Table 1.
Results of linear effects models examining factors which influenced Travis Kelce’s performance in the Pre-Swift era (2014-2022).
Chiefs’ pre-game Elo was the only predictor that was entered into the model. Opponent Elo score, Chief’s win probability, and home/away game status were not selected for inclusion in the final model.
Fig 1.
Matching Pre-Swift and Swift era games.
Each vertical column of data points along the x-axis represents one “matched set”, consisting of one game from the Swift era (2023 season, colored squares), and five games from the pre-Swift era (2014-2022, grey circles). Red squares indicate games that Swift attended in the 2023 season, and blue square represent games she did not attend in the 2023 season. The y-axis represents the Chiefs’s pre-game Elo value, which quantifies team’s strength (higher values indicate better performance) and is also related to Kelce’s performance (see S3 File – Supplemental Results, for details regarding Elo). The similar pregame Elo values within each matched set demonstrates the matching procedure was effective. This allows Kelce’s Swift era performances to be compared to similar games from the pre-Swift era. Note: Kelce did not play in Week 1 or 17 of the season, thus these numbers are missing from the x-axis.
Table 2.
Results of linear effects models examining the Swift effect on Travis Kelce’s performance.
Taylor Swift’s presence or absence did not have a statistically significant effect on Travis Kelce’s performance during the Swift era, relative to Elo-matched games from the pre-Swift era.
Fig 2.
Comparison of Kelce’s performances between Pre-Swift and Swift eras.
Each vertical column of data points represents one “matched set”, consisting of one game from the Swift era (2023 season, colored squares), and five Elo-matched games from the pre-Swift era (2014-2022, grey circles). Red squares indicate games that Swift attended in the 2023 season, and blue square represent games she did not attend in the 2023 season. The y-axis represents Kelce’s yards per game (higher values indicate better performance). There is substantial variability in Kelce’s game performances in both eras. When Swift is present, there are examples of him playing unusually well (i.e., Week 7) and unusually poor (e.g., Week 16). Note: Kelce did not play in Week 1 or 17 of the season, thus these numbers are missing from the x-axis.
Table 3.
Results of binary logistic regression model examining the Swift effect on Chiefs’ game outcomes.
Parameters are exponentiated (i.e., exp(β)) to achieve odds ratios. Taylor Swift’s presence in the 2023 season did not have a statistically significant on the Chiefs likelihood of a victory. The best model fit included Chief’s pregame win probability as a continuous variable in the model, but this was also not statistically significant.
Fig 3.
Sensitivity Analysis of the “Swift Effect” Across Matching Algorithms.
Panel A (left) shows results for games in which Taylor Swift was present, and Panel B (right) shows results for games in which she was absent. On the x-axis, the matching ratio (1:1, 2:1, 3:1, 5:1, and 8:1) is plotted on a continuous scale. For each matching ratio, results from two matching methods are displayed: replacement (orange markers) and non‑replacement (purple markers). The y-axis represents the estimated effect in Kelce's yards (with corresponding 95% confidence intervals shown as error bars) from linear mixed effects models comparing Swift-era games to matched pre‑Swift controls. In Panel A, the effects indicate that Kelce’s performance does not differ significantly when Swift is present, while Panel B reveals that any performance decline when she is absent is sensitive to the matching algorithm employed. These findings illustrate that, despite some minor variability in point estimates and confidence intervals across different matching specifications, the overall conclusion regarding the lack of a robust “Swift effect” remains stable.