And your post was barely worth that.
‘Wardec changes weren’t a nerf’ says enough.
Go back to the csm thread, it’s overwhelmed by people posting;
‘my god, the ALLY mechanic!’
1 Like
Let’s say I run a tire company. Further, let’s say in a simple statistical analysis I find that 98% of the tires I sell suffer a blowout before they are replaced. This correlation does not imply causation. Perhaps all of my customers run every tire they buy until it explodes, and my tires actually last longer and run safer than any of my competitors.
That said, with a correlation that high I would be negligent (perhaps criminally so) if I didn’t investigate further. Correlations give us possible candidates for causation. A 98% correlation does not mean causation … but it probably means the factors are related (causation, shared cause(s), etc.).
I feel some of you are being more than a little disingenuous here. If we examine 3 potential factors in our search for causation candidates, with respective correlations of .98, .43, and .1, which would you investigate further first? A high correlation is not meaningless…
Dismissing a correlation of .98 requires more than just “that doesn’t mean causation.”
Nope. You were right until this point. “probably” means nothing
You have a correlation, but the probability of a causation is the same as for other random events. Because a LOT of events are correlated without a causation, and correlations can not appear even when there is an actual causation - depending on the study.
Did you consider that maybe there is a defect on the model of cars you sell your tires for ? or the road out of your garage is bad and actually damages the tires ? Or that your customers drive too fast for your tires ?
Those possibilities are none of your concern, so you can discard them. What’s more if you can’t imagine other reasons why the tires would blow out, then of course you can only investigate on this specific correlation. And of course it’s your duty to investigate a possible defect in the tires you sell. But just because there is a possible issue with your tires does not mean there we can skip the investigation. What’s more if someone affirms that your tires are bad that’s defamation.
I am not dismissing the correlation. I am refuting the affirmation that your tires have an issue without further investigation.
I gave you one in the previous post.
BTW I did not say a correlation CAN NOT, I said with no correlation being apparent. Again, I am talking about an experiment, not all the possible experiments.
1 Like
I hesitated to use the word “probably” there, so I’ll give ya that. But “relationship” is a synonym of “correlation”… so I kinda said " correlated factors are probably correlated"… moving on…
The “etc.” covered correlation without any causal relationship, even existing in separate causal trees altogether (rather than different branches of the same tree).
And my one example of how it could be correlation without causation was meant to be just one of many possible.
With your statement that:
I think we can agree to agree to agree?
You know you’ve literally agreed with those of us saying ‘it doesn’t mean causation’. Because none of us have dismissed that there could be causation, we’ve just said that that particular CCP study is useless for trying to assign causation to anything because it was too broad a look and didn’t isolate any factors. Which is what the other posters have been trying to do, which is say ‘correlation means causation’
I’ve literally said that it provided a number of good points to do a more in depth look into.
1 Like
You can but I am not willing to agree that you are right, even if you actually agree with me 
@Nevyn_Auscent yeah that’s exactly what he just said.
1 Like
This argument started out as us referring to a correlation. To which someone said statistics are useless.
That’s how we got here.
No… the argument started when someone took that correlation and tried to use it as a causation.
Rather than ‘Hey here’s an interesting data point, it might or might not mean something’
And said people then objected to being called out on using the data badly.
Statistics isn’t useless when used right, but when used badly it can be extremely harmful even.
1 Like
So, I’m arguing that correlation at the 98% level is strong to the point that any rational being would go all-in.
My turn to train (a source covering the basics of correlations and relationships):
I know that that causation and correlation are different, a fact that is generally irrelevant to this discussion.
The fact is, coincidence of these factors allows for prediction… which is the purpose of correlation… and is absolutely inarguable at the 98% correlational level.
Here is a quote from a “basics of correlation site” which may help illustrate the point.
“A correlation between variables indicates that as one variable changes in value, the other variable tends to change in a specific direction. Understanding that relationship is useful because we can use the value of one variable to predict the value of the other variable.”
That is to say, I can rationally postulate that any noob who joins the game and is not subjected to any form of PvP has a 98% probability of quitting if:
There is a 98% correlation between not experiencing PvP and quitting.
You said that correlation exists. Causation is irrelevant. It could be some factor of swamp gas and sock preference… but I still know they’re going to quit if they don’t experience PvP.
That is what a correlation is.
Read the site. Causation doesn’t matter. A correlation that strong is enough to take action.
We can predict with correlation… and CCP should be predicting with it because it’ll save them a boatload if money.
CCP… if you send me the data or let me design the queries from your available data pool, I’ll be happy to provide a thorough analysis which will withstand a statistical review.
Mo
And this is why you don’t go all in even at a 98% correlation. Yes these are deliberately silly examples. But they highlight the point just fine.
No, that isn’t a correlation, you have turned it into a causation statement.
Because the causation to create that correlation could also equally be “People who quit early don’t play EVE long enough to encounter PvP”
So you are misusing correlations and stating them as if they are causations by ignoring other potential causations.
Edit: So yes, Causation matters, and you should not act simply on correlations.
Those are great examples… most probably of outside relationships where either random variation dropped two things together… or where a higher level variable controlled both.
I’ve long thought there was a link between Nicholas Cage films and suicides… that one is also causation.
That being said, the purpose of correlation is in predicting future outcomes.
