous threads as possibly being indicative of dedicated servers.

Data Rules for Data to be Analyzed:

1. Data must indicate which games had the suspected UI indication for dedicated servers and which did not.

2. Only Slayer games may be included (if using FFAs, ignore the "Own Team Kills" category)

3. Games must be a continuous sequence of at least 30 completed games. No games are allowed to be excluded. If a session is known to be P2P due to host migration, note that as in my data below.

4. You may not quit any of the games during the data taking period. No matter how badly the game is going, you may not quit until at least 30 consecutive completed games are attained.

5. The consecutive games may be played over the course of several days (below is 4 days of playtime from me).

6. No censoring games. If anyone from 343i checks (and I really hope they do), the data must be correct. If it is not, all credibility is lost, and nothing that comes out of this will be taken seriously. It may not have an impact regardless, but faking data or censoring games ensures that it will be ignored.

7. Data must include all fields listed below. If any are excluded, the set will not be used.

____________________________________________________________________________________________

Format is Lobby Host GT - Lobby Host Kills, Lobby Host Place, Own Team Kills | Enemy Team Kills (Playlist) [UI Indication of Dedicated / P2P]
____________________________________________________________________________________________

1. BradtheBadz - 18, #1/6, 15, 10 | 12, 10, 7 (Snipers) [N/A]
2. A7X xNiGhTmAr3 - 12, #3.5/6, 14, 6 | 30, 12, 8 (Snipers) [N/A]
3. Confirmed P2P (Big Girl Dookys, mid-game quit resulted in blackscreen for all players)
4. LucidIntegral31 - 29, #1/6, 13, -2 | 20, 9, 4 (Snipers) [N/A]
5. Solid Joe - 14, #2/6, 36, 10 | 9, 8, 3 (Snipers) [N/A]
6. Unknown (IFoose quit @ end of voting or beginning of game; replacement host or server unknown)
7. Confirmed P2P (Its Champii, mid-game quit resulted in blackscreen for all players)
8. ProFlaco23 - 2, #6/8, 28, 13, 7 | 7, 5, 1, 1 (Snipers) [N/A]
9. ZkkZ37 - 9, #5/8, 10, 3, 2 | 16, 14, 12, 8 (Snipers) [N/A]
10. DamuRider187 - 18, #1.5/6, 18, 14 | 9, 5, 4 (Snipers) [N/A]
11. BTD Lik - 19, #1/6, 16, 12 | 11, 10, 9 (Snipers) [N/A]
12. Prodigous Baron - 9, #5/8, 14, 7, 0 | 19, 13, 11, 7 (Snipers) [N/A]
13. Prologue - 15, #1/6, 9, 4, | 7, 1, 1 (Snipers) [N/A]
14. You Get Shrekt - 15, #2/8, 11, 9, 4 | 23, 11, 8, 8 (Snipers) [N/A]
15. 7 consecutive fail to connects, 1 session full, 2 MCC restarts, 1 Xbox restart
16. Cold 42o - 28, #1/8, 12, 6, 4 | 19, 13, 12, 6 (Snipers) [N/A]
17. JohnnyP218 - 14, #1/10, 13, 12, 7, 4 | 12, 11, 9, 7, 4 (H2A) [N/A]
18. ThinkingBaton51, 6, #7/8, 17, 16, 6 | 23, 12, 9, 6 (H2A) [N/A]
19. NorCalKnockout - 12, #3.5/8, 12, 6, -1 | 19, 15, 9, 7 (H2A) [N/A]
20. SharedWriter13 - 16, #2.5/8, 8, 7, 3 | 23, 16, 7, 4 (H2A) [N/A]
21. Unknown (Kevin Klutch, quit @ end of voting, new session leader unknown)
22. I IZ CAB0OSE - 9, #2/6, 19, 4 | 6, 4, 1 (Snipers) [N/A]
23. Two consecutive lobbies stuck on searching for more players, lobbies dissolved, MCC restarted
24. Confirmed P2P (DCrane02, went on suicide/betrayal spree, quit @ -5 caused host migration)
25. Y29k - 17, #2.5/6, 21, 5 | 17, 11, 7 (Snipers) [N/A]
26. PhAnToMIII - 11, #4/8, 13, 10, 9 | 24, 18, 5, 3 (Snipers) [N/A]
27. TheTrashMan8 - 17, #2/8, 10, 9, 6, 4 | 24, 15, 11 (game started in lobby as 5v3) [N/A]
28. EndofzWorld240z - 7, #3.5/6, 6, 2 | 23, 21, 7 (Snipers) [N/A]
29. ElImpostor - 15, #2.5/8, 15, 5, 3 | 17, 12, 11, 10 (Snipers) [N/A]
30. KSI V3GAS - 12, #3.5/6, 13, 6, 3 | 16, 12 (game started in lobby as 4v2) [N/A]
31. Alphagamer 359 - 7, #4/8, 18, 15, 4 | 13, 13, 5, 2 (Snipers) [N/A]
32. DannyARTJ - 10, #4.5/8, 28, 7, 5 | 21, 12, 10, 6 (Snipers) [N/A]
33. Simba Suess - 11, #4.5/8, 15, 14, 10 | 19, 11, 8, 6 (Snipers) [N/A]
34. Oracion6 - 3, #6/8, 20, 19, 8 | 13, 13, 2, 0 (Snipers) [N/A]
35. R3v3ng - 9, #7/10, 18, 12, 10, 1 | 14, 13, 13, 6, 3 (H2A) [N/A]
36. Tacherr - 26, #1/10, 13, 9, 2 | 17, 16, 9, 8 (H2A) [N/A]
37. Two consecutive lobbies stuck on searching for more players, lobbies dissolved, MCC restarted
38. GrapeMe25 - 27, #1/6, 16, 9 | 19, 8, 5 (Snipers) [N/A]
39. VOID ViVEK - 23, #1/8, 9, 8, 6 | 15, 14, 12, 9 (Snipers) [N/A]
40. Unknown (missed recording lobby host)
41. Kakashi915 - 18, #1/6, 15 | 16, 14, 13, 7 (game started in lobby as 4v2) [N/A]
42. Blaiseeeeeeee - 16, #4/8, 17, 11, 0 | 23, 19, 5, 3 (Snipers) [N/A]
This is excellent information! We have all known we are on PvP servers 99% of the time. It truly is frustrating to see saying most of the games are dedicated servers.

