Linear vs curved dice mechanics

[Irish Wolf]

Originally Posted by whswhs

Actually, that's not true. Consider a very simple rule: Before you can try to learn from experience, you have to succeed in using the skill (roll below); when you try to learn from experience, you have to roll higher than the skill. That's easily understood by people who don't do probabilities. If you are comfortable with math, you can work out that at 1% skill your chance of learning is 1%x99% = 0.99%; at 10% it's 10%x90% = 9.00%; at 50% it's 50%x50% = 25%; and so on up to 99% is 99%x1% = 0.99%.

Bill

Except that I, for one, also learn from my failures; each time I fail a "skill check" in RL, I learn one more way that it doesn't work, and discard that approach, increasing my chances of success next time. (Admittedly, there are some fields you don't want to use this method for - driving, for instance, or operating demolition equipment...)

[Nymdok]

Originally Posted by Rocket Man

To each their own, but I've never been in a situation where I wanted to assign a percentage modifier. I always think in terms of incremental ones ... and I base it on the situation at hand rather than "What bonuses/penalties will give this character X chance of success?"

Granted, in a "curved system," incremental modifiers will have a varying impact depending on your skill level, as already discussed. But I think that's a good thing. I cannot readily conceive of any real-life situation where adding a new factor would add 5 percent to the success chances for everyone attempting it, regardless of skill.

And this is largely the difference. Im not really looking for a 'real world' example or analog. I like the flat distros because they are more visibile in my head and therefore more easily gamable.

For our group, we're not really interested things like simulationism, suspension of disbelief or other things that require 'real world' anchors.

Nymdok

[whswhs]

Originally Posted by Irish Wolf

Except that I, for one, also learn from my failures; each time I fail a "skill check" in RL, I learn one more way that it doesn't work, and discard that approach, increasing my chances of success next time. (Admittedly, there are some fields you don't want to use this method for - driving, for instance, or operating demolition equipment...)

Hardly any RPGs use that approach. It's certainly not a feature of GURPS.

But in any case, it's a different question than whether the BRP experience rule involves arcane calculations or not, which is all I was commenting on.

Bill Stoddard

[Ze'Manel Cunha]

Originally Posted by Nymdok

For our group, we're not really interested things like simulationism, suspension of disbelief or other things that require 'real world' anchors.

That's exactly the point though, if you want a game with willing suspension of disbelief a curved distribution tends to fit the bill, but it can be too staid for some groups.

If you want more random chaotic results which is what some groups consider exciting, then a linear distribution fits the bill, but it can break suspension of disbelief for some groups.

Personally, I prefer my random chaos is specific cases, not in the normal actions taken by characters in a game, I like verisimilitude for suspension of disbelief, which means I find the results from linear distributions silly and unwieldy in play.

[Joseph Paul]

Originally Posted by roguebfl

Can, but it not easy, it not a 'natural fit' and actully requires the GM to have a working knowledge of normal-like curves

Roguebfl - if you can get a copy of RQII or even RQIII. There are even reworked RQ rules sets <cough> GORE<cough> out on the web that will let you see that it was in fact elegant and simple to use like so many of Chaosium's rules.

What I still don't see in this discussion is any evidence for the assertion that a gaussian distrubution is actually models real life. Any one want to tackle that or give a link to past discussions about it?

[teviet]

Although Gaussian distributions come up a lot in real life, the way that GURPS converts skill levels and modifiers to a target number in a Gaussian distribution is completely arbitrary, done for game-mechanical simplicity not for agreement with the underlying distributions. The same is true for systems with flat distributions.

E.g. bullets fired at a target may be approximately Gaussian-distributed. But skill or difficulty is measured as a multiplicative factor on the width of the distribution, not an additive term applied to the target cutoff.

[Aside]
Once when I was young and foolish I decided to design the perfect game system that put an end to this ambiguity and defined exactly what "+1 to skill" actually meant: I defined each +1 to represent a doubling of effort (making twice as many attempts, firing twice as many shots, taking twice as long, etc.). This led to the straightforward probability of success for modified target number n (which could be positive or negative):

P(n) = 1 - 2^(-2^n)

Of course no simple combination of die rolls would generate this, so I made it a table to be compared to a percentile die roll. I then stepped back and said, "Teviet, this is why you're a physicist and not a professional game designer."
[/Aside]

TeV

[ak_aramis]

Originally Posted by Nymdok

I still dont think Im understanding you correctly Ze, Brandon has the right of it.

74% is 74%.

Nymdok

Unless you start looking at modification lists.

In a d% system, a +5 is always +5%

In a 3d6 system, a +1 to a 10 is about +5%... but to a 3 it's +10%, and to a 16, it's only about +1%

You can't simply look at the chance of success in the abstract - almost all games include modifiers. (For example, in my Rogue Trader campaign, the Navigator is routinely having a +40% or more modifier to his 52% navigation... and sometimes as much as +150. (+10 from warbsbane hull, +10 from Navis Prima, +10 from charts, +10 from the RT, ±20 from route chosen, and up to +90 from success on finding the astronomicon.)

And a +1 modifier on a
+1 on2d10 ranges from +10% (10- to 11-) to +1% (19- to 20- or 1- to 2-).
+1 on 3d6 from 1% to 13%,
+1 on 4d6 from 11.1% to 0.1%

[Joseph Paul]

+1 to a score of 10 is a 12.5 % increase (50% >62.5%. Going from 3 to 4 is about a 2% increase (.046% >2%). I am not at all sure that I understood your example since I got very different values.

[Ze'Manel Cunha]

Originally Posted by Joseph Paul

What I still don't see in this discussion is any evidence for the assertion that a gaussian distrubution is actually models real life. Any one want to tackle that or give a link to past discussions about it?

This is a simplification, but if you do something and then repeat that action the exact same way in real life with all conditions the same, then the result is the same.
Skill is about the ability to repeat the same action in the right way to get the same result consistently, the more skilled someone is the more consistently they get the same result and the more varied the modulating conditions they can overcome.

Skill level is about depicting what someone is normally capable of doing, consistently, day in, day out, that capability is a curved distribution which approximates a normal distribution.

Skill, abilities, testable IQ, and just about any other human trait and action, (height is the standard example given in statistical textbooks for this), is found to be a curved distribution, again, real life approximating a normal distribution.

Curved distributions do break down at the extremes, the ends, which is where chaotic random tables can come in perfectly fine, but if your every skill action is inconsistent, if your performance is all over the place, if you're using linear distribution, then you aren't describing skill use, you're describing random actions.

Granted, when you watch someone with no skill doing something you maybe tempted to used linear dice mechanics to describe their "skill", but that's just the thing, that's not skill use, that's someone without skill randomly flailing about.

[roguebfl]

Originally Posted by Joseph Paul

Roguebfl - if you can get a copy of RQII or even RQIII. There are even reworked RQ rules sets <cough> GORE<cough> out on the web that will let you see that it was in fact elegant and simple to use like so many of Chaosium's rules.

Not thanks, I've played both RQ and SpaceQuest, the system was *shudder* it work ok with the group i played it with, but we were all younger high school students then, and all of math geeks in compensation with each other for the school's math award, ever other group I've played it with a much more 'diverse' grouping (but still mostly dominated by various flavor of geek) it fell over like lead balloon... this includes a group that USED the weapon modes for THACO in 1st Ed AD&D

The main thing that made RQ interesting was the flavour text in the crit tables, but that got old once the novelty wore off.