A statistical analysis of which start conditions improve the likelihood of winning

beetle said:

OTOH, you are not controlling for a huge variable, namely opponent skill. Does FilthyRobot face the same set of opponents with enough frequency to use only those games?

Failing that, how about using only games where FilthyRobot wins by a certain VC, but then looking for a correlation between terrain and win time?

Click to expand...

Controlling for opponent skill would be very difficult, almost impossible

One upside of the fact that he plays in a no-quitters group is that it tends to attract at least mostly serious players (people who have no experience aren't likely to commit to a 4-8 hour game). Also, there was a long stretch of time that encompasses a lot of games when he tended to be quite selective and mostly play with other very skilled players (like zempt, yoruus and purify).

So I don't think there is a vast range in opponent skill, but the variance is definitely there, and there isn't much we can do about it :-\

As for specific win conditions and times, it's a nice idea but sadly I don't think my dataset will be helpful. In multiplayer the "wins" are usually wrapped up by popular vote/consensus when a player has got so far. Nobody is going to wait for someone to take all the capitals. Occasionally there is a genuine space race and someone actually finished, but just as often there is a consensus "player X is an era ahead and just finished Apollo/got nukes, let's concede"

beetle said:

I disagree. The conventional wisdom has long been:
Player Skill >>> Starting Dirt >>> Civ Choice

The frequent civ tier discussions never come to any consensus, so that implies that such tiers are largely arbitrary. The best attempt at an objective numerical ranking ended with only a single point difference between tiers, so again more implication that tiers are imaginary.

Still, it is nice to some objective evidence that the civs are pretty well balanced!

Click to expand...

We have only extremely weak evidence that civs are "pretty well balanced". We have reasonably sound evidence that the basis used for placing civs into "tiers" is poor, however.

It could be there really are tiers in terms of "expected win %, all else equal", but that the factors influencing are different from what is being used even in the "objective" lists.

To illustrate an extreme example, pretend for a moment that coastal start bias was the only attribute between civs that made them more likely to win on average. You would then have a very clear division into two tiers; nations with or without that start bias. However, if the community lacks this information and believes other factors are more influential, there could easily be too much noise to detect the only factor that matters.

So I don't think there is a vast range in opponent skill

Click to expand...

Skill can be separated from style too unfortunately. Two different players with nigh-identical expected win % going into the game could have very different strengths/weaknesses such that they will over or underperform in different situations. You are correct that attempting to account for this is not viable; even for one person skill is transient over 100's of games. If I started playing really seriously today, I would be much stronger as a player by the end of a 200 game sampling, causing enough noise that civs I picked near the end of that time period would appear better on average even if they were identical.

Win times are for SP, I think you are correct to use a measure of whether someone wins at all in MP environment.

HoF is a good test case for showing which civs perform under optimized conditions and is probably your only other available dataset, useful only for making conclusions about that type of game.