Gms Nx Cube tier-up rates fluctuate
This weekend I decided to put the "labor" in Labor Day Weekend. I wanted to answer the age old question of "are red cubes or black cubes more cost efficient during miracle time?" Long story short, I watched a bunch of youtube videos, logged the results, and watched the answer flip flop from "red cubes are better!" to "black cubes are better!" and back. It got to this point: http://i.imgur.com/LQaqJ4N.png Since I wasn't really getting anywhere with the answer, I decided to do a comparison of March miracle time tier up rates versus July miracle time tier up rates. Here is what I found: http://i.imgur.com/XtciHB0.png. I ran an A/B test to determine whether my results were statistically significant. For those who do not understand what these numbers mean, the confidence level is basically how confident I am that the tier up rates changed from March to July. Ideally, people generally prefer a threshold of 95%, but it varies. So basically, I am pretty sure that some of the tier up rates changed.
I guess I did answer the question I was originally motivated by. The cost efficiency of red cubes versus black cubes depends on 1)how much the tier-up rates will change for next miracle time and 2)how much the cubes will cost. It's a "let me look into my crystal ball" sort of answer.
Anticipated questions:
Q: What does R, E, U, and L stand for?
A: rare, epic, unique, legendary
Q: Why is the [Rare -> ?] tier up rate for red and black cubes grey?
A: Because there is an insufficient sample size, and because it is way more cost effective to use epic pot 50% scrolls rather than cubes to tier up.
Q: How did you factor in the possibility of tiering up R-->E-->U vs R --> U?
A: (chance of tiering R->E instead of R->U)*((Avg # of cubes to tier R->E)+(Avg # of cubes to tier E->U))+(Chance of tiering R->U instead of R->E)*(Avg # of cubes to tier R->U)
Q: How accurate were you in logging the actual data?
A: I don't have an actual answer to this, but I spot check my recording, and I rarely make mistakes. If anything seems abnormal (60 black cubes no tier ups), I go back and relog the data. If I were to give a number it would be maybe 1%. Because it is a youtube video, I pause and rewind or play at 0.5 speed if I am unsure of something. If I can't see the start potential, I don't log that cube.
Q: How did you choose the videos?
A: I looked up "Maplestory miracle time cubing 2015" sorted by relevance and went down the list. Then I sorted by upload date and went down the list.
Q: Can I see your excel document and not just screen shots?
A: If you can tell me how to upload it. And no I can't manage to upload it to google drive. I tried. EDIT: Yes you can! http://1drv[.ms/1Nhjlpd (remove the [ ).
Q: Is all data GMS cubes?
A: Yes
Q: Is all data miracle time cubing?
A: Yes
Q: Did you watch ALL the videos?
A: No, but I watched a good chunk of the ones available
Q: What if you made up all of this?
A: I didn't, and if you don't believe me you can go through all the videos I went through and compare your results to mine.
Q: Do you have no life?
A: I only have the one I am living.
A huge thank you to all the wonderful youtubers who provided the source material. Thank you Essentiate, imYourSlayer @fearofoneman, DreanLuminos @dreanburster, xxrangerx9 @miked6969, fuwa1122 @constantlyme, WhiteWidower @fiercepaingwin, iJayZero @ijayms, DarkCovenant, SubtractTime, Aldrahil @drager260, Noxuna @noxuna, iTimmix, HugsxXKisses, AcerBeast @drago0nboy, Animefan920 @animefan920, DexlessBucc, Xtremists @hewhoisgod, SatanWings9, DelusionDash @masterobert, McSwagerSon @khainiswager, Timzer, oBec @oBec, Jzurc, Keyan22 @keyan22, BICBOY @multik3d808, XSeu @aeynx, Misusing @misusing, and ForeverLuvMC @night0fsorrow.
TLDR: Tier up rates between the March 2015 miracle time and the July 2015 miracle time have likely changed for certain categories. Namely using black cubes to tier up from U -> L, using black cubes to tier up from E -> U, and using bonus cubes to tier up from R -> E/U.
33 Comments • Newest first
From my own experience, I find black cubes to be better, I also like how you can choose to keep which pot you want, sometimes I don't get amazing things with black cubes, but i'm sometimes lucky, with red cubes i've never gotten lucky though.
Oh ya, I also make videos regardless of the outcome, however its usually lucky or very bad, that's just how it is for me I guess, I also usually cube during miracle time.
ayyyyyyy
@elufu and to everyone else interested. I finally managed to upload my excel sheet http://1drv[.ms/1Nhjlpd (removed the [ in the url). It's a view-only file on skydrive but you can download the file to fiddle with it yourself. You can look at each page for every person's cubing spree. There are definitely some people whose entire youtube video was pretty darn lucky or unlucky. I removed those people from the sample size (Noxuna, Darkcovenant, SubtractTime, iJayZero, DreanLuminos, SatanWings9, Animefan920, keyan22, xseu, BICBOY, Jzurc). I ended up with a 98.5% confidence level for the unique --> leg black cube, which still says something. Of course lucky/unlucky is subjective so feel free to fiddle with the document and remove people from the sample size.
