Analyzing the Impact of Songs on Instagram Reel Performance

Based on @thorfinn_wisdom's content

---

Intro

This research explores how background music choice affects the performance of motivational reels on Instagram. The dataset is based on 196 videos from the creator @thorfinn_wisdom. While some characteristics (like font use) are not present across all videos, the main attributes—song choice, likes, play counts, and view counts—remain constant.

We apply multilinear regression and ANOVA to determine whether song choice significantly influences:

• likesCount (number of likes a video receives),
• videoPlayCount (total plays including replays), and
• videoViewCount (unique views).

---

Part 1: Likes Count – Song Regression Analysis

Regression Table Interpretation

The regression for likesCount shows an R² of 0.078 and an adjusted R² of 0.033, suggesting that song choice explains only a small fraction of the variance in likes. Most coefficients are not statistically significant, with the exception of a few cases.

Top 2 Songs

• T.thorfin1.mp3: Coefficient = 3.263, p = 0.006 → Significant positive impact on likes. On average, reels using this song received about 3.3 more likes compared to the baseline, a substantial result.
• T.thorfin4.mp3: Coefficient = 2.588, p = 0.034 → Statistically significant at the 5% level. Adds about 2.6 likes per video.

Bottom 2 Songs

• T.bossavibez.wav: Coefficient ≈ 0.037, p = 0.973 → Essentially no effect on likes; not statistically significant.
• T.tavernahoy.wav: Coefficient = 0.286, p = 0.825 → Insignificant and close to zero effect.

ANOVA Results

C(song_name)   sum_sq=3.19e+04, df=9, F=0.50, PR(>F)=0.873
Residual       sum_sq=1.32e+06, df=186

ANOVA indicates no overall significant effect of song choice on likes (p = 0.873). Although individual songs like T.thorfin1.mp3 and T.thorfin4.mp3 showed significance, the broader model suggests most songs do not meaningfully alter likes.

---

Part 2: Video Play Count – Song Regression Analysis

Regression Table Interpretation

The regression on videoPlayCount yields R² = 0.049, adjusted R² ≈ 0.003. This suggests very little explanatory power. Most songs are statistically insignificant.

Top 2 Songs

• T.thorfin2.mp3: Coefficient = 178.356, p = 0.022 → Significant positive effect, with videos using this song averaging ~178 additional plays.
• T.thorfin1.mp3: Coefficient = 72.569, though not statistically significant (p = 0.325), shows a relatively large positive estimate compared to most other songs.

Bottom 2 Songs

• T.tavernahoy.wav: Coefficient = -49.385, p = 0.540 → Negative but not significant, suggests it may slightly reduce plays.
• T.thorfin4.mp3: Coefficient ≈ 1.573, p = 0.983 → Almost no measurable effect on plays.

ANOVA Results

C(song_name)   sum_sq=3.19e+04, df=9, F=0.50, PR(>F)=0.873
Residual       sum_sq=1.32e+06, df=186

Again, ANOVA shows no significant overall effect of songs on total plays (p = 0.873). The strong coefficient for T.thorfin2.mp3 remains notable, but the model as a whole lacks predictive strength.

---

Part 3: Video View Count – Song Regression Analysis

Regression Table Interpretation

The regression for videoViewCount has R² = 0.024, adjusted R² = -0.024, indicating the model performs worse than a simple mean prediction. None of the coefficients reached statistical significance.

Top 2 Songs

• T.thorfin2.mp3: Coefficient = 38.462, p = 0.162 → Positive but not significant; suggests some potential effect.
• T.springtheme.wav: Coefficient = 26.372, p = 0.301 → Positive estimate but insignificant.

Bottom 2 Songs

• T.tavernahoy.wav: Coefficient = -7.341, p = 0.797 → Negative effect, not significant.
• T.bossavibez.wav: Coefficient = 13.687, p = 0.563 → Low effect, not significant.

ANOVA Results

C(song_name)   sum_sq=3.19e+04, df=9, F=0.50, PR(>F)=0.873
Residual       sum_sq=1.32e+06, df=186

ANOVA confirms the regression: no significant variation in views by song choice. The F-statistic is low, and the p-value (0.873) indicates non-significance.

---

Conclusion

Across the three regression analyses, the following patterns emerge:

• Likes Count: T.thorfin1.mp3 and T.thorfin4.mp3 stood out with significant positive effects, while songs like T.bossavibez.wav and T.tavernahoy.wav showed negligible influence.
• Video Play Count: T.thorfin2.mp3 had a strong and statistically significant positive effect, while T.tavernahoy.wav and T.thorfin4.mp3 were weak performers.
• Video View Count: None of the songs had a significant effect, though T.thorfin2.mp3 showed the highest positive coefficient.

The ANOVA results consistently show no overall significant song effect across all three performance metrics. This means that while individual songs (notably T.thorfin1, T.thorfin2, and T.thorfin4) can have notable effects, the general choice of song does not systematically change outcomes.

Recommendation

For influencers, the data suggests focusing on songs like T.thorfin1.mp3, T.thorfin2.mp3, and T.thorfin4.mp3(albeit this effect only significant for likes), which showed the most promise in boosting performance. Conversely, songs such as T.bossavibez.wav and T.tavernahoy.wav should likely be avoided as they provide no meaningful engagement uplift. However, since the ANOVA shows weak overall effects, song choice should be seen as a minor optimization, while other factors (visuals, captions, timing) may play a much larger role in reel success.