The first aspect I compared in my games was to see whether gold collection or XP collection actually produce a good indicator of which team will win a game. I used the total gold accumulated to make a ratio where I divide gold accumulated by the winning team by the gold accumulated by the losing team. A ratio of 1 means both teams accumulated the same, anything above one represents the winning team having more gold than the losing team.
All the data below is based on my last 131 matches. As you can see it appears pretty obvious from looking at these ratios that both are indicators on average to what team will win, and it is obvious that the team to get the most will tend to win. However looking at the individual numbers, I do see that there are 8 out of the 131 matches where a team with less XP actually managed to win. Most of them are quite close to 1, however there is a 0.84 and a 0.86, which sort of suggest that in general the rule works but there are circumstances where the cumulative XP isn't the only indicator.
Interestingly however, is the fact that the gold ratio is a much much better predictor. Only one game in my records has a 0.99 ratio, which is basically 1. While it does appear that gold might be the more important number here, it did make me wonder what kind of effect towers have in this. How much of this gold is actually coming from towers, in which case the late game, especially when the other team is doomed in their fountain, produces large amounts of gold, that are hardly useful for the purpose of predicting a winner. I suppose it would be much more useful to look at numbers excluding tower gold, or numbers up to the last 5-10 minutes. Another option which is probably more verifiable is to look at last hits, since this will potentially be less biased, however the late game can offer a large amount of creep killing as well.Another cool little graph I produced was looking at the same gold ratio vs the duration of a game. I figured that games that last longer will generally have a more even gold distribution as teams are probably more balanced. From plotting these 131 games, it is pretty clear that this trend is accurate. For the most part games with a gold ratio close to 1 tended to last 45-65 minutes. As an aside, the average duration of my 131 games was 40.6 minutes.
My last comparison involves looking at my performance and how much I contribute to team gold in my wins and losses. Basically I divide my total accumulated gold by the gold accumulated by my team. Initially this sounded like a good comparison and I have a 20.7 ± 0.6 % during my losses, and a 19.0 ± 0.4 % participation during my wins. While there may even be a statistical difference suggesting that I do seem to be less important of a player during games I lose, I think not accounting for my role adds a lot of noise to the data. I mean this is an obvious stat to look at, but I wanted to see how much of an impact I actually have. A much more meaninful comparison would be to isolate only the gold I contribute in games I play a particular hero or a particular role.
