<< Back to Warzone Classic Forum   Search

Posts 1 - 20 of 33   1  2  Next >>   
Global Chat Raffle Analysis: 10/2/2021 23:48:16


l4v.r0v 
Level 59
Report
TL;DR: https://bit.ly/raffle-analysis has some graphs, enjoy
Raffle Analysis

Back in July, functor released some raffle analysis (https://www.warzone.com/Forum/564739-secrets-rafflebot) based on RaffleBot chats courtesy of JK_3. I got a hold of similar chat logs (again thanks to JK_3) for RaffleBot from April 13 to September 25 of this year and did some more analysis.

Some highlights:
  • 5 players- awaythro, krinid, John Smith, faxfox, and JK_3- made up 25.2% of RaffleBot activity. A slightly different group of 5- krinid, John Smith, awaythro, JK_3, and Gunk- took home 25.1% of RaffleBot winnings. There's 784 unique accounts that attempted to raffle during this time period, but just 15 took home over half the winnings.


  • The timing of your raffle entry has no impact on your odds of winning. There's a few times with statistically significant higher-than-expected (11 seconds remaining) and lower-than-expected (18, 15 seconds remaining) win rates, but this significance is spurious due to volume (if you check random 20 p-values, you'd expect 1 to be <5% even when the null hypothesis holds. We checked 54 and got 3 seemingly significant ones where we'd expect 2.7, so overall it's consistent with randomness.). Sorry to anyone who got their hopes up about sniping at the last second! functor's analysis had a graph showing that more raffle wins happen for players who join within the first 10 seconds, but that's because people tend to join raffles earlier rather than later (so it's just a higher base rate). But if you're really superstitious, try to join the raffle with 11 seconds remaining!



  • IMaq AQ was the luckiest player- winning 27 raffles instead of an expected 19. On the other side, Rick Sanchez won only 4 raffles instead of an expected 10. (see "Player Luckiness"). Overall, of course, no one's really lucky or unlucky- these are just the results of randomness, and our findings- a uniform luckiness score distribution & finding 8 lucky/unlucky players when we would expect 9- are highly consistent with this raffle randomness:


  • I visualized when players tend to raffle the most, both by week (of the year) and time of day. You can see that functor was super-active for a few late summer months and then dipped, and that some players raffle all day while others have schedules (see "Player Activity").



  • There's ~154 raffles/week, and you can expect to win 1.01 coins per raffle entry. To earn a Lifetime Membership purely from raffling, the fastest you can expect is about 39 weeks if you join every raffle. More realistically, if you join about 20% of raffles throughout the year, you should be able to fund Yearly Memberships from just raffle winnings.

    The activity stuff might come in handy when planning community events since the time of day activity (see "Activity Trend (Time of Day)") gives a rough snapshot of when players are most active on Global Chat.

    If you're interested in further analysis, you can just take a 'chat log.txt' file (like the ones JK_3 generates), convert it to a CSV, and then import it in place of the very first column on the Raffle Events sheet. You might need to copy and paste some formulas (because there's more raffle events and not everything in the sheet is ARRAYFORMULA'd), but this spreadsheet is extensible to new data if anyone wants to further analyze raffles.


Edited 10/3/2021 13:32:55
Global Chat Raffle Analysis: 10/2/2021 23:48:26


John Smith
Level 56
Report
10/10
Global Chat Raffle Analysis: 10/3/2021 00:51:45


Gunk
Level 56
Report
Fantastic work l4v.r0v
Thank you for sharing your insights into these data sets.
Bravo!
Global Chat Raffle Analysis: 10/3/2021 01:09:51


krinid 
Level 60
Report
And thus 5 Raffle Kings were named.

"Kings awaythro, krinid, John Smith, faxfox, and JK_3".

Pls show respect to these kings in global chat & you might be honoured by the spirit of RaffleBot.
- downvoted post by Loxiiv
Global Chat Raffle Analysis: 10/3/2021 04:00:41


Jacob the Restless 
Level 63
Report
Great work.

To the best rafflers, may god have pity on your souls
Global Chat Raffle Analysis: 10/3/2021 04:46:56


John Smith
Level 56
Report
hey l4v.r0v, can you explain what's going on with the "Player Luckiness" tab? Like why do the numbers no match up?
Global Chat Raffle Analysis: 10/3/2021 09:09:26


l4v.r0v 
Level 59
Report
Good catch, John Smith. There was a formula error causing the sheet to match the wrong player names with data. It's fixed now, as is a logical error in how Luckiness scores were calculated.

