divine-s_portfolio

The Survival Formula: Predictors of EPL Consistency

Soccernomics famously argues that a club’s wage bill is the single best predictor of its league position. However, this project challenges that consensus by analyzing the “Non-Relegated 9”—the specific cohort of teams that have avoided relegation for 10 consecutive seasons (2014–2024).

Research Question: “For Premier League clubs with long-term survival status, is Wage Expenditure truly the dominant predictor of success, or do Managerial Stability and Squad Experience play a more significant statistical role?”

By isolating this unique group, this study uses regression analysis to determine the most important predictor of points accumulation among three competing variables:

1. Financial Power (Wage Bill)
2. Managerial Stability (Tenure Duration)
3. Squad “Game Intelligence (Relative Experience Index)

Scope

I will be collecting the following data points: Club, Season, Total Wage Bill, League Position, Manager Tenure(In Days), Matches Played and Points. Additional Data Points: Squad Age, GD, XG, Wage Spent Per Position.

Teams

• Arsenal
• Everton
• Liverpool
• Manchester United
• Tottenham
• Chelsea
• Manchester City
• Crystal Palace
• West Ham

Methodology

1. Manager Tenure(Stability):

To strictly measure “Ex Ante” stability (stability prior to performance), manager tenure is calculated using a September 2nd Snapshot Rule:

-Definition: Tenure is defined as the consecutive days a manager has held the position as of September 2nd of the target season.

-Rationale: This date aligns with the closure of the English Premier League Transfer Window. It effectively attributes the squad’s construction and pre-season preparation to the manager in charge at that deadline.

-Noise Reduction: This method isolates “Preparation Stability” from “Survival Tenure,” preventing reverse causality where match results influence the tenure variable (e.g., mid-season sackings).

2.The Relative Experience Index (REI)

Rather than using average age, which fails to capture game intelligence, I engineered a Relative Experience Index.

-Metric: Cumulative career appearances in Europe’s “Top 5 Leagues” (EPL, La Liga, Bundesliga, Serie A, Ligue 1) for all players with >5 apps in the season.

-Normalization: Raw experience counts were normalized against the 9-team cohort average for each specific season. * Formula: `Team_Avg_Experience / Cohort_Avg_Experience

• Why Normalize? This controls for the secular trend of increasing career lengths over the seasons, ensuring that a “highly experienced” team in 2014 is comparable to one in 2024.

3. Data Integrity & Constraints

-Leap Year Adjustment: The algorithms account for the extra calendar day in 2016, 2020, and 2024 to ensure precise tenure calculations.

-The “Unified Season” Approach: January transfers are included to reflect managerial adaptability/correction, with their experience metrics retroactively baselined to the season start to maintain temporal consistency.

Project Roadmap

• Phase 1: Scope Definition & Team Selection
• Phase 2: Data Collection(Web Scraping Wages & Match Stats via Selenium/Soccerdata API)
• Phase 3: Feature Engineering
  • Calculate Manager Tenure (Sept 2nd Rule)
  • Calculate Relative Experience Index (Top 5 Leagues)
• Phase 4: Exploratory Data Analysis (EDA) (Correlation Matrices & Visualizations)
• Phase 5: Regression Modeling (OLS / Panel Data Analysis)
• Phase 6: Final Report

Key Findings

• Structure > Strategy: “Club Identity” (Fixed Effects) is the primary driver of points ($p < .001$). For example, Manchester City starts with a +13 point structural advantage over the baseline, holding all else constant.
• Money Has Limits: Once club structure is controlled for, marginal increases in wage spending (Log_Wage) do not statistically predict higher point totals ($p = 0.38$).
• The Myth of Stability: Managerial tenure has no significant linear or quadratic relationship with performance, disproving the “stability dividend” hypothesis for this cohort.

Technical Implementation

• Languages: R, Python
• Libraries: tidyverse, ggplot2, car,e1071,ggcorplot
• Statistical Methods:
  • Fixed Effects Modeling: To control for unobserved club heterogeneity.
  • Nested F-Tests (ANOVA): Used to scientifically reject complex quadratic/interaction models in favor of Parsimony.
  • Multicollinearity correction: Applied Mean Centering to resolve structural multicollinearity (VIF reduced from >50 to <5).

Visualizations

• Functional Form Analysis (Wages vs. Points) Testing for diminishing returns using Loess, Linear, and Quadratic fits