before you go reinventing the wheel, here are the results from running your examples through the formula suggested by Evan Miller in the link... 10 up 10 do

before you go reinventing the wheel, here are the results from running your examples through the formula suggested by Evan Miller in the link... 10 up 10 down 0.327403766068315 50 up 50 down 0.418847795168265 100 up 0 down 0.973657278603792 70 up 30 down 0.620167865061336 2 up 1 down 0.253533865071156 10 up 20 down 0.210836926288844 and a few more to see some other interesting cases... 100 up 30 down 0.703333001615963 10 up 3 down 0.541934244292211 30 up 100 down 0.175849385303401 1 up 0 down 0.269865944074627 1 up 1 down 0.120866317496523 2 up 0 down 0.425030603165412 100 up 0 down 0.973657278603792 100 up 100 down 0.442235043128361 200 up 0 down 0.986652839030471 as you can see. as there are more votes cast, the confidence that this represents the ultimate ranking improves. with less votes, a high up/down ratio is good, but not as good as a high up/down ratio with more total votes (because there is more statistical confidence with more data points). finally below are the test cases that broke your initial implementation. 200 up 1800 down 0.0895005864992562 300 up 4200 down 0.0608069702294983 400 up 7600 down 0.0461419135745087