For how Luckiness scores work in general, even though it's not your question:

1) Not all raffle entries are equal. Some raffles have more players, and a player who plays during low-activity times (entering small raffles) would expect to have a higher win-rate than someone who enters mainly large raffles.
2) A 1% deviation from expected win rate could be nothing or a big deal, depending on how many raffles you've entered (and the specifics around them, but mainly how many). A player who's entered 7 7-person raffles and won 2 instead of 1 is less surprising than someone who's entered 70 7-person raffles and won 20 instead of 10.

functor's analysis compared actual win rate vs. expected win rate in a random raffle, for high-activity players. Luckiness Score instead uses a Normal approximation of the binomial distribution to look at players' win expectancy and the surprising-ness of their actual results for the specific raffles they entered. Note that we'd expect some players' Luckiness to be significantly high (+90) or low (-90) just due to volume, because an event with a 5% chance should occur about 1 in 20 times:

(https://xkcd.com/882/)

Luckiness scores are roughly uniformly distributed, overall suggesting the obvious- that over/under-performance in raffle results is random, because raffles are random. Overall, we found about as many "lucky" (5) and "unlucky" players (3) as you would expect (9) among the 90 players with good-enough data for us to compute Luckiness scores for.

Edited 10/3/2021 11:59:53
Global Chat Raffle Analysis: 10/3/2021 09:28:56


Shin
Level 59
Report
funny comic about wrongly using p-value
Global Chat Raffle Analysis: 10/3/2021 11:33:02


l4v.r0v 
Level 59
Report
And thus 5 Raffle Kings were named.

"Kings awaythro, krinid, John Smith, faxfox, and JK_3".
There's only 4- awaythro, krinid, John Smith, and JK_3. The remaining member of the top 5 depends on what you look at- for attempts and entries, it's faxfox; for wins, it's CAB; for coins won, it's Gunk.

faxfox under-performs on wins (#7) and winnings (#7) because of bad luck (won 13 fewer raffles than expected) and lower-coin wins (won only 73% of the coins they should expect), on top of entering slightly lower-value/higher-participation raffles than the average player (0.99 expected coins per entry vs. average of 1.01). Maybe RaffleBot is punishing him for saying "!raffle" outside the designated raffle time.

CAB slightly over-performs on wins (#5 on wins vs. #6 on attempts) and outshines faxfox on winnings because CAB joins lower-participation and higher-coin raffles at disproportionate rates (expected win rate of 17.22% vs. the average 13.94%, 1.28 expected coins per entry vs. an average of 1.01), which makes up for him having about the same bad luck as faxfox in terms of winning raffles (relative to his expected win rate; CAB's win rate is still substantially higher at 16% vs. fox's 12%). At about the same Luckiness Rating, CAB still manages to win about 86% of his expected coins, which is great since CAB's expected winnings are much higher than faxfox's even though CAB joined ~10% fewer raffles.

Gunk (#7 on attempts, #6 on entries) rises to #6 on wins and #5 on winnings through a mix of a lucky win rate (unlike CAB and faxfox's bad luck, Gunk has a decent positive Luckiness Score and wins 118% of his expected coins) and high value wins (7.78 coins/win vs. CAB's 6.99 and faxfox's 6.05; the average raffle win gets 7.24 coins).

You can see how relatively lucky Gunk is if you look at only expected winnings, taking away the luck element but keeping factors like timing (joining low-participation or high-reward raffles) and, of course, volume of activity. This is what we would get if raffles just distributed coins evenly to all participants:


In general, the top 10 rafflers all fare badly on luck in terms of how their actual winnings compare to expected winnings. Only Gunk and Lefty win more coins than they should expect to, while krinid rises to the top in terms of winnings not because he's lucky but because his competition is unlucky:


Poor faxfox has lost nearly 300 (expected) coins to sheer chance:


Edited 10/3/2021 13:27:16
Global Chat Raffle Analysis: 10/3/2021 23:38:42


John Smith
Level 56
Report
Truly incredible!
Global Chat Raffle Analysis: 10/4/2021 10:23:29


Kenghis Ghan 
Level 60
Report
The average raffle happens every 65 minutes; the average return to joining a raffle is 1.01 coins (excluding idle rewards of course — we all know how much those are worth). This makes for an average return of 0.9323 coins per hour.

At an exchange rate of 150 coins per dollar (based on black market data from August, 2020), the hourly return from "raffling" is equivalent to $0.006215 per hour. Given a 40 hour work week, this is equivalent to $0.04972 per work day (the global poverty line is set at $1.90 by the world bank based on 2011 prices) or approximately $12.93 per year.

The top 5 raffle participants entered an average of 1359 raffles in a 165 day period. From this, we can expect that the top 5 participants spend 8.922 hours raffling daily. The top raffle participant is expected to spend 10.64 hours raffling daily.

