Let's talk numbers

Gameplay Chat (PC/Mobile)
43

I’ll speak for my numbers, while in agree in general to what you say.

I already stated that I’m not using any Impervious troop, so this eliminates this problem. There’s one point that is very true, is that errors are possible. I’m not a robot, I could have missed a few things. What you pointed is true, the most probable way to do a mistake when tracking this is to forget a missed devour. I likely not missed any successful devour since I can see the animation with the fangs and the increase in the stats. So if we consider that the amount of successful devour is true, to reach a devour rate of 45% means 30 mistakes, so 1/4 of my data. I’m pretty confident that I didn’t do that much mistakes.

But again, and as stated in my first message, I would like to hear more from the others to, as you said, help smooth these inconsistencies. My numbers and the numbers of my friends are showing something, but that’s maybe just for us.

44

This is an excellent thread and a very good approach.

However - if you are the only person doing the tests - we won’t be getting enough data to work with. How to say it … If there are 10,000 people playing this game and each of us experienced 10 enemy devour casts with 40% chance to succeed … There’s a 1% chance that at least 1 of these people will experience all 10 devours in line. If we are talking about 100 enemy casts … you’ll get to much higher chances.

Which means - if the devour has 40% chance - data from an experiment showing 57%-ish are well withing the high probability bracket.
Plus - there is one thing that you haven’t considered when talking about other people’s opinions.

People are always more eager to state their opinion and experience if their experience is negative. If if is positive - they will most certainly keep it to themselves unless asked directly.

For that purpose - I’ll do my best to keep track of enemy devours, both kerberos and kraken ones (I don’t expect to meet type-specific devours anytime soon). When I reach 100, I’ll come share the results.

1 Like
45

I just want to expound upon this. Just because none of your arbitrary 10 casts had an amount less than 4, doesn’t mean that if you pluck out some random groups of 10, you’ll never find a group of 3/10. You could have two sets like this:
x=devour o=no devour

xxxoooxoxo oooxoxxxxx

you’ll tell me you’ve got a 5/10, and a 6/10, so it’ll never go down to 40%. But I’ll show you the middle are where it’s 2/10, and show that it is possible.

When it comes to this data, I’d need to dust off my stats books (or google) to find the confidence level in this set of data. Yes, it is high, but it’s still within the realm of possibility.

And when it comes to the assertion: something is wrong with the code, and devour casts more often than it should…I am doubtful, personally. It’s not hard to code a 4/10 chance, as seen by discobot here.

1 Like
46

I know there’s a lot of uninteresting stuff, but you should read the whole topic. I stated that I said this because I formatted my numbers like this and mainly because I wanted my answer to be a taunt. That’s 5 posts above yours. I was tired of reading crap so I came up with this. This is an answer to someone noticing a downward trend while the last 50 events show the same percentage, more or less. The idea is that if the percentage of activation doesn’t change, then it’s never going to go to 40%.

Sorry for the confusion, you lost your time, I already know all of this.

47

I read the whole topic. Every single post. I was explaining that your logic wasn’t necessarily true. It is a downward trend regardless of how you reach your own conclusion.

But now that you’ve decided to have a snide attitude to anyone entering the conversation, or having any reasonable discourse, I’m not sure there’s much else that will occur in terms of dialogue within this topic.

2 Likes
48

My attitude is perfectly fine, I just wanted to be sure that we understand each other. Seems like it’s still not the case since you’re speaking again of “my logic” when this is nothing more than a taunt. I totally agree with what you said and I’m perfectly aware of this, I never meant to say something different than this, at least when I was serious. And thus, I apologized for the time you spent writing the message.

I simply formatted my numbers like this to show multiple steps and used this for a joke.

Did I say something wrong?

Edit: And why did I choose to format the numbers like this? When you are looking for an average value like I did, you need a good amount of samples. The fact that there’s a downward or upward trend is irrelevant. If my numbers were in a different order, we could have noticed an upward trend, or a curve very horizontal. What is important to note, and that’s what I’m fighting to explain, it’s that these numbers could start to tell something. Over the last 50 events, the percentage varied up and down by less than 5% for every bunch of 10 devours, which seems to prove that I was maybe getting closer to something interesting. It’s normal to expect a lot of variation in the first numbers because of the lack of samples. But to be sure, I could use more numbers from more players. That’s it.

49

Because if you are using your numbers as “a joke” or “to taunt” then you are weakening your value as a position of authority. @Vangor, crass as they may be at time, was still making valid points. So to then devalue your observations through a series of taunts is not the road you want to take.