[quote=imaplenolifer]it did go over my head... Thanks for the link. I'll make sure to read the entire thing... Hey wait, arent you the spoiled rich kid who complained how u spent 500k-2000k nx cubing and then prices drop and nexon released violet cubes?[/quote]
You're welcome, but your flattery will get you nowhere.
[quote=lecarde]I'm not doubting your math, but I do think that your sample size was not only small, it was also not a very accurate sample size. People who post videos of miracle time cubing fall into 3 categories. Success videos (look at how awesome I did with so few cubes!), fail videos (look at all these cubes and still no L/L gears ) and people who just upload everything.
On top of only getting videos from a small number of people participating in miracle time, you have to ignore the videos that are specifically [I]"Look at how good/bad miracle time was for me!"[/I], and it doesn't sound like you did that. Once you accoutn for that, your sample size is so small that it is not likely to be representative of real tier up rates at all. Factoring that in, I just can't take your results as accurate, and I have to view them as just random data[/quote]
Excellent criticism! That thought definitely has occurred to me. I am looking at ways to control for that factor. I thought that maybe I could just take people who have cubed during both miracle times. But frankly there aren't enough people who cubed during both miracle times. Literally the only person I can find who has videos for both miracle times is iJayZero. So I have ruled out that method.
Another method is to straight up ask the people "Would you have uploaded the video regardless of the outcome?" Those who say no can be removed from the sample.
Another method is to keep going and collect more evidence.
A couple personal observations though: none of the video titles had the word "lucky" or "unlucky". Some people who thought they were unlucky were around the average tier up rates, and vice versa. Also, people used a good number of cubes. From a glance, the people who consistently beat the odds for every category was fuwa1122 and noxuna. For everyone else,
I will try to do a comparison of how much luckier each person was compared to the average and see if that yields anything.
@loxiona None of the videos were highlight reels. Some videos were cut, but the cut portions were of the person entering and exiting cash shop, and buying the cubes. I guess they didn't want everyone to see their PIN. Either that or they cut scenes of them moving to another lucky cubing place. I am pretty sure that all my videos weren't an amalgamation of cherry-picked tier-up instances. Videos usually started off with the person waiting for miracle time, then going into cash shop to buy cubes, then cubing, then equipping all their gear and checking their range improvement at the very end. From this, I can surmise that they recorded the entirety of their miracle time cubing.
Honestly, the sample sizes aren't big enough to say these numbers are entirely accurate. The best we can say is that while the tier up rates are low in general, the black cubes did give slightly better tier up rates overall for the people in the videos.
[quote=lecarde]People who post videos of miracle time cubing fall into 3 categories. Success videos (look at how awesome I did with so few cubes!), fail videos (look at all these cubes and still no L/L gears ) and people who just upload everything.[/quote]
Yea I agree, for the results to be most accurate, the videos used would have to be uncut and include all successes and failures the person experienced. Given that many try to make cubing look entertaining, they sometimes cut out those long runs where they got nothing interesting
Thanks for using my vid as part of your research! I am Essentiate & I also cubed for fuwa1122
#APStats
[quote=imaplenolifer]Seems pretty valid to me given the sample size and confidence level. In-case if u havent taken stats in uni kid, when the sample size is big, things should be normally distributed via the central limit theorem. And given the high percentage of confidence levels. I trust the data 98%.[/quote]
And the point goes right over your head. It seems you've never learned the definition of validity.
https://en.wikipedia.org/wiki/Validity_(statistics)#Construct_validity
@mckelway: This is great stuff It'll surely be useful to anyone wanting to cube the next Miracle Time in about 2 months (or even just cube a lot during any other time)! Nexon did say that the cube tier up rates have increased before July's DMT though! Whereas they were untouched before March's DMT. Thanks for the work and effort you put into this It's appreciated.
[quote=mckelway]@yuzesun looking at it, reds would have been more cost effective during the July miracle time and blacks would have been more cost effective during the March one. But for the next miracle time, who knows? Rates may change, flipping the tables. What I want people to take away from my post is the notion that NX cube tier up rates can change and HAVE changed.[/quote]
I'm not doubting your math, but I do think that your sample size was not only small, it was also not a very accurate sample size. People who post videos of miracle time cubing fall into 3 categories. Success videos (look at how awesome I did with so few cubes!), fail videos (look at all these cubes and still no L/L gears ) and people who just upload everything.
