Around the goal: examining the effect of the first goal on the second goal in soccer using survival analysis methods

Daniel Nevo, Ya’acov RitovJournal of Quantitative Analysis in Sports 2013 · DOI · arXiv

Abstract

In this paper we apply survival techniques to soccer data, treating a goal scoring as the event of interest. It specifically concerns the relationship between the time of the first goal in the game and the time of the second goal. In order to do so, the relevant survival analysis concepts are readjusted to fit the problem and a Cox model is developed for the hazard function. Attributes such as time dependent covariates and a frailty term are also being considered. We also use a reliable propensity score to summarize the pre-game covariates. The conclusions derived from the results are that a first goal occurrence could either expedite or impede the next goal scoring, depending on the time it was scored. Moreover, once a goal is scored, another goal scoring become more and more likely as the game progresses. Furthermore, the first goal effect is the same whether the goal was scored or conceded.

Introduction

Predicting soccer scores is notoriously difficult because goals are rare, discrete events. Traditional models often rely on Poisson distributions, which assume that goals happen independently of one another. However, in practice, a goal changes the psychological state of the players and the tactical approach of the coaches. This paper addresses the dependency between the first and second goals, shifting the perspective from how many goals are scored to how much time passes between them.

Methodology

The study employs Survival Analysis, specifically the Cox Proportional Hazards Model. In this framework, the "event of interest" is a goal being scored, and the "hazard function" represents the instantaneous probability of a goal occurring at time t, given that it hasn't happened yet.

To manage the complexity of pre-game variables, the researchers used a Propensity Score called ProbWin—the probability of a home win derived from betting market odds—as a single proxy for team strength and other pre-match information. The model also incorporates time-dependent covariates and tests for "frailty terms" (random effects) to account for unobserved game-specific factors.

Variable definitions used in the survival analysis models.

Data and Results

The analysis was conducted on 760 games from the English Premier League (2008–2010). The dataset includes "censored" observations for games where a second goal never occurred, a scenario survival analysis is uniquely equipped to handle.

Key Statistical Findings

The regression results reveal several critical insights into match dynamics:

  • Team Strength as a Primary Driver: ProbWin was the most significant predictor ($p < 0.001$). For every 0.1 increase in the home team's win probability, the hazard of them scoring increases by approximately 21%.
  • The "Who" Doesn't Matter: Surprisingly, the identity of the team that scored the first goal (FirstGoalTeam) was statistically insignificant ($p = 0.649$). Whether a team scores or concedes, the tactical reaction and its effect on the timing of the next goal are essentially mutual.
  • The 52-Minute Pivot: The model identified a "break-even" point at the 52nd minute. A first goal scored before this time tends to reduce the hazard of a second goal (as teams may tighten up defensively), while a goal after the 52nd minute increases the likelihood of a quick subsequent goal.
  • Increasing Intensity: As the match progresses, the "hazard" of a second goal increases. By 30 minutes after the first goal, the hazard is 1.72 times higher than at the moment of the first goal, likely due to more adventurous tactics and substitutions in the closing stages.
Cox Proportional Hazards Model results showing coefficients and significance levels for match variables.

Conclusion and Limitations

The study suggests that once team strength and goal timing are accounted for, there is little evidence for "game magic" or unobserved frailty terms driving high-scoring matches. However, the model faces limitations in the first 15 minutes of play due to sparse data and struggles with "stoppage time clumping" where goals are recorded at the 45th or 90th minute. Future work could extend this analysis to third or fourth goals, though this would require managing increasingly complex game states and heavier data censoring.

Summary updated

Abstract

In this paper we apply survival techniques to soccer data, treating a goal scoring as the event of interest. It specifically concerns the relationship between the time of the first goal in the game and the time of the second goal. In order to do so, the relevant survival analysis concepts are readjusted to fit the problem and a Cox model is developed for the hazard function. Attributes such as time dependent covariates and a frailty term are also being considered. We also use a reliable propensity score to summarize the pre-game covariates. The conclusions derived from the results are that a first goal occurrence could either expedite or impede the next goal scoring, depending on the time it was scored. Moreover, once a goal is scored, another goal scoring become more and more likely as the game progresses. Furthermore, the first goal effect is the same whether the goal was scored or conceded.
BibTeX
@article{nevo2013around,
  author = {Daniel Nevo and Ya’acov Ritov},
  title = {{Around the goal: examining the effect of the first goal on the second goal in soccer using survival analysis methods}},
  journal = {Journal of Quantitative Analysis in Sports},
  year = {2013},
  doi = {10.1515/jqas-2012-0004},
}
Figure