Simulationist Stat Generation

[MukkTB]

If your goal is to create a world with the most verisimilitude, what stat generation method would you use for basic NPCs?

#1 3d6 in a row 6 times. This stat generation method does not account for inheritance. Furthermore, you will get a higher number of gimps and supermen in a population than seems to be appropriate.

#2 Standard stat array. This doesn't allow for the edge cases. It doesn't allow for any diabled in the population, or anyone who 'rolled' above or below average.

#3 Reverse engineered stats. The banker should have at least enough wisdom to see through obvious scams, and a moderately good intelligence; maybe 13 int 11 wis. If it was lower than that he couldn't reasonably hold that position without something else going on. The guardsman must have a decent strength and a non gimped constitution, ect.

I would like to emphasize that I'm not talking about important NPCs. Those are normally handcrafted. I'm talking about NPCs that are background characters who are suddenly relevant for some reason. How do you deal with them in a way that brings verisimilitude to the setting?

[Blarmb]

Basic NPCs don't interface with the PCs or world on screen enough to need concrete stats. If anything micromanaging every NPC in any fashion would make keeping verisimilitude more of a challenge than it needed to be.

Basic/Background NPCs just sort of do what they need to without really interfacing with the rules in any capacity. They of return information and take actions they need to.

Unless a particular individual is going to be on screen enough where the PCs could sort of mentally reverse-engineer what's going on under the hood by watching it how it behaves It doesn't need stats.


Barring that, #3.

[SowZ]

How about you do 3d6 10 times and take all the numbers closest to the median, disregarding extreme numbers? 18s and 3s will only happen if, say, three 18s and two 3s were rolled. Then assign the stats in a logical way as opposed to random, taking into account the stats of the parents when assigning best and worse scores. (Typically, someone with high Str. has high Con.)

It wouldn't be worth the trouble for an actual campaign, but solves some realism issues.

[Douglas]

If I really wanted to be fully simulationist, I'd do something like this:
Determine total size of population.
3d6 6 times in order once for each NPC.
If I care about family relationships, try to group similar result sets together.
Determine approximate total numbers of each occupation.
For each NPC, assign an occupation that is at least not wildly unsuited for his stats, with higher weighting towards the occupations his stats are best suited for.

I would, of course, write a program, script, macro, or something to automate the whole process for me, as this is way too much work to do manually for any significant community of NPCs.

In practice, I think the final step would cause this approach to produce results similar to your #3, though perhaps with a greater degree of variation.

As for getting "too many gimps and supermen", 3d6 is pretty much the official definition for what the simulationist distribution should be. It only strikes you as being off because it is so rarely actually applied.

[NotScaryBats]

I think worrying about stats is far less important than giving npcs personalities and histories when regarding simulationism. It matters far less who has a 16 str in town than whether or not 'the strong guy' is having an affair with the innkeeper's wife when trying to make a simulationist town or village.