A STATISCAL ANALYSIS OF THE CORRELATION BETWEEN

WIN PROBABLITY AND OUTCOMES OF THE PREMIRE

LEAGUE 2023-24 SEASON

The 2023 Premier League season offered an intriguing opportunity to evaluate the

relationship between pre-season probabilities of winning and actual points earned by

teams. Probability forecasts often serve as predictive tools to gauge team performance,

yet their accuracy and association with real outcomes merit statistical examination. This

study employs various statistical tools to analyze this relationship, focusing on

correlation analysis and the chi-square test while addressing key limitations, such as the

presence of outliers and data distribution.

Why Raw Data is Not Appropriate for Direct Use

The direct use of raw probabilities and points is inappropriate for this analysis due to

two primary reasons:

The Influence of Outliers

Manchester City represents a significant outlier, with an overwhelming probability of

winning (65.4%) compared to other teams, whose probabilities mostly fall below 10%.

Similarly, Manchester City scored 91 points in the league, far exceeding most teams.

Such outliers can disproportionately skew statistical analyses, particularly linear

correlation measures like Pearson’s correlation, leading to biased interpretations

(Howell, 2012).

Non-Normal Data Distribution

The dataset exhibits a non-normal distribution, as probabilities are heavily clustered at

lower ranges and points are concentrated below 70. Statistical tests like Pearson’s

correlation assume normality, and deviations from this assumption reduce the reliability

of the results (Field, 2013). These factors necessitate alternative methods and

adjustments, such as rank-based correlation metrics and visual outlier detection (Siegel

& Castellan, 1988).

Statistical Methodologies

Spearman vs. Pearson Correlation

Correlation analysis was employed to determine the relationship between probabilities

and points, using both Spearman and Pearson correlation coefficients. Spearman’s

correlation is particularly robust to outliers and does not assume a linear relationship
(Siegel & Castellan, 1988). Pearson’s correlation measures linear relationships but is

highly sensitive to outliers, making it less reliable for this dataset (Howell, 2012).

Outlier Identification and Treatment

To address the influence of outliers, visual analysis using box-and-whisker plots was

conducted, revealing Manchester City as a significant outlier in both probabilities and

points. Removing this outlier allowed for recalculating correlations, offering a clearer

view of underlying trends (Barnett & Lewis, 1994).

Chi-Square Test of Independence

The chi-square test was applied to examine the relationship between team rankings

(based on points earned) and match outcomes (win-draw-loss records). This test

evaluates whether observed distributions differ significantly from expected distributions

under the assumption of independence (Agresti, 2013).

Results

Spearman and Pearson Correlation Results

The correlation analysis yielded the following results:

Correlation Method Including Manchester City Excluding Manchester City

Spearman 0.86 (strong) 0.83 (strong)

Pearson 0.50 (moderate) 0.67 (strong)

The Spearman correlation remained strong regardless of whether Manchester City was

included, confirming a robust positive relationship between probabilities and points.

However, Pearson's correlation improved significantly upon excluding Manchester City,

highlighting the outlier's influence. This finding underscores the importance of using

rank-based metrics like Spearman’s for such datasets (Siegel & Castellan, 1988).

Correlation analysis
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.8 1 1.2 1.4 1.6 1.8 2 2.2

Spearman Pearson

Chi-Square Test Results

The chi-square test examined the association between team rankings and match

outcomes. The calculated chi-square statistic was 29.40, with a p-value of 0.021 and 16

degrees of freedom. The low p-value indicates a statistically significant association,

suggesting that higher-ranked teams (e.g., Manchester City, Arsenal, Liverpool) were

more likely to achieve favorable outcomes (wins) than lower-ranked teams (Agresti,

2013).

Discussion

The analysis reveals a strong positive correlation between pre-season probabilities and
points earned, affirming the predictive validity of probabilities in evaluating team

performance (Lane, 2014). The chi-square results further emphasize the influence of

team rankings on match outcomes. Higher-ranked teams consistently outperform lower-

ranked teams, reflecting the inherent disparities in team quality and resources (Dixon &

Coles, 1997). However, the presence of outliers like Manchester City demonstrates the

need for robust statistical methods that account for such anomalies.

Limitations

Outlier Impact

Although outlier management improved the results, the exclusion of Manchester City

may oversimplify the analysis by ignoring the team's legitimate dominance. Future

studies could explore methods to integrate outliers more effectively without excluding

them entirely (Barnett & Lewis, 1994).

Data Scope

The analysis is limited to the 2023 season and does not account for historical trends or

other influencing factors, such as injuries, transfers, or managerial changes. A broader

dataset could provide a more comprehensive view.

Statistical Assumptions

While Spearman’s correlation mitigates some limitations, the chi-square test assumes

independence among match outcomes, which may not fully capture the complexities of

team interactions and strategies (Agresti, 2013).

Conclusion

This study demonstrates a strong positive relationship between pre-season probabilities

and points earned in the 2023 Premier League season. Spearman’s rank correlation

emerged as the most reliable measure, effectively addressing issues of non-normality

and outliers. The chi-square test further highlighted the significant association between

team rankings and match outcomes, underscoring the predictive power of probabilities.

The findings reinforce the utility of statistical tools in sports analysis, while also

highlighting the challenges posed by outliers and data distribution. Future research

could expand on these insights by incorporating additional variables and exploring

alternative statistical models to enhance the accuracy and applicability of predictions

(Lane, 2014).

References:

[Link]

opta-predictor

[Link]

[Link]

2023-2024-odds-title-winner-epl/e6p6xxtp4xymrigpmclwzxqc

 Agresti, A. (2013). Categorical Data Analysis (3rd ed.). Wiley-Interscience.

 Barnett, V., & Lewis, T. (1994). Outliers in Statistical Data (3rd ed.). Wiley.

 Dixon, M. J., & Coles, S. (1997). Modelling Association Football Scores: A

Probabilistic Analysis of the 1994–1995 Season of the English Football League. Journal

of the Royal Statistical Society: Series D (The Statistician), 46(2), 265-280.

 Field, A. (2013). Discovering Statistics Using IBM SPSS Statistics (4th ed.). Sage.

 Howell, D. C. (2012). Statistical Methods for Psychology (8th ed.). Cengage

Learning.

 Lane, D. (2014). Advanced Sports Analytics: Predictive Models for Football and

Other Sports. Springer.

 Siegel, S., & Castellan, N. J. (1988). Nonparametric Statistics for the Behavioral

Sciences (2nd ed.). McGraw-Hill.