In the course of doing something entirely different, I ended up mucking about with D&D's attribute system. It could be argued - and I would somewhat agree - that original D&D started out with a 'not bad at all' attribute system, although they didn't end up doing that much with it mechanically, at least to start with. 3d6 gives a nice distribution that should covers about 99% of a population very simply.
But there are a couple of issues I have with it. Firstly, the attributes they picked find it hard to distinguish between different types of physically competent characters. And secondly, the attributes aren't correlated but they aren't in reality entirely orthogonal to each other (I ended up here because I was looking at orthogonality and overlap in a different context).
The system I grew up with was MERP (so based on the Rolemaster system). In principle, this worked okay if you considered that the percentage-based attribute was the percentile of performance. Mechanically it didn't quite seem to work like that though, plus it was possible to get scores of over 100 in certain conditions (which would imply literally super-human performance; or that like D&Ds 18/51 or whatever, they would be differentiations in the top percentile. Given that all attributes had an independent racial modifier, it implied (in principle) that there was another undefined measure of performance. How to explain? A normal human has no modifier to any attribute except Strength, where they have +5. An Elf has a modifier of 0. So in some senses a human with '100' strength is stronger than an Elf with the same. But that 'percentile+bonus' was never defined, except that the percentile itself had a bonus, so in this case the Human would be '+30' and the Elf '+25'. But then in practice most characters will have strength of +5 or 0 without further differentiation, so what was the point of expressing it in percentile terms to begin with? However, in principle it could have worked great, if there were a key to what those levels would imply in terms of performance. They tended to mainly just get used as modifiers to skills though. Still, I might revisit the system one day, it does have some promise.
The next system I really learned was WFRP. And much as I love it, the attribute system was just worse than both D&D and MERP. It was a percentage based system, but not a percentile - it is the percentage chance of passing a relevant attribute test i.e. a character with Strength 6 succeeds in an average Strength test about 60% of the time, a character with Strength 3 succeeds about 30% of the time. The percentage chance of succeeding in an attribute check just doesn't reflect how things like this really work, nor was it easy to derive real-world outputs from the numbers. The only good things about it were that it was more dynamic and tended to improve over time, and it was mechanically very easy in play. Very weirdly to me, some tables in D&D tend to replicate the badness, using the D&D attributes in a WFRP way and using them as the base number for a probability check.
In any case, back to my D&D modifications. What I came up with was:
Latent Factors:
P = Power (2d6) p1 + p2
F = Finesse (1d6) f1
L = Logic (1d6) l1
I = Intellect (1d6) i1
A = Affect (2d6) a1 + a2
W = Will (1d6) w1
Visible Attributes:
Strength = P+1d6
Constitution = P+1d6
Agility = F+p1+1d6
Dexterity = F+2d6
Reasoning = L+I+1d6
Memory = I+W+1d6
Awareness = L+2d6
Intuition = 3d6
Charisma = A+1d6
Empathy = A+W
Willpower = W+2d6
This
gives strong correlations between Strength and Constitution, and
Charisma and Empathy; and mild correlations between Agility and
Strength/Constitution; Agility and Dexterity; Reasoning and Memory.
Clearly this still doesn't catch all the real world trade-offs,
particularly at the very high and low levels, but it gives much more
balance than in the original set. Without any great increase in
complexity, it allows a wider range of characters to be described.
We might think of attributes as having 3 possible implementations - an absolute value, a probability and a modifier. The absolute value is good in physical tests, so strong people can lift stuff, fast people can just run fast etc. The Shadowrun CRPG works on this basis outside of combat, and I liked it. It was mercifully quick but more importantly characters are competent, they can do what they can do. In principle any system can work this way, as long as the outputs are worked out in advance (what can a Strength 20 character actually do).
It worked okay in social tests too but it didn't seem to describe intelligence and memory tests as well as other systems. They genuinely seem to work better as the target for a random chek, the reason being that there are ~infinite things to know, so a character with Intelligence '9' doesn't automatically know everything that a character with Intelligence '4' does, they just have more chance of knowing any random thing. Now, I am not saying that you couldn't use these in an absolute value system, but I think you would have to have a somewhat more detailed character record system (e.g. if a character has INT => x, and History skill >= y OR Heraldry skill >=z, then the character knows the thing, otherwise not. And in effect, if you have to improvise at the table, then the GM is just deciding whether the character knows it or not (since the test is fully deterministic, deciding the conditions on the fly is just the same as deciding the end result, assuming the GM knows what is on the player's character sheet).
In combat sequences (and other skill-in-adversity environments) I think attributes work best as skill modifiers - a stronger might deal more damage than a weaker one, all other things being equal - but if the weaker character is the more skillful fighter, then their quicker and more accurate blow will trump the raw strength of the other, to some degree. The point of the above then is to argue that attributes become more useful if they are in a form where they can be used for all 3 implementations.