So frustrating that we the community are still beta testing this game. Absolute nightmare. I honestly quit playing the mp almost a month ago. I refuse to play until there is another major update. Which I know still wont fix all the issues.
Frankie, like the rest of 343, is very deceitful with how he uses words. What he said was not a lie. He said the vast majority of the time it is a UI bug. So what he means is, it is possibly useful for determining host, but it means nothing when you connect to a dedicated server.

If you fail to connect to a dedicated server, it is very likely that the session leader will indeed become host. It would explain why such garbage connections sometimes get host. But that's still not a proven certainty. I posted in the previous thread how to use Cain and Abel -- have you done that? It would be far more concrete. The hard part would be tying the host IP to a gamertag.
This is excellent information! We have all known we are on PvP servers 99% of the time. It truly is frustrating to see Frank lying and saying most of the games are dedicated servers. That is clearly a flat out lie.
That is in no way at all what this proved.
Quote:
Also the fan Dedi/P2P "research" is wildly off (not their fault, as you will see). You will fall back to P2P in certain circumstances, but (embarrassingly) the vast majority of times people are seeing that is actually a UI bug and they are in fact on dedis.
So, did you pretend to not read that part? Plus, the MCC has proven to widely vary in performance for different people based on your location. I, for example, have had little major problems, and rarely do I go blackscreen because the P2P host quit.

So, while your data does support that you in particular have played WAY too many P2P games vs Dedis games, you post is misleading. Not everybody matches those results, in fact it's not even close. You can't prove that everybody plays on P2P more than they should, so I would edit your OP to make it very clear that this is YOUR data that confirms YOUR experiences.
Quote:
Since my ability to connect to a dedicated server may not be representative of all players, any result here applies to my situation alone and cannot be generalized to all players.
This statement should have been at the top rather than at the middle to protect you from the coming -yoink!-storm of people who can't understand numbers.

You seem to have put a LOT of time and energy into your post, so I commend you on that. Thanks for being rational in a sea of irrationality.
This is excellent information! We have all known we are on PvP servers 99% of the time. It truly is frustrating to see Frank lying and saying most of the games are dedicated servers. That is clearly a flat out lie.


That is in no way at all what this proved.
Turtle is 100% correct.

I proved only that I myself am on P2P most or all of the time. In the very first post, I stated that the P2P vs. dedicated server answer applies only to me. It cannot be generalized to other players.

