The impact of statistical applications on sports journalism: comparative study between algorithmic and human evaluations in elite football

Abstract

This study compares journalistic and algorithmic evaluations of elite football players’ performance in a context of growing influence of platforms such as Sofascore and Flashscore. Using a cross-sectional observational design, a purposive sample of 60 players across 10 high-profile matches (2024–2025 season) was analyzed. For each player, ratings were collected from both algorithms and two European media outlets, selected from a total of 13 (7 traditional and 6 digital native). Descriptive and inferential analyses were applied to examine central tendencies and dispersion, convergence between systems, the impact of match outcome, variation by player position and role, divergences by media type and country, patterns by specific outlet, and paradigmatic qualitative cases. Results show that media assign significantly lower ratings (mean 6.76; p < 0.001) and with greater variability (CV = 26.4 %) compared to algorithms (Sofascore: 7.53, CV = 11.8 %; Flashscore: 7.33, CV = 12.4 %). The correlation between algorithms is very high (r = 0.858) and higher than between media and algorithms (r ≈ 0.72). All systems display statistically significant sensitivity to match outcome (p ≤ 0.01), but the magnitude of bias is markedly greater in media (-2.01 points between win and loss) than in algorithms (-0.75/-0.96). Convergence is significantly stronger for standout players (d > 0.90). No significant differences were found by position or media type. Exploratory analyses suggest that each outlet’s editorial culture is more decisive than its country or media type. It is concluded that both systems are epistemologically complementary, and their integration—not substitution—enriches sports performance evaluation.