The problemI did a tally of the votes on the first page of ranked maps:
Actual quantities:
SPOILER1 42
2 4
3 9
4 10
5 29
6 40
7 108
8 330
9 402
10 1353
This should be plain to see. The modal value of votes is 10, with a huge tendency towards upper votes. 59% of all votes are 10s.
Are all these users really finding all these maps to be perfect? And 1 is the most effective way to bring down a map's rating, so that's what a user will do when they think a map is in any way overrated. Sure, people are opinionated sometimes, voters need to make greater use of the entire spread, maybe with 5 or 6 representing "this map is pretty average compared to other maps".
Fixing itWe need to establish that 8, 9 and 10 are not the votes to be given to maps ranging from average to good. A first step would be to tell users this on the rating screen. However, that requires faith in the voters, so it might be nice to implement a more complex calculation for the rating system. Weighted votes would be the best way to go about this, using mean to encourage varied votes, and perhaps even vote count to encourage more people to vote to avoid small sample size bias and non-response bias. I posted a similar formula in a previous thread that suggested thumbs up/thumbs down (which wouldn't really work, it would be like allowing users to only vote 1 or 10), but here it is adapted for any sort of numerical system (including the current "out-of-ten"):
vote_weight = abs((vote_maximum/2)-voter.vote_average) * (1+0.1 log10(voter.vote_count))
This still allows users to give out only 1s and 10s (or minimums and maximums) and end up with heavy vote weight, though a standard deviation component could be factored in.
While maps may not see the entire spectrum of scores from 1 to 10, a system such as this should allow for a larger range of scores. As for the old votes, they would be useless in the new system unless a bell curve approach or something similar were used to give current maps a more diverse score range.