The result that can be generalized to other players is the UI function, as that is generic. It may be that 99% of the games are on dedis and that I am one of eleven people on the planet that cannot connect to them. I doubt that is the case, but nothing I have stated proves otherwise.

Lastly, if everyone could refrain from saying or implying that Frank is lying, I would appreciate it. We have no call to make that accusation. Frank is quite possibly relaying information from the programmers that was either accidentally misstated, accurately stated by the dev team but partially misunderstood by Frank, or any other possibility that does not involve deliberate deception.

This thread is about proof. Unless anyone can prove that Frank deliberately made a false statement, please refrain from making that accusation.
ZnaZanZ wrote:
Quote:
Also the fan Dedi/P2P "research" is wildly off (not their fault, as you will see). You will fall back to P2P in certain circumstances, but (embarrassingly) the vast majority of times people are seeing that is actually a UI bug and they are in fact on dedis.


So, did you pretend to not read that part? Plus, the MCC has proven to widely vary in performance for different people based on your location. I, for example, have had little major problems, and rarely do I go blackscreen because the P2P host quit.

So, while your data does support that you in particular have played WAY too many P2P games vs Dedis games, you post is misleading. Not everybody matches those results, in fact it's not even close. You can't prove that everybody plays on P2P more than they should, so I would edit your OP to make it very clear that this is YOUR data that confirms YOUR experiences.
Quote:
Since my ability to connect to a dedicated server may not be representative of all players, any result here applies to my situation alone and cannot be generalized to all players.


This statement should have been at the top rather than at the middle to protect you from the coming -yoink!-storm of people who can't understand numbers.

You seem to have put a LOT of time and energy into your post, so I commend you on that. Thanks for being rational in a sea of irrationality.
The italicized, bolded part you have is precisely what the post is about, so I'm not sure I understand your objection. The fan dedi/P2P research is based on the information from the roster. Frank implies that information is inaccurate. His implication is false.

The issue about percentage of games on dedicated servers is not the main point of the thread. I cannot prevent people from failing to read or understand, and adding the caveat at the beginning makes no sense. I add the caveat at the point where I discuss the relative percentage of P2P vs. dedi games, and that is the appropriate point. Furthermore, the caveat is in big, bold letters in the results post. Since we can no longer change colors or font sizes, I'm not sure how to make that more plain.

However, I'll go ahead and add it at the top to attempt to stave off at least a portion of the inevitable misunderstandings.
This is excellent information! We have all known we are on PvP servers 99% of the time. It truly is frustrating to see Frank lying and saying most of the games are dedicated servers. That is clearly a flat out lie.

So frustrating that we the community are still beta testing this game. Absolute nightmare. I honestly quit playing the mp almost a month ago. I refuse to play until there is another major update. Which I know still wont fix all the issues.
We just played through the campaigns. At first we couldn't even do that with all the crashing and party disconnections but updates seemed to help.

We only had one crash while playing co-op through Halo 4. The game crashed to the dashboard after completing Infinity. Our party stayed connected though (thank goodness)

May just finish the Spartan Ops while waiting for another content update. Hopefully they actually fix things.

While I anticipated issues with the MCC I'm pretty shocked it turned out like this.
NNMS MXMS wrote:

3. The mean host advantage turns out to be 1.75 places on the post-game leaderboard and 3.23 additional kills (more if any of the games I played were actually on dedis).

Would be very curious to know the methodology behind these statistics.
I think I too play way more on P2P than dedicated server.

It's just that I get host ;)

I have been checking that regularly by shooting the birds on guardian for example.
How many games were Halo 2 Classic? Games with CSHD (CE, H2A, H3, H4) should have minimal if any host advantage. Can you apply the same tests to H2C sub-sample?
NNMS MXMS wrote:

3. The mean host advantage turns out to be 1.75 places on the post-game leaderboard and 3.23 additional kills (more if any of the games I played were actually on dedis).


Would be very curious to know the methodology behind these statistics.
Post #2 and the attached Scribd document describe this.

Conceptually, if being host has no relationship to performance, then we would expect the host to appear about equally in each place on the leaderboard. We can calculate the actual percentages vs. theoretical. The difference is the host advantage. The method in post 2 is mathematically identical to doing this, but less straightforward to explain.
Great to see proof... But everyone knows 343 is lying about the dedicated servers. This has been 'proven' (just via Twitch, playing ourselves, posts) many times over. This post is more 'legit' than most, but P2P is the vast majority of the games and this is known. I actually have a friend that works at 343i that has confirmed this to me, however, many other posts like this drive the point home.

