Rehgar - Study
By David Connell
1 Introduction:
In this study we set out to find what playstyle differences between top tier players affect winrates as Rehgar. The first thing we did is obtained data on one most recent game played within the last month from top tier players. Our total sample is 120 games from 200 players. In our study we use many abbreviated notations and this key can be used to understand each.
TD: Takedowns (K + A)
K: Final blows
A: Assists
D: Deaths
TiDGT: Time dead as a percentage
HD: Hero Damage Total
SD: Siege Damage Total
H: Healing
XP: Total XP
Win: Dummy variable 1: Game was won 0: Game was lost
GTnumeric: Game time in minutes
Many of the variables are combined in some form to test different hypothesis. One example would be SDGT which is the amount of siege damage per minute of game time. Please note that a few of the variables specifically “TiD” and “GameTime” were further converted, and do not have meaningful numbers in this chart. A chart of summary statistics is below and some findings from this chart are as follows:
TD: Takedowns (K + A)
K: Final blows
A: Assists
D: Deaths
TiDGT: Time dead as a percentage
HD: Hero Damage Total
SD: Siege Damage Total
H: Healing
XP: Total XP
Win: Dummy variable 1: Game was won 0: Game was lost
GTnumeric: Game time in minutes
Many of the variables are combined in some form to test different hypothesis. One example would be SDGT which is the amount of siege damage per minute of game time. Please note that a few of the variables specifically “TiD” and “GameTime” were further converted, and do not have meaningful numbers in this chart. A chart of summary statistics is below and some findings from this chart are as follows:
- 52.5% of all games in our sample were won as Rehgar
- 14 The approximate average takedowns per game
- 2.5 The approximate average deaths as Rehgar per game
- 8% Average time dead as Rehgar
- 19.68 Minutes was the average game length
2 Key Findings:
Key findings from this model include:
- Securing 1 takedown per minute was correlated with an increased chance of winning a game by 82.5%. Assuming an average game time of 20 minutes this means 1 takedown increases your chance of a winning game by 4.125%
- For every 1% of the game you spend dead as Rehgar, your chance of winning decreases by 1.86%
- If your average death timer is 30 seconds each death will decrease your chance of winning by 4.65%
- Higher healing resulted in higher probability of a win
- The more damage you do the less likely you are to win. This can be explained by a risk-reward situation. Purely doing hero damage that doesn’t result in a takedown puts Rehgar at unnecessary risk. However, if you get a takedown, you get rewarded for taking the risk of dangerous positioning.
3 Models
First off we worked under the logic that many of the statistics obtained from our data were relevant. This included Healing, Siege Damage, Hero Damage, Takedowns (all adjusted for time), Hero Level and % of the game spent dead.
Our original thinking produced the model as follows.
Our original thinking produced the model as follows.
Through this original model we were able to explain about 74.45% of the probability of winning or losing a game, however more improvements allowed us more accuracy. Both Hero Level and Siege Damage were highly insignificant so we dropped them from our model. We also made an adjustment for in-game scaling in which we squared game time for relevant variables. Both of these changes very slightly improved our model.
We then ran a RESET test for specification choice and found that we had adequate specification choice (variables are correlated in the correct functional form (log v linear v exponential etc) and that we are using the correct variables).
However we then wanted to know if it was possible to carry a game through superb healing numbers. We took HGT2 and transformed it using a logarithmic function. This improved our model and suggests that especially high healing numbers impact the chances of winning a game. Our next model is as follows and it explains 74.68% of the chance of winning a game in our sample.
We also wanted to know how accurate our model was, so we ran evaluative statistics against it. This is our final model that we take many of our key findings from.
This model accurately predicted about 80% of the games in our data, with an average absolute error of 29%.. More improvements to our model can create better forecasts. One useful transformation would be to switch this to a LOGIT model that correctly predicts 80% of our sample with a mean absolute error of only 24%:
4 Application:
A quick review of our findings:
In theory optimal Rehgar play includes doing damage often to get many resets from his level 7 talent "Blood and Thunder". The model suggests that more cautious Rehgar players achieve better results. You will still want to get the cooldown reduction and extra damage from Ghost Wolf, however you should do so as safely as possible. One method of thinking, in the professional scene, is that Rehgar should only jump in to make a killing blow. However, this seems to be disproved by our model where killing blows turned out to be insignificant. If you take Blood and Thunder, look to get as many safe Ghost Wolf attacks as possible (even if you choose to hit a safe minion instead of a Hero). Hero damage can also contribute to making a takedown, in which the reward for a takedown is far larger than the risk of dealing hero damage.
Trading damage in lane is inadvisable, especially in the early game. The reward for achieving a takedown is smaller and you'll leave yourself out of mana before early objectives.
Look to maximize your healing throughout the game, the more healing you do the higher your chances of winning are (even though the effect is small). A good strategy is to heal early, often, and the person currently taking damage first.
Because a death is nearly equal to a takedown in the change in win percentage, there is no need to attempt to trade 1 for 1 with Rehgar. To win a game it is much more advisable to try to take kills without dying. This is supported by both theory and our model, and goes hand and hand with our first point, high level Rehgar players are likely taking too much risk with their aggression.
- Higher hero damage meant lower chances of winning
- Higher healing meant higher chances of winning, data suggests that it may be possible to carry a team to victory through healing.
- Deaths and takedowns had similar impacts on chances of winning the game, but opposite direction.'
In theory optimal Rehgar play includes doing damage often to get many resets from his level 7 talent "Blood and Thunder". The model suggests that more cautious Rehgar players achieve better results. You will still want to get the cooldown reduction and extra damage from Ghost Wolf, however you should do so as safely as possible. One method of thinking, in the professional scene, is that Rehgar should only jump in to make a killing blow. However, this seems to be disproved by our model where killing blows turned out to be insignificant. If you take Blood and Thunder, look to get as many safe Ghost Wolf attacks as possible (even if you choose to hit a safe minion instead of a Hero). Hero damage can also contribute to making a takedown, in which the reward for a takedown is far larger than the risk of dealing hero damage.
Trading damage in lane is inadvisable, especially in the early game. The reward for achieving a takedown is smaller and you'll leave yourself out of mana before early objectives.
Look to maximize your healing throughout the game, the more healing you do the higher your chances of winning are (even though the effect is small). A good strategy is to heal early, often, and the person currently taking damage first.
Because a death is nearly equal to a takedown in the change in win percentage, there is no need to attempt to trade 1 for 1 with Rehgar. To win a game it is much more advisable to try to take kills without dying. This is supported by both theory and our model, and goes hand and hand with our first point, high level Rehgar players are likely taking too much risk with their aggression.