Edited 10/4/2021 10:25:33
Global Chat Raffle Analysis: 10/4/2021 13:48:24


krinid 
Level 60
Report
LOL, KG, you make it sound like the top rafflers have forsaken their jobs in order to try to make a living or minimally try to augment their income with WZ coin earnings.
Global Chat Raffle Analysis: 10/4/2021 14:03:21


Athena's Child 
Level 60
Report
Hey, you just proved that raffles are random!
Global Chat Raffle Analysis: 10/4/2021 16:21:48


l4v.r0v 
Level 59
Report
Well, there's two parts to this: one literally is just analyzing randomness, the other is looking at activity patterns. We've learned, for example, that the best time to raffle is on Sundays around 2:30-3:30 AM US-Pacific time, when you can expect ~80% higher returns per entry than during an average raffling time. And that roughly 11 people (John Smith, awaythro, krinid, JK_3, CAB, faxfox, Gunk, Z, Shin, Kcebnroh, and Lefty) participate in raffles frequently enough to expect to afford a Yearly Membership every year purely from raffle winnings.
Global Chat Raffle Analysis: 10/5/2021 19:47:02


krinid 
Level 60
Report
@Knyte
Perhaps some more stat?

# of chat messages
# of chars entered into chat (measured in KB or MB, whichever makes more sense)
average time between chat messages
average # of msgs entered per raffle entry
average # of chars typed per raffle entry
Socialite chart 1 - order by high (# of chat messages) [primary] & average (# of msgs entered per raffle entry) [secondary]
Socialite chart 2 - order by high (# of chars entered into chat) [primary] & (average # of chars typed per raffle entry) [secondary]

This is meant to show GC activity/participation levels, perhaps to show social rafflers vs just pure rafflers. Socialite 1 vs 2 would show people who talk a lot (# msgs) but don't necessary type a lot each time (# chars in each message, long messages). Socialite 2 would highlight the verbose among us, those who enter lots of characters in their messages.

So I would expect (but the numbers would tell) that of the 11 people you mentioned, JS, k, JK, G, Z, Sh, KCE are gonna be super socialites; L, AT, Z mediums socialities; ff, CAB gonna be close to pure rafflers.

And of course it'd give stats on everyone outside of the 11 as well. And also highlight pure socialites (frequent chatters who don't or only infrequently enter raffles). Surely l4v would be a high socialist medium raffle winner.
Global Chat Raffle Analysis: 10/6/2021 04:40:06


John Smith
Level 56
Report
That'd be neat krinid although the log is strictly raffle chat, we don't have the full chat log (as far as I'm aware).
Global Chat Raffle Analysis: 10/6/2021 05:08:42


l4v.r0v 
Level 59
Report
show GC activity/participation levels
The event logs in the "Raffle Events" tab of the spreadsheet contain all the raw data available to me- only messages from RaffleBot. If general chat logs become available to me, then I could perform this analysis (as long as I could confidently do so without risking a suspension).

Do you have access to general chat logs? If you're willing to share them (and have some confidence that doing so does not risk a perma-ban), then we can proceed with your proposed analyses (and much more). If you have them but are not willing to share them, I could also just build the tools for you that you could use to generate the analysis you seek.

A lot of other interesting analysis could also be done.
Global Chat Raffle Analysis: 10/14/2021 02:00:18


l4v.r0v 
Level 59
Report
TL;DR: If you or a loved one joined the Global Chat raffle on or after 2021/08/23, you may be entitled to coinancial compensation. https://bit.ly/raffle-analysis also has some new stuff if you're curious about which players tend to raffle together.
We've finally got something actually useful out of the Global Chat Raffle Analysis: a true crime story!

Our story begins on May 25, 2021, at 3:25 PM in Central Europe. On that day, a new player- nonolet (https://www.warzone.com/Profile?p=68138926223), who'd joined only 5 weeks earlier on 04/19- got his first taste of the raffle. He joined and won his very first raffle, beating 5 other entrants to win 5 coins and a Supercharge Mine raffle.

Over the next 3 months, nonolet would go on to win 61 more raffles, earning 413 coins, 1 Fog Buster power, 19 Free Cache powers, 2 Inspire Mercenaries powers, 2 Market Raid powers, 9 Supercharge Army Camp powers, 7 Supercharge Mine powers, 11 Time Warp powers, and 5 poor artifacts- altogether, a combined value for $35.93 in raffle winnings. But this wasn't enough!

Enter the criminal instinct!

On August 23, 2021, nearly 3 months after his first raffle, nonolet created his second shot- unoturbo, later #Uno, and now JennyA (https://www.warzone.com/Profile?p=31141482252). Almost immediately, at 10:02 PM Central European time, unoturbo entered and, like nonolet's main, won its first raffle, beating 15 other entrants to gain 3 coins and a Free Cache power. Setting the stage for the future, nonolet also entered the same raffle on his main, nearly doubling his odds from 1/15 (6.7%) to 1/8 (12.5%).

