Research Question

Does taking more objectives increase a teams chance of winning in League of Legends? For this presentation all you need to know about the game is, the main goal for each team is to win via breaking the enemy teams nexus. There are multiple neutral objectives that give team wide buffs if taken, which allows the team to become stronger, making breaking the enemy teams nexus easier, this presentation looks into how important / impactful are these objectives really

Objectives included:

  • Dragons
  • Barons
  • Towers
  • Inhibitors
  • Rift Heralds

The response variable is whether the team won or lost

Objectives and Winning

Teams that take more objectives tend to win more often.

This graph shows a strong positive association between the number of objectives taken and a teams winrate

Why Logistic Regression?

The outcome we are trying to predict has only two possibilities win or lose (1 or 0)

I originally planed to do something with simple linear regression, but the data I was looking for wasnt publicly available, so I had to go with something different.

Because the response variable is binary logistic regression is more appropriate than simple linear regression.

Logistic regression models the probability that a team wins based on the number of objectives it takes.

\[ P(\text{Win}) = \frac{1}{1 + e^{-(\beta_0 + \beta_1x)}} \]

Where:

  • \(x\) = total objectives taken
  • \(\beta_0\) = intercept
  • \(\beta_1\) = effect of objectives on the probability of winning

Data and Variables

The dataset contains ranked League of Legends matches.

Each match was split into two observations:

  • One observation for Team 0
  • One observation for Team 1

This resulted in 203,686 team observations.

For each team, the following objectives were counted: Dragons, Barons, Towers, Inhibitors, Rift Heralds

Total objectives were calculated as:

\[ \text{Total Objectives} = \text{Dragons} + \text{Barons} + \text{Towers} + \text{Inhibitors} + \text{Heralds} \]

The response variable was:

\[ \text{Win} = \begin{cases} 1 & \text{if the team won} \\ 0 & \text{if the team lost} \end{cases} \]

Logistic Regression Results

The logistic regression found a strong relationship between objectives taken and winning.

  • Objective coefficient: 0.380
  • Odds ratio: 1.46
  • p-value: < 2e-16

An odds ratio of 1.46 means that for each additional objective taken, the odds of winning were about 46% higher.

This is quite shocking to me the winrate going up that sharply, is wild.

Objectives Taken by Winners and Losers

Winning teams take substantially more objectives on average than losing teams.

One caveat is towers teams that destroy the Nexus normally must destroy several towers along the way, which helps explain the large difference in tower count.

R Code Used for the Analysis

The logistic regression model was created using the glm() function in R.

model <- glm(
  win ~ totalObjectives,
  data = teamData,
  family = binomial
)

summary(model)

exp(coef(model)["totalObjectives"])

family = binomial is used because the outcome has only two possibilities win or loss.

The coefficient is converted into an odds ratio so the result is easier to understand.

Interactive Win Rate Plot

Hover over each point to see the exact number of objectives and corresponding win rate.

Conclusion

As an avid league of legends player, I was very surprised to see how strong objectives impact winrate, I have been wondering if its really the objectives increasing winrate this much, or if teams that are already stronger then their opponent simply have an easier time taking the objectives.