On top of only getting videos from a small number of people participating in miracle time, you have to ignore the videos that are specifically [I]"Look at how good/bad miracle time was for me!"[/I], and it doesn't sound like you did that. Once you accoutn for that, your sample size is so small that it is not likely to be representative of real tier up rates at all. Factoring that in, I just can't take your results as accurate, and I have to view them as just random data
i actually believe this is the case.
on another note, this investigation doesnt seem bad, i mean you could definitely used it for some kind of uni or school investigation for sure (although it would be weird having the reserch question being something like is there a difference in cubing for an online game) but it could work i guess
inb4 Nexon mafia makes mckelway "disappear"
@axnslicer I don't want to believe my data either, but I can't seem to come up with a plausible explanation to explain the difference. What can explain the massive difference? Here are the standard errors for unique to legendary blacks: March is 10.02% +/- 1.39% and July is 5.89% +/- 0.75%.
[quote=mckelway]@Itachifire Stat tests take into account sample sizes. The lower your sample size, the lower the confidence level. The higher your sample size, the higher confidence level. The closer your percentage tier up rates are, the more samples you will need to have to achieve the same level confidence. Some categories have pretty small sample sizes, but for the black unique to legendary category, I have hundreds of samples. I went back and fudged with the data a little. Even taking out an outlier (the guy who went 60 black cubes with no tier ups), I still have a 98% confidence level.[/quote]
What is wrong isn't your test, your test is fine. What's wrong is there's something strange about your data.
It's very clear that your data and A/B test demonstrate tier up rates have been lowered. The statistical significance is obvious with differences this large. That's not what we're discussing, in fact it's the opposite.
What we're discussing is that your data shows tier up rates almost halved, which is the source of your ridiculously high confidence values. To put it concisely, your data is too extreme.
A 10% stealth nerf to tier rates still falls under Weber's Law. It would be slightly unethical, but not completely unreasonable. In contrast, a 50% nerf is just insane. At this point we need to question the validity of the data.
[quote=mckelway]@Itachifire Stat tests take into account sample sizes. The lower your sample size, the lower the confidence level. The higher your sample size, the higher confidence level. The closer your percentage tier up rates are, the more samples you will need to have to achieve the same level confidence. Some categories have pretty small sample sizes, but for the black unique to legendary category, I have hundreds of samples. I went back and fudged with the data a little. Even taking out an outlier (the guy who went 60 black cubes with no tier ups), I still have a 98% confidence level.
@littletlk That is an excellent idea for something that could be unintentionally influencing the data, so I took out all videos who used less than 50 cubes. 50 being an arbitrary number. This is what I got: http://i.imgur.com/dyHNPTf.png. As you can see, things didn't really change and the ones that were above 95% confidence level are still above the 95% confidence level.[/quote]
Interesting results.
Though, it's unlikely we'll get any answers from Nexon on the rates.
IDK, it seems to be doing them more harm than good to hide them.
[quote=ltachifire]Er I worded that wrong. I meant that I feel like it's only representative of miracle times and not cubing as a whole and there's probably way too many variables to accurately calculate tier up chances.[/quote]
I guess I didn't make it obvious enough but ALL my data is miracle time cubing. Thus conclusions drawn from this data set can only be applied to miracle time cubing. After all, I thought "why would you cube when it isn't miracle time?" Frankly, I wouldn't bother to tier up if it isn't miracle time. So yes, you are absolutely right in that it is only representative of miracle times. And yes, this is miracle time tier up data, and not normal tier up data.
[quote=mckelway]@Itachifire Stat tests take into account sample sizes. The lower your sample size, the lower the confidence level. The higher your sample size, the higher confidence level. The closer your percentage tier up rates are, the more samples you will need to have to achieve the same level confidence. Some categories have pretty small sample sizes, but for the black unique to legendary category, I have hundreds of samples. I went back and fudged with the data a little. Even taking out an outlier (the guy who went 60 black cubes with no tier ups), I still have a 98% confidence level.[/quote]
Er I worded that wrong. I meant that I feel like it's only representative of miracle times and not cubing as a whole and there's probably way too many variables to accurately calculate tier up chances.
@Itachifire Stat tests take into account sample sizes. The lower your sample size, the lower the confidence level. The higher your sample size, the higher confidence level. The closer your percentage tier up rates are, the more samples you will need to have to achieve the same level confidence. Some categories have pretty small sample sizes, but for the black unique to legendary category, I have hundreds of samples. I went back and fudged with the data a little. Even taking out an outlier (the guy who went 60 black cubes with no tier ups), I still have a 98% confidence level.
