Balance: Causal Effect of First Turret on the Probability of Winning

Introduction

Initially released in Patch 6.15 the First Turret bonus has been associated with a high positive effect on the probability of winning. Statistics released by Riot in here suggest that 71.1% of the teams that acquire the First Turret bonus also win the game. My purpose is to estimate this effect while taking into account concerns over sample selection bias and hopefully obtain a proper causal effect estimate.

PDF Link in here for details on methodology, tables, and figures.

Data

To determine this effect I have collected data on 32,603 matches that took place in seasons 6 and 7. The data was collected by extracting match histories from summoner IDs contained in RIOT's seed match list. Please visit here for more information on how to use the API to extract data and to find download links for the seed matches.

Results

The baseline model simply compares averages across the group that obtains the First Turret and the group that does not. Estimation over the full-sample shows a 26.9% higher probability of winning when you obtain the first turret. When breaking the sample into pre-patch 6.15 and post-patch 6.15 we find the effect to be 25.2% and 28.4%, respectively. As such, the introduction of patch 6.15 led to an increase in the estimate of about 3.3 percentage points. This estimate is affected by time related events so we cannot guarantee that it solely captures the first turret bonus.

The regression discontinuity design model departs from the assumption that teams cannot perfectly control the time gap between the moment each team destroys their first turret. Imagine that the two teams are mutually attacking the first turret. The time difference between the moments in which the red and the blue team acquire their turret cannot be precisely determined by players as it depends on many factors. When control is imprecise then we can define an arbitrarily small interval of time in which the outcome becomes random. For example, if we collect data on matches where teams obtained the first turret by 1 second, then we are bound to find teams that possess more comparable characteristics since the outcome becomes random at that point. This provides a quasi-random setting where we can compare the red teams who marginally obtained the first turret with the red teams who marginally lost it. Results from **parametric **estimates suggest that the first turret bonus produced an effect between 1 and 3% on the probability of winning. Results from non parametric estimates suggest that the effect is almost non-existent.

Conclusion

Analysis based on simple averages is ineffective since it fails to take into account endogeneity present in the relationship between winning a match and obtaining the first turret. Skilled teams are more likely to acquire both, thus establishing a link outside of the causal effect of the first turret bonus. To correct for this we rely on the lack of precise control over a "first turret rush" to match on unobserved characteristics. Comparing teams that barely get the first turret with those that barely lose it shows effects ranging from 0% to 3%, much smaller than any other number shown so far. First turret bonus appears to be balanced and its main purpose is to simply accelerate games which are a priori unbalanced. Details on the analysis are available on the PDF link.