Over the next month, 768 raffles would give away 5,576 coins and an additional 38,050 coins' worth in non-coin rewards.


During this time, exactly a quarter of raffles not only had nonolet enter but had nonolet enter on two accounts- tainting 1,615 worth of coins and another 9,360 worth of non-coin rewards. With his second chance, nonolet increased his expected winnings during these double-raffles by a whopping 73%- and he managed to win 476 coins and another 1960 in coin-equivalent value from non-coin rewards during this time. His double-raffle winnings wildly eclipsed what he got from playing fair- by a factor of 6.

nonolet couldn't help himself- he double-joined 192 raffles over the course of a month.

The Victims

But for nonolet to gain better odds, someone had to pay... those 1,070 expected coin-equivalents gained from entering raffles twice came out of others' fair chances.




The bulk of them came from just 9 players- JK_3, krinid, awaythro, John Smith, Gunk, Kcebnroh, faxfox, Shin, and Lefty. The 4 "Raffle Kings" alone lost 306 expected coin-equivalents to nonolet's double-entries.

How to (almost) get away with raffle fraud

Warzone, as many of you know, has its own alt detection system, which uses device fingerprinting (see https://www.amiunique.org/fp) to detect accounts that play not only on the same network but on the same device. That's right- Warzone doesn't just keep track of where you play from; they also try to track your device. So how, for all this time- up to now, even (nonolet and JennyA still double-raffle, and John Smith's report against this cheating is still in review)- did he hide it?

It's simple: He plays from an iPhone! Apple phones have anti-fingerprinting measures that make them considerably harder to fingerprint than desktop-based Web browsers and other mobile devices, stumping Warzone's rudimentary device fingerprinting. Per an inside source, even today the system doesn't authoritatively flag nonolet and JennyA as the same person.

But he didn't cover his tracks...

nonolet's fatal mistake? He didn't bother disguising his activity. Even today, JennyA and nonolet log on at roughly the same times. :O

Once he started double-raffling, he didn't control himself. When his second chance came onto the scene, nonolet joined the vast majority of his raffles on both accounts:


As a result, JennyA and nonolet had obviously the same activity pattern:



When the players noticed, it became painfully obvious that nonolet and #Uno/unoturbo/JennyA are the same person.



When Jenny's active, nono's active. You can predict over 90% of JennyA's activity schedule just off of nono's. Line them up and it's almost hilarious:


Even playing in the same time zone doesn't come close to explaining this. I checked activity correlations between all frequent RaffleBot players in the dataset, combining #Uno and unoturbo (same account, different usernames, so they show up separately) into one "JennyA" entry. Unlike the earlier graph, I didn't restrict to only weeks when both accounts were active, so the correlations here are weaker.

The strongest correlations?
Players                        | r^2
-------------------------------|-----
JennyA and #Uno                | 0.70
JennyA and unoturbo            | 0.67
JennyA and nonolet             | 0.61
That's right- the only comparison point for the nonolet & JennyA correlation are correlations between JennyA and JennyA under a different username. No other correlation even cracks 0.5. The strongest activity correlations between players are only between nonolet and his alts.

Once nonolet's activity raised red flags, it became obvious nonolet and JennyA are the same person playing from the same location, on the same network, likely using the same iOS device. riskboy88 even recorded him in the act (and cheating in an Idle Battle on top): https://www.youtube.com/watch?v=QF6S_dPVqrQ

Some measure of justice...

nonolet's report is still in the system and, if past handling of alt abuse is anything to go by, the most likely outcome is a warning (and that's if Warzone can overcome the apparent technological limitations in its device fingerprinting).

Meanwhile, at least 141 other players were impacted by nonolet's double-raffling (and that's only during the time period in the Global Chat Raffle Analysis dataset). What happens to them and their lost 1070 coin-equivalents?

For now, I've stepped up to the plate myself to thank nonolet and Warzone.com's rudimentary device fingerprinting for generating this level of entertainment value. It would cost 1,364 coins to make nonolet's victims whole, so I've only provided coin restitution to the 32 victims (excluding myself) who have lost over 1 expected coin to nonolet's raffle cheating. This restitution requires a much more modest sum of only 172 coins to enact.

Were you impacted? How much did you lose? You can find an incomplete list of the victims and their precise expected losses at https://bit.ly/nonolet-victims.

Let us all thank nonolet for creating this level of entertainment value by stooping to alt-cheating on a Global Chat raffle, all to gain illicit winnings worth at most $19.60 in a month. Was it worth it?

In a poetic twist, nonolet's biggest victims are the two active moderators on Global Chat.

Edited 10/14/2021 02:37:14
- downvoted post by Rick Sanchez
Posts 1 - 20 of 33   1  2  Next >>