Just wanted to bring this back around as we head into season 10. Riot, any chance you could implement a statistical metric system for LP increases/decreases in ranked queues? Then you could individually reward players for their contribution to wins and/or losses.
To reply to all of the commenters that said this system wouldn't be accurate as to real game play, you need to look at how these systems are used in all of the major team sports. These systems are built by studying wins and losses and determining what actions during a game contributed to those wins or losses. These systems typically look at thousands or tens of thousands of events in a single game and can take into account hard to quantify actions like proximity to specific events. Look into how soccer, football and basketball models work and you will see how even actions off the ball (blocking, positioning for defense, etc.) can be taken into account. Many of the big money sports teams employ multiple data scientists to build these models to help them evaluate players, determine what to pay players, how to approach game matchups, etc.
(op.gg's model is probably very minimal, and thus not as accurate, since op.gg doesn't have all of the game events statistics like Riot does. Riot knows every action taken at every time throughout the game. That is a gold mine of data to a model like this.)
For another concern, these systems certainly are exploitable: by performing the actions that have been determined to help win games. So in basketball, if making a basket contributes +0.3% chance to your team's win percentage, you can exploit the system by making baskets. If a turnover results in -0.8% chance to your team's win percentage, you can exploit the system by not turning the ball over. If forcing an opponent to drive to his left results in +0.6% chance to your team's win percentage (because your opponent doesn't shoot well going to his left), then you can exploit the system by forcing your opponent to drive to his left. All of those are certainly exploitable actions and all of them should result in increasing your team's win percentage. So exploiting the system is accomplished by actually playing well.
Third, the system can take into account matchups if they wanted to include that in the model. They could take into account champion differences, lane differences, ranked level differences, etc. All of those would just have to be included in the model. The mathematics would sort out how much each of those factors mattered in game wins or losses. (Incidentally, Riot could use these numbers as additional input into champion balance questions.)
Finally, I'm not suggesting that LP increases/decreases for wins or losses should end. Rather, the LP increases/decreases should be apportioned out based on how each of the players contributed to the win or loss. So the losing team would still suffer a team total of 110 LP decrease (22 LP X 5 players) but the ADC and support who did well in the bot lane could each take -10 LP, while the inting/feeder solo laner could take a -35 LP. I think this would not only help players reach and stay in their proper rank, but it would also make inting/feeding players think twice as they will suffer a much larger LP hit.
The bottom line is Riot has tons and tons of raw data on their games that could be looked at an analyzed to build a very robust model of performance. (Much easier than the major sports where they have to decipher actions from videos of the games) Once they have the model, Riot could then make a video explaining the good things and bad things that players should do to generate wins. Then, new and low ELO players would understand the game better and the maybe overall level of play would improve.
But at a minimum, handing out LP increases or decreases on a proportional basis would be a much needed addition.