@littletlk That is an excellent idea for something that could be unintentionally influencing the data, so I took out all videos who used less than 50 cubes. 50 being an arbitrary number. This is what I got: http://i.imgur.com/dyHNPTf.png. As you can see, things didn't really change and the ones that were above 95% confidence level are still above the 95% confidence level.
A further explanation of what I mean:
It's entirely possible you're using biased data. The videos that end up being uploaded might possibly be runs that were only particularly unlucky or runs that were particularly lucky.
For example, if I happened to get 6 tier up from epic to unique in a single black cube package, I'd be much more likely to upload that run to youtube. And, if I got no tier ups in 6 black packages, I'd also upload that. Basically, the more extreme the results of my cubing, the more likely I'd upload it. So it's entirely possible that all the data you used was skewed.
Not saying that this IS the case, but there's certainly the possibility of cherry-picked data not by your fault but by your sources.
Amazing find
Small sample size. But regardless, didn't Nexon specifically state that the tier up rates were changed? So I don't necessarily see how this is surprising or anything.
Also I wouldn't be surprised if tier up rates changed day by day or even hour by hour. It doesn't seem like something that should be kept static.
@imaplenolifer: Go back to school and learn basic reading comprehension kid. We're talking about the validity of the results, not the number crunching. Maybe the difference is something your brain can't comprehend.
[quote=lunaracid]What you have to understand is that with the sample size and the amount of data she was able to gather, these are the cold hard numbers showing the results. Nexon doesn't release any data regarding the changes in cube tier up rates, on or off of miracle time. With the samples available, i.e. youtube videos, the results show that the black tier up rate did drastically decrease between the March and July MTs.[/quote]
What you have to understand is that if someone comes up to you saying "I've proven the existence of aliens", your first response should be skepticism, not wide-eyed acceptance.
I should not have to explain that a nearly 50% nerf in tier up rate is insane. The data does indeed show this, but for a result this extreme I have doubts about the validity of the data.
[quote=mckelway]@yuzesun looking at it, reds would have been more cost effective during the July miracle time and blacks would have been more cost effective during the March one. But for the next miracle time, who knows? Rates may change, flipping the tables. What I want people to take away from my post is the notion that NX cube tier up rates can change and HAVE changed.[/quote]
I'm going to hold off on saying these results are wrong, but I'm very very skeptical. Mostly about the accuracy of the data.
I might go and do my own calculations.
I would like to hear an explanation from Nexon.
Honestly, they could completely clear the air if they'd just release the rate of tier ups.
I don't think they'd lose anything. We'd still buy them even if the rates were low.
[quote=axnslicer]Honestly, I find this questionable simply because of how massive the changes are. Your results suggest that the Black (U->L) tier up rate was nearly halved between March and July, that's an insane difference.[/quote]
What you have to understand is that with the sample size and the amount of data she was able to gather, these are the cold hard numbers showing the results. Nexon doesn't release any data regarding the changes in cube tier up rates, on or off of miracle time. With the samples available, i.e. youtube videos, the results show that the black tier up rate did drastically decrease between the March and July MTs.
@elufu It's an .xlsx @axnslicer The rates you see there aren't the actual tier up rates. There will obviously be a margin of error. But what the A/B statistical test does is tell you whether A or B is better, and how sure you are of that result. So I don't know if July black cubes are ~2x better than March black cubes, but I am 99.5% certain that July cubes are worse than March black cubes.
[quote=mckelway]@yuzesun looking at it, reds would have been more cost effective during the July miracle time and blacks would have been more cost effective during the March one. But for the next miracle time, who knows? Rates may change, flipping the tables. What I want people to take away from my post is the notion that NX cube tier up rates can change and HAVE changed.[/quote]
Honestly, I find this questionable simply because of how massive the changes are. Your results suggest that the Black (U->L) tier up rate was nearly halved between March and July, that's an insane difference.
Your sample sizes are fairly small too, the number of successful tier ups in any category is less than 50. So a lucky streak of even 5-10 tier ups could result in a massive shift in your %s.
^ You mean appears to have changed, not "HAS" changed. Only the devil that is Nexon will know.
[quote=mckelway]If you can tell me how to upload it. And no I can't manage to upload it to google drive. I tried.[/quote]
? Is it an xlsm?
@yuzesun looking at it, reds would have been more cost effective during the July miracle time and blacks would have been more cost effective during the March one. But for the next miracle time, who knows? Rates may change, flipping the tables. What I want people to take away from my post is the notion that NX cube tier up rates can change and HAVE changed.
[quote=yuzesun]So basically in terms of cost effectiveness, tier using red during the july miracle time?[/quote]
Tiered 6 gears legendary from epic-rare with 2 packs of red cubes. I'd go for it.
So basically in terms of cost effectiveness, tier using red during the july miracle time?
Friendly reminder people will actually defend these business practices.