Yuku free message boards

chaperone wrote:

---

I always felt that a notable (but not huge) weakness in our computation was that one N vote with 1 P vote avg'd to neutral. Three responses: First: The addition of a mixed tone makes this phenomenon less of an issue as (though not mathematically) because when you see a handful of N votes and a handful of P votes it's pretty clear that it should be mixed neutral.

Second: Those P and N votes averaging to 0 (neutral) actually does in some ways line up to exactly what the rating should have been (mixed neutral).

Third: I took an approach 2 seasons ago where I had an alternative MORDiff that I calculated retroactively back to Guatemala where I used the absolute value of the tone so that a negative and a positive equally weigh toward the presence of tone and did not cancel one another out and some seasons it brought the winner closer to the MORDiff chart and in other seasons it moved the winner further away.

See, I don't understand why Mixed is seen as equivalent to neutral in the first place. This assumes that:

A) Having no tone is, for the purpose of MORDiff and figuring out tone averages, the exact same thing as having both P and N tone.
B) The mix of tone was evenly distributed (ie: the exact same amount of N and P). It's going to be close, but is close close enough?

To me, a mixed person is about as far from neutral as you can get. Neutral denotes absolutely no tone, which is a possible indicator that the editors are ignoring or marginalizing a character. M indicates that BOTH tones are present and likely indicates attention from the editors.

These are clearly not the same thing, so why do the math like they are? If you throw enough statistics at a concept, you're going to get some spurious connections. The true test of effectiveness is whether your statistics conform to an a priori theory (that is, you predict that a connection will be seen for such and such a reason). If you change the formula in a way that makes more theoretical sense and it doesn't yield something, you still have to ask how sure you are of your statistics.

Overall, you can't rely on the math as much as I would like to hope, but to throw the baby out with the bathwater is a bit extreme. The winner is always in the top half of the MORDiff and in most cases in the top 4-5 players (at the merge there are usually a few left around the top) which allows human analysis to come into it's proper place rather than just seeking a mathematical formula to find the winner...which is only fun for the mathematician and for nobody else.

You don't need to find the winner absolutely with math, but you should work to make it as close as possible. If nothing else, you need to guard yourself from the accusations people will (and do) make that Edgic is just a guessing game where people get lucky sometimes and don't get lucky other times. Your best weapon, the MORDiff, is something no one even knows why it works. The scale it is working from also has questionable validity, as we don't even know what single variable it's measuring.

No matter what you do, you're never ever going to come up with a mathematical formula to perfectly predict the way human beings create and edit a story from unique and independent events. So no one should be afraid that Edgic will turn into a math game. However, if you want to make a strong stand against naysayers, you should (in my opinion) do everything in your power to put together a statistic that can confirm a priori predictions based on Edgic theory. I have no doubt that MORDiff works and Edgic should continue to use it until something better comes along. However, by foregoing progress in Edgic theory in the name of the chase, debate, and uncertainty, you will continue to invite skeptics and naysayers.