It has no other purpose.
To bother observing the correlation and to choose not to react to it negates the value of doing it.
Furthermore, correlation can exist where there is no relationship, which is what your site demonstrates… but the question of relationship is vital to the discussion.
Would you say there is no relationship between noobs being pvp’d in their first 4 months and noobs quitting?
I would have suspected you would argue that there is a strong relationship… but that it doesn’t look like the observed results of CCP’s study.
Can a correlation be wrong? Sure. Is a 98% correlation strong? Yes. Do I think CCP has enough data for that correlation to be bulletproof in a predictive model?
Yeah, I do.
I use statistics pretty regularly. They’re black magic, but they work.
You and Anderson_Geten say it’s using the data badly because it doesn’t say what you want it to say. What was it about cherry picking you said before? And one of you was in such denial about it they said statistics is useless.
I’m offering an explanation for data. If you’ve got better data, then please show it. If you’ve got better explanations for the data being shown to you then go ahead and give them. But the truth is, it doesn’t seem to exist. Even when ccp set out to see if pvp affects new player retention, they come back with data saying the opposite of what they expected.
2 Likes
Which they failed to isolate out things such as hour 1 quitters from the statistics who may have not even gotten as far as pressing the undock button before deciding they didn’t like EVE. Now yes, isolating out factors you may still end up with the same results at which point you can start to assign a possible causation factor to PvP as a retention tool, but depending on how many hour 1 quitters we have, you may also get the opposite data once you actually isolate factors out.
It’s simply a correlation, it’s certainly an interesting data point to take a deeper look at, but it’s not a meaningful data point that you can draw a solid conclusion from.
Offering an explanation for data does not make it a causation, and without both the data set and the ability to get additional data sets I can’t offer a ‘better’ explanation, because any explanations you or I offer are pure guess work, but I can (& just did) offer alternative explanations that could be possible, but without further study we have no idea which are or aren’t accurate.
Hence, Causation != Correlation, and you can’t use a correlation as evidence.
I understand the survivorship bias and all, and possibility Eve’s overall retention is generally low.
But this says that 98% of players who join Eve, and are not killed by another player, will quit before 4 months.
If that was a great retention rate, one would guess they would advertise that.
I’m making the assumption that the retention results are better for people who are killed by other players. I guess it’s possible that they quit 99% of the time, making the 98% probability of quitting a 100% improvement in retention.
If that’s the case… Eve Online is going to soo. Be Eve Offline… because I’m thinking that retention is too low…
I’m really trying to see why you guys think a 2% retention rate on PvE only players at the 4 month mark is just an interesting data point… it sounds like time to panic to me.
I’m pretty sure the 98%/4 month stat was that 98% players who did literally nothing but mine would quit before 4 months… (as in no exploration, no missions, no industry, not even any of the other forms of PvE)
Can you double check your reference because I really feel you are mixing up 2 different stats here.
no.
This is not a fact. This is your personnal interpretation, and something that is widely known as false. Exhibiting a correlation from a statistical analysis is not making a prediction, it’s looking at the past.
No. Just, no. Unless you replay the same exact experience. And since we are talking about taking actions, the experience will be different.
If you want to take actions to reduce an event, you NEED to consider what CAUSED this event and not what it is correlated to.
What can we do to get rid of an illness that makes the victim vomit ? According to you, since illness and vomit are correlated, preventing people from eating will reduce the vomit, thus the illness. Sorry, that’s stupid.
Even worse, in a town people who were ill went to the church to be blessed. The water contained bacteria, and people who were already weak because of the illness developed benign symptom that made medics misunderstand the illness, and thought there was a new illness. People kept becoming ill even after healing (illness was not deadly) because both illness were weakening the body - until they found out the blessed water had bacteria.
So even if you have a very good correlation between two events in a case and there is a statistical causation, in some cases your predictions are false.
No. It’s getting hint on what we can further experiment to find out causations.
Just because we say your opinion is based on fallacies is false, does not mean the opposite opinion is false.
I personally think that players who don’t PVP lack a lot of knowledge about the game and so, they can’t really choose to do what is interesting and thus it is more likely they will leave (less possibility to find something interesting). But it is my opinion, and not backed by any study.
No. “strong” means nothing in that sentence, and especially since you want to slide the meaning to “strong enough to use as a causation”. It is NEVER a causation, even with 100%.
Then you are wrong pretty regularly with this mindset of yours. And no, they are not black magic.
No it does not, because your formulation implies a causation. It says there is a correlation, and you translates that as a causation, hence : no.
Or banned bots, or spy alts, or alts for an event (I made some for the gallente race event, in a shuttle it was boring but actually easily doable)
Who is saying it means causation?
No you didn’t. You made an unsubstantiated argument. Now substantiate it. Give an example.
Both Ghost & Daichi for a start are saying it does.
I’m saying it does not equal causation… but that it doesn’t matter.
I’m saying a 98% rate predicated on a good sample, assuming the system hasn’t changed, would be expected to stay at 98%.
i.e. the use of correlation for prediction of future results.
1 Like
For it to predict future events, it has to be causation.
Correlation simply means A & B have some similar values.
For it to make any kind of future impact, it means something has to be causing it.
Therefore, you are saying it’s causation.