I can also confirm that UNDER 20% of 343i works on MCC after the debacle of a launch.

343 needs to be more transparent for Halo not to die.
How many games were Halo 2 Classic? Games with CSHD (CE, H2A, H3, H4) should have minimal if any host advantage. Can you apply the same tests to H2C sub-sample?
I'm assuming CSHD = client side hit detection... if so, H3 most certainly does not have this.
How many games were Halo 2 Classic? Games with CSHD (CE, H2A, H3, H4) should have minimal if any host advantage. Can you apply the same tests to H2C sub-sample?
I would have thought so going into this, which is why I did it primarily with snipers. Any host advantage would be easier to see. However, I don't really think that's the case.

H2C definitely has more off-host hit detection issues, but there was a noticeable host advantage in the others as well. Game 38, for example, was H2A snipers. We would often die without any ricochet sound or bullet trail (though those would sometimes happen after death). Game 36 was in the H2A playlist and was one of the most frustrating games to play because host was seemingly 1-shotting everyone with a BR.

I don't remember specifics about the prior games, but the general impression was that host advantage in H4 was the least, followed by H2A, then CE, then H2C . . . but without as large a gap between them as I would have thought. I have no H3 games in there, as H3 isn't in snipers, and the only other games on the list are in the H2A playlist.
J Teeps wrote:
Great to see proof... But everyone knows 343 is lying about the dedicated servers. This has been 'proven' (just via Twitch, playing ourselves, posts) many times over. This post is more 'legit' than most, but P2P is the vast majority of the games and this is known. I actually have a friend that works at 343i that has confirmed this to me, however, many other posts like this drive the point home.

I can also confirm that UNDER 20% of 343i works on MCC after the debacle of a launch.

343 needs to be more transparent for Halo not to die.
Please do not use the word lying. Especially with respect to dedis . . . all I proved is that I cannot access them. The results above say nothing for anyone else.
How many games were Halo 2 Classic? Games with CSHD (CE, H2A, H3, H4) should have minimal if any host advantage. Can you apply the same tests to H2C sub-sample?

I'm assuming CSHD = client side hit detection... if so, H3 most certainly does not have this.
Neither does CE, H2A, or H4. The animation of a hit (including hit markers and shields) is done client-side, but actual damage is given by the host. That's why in particularly bad games, you could pump 5 headshots into someone in SWAT and still not have them die. Or you can melee someone 5 times in the back on CE only to have them beat you down after having taken no damage.
NNMS MXMS wrote:
TL/DR Summary: The claim that P2P hosts can be determined via the pre-game session leader listed in the roster was termed a "glitch" by Frank O'Connor at Neogaf, which implies that the information is inaccurate. Frank's implied claim is demonstrably incorrect. With a statistical certainty of greater than 99%, the pre-game session leader is the host during P2P matches and will place an estimated 1.75 places higher with an estimated 3.23 more kills than when off-host.

Statistical dork stuff

First, the claims to be examined:

1. The pre-game lobby host listed is a UI glitch, implying that the listed information is not accurate and/or bears no relation to the in-game host. Since everyone has the same UI, any result here applied generally - meaning to all players.

2. Most of the lag that it currently attributed to P2P connections is actually the result of netcode issues, and most of the games are actually played on dedicated servers. Since my ability to connect to a dedicated server may not be representative of all players, any result here applies to my situation alone and cannot be generalized to all players.

Since we know that the host in P2P connections has an advantage compared to the other players, then if the pre-game session leader is actually the game host we should be able to detect the host advantage by accumulating enough post-game performance data to see if the session leader tends to perform better on average. If the pre-game session leader is not the host, then being listed as session leader confers no advantage, and the long-term performance of the session leader should be average. This lets us test both claims at the same time. So our null hypothesis, then, for our statistical test is:

Null Hypothesis: Being listed as the pre-game session leader confers no in-game advantage (Frank O'Connor's implied claim).

In order to reject the null hypothesis, we will require a confidence interval of 95%, which is fairly standard for engineering and scientific usage. In other words, we will not conclude that Frank's claim is false unless we are 95% certain that Frank's claim is, indeed, false.

