PSA: Reasonable probability that charm has been nerfed

[SirAlvarex]

On the inverse, when I was pulling in Seb on Sunday I was resisting about 70% of the spells the lower level frogs were throwing at me. I eventually stopped charm pulling casters because "whats the point?"

To point you at your own thread, I think they may have implemented the 255 cap without telling us.

[ripwind]

Originally Posted by SirAlvarex [You must be logged in to view images. Log in or Register.]

To point you at your own thread, I think they may have implemented the 255 cap without telling us.

Sometimes code changes go up to production by accident, too.

[falkun]

No, that would artificially alter the data.

Take all data with the same controls (CHA, player_level, mob_level, debuffs (tash & malo), "natural" charm break, zone, mob), and only discard data that does not match the controls. If you start discarding data because it doesn't fit what you expect, then you are fitting the data to your conclusion(s), not the conclusion(s) to your data.

[Splorf22]

Imagine I am charm soloing for XP. I charm something and send it at thing two. If charm breaks, I recharm it. Finally after it kills the enemy I break charm and kill it. The charms I break are disproportionately the longer ones. Example:

Fight #1: Charm (100s), break, Charm (100s), self break, kill
Fight #2: Charm (200s), self break, kill

If we simply average those we get 133s (and the actual duration is higher!). If we throw out the two that were broken by the caster, we get 100s, which is clearly wrong. And this bias will persist as you add more samples.

I am suggesting "pretend the charm was allowed to continue until it broke on its own by replacing it with one that broke naturally that lasted at least that long"

[falkun]

So you are artificially lowering the mean by instigating the charm break. Wouldn't these be offset by the super short natural charm breaks to get your true distribution? Either way, let the data speak for itself. Unbiased data cannot be right or wrong, it just is, only analysis can be correct or incorrect.

[Splorf22]

Yeah I'm starting to believe I was just pretty unlucky [You must be logged in to view images. Log in or Register.] I had fun doing the math though!

[lecompte]

Originally Posted by Splorf22 [You must be logged in to view images. Log in or Register.]

Yeah I'm starting to believe I was just pretty unlucky [You must be logged in to view images. Log in or Register.] I had fun doing the math though!

And that is what is important. /me pats Splorf22 on the head.

[koros]

Last night in Mistmoore, on my 28 enc, I soloed the last half of the level using the same mobs I had used to previously solo the first part. Charm seemed no different (better if anything)

[koros]

Also your math looks way off in a few areas - 4 of your samples were well over 3 minutes, with 1 (the 2:31 where you broke charm) having the potential to be. Where did the <1% chance for 3 min charm duration come from? Assuming a normal distribution of charm durations (which I doubt from this, although the sample size is small) the standard deviation would have to be huge.

[Splorf22]

Originally Posted by koros [You must be logged in to view images. Log in or Register.]

Last night in Mistmoore, on my 28 enc, I soloed the last half of the level using the same mobs I had used to previously solo the first part. Charm seemed no different (better if anything)

I think its extremely hard for the human brain to measure small differences in the average value of highly random processes. How about you check your logs and post? This goes for you too Writ3r.

Also your math looks way off in a few areas - 4 of your samples were well over 3 minutes, with 1 (the 2:31 where you broke charm) having the potential to be. Where did the <1% chance for 3 min charm duration come from? Assuming a normal distribution of charm durations (which I doubt from this, although the sample size is small) the standard deviation would have to be huge.

I said the average has a 1% chance to be over 3 minutes, not an individual charm. I resampled the data many times to create a distribution for the mean, which should be more or less gaussian even though the distribution for an individual charm is not.