Thursday, August 11, 2011

Are Nikephoros' Metrics Worthless?

Some people think so but I actually think they have value. Go read the link (to the left, which obviously some people missed), and then come back.

I think Nurglich's main points are entirely valid, and he has written up something that had been bouncing around in my brain stem for some time, with a few minor variations.

Let me go a bit further on his second main critique, which is that the system ignores factors such as positioning, and unjustifiably grants "optimal" circumstances to the attacker.

I think this is a fair justification, but I do think that Nike's system has some value, and let me explain why.

By granting the measured army "optimal" conditions, and then determining what the army can do with those conditions, you set a "high bar" for what that army can POSSIBLY achieve. If that "high bar" is still not high ENOUGH, we know the army cannot succeed.

It's sort of like seeing if someone can shoot a target while in the prone, shooting off a rest, with a scope, from a hundred yards. If they can hit the target under these "easy" conditions, we might hope they MIGHT be able to hit the target under more difficult ones. On the other hand, if they can't hit the target even when given every advantage, we KNOW they will not hit it in a "live fire" situation.

Thus having good metrics is a necessary but not sufficient condition for killing sufficient amounts of things in the game. Real life games don't give you five uninterrupted turns of shooting at uncovered targets.

My biggest concern is actually not that the metric doesn't accurately reflect what your army will accomplish on the battlefield (it doesn't) but that it's INTERNALLY unbalanced. I think that close combat is SEVERELY over-emphasized through this system, and thus you will see close combat armies do really well in the metrics, provide a false expectation of success, and fail miserably in the field.

I do think that people are taking these metrics far too seriously. All they are is a way of measuring the killing power of your army, and really it's something we've all done in one form or fashion for, well forever. Nike's just laid out some rules for standardization so now we can do these measurements with a common metric. I'm not sure that they standardization is good or even great, but that's the value, having a common metric which we can use if we feel like it, and compare ourselves to other armies.

Finally, I have called Nike a good scientist, because I believe in general he is. However, his approach to this particular problem was flawed not because an empirical approach is doomed but because of the way he constructed his experiment.

The way I understand it, he took some known "losers" and confirmed that their metrics were low, and then took some known "winners" and confirmed that their metrics were high. He then asserted that a high metric was positively correlated to wins. Again, all he's really confirmed here is that having high metrics are a necessary but not sufficient condition for success. What he ought to have done is selected a random sample of players, assessed their metrics, and then done a test to determine whether the metrics were strongly correlated with their success. This would allow him to get a REAL sense of how much, if any, the metrics have to do with winning.

At the end of the day, I applaud Nike for his efforts and Nurglich for his insightful criticism of them. Without people trying new things, we all get dumber. Without people checking to make sure those new things are really good ideas, we all do stupid shit.

Keep it up, gents.