A feature of most role-playing system designs is that they run into the classic issue of "Nature vs Nurture". i.e. How much of what a person is like is from inborn genetics, and how much from environment and training? RPG systems tend to imply answers to this question in how various character stats are related and cost. The problem is that this question is not just scientifically unclear, it is politically and emotionally charged, and there are widely varying opinions on the answer.
This article examines how RPG systems run into this issue, and the consequences. My main suggestion is to try making the answer an optional or scalable rule, so a gaming group can choose the option which it is most comfortable with. I don't think there is such a thing as a neutral answer to the question. Just as many people will still find fault with a middle-of-the-road answer as with an extreme one.
A key issue in most role-playing games is the division between "attributes" and "skills". Most RPGs take nearly the same approach to these, although they use different terminology. There is a set of 3 to 9 basic "attributes", possibly some derived stats, and somewhere between 40 and 200 or more "skills".
Attributes are viewed as inborn qualities that all characters have, such as "Strength", "Intelligence", and so forth. They are usually defined as some essential quality of the person, rather than in terms of an activity the person does. i.e. "Speed" rather than "Running", and "Empathy" rather than "Persuasion". They can change in some limited fashion, but are comparatively static.
Skills are a list of potential stats for specific trained fields, such as "climbing" or "lockpicking. The list of skills can range from around 40 skills to 200 or even more, depending on the game. Each character may develop a subset of these, listing only those skills that she has developed.
Game mechanics imply an answer for nature vs nurture in the relationship between attributes and skills. In general, base effectiveness at any skill depends on a single controlling attribute or perhaps the average of 2 or 3 related attributes. Usually a effective skill for trained characters is defined relative to this controlling stat. In many systems (including Storyteller, GURPS, HERO, and d20), effective skill is defined by addition/subtraction from the controlling stat (i.e. "stat - 1"). In others, such as HârnMaster, effective skill is based on a multiple of the stat (i.e. "stat x 3").
In a system where it is easy to increase skill relative to controlling stat(s), it favors nurture over nature. The true capability of a character will be clear when looking only at skills. This conveys the idea that a person is primarily a product of his training and society.
In a system where it is more effective to increase the controlling stat(s), it favors nature over nurture. This conveys the idea that a person is primarily a product of his genetics.
To illustrate this, I will compare the skill bases in GURPS and the HERO System which both define skill as a number based on attribute on 3d6. GURPS falls strongly on the side of nature -- i.e. your base attributes dominate over points in training. HERO falls mildly on the side of nurture -- i.e. points in training are at least equal to points in attribute, and often arguably better. This is tricky to show quantitatively, however. The difficulty in comparing systems is in how to convert attribute and skill numbers between systems.
GURPS and HERO both use 3d6 rolls for skill, which simplifies comparison. However, one still has to determine how to convert attributes. I use two benchmarks for comparing attribute scales: human average and "realistic" human maximum (i.e. the highest non-legendary attribute level, or the recommended maximum for non-cinematic campaigns). Since systems differ in how cinematic or legendary characters are allowed to be, the true attribute maximum is not as reliable. The lower end of the scale is not useful since it is rarely clear what that corresponds to. My two benchmarks for skill are minimal training (i.e. a few days, or the first mechanical tier) and a "realistic" expert skill (i.e. the highest non-legendary level). Because systems differ in how they treat unskilled characters, I choose skill one rather than skill zero as my lower benchmark. The upper skill benchmark is for similar reasons to the attribute maximum.
HERO has average human attribute is 10, and a "realistic" maximum of 20. For average difficulty skills, minimal skill is defined as 8, for a cost of 1 point. Expert skill is around 15, which costs either 6 or 11 points with an average base attribute and costs either 4 or 7 points with a 20 base attribute.
GURPS has average human attribute is 10, and a "realistic" maximum of around 16. For average difficulty skills, minimal skill is defined as attribute - 2, for a cost of 0.5 points. Expert skill is around 16, which costs either 14 or 40 points with an average base attribute and costs 2 points with a 16 base attribute.
The mechanical issue is that the nature-vs-nurture answer determines the relative point cost of attributes and skills. To vary the answer while keeping the same total number of points, you must change both the cost of attributes and skills. Few games seem willing to do this, even in optional rules.
FUDGE takes the stance that the connection is entirely up to the players and GM. i.e. If a character has very high Sword skill, the player should take a high Dexterity as is appropriate. This works in theory for initial character creation, though there is a difficulty that this is identical to an extreme nurture answer. Unfortunately, it utterly fails to work as characters advance during play. A genius character has just as much difficulty learning a new language as a dull one.
A possible option is to dispense with attributes entirely as a separate category. The idea is to simplify the issue by eliminating the divide between nature stats and nurture stats. Most games have skill groups such as sciences, languages, social skills. In many games, characters have ratings in skill groups as well as in individual skills. What is not done, but easily possible, is to eliminate use of attributes in play and use skill groups instead. The nature-vs-nurture question can then be changed into two parallel ratings: natural aptitude in a skill group, and effective level in that skill group.