And to your last point, whether it trends (key word there) up or down is not irrelevant. It’s a term called regress to the mean. And if you tell me you know that, then you wouldn’t be saying it’s irrelevant. Yes, there can be ups and downs. Obviously. We get it. But when your data begins to regress toward a particular direction, and continues to do so, you haven’t reached any conclusion.

By saying I’ve wasted my time is the attitude part. You assume I value my time. You are highly incorrect :wink:

1 Like
50

It seems like it’s difficult for me to explain what I mean in english, which is not my native language, or maybe my edit came after your message.

I don’t devalue my numbers or anything else, I was just bored of reading that guy, we both stated our opinion and he continued with what’s looks irrelevant to me, that’s why it ended up like this. We should not worry about this anymore and focus on the interesting part because it seems like we have the same idea but we don’t understand each other.

Let’s imagine my numbers were in a different order? In an order that would have led to an average of 55% every bunch of 10. I chose 10 as an arbitrary value, to have a lot of steps to see the evolution. What conclusions should we have of this? The curve would have been horizontal all the time, so that would have proven that the results were accurate? In my opinion, and for this case, I think that an accurate value, and you seem to agree with that if I understand you, is a value that tends to be constant, with little variations, over the time, so not going up or down constantly bu a huge amount. This is exactly what I have for the last 50 events, and this is why I decided to stop the tracking, for now. I’m not saying these numbers are showing the reality, but they started to show something, that they can be a good enough reason to start the discussion and ask for more data or impressions.

Edit: I think I understood what I maybe explain in a bad way. I totally agree that to start talking about an accurate value, the curve of the result must be horizontal, more or less. What I mean is that the fact that the first bunch of results is going from 70% to 55% does not matter, it does not mean the curve is on a downward trend. It’s just the lack of samples that cause these. That could have been different with the values in a different order, we could have noticed a curve going upward, or as I said, a perfectly horizontal curve. What is important is to be able to notice a curve that is close to horizontal over a good amount of samples, and this is what I started to notice on my results. But who knows, with 100 more that could go down, or that could go up. But the current state is enough to start the discussion, in my opinion.

51

I think the “simple” solution to all this is for more people to run these numbers and see what we can come up with (as others have stated further up the thread). I intended to do just that myself, and urge others to join me and see what we come up with (as we’ve done in some of the Data Collection threads in the past).

Just to make sure everyone is on the same page, @Stoi84 am I correct in assuming the following points?:

  • Ranked PVP battles
  • Tracking Devour changes AGAINST the player (ie enemy devouring you)
  • No Impervious troops allowed on the player team

And are there any other rules that we can add? While I agree it’s worth trying to keep track of cases where damage killed, it’s a bit more difficult when we have people testing on both PC and Mobile, since on PC currently it’s possible to devour even if the damage kills (you devour the next living troop available). So if people want to track this they can, but in terms of keeping a concrete base I’m gonna leave it out for simplicity sake.

1 Like
52

Yes, I tracked the numbers against me, and I tried to restrict to the same game mode and the same strength for the teams in order to limit any improbable thing that probably does not exist (like thinking the percentages are different between the various game modes, or based on the difficulty, etc).

Now that you mention it in a very understandable way, I probably have tracked the devours on the second to last troop without noticing it due to the changes with Unity on PC. That probably creates a little error margin in my data since I probably didn’t count the missed devours when the last unit was killed by the damage. That is for sure very few samples because I think this only affects the Kraken and only if the damage kills the last troop, and I mostly faced Kerberos, but that must be considered.

As for the impervious trait, I didn’t use one to be sure that I’m not missing anything. But as far as I remember, the animation was played when devour was triggered on a troop with impervious, so that would still be possible to track this. Well, that was true before the Unity update.

53

Yes, there’s a little error margin that is now appearing to me, based on the Kraken and the fact that killing the last troop on PC should still trigger the devour. Based on the teams I usually play against, I estimate the missing “missed devours” to be something like 6% or 7% at maximum. Unfortunately, the values are less precise now, the difference could even be a little bigger, but in any case that would make the main idea obsolete. To reach a 40% average devour rate with my numbers, I should have had 43% more missed devours (63/157)

So the final percentage should be closer to 53/54 than 57 if I consider this. This puts the value in a “more probable” block of your graph, but keep the discussion open to further testing.

54

Should it matter that it is a ranked PvP match?

For example, couldn’t one set a Kerberos team up as their own defence, fight against the self, and get the results quite quickly?

I thought that the devour rate would be the same in all areas of play.

1 Like
55

It shouldn’t matter no, I was mostly just trying to keep the same criteria as the OP just in case. The rate should be the same in all areas of play.

56

Well it is obvious to me that according to the op i am too stupid to understand statistics i shall abstain from this endeavor

57

Actually, I started tracking the 25% on TDS, it’s way faster to set a one troop defense and kill it, and there’s no room for mistakes.