The data we will use comes from 42 consecutive attempts to play matches. No game attempts have been censored (lest we bias our data to the answer we think is "correct"). These 42 consecutive attempts led to 3 games in which P2P was confirmed via the pre-game session leader quitting and causing a host migration, 3 cases in which the pre-game session leader quit prior to the start of the match, but too late for the replacement session leader to be noted, and 3 sets of failed lobbies that required an MCC or Xbox restart to resolve. The data taken during the 33 remaining sessions was pre-game session leader GT, post-game GT score (in kills), post-game GT placement, GT own team kills, and opposing team kills.

The placement (rank) data was processed for analysis as follows:

1. The data was centered such that the median placement was zero, better-than-median placement was positive, and lower-than-median placement was negative. In other words, for a 3v3 game, first place would be "3" and last place would be "-3". This was done to allow combining data from games with a differing number of participants.

2. An additional data set was created using #1, but scaling the data such that first place was "1" and last place was "-1". This data set sacrifices sensitivity in favor of ensuring that a single outlier cannot unduly affect the outcome of the analysis.

The kill data was processed for analysis as follows:

3. The data was centered such that the median number of kills was zero, better-than-median kills were positive, and lower-than-median kills were negative.

4. An additional data set was created using #3, but scaling the data by dividing by the maximum number of kills achieved by any player in the match. This gives the player with the most match kills a "1". This was done for the same reason as #2 above.

The following analyses were performed:

A. T-test: This tests whether the mean of the data is greater than zero with a confidence interval. Since our data is just a sample, the confidence intervals are necessary to avoid us accidentally reaching the wrong conclusion due to purely random factors. This was performed for all 4 data sets.

B. Wilcoxon Signed Rank test: Since the data sets are not pulled from a Gaussian population (ask if you want an explanation), the T-test may yield erroneous results. To guard against this, the non-parametric Wilcoxon test was also performed. This test has the advantage that its results are valid for all distributions. It has the disadvantage that it is less sensitive than the T-test.

The results:

Data set 1:

The T-test yielded a 99.99999914% chance that the pre-game session leader has a real in-game advantage in terms of placement on the post-game leaderboard. The mean benefit is 1.75 places higher.

The Wilcoxon test yielded a > 99% chance that the pre-game session leader has a real in-game advantage in terms of placement on the post-game leaderboard. The median benefit is also 1.75 places higher.

Data set 2:

The T-test yielded a 99.9999999306% chance that the pre-game session leader has a real in-game advantage in terms of placement on the post-game leaderboard. The mean benefit is (0.4876 * # players per team) places higher.

The Wilcoxon test yielded a > 99% chance that the pre-game session leader has a real in-game advantage in terms of placement on the post-game leaderboard. The median benefit is (0.5000 * # players per team) places higher.

Data set 3:

The T-test yielded a 99.9% chance that the pre-game session leader has a real in-game advantage in terms of kills achieved in the game. The mean benefit is an additional 3.23 kills per game.

The Wilcoxon test yielded a > 99% chance that the pre-game session leader has a real in-game advantage in terms of kills achieved in the game. The median benefit is an additional 2.88 kills per game.

Data set 4:

The T-test yielded a 99.99% chance that the pre-game session leader has a real in-game advantage in terms of kills achieved in the game. The mean is an additional (0.3527 * highest # of kills) kills per game.

The Wilcoxon test yielded a > 99% chance that the pre-game session leader has a real in-game advantage in terms of kills achieved in the game. The median is an additional (0.3500 * highest # of kills) kills per game.

Full statistics on Sribd.

It is absolutely false that the in-game session leader has no relationship to the host in a P2P game. This is a universal statement, and it applies to anyone who plays MCC. It is also absolutely false that the lag I experience while gaming is due to poor netcode on dedicated servers. Rather, it is a near certainty that the lag I experience is P2P lag. This is a specific statement, that applies only to myself. It does demonstrate, however, that 343i’s statements about dedicated server availability do not necessarily apply to even players in North America who are located next to large population centers (I live about 1:30 from Chicago and 2:00 from Detroit along the I-94 corridor).
I am so happy to see that someone else knows statistics! Did you use Excel or SPSS to analyze?

However, given the low amount of instances analyzed you do not have strong validity nor statistical power. Bring up the instances of actual games played and this will be more accepted (at least by anyone that knows data). Regardless, well done. Do you happen to have the raw data? I would like to run it on SPSS if possible.
  1. 1
  2. ...
  3. 2
  4. 3
  5. ...
  6. 5