For 200 kills, I have a 31% chance of activation, 62/200. This is 24% above the expected number. To reach 25%, the next 48 kills should not trigger a revival.

During a single battle, I had the following numbers of revivals in a row.

1: 30 times
2: 08 times
3: 03 times
7: 01 times (yes, 7 in a row).

This one is very fast to track, and percentages are percentages. I don’t think it matters what is tracked. I’ll try to add more over the next days to see if the value is decreasing.

2 Likes
58

I think you are misunderstanding numbers of statistics for what they are. And it’s getting kind of obvious to me.

31% of TDS revival is totally well within the realm of possibility. Percentages do not tell you how many times out of X an event will occur. It tells you that if your event repeats to infinity as a limit - your results will be getting closer to the stated Percentage number.
Which all of us could witness in the 1st set of data. Your data did follow a reducing pattern.

And this is an indicator that you are already under influence of a bias. “Should not” ? It is still possible to get a 150 revives out of 200. Highly unlikely - but still possible.

Your perception of statistics and chances is not 100% correct. Which is the reason why you feel a difference.
…
In the meantime, I was able to test 62 battles. 50 against Kraken (roughly 45 tested specifically, with set defense) and 12 against Kerberos (all from PvP/GW). And while out of 50 Kraken casts 29 were devours… Just. A. Single. Kerberos managed to devour. Out of 12 casts. All others were missed.
Is that a call-out for one thing or another? No! It’s all statistics. And I can bet 2 fingers on my hand that if you ran an infinite-cast simulation - you’d get to your stated devour/revive chances.

So why do people “feel” the real chances are higher? Because negative experience will always leave a deeper emotional aftershock.

3 Likes
59

If you really want to collect data then I think the fastest way would be to set up an FG/Kerby team as your defense then attack yourself with a Brown/Purple gem creator team like Gorg / Deep Borer / 2x Giant Spider. You take all the Y/R/U/G matches to create B/P gems and then let Kerby have the B/P. Gorg/Deep Borer help slow down death-by-skull-spam from FG while Giant Spider gives some summons for more Devours per match. I’m guessing you could get 10-20 devours per match and do so pretty quickly.

1 Like
60

That said, crowsdourced statistics have uncovered at least two issues that I can recall: the tribute table only drew from the first 25 kingdoms (definite bug), and the red statue bonus was multiplicative instead of additive (possibly design, but poorly communicated if so). If people want to track this stuff, more power to them – as long as the actual analysis is done by people with training.

61

I think you are misunderstanding what I’m saying or trying to do.

I did not say, not a single time, that this 31% was out of the realm of possibility. It’s just a report of what I experienced until now. I even did not draw any conclusion, nothing, I just reported the numbers. I’m trying to track and see if the numbers are constantly far to what is considered the “normal” value to see if there could be some kind of issue.

And please, do not take what I said out of the context. The 48 is just a number to show the current gap that exists between my numbers and the numbers we would expect in the best of the worlds, that’s just another way to express the fact that it’s 24% more than expected at this state of the tests, to be sure that everyone gets the idea behind the numbers. This 24% seems to not bother you, so the rest should not since it’s the same thing, just expressed in a different manner. You can call this popularization if you want. This is nothing more than an illustration of the numbers. And this is why I ran 200 tests, and that I plan to run more to see what’s going to happen. This is exactly what I’m saying, I’ll try to add more over the next days to see if the value is decreasing.

This is exactly the purpose of this thread, to see if we reach this point. But since the early results show some kind of similarities, it was worth to ask for more impressions and data. And to be honest, I did not wait for anyone to test this. I’ll share the resource here since it’s in the discussion now.

That’s a little script I created using JavaScript, simulating TDS revival. It’s a simple html file that you can launch in your browser.The only thing you have to change in the file is this line, to change the number of samples.

var maxSamples = 300;

I found that around 1000 samples the numbers were pretty accurate, that’s what I’m trying to replicate here, but 1000 samples is a lot, thus this thread, and thus the initial question. But what if a lot of numbers were going in that direction?

Edit : Forgot this

This makes absolutely no sense to me. Maybe I’m missing something and you could tell me what. But here, you’re taking into account the order in whom the numbers came to calculate a percentage to say that it follows a reducing pattern, while calculating a percentage is unaffected by this. In other words, If the numbers came in a different order, you would not have been able to say this, and thus, this is not a valid argument. Look at these numbers.

That’s the same amount of devour, just in a different order. Am I missing something?

62

That’s exactly what I started to do, but I used TDS as the opponent. 200 samples so far, I hope for up to 5 times this number over the next days if I can find the time for this or if others participate.