Analyzing the Impact of Songs, Backgrounds, and Fonts on Instagram Reel Likes

Based on @thorfinn_wisdom's content (Batch V1)

---

Intro

In this short analysis, we explore how video production choices, specifically the song used, background visuals, and font style influence the likesCount performance of Instagram Reels from the creator @thorfinn_wisdom. Check out the content of the account to get a better gauge of the content being analyzed.

The dataset consists of 93 videos, with each video using one of several possible background visuals and one of several possible songs. The variable of interest is likesCount, representing the number of likes a Reel receives.

To quantify the effects of these creative choices, we ran a multilinear regression, where each song and background is represented by a dummy variable (1 if used in the video, 0 otherwise). Font choice was also included with Roboto-Regular as one dummy; where Roboto-Medium serves as the reference baseline.

---

Method

A multilinear regression model was used:

likesCount = β₀ + Σ βᵢ * Songᵢ + Σ βⱼ * Backgroundⱼ + βₖ * Font + ε

• Dependent variable: likesCount
• Independent variables:
  • Dummy variables for songs
  • Dummy variables for backgrounds
  • Dummy variable for font
• Reference categories:
  • The song and background not shown in the regression table (implicitly captured by the intercept).
  • For font, Roboto-Medium is the reference.

Note: New backgrounds were introduced at a later point, hence half the sample (around 45 reels) had the chance of using more video backgrounds. The most commonly used background was vin6 (13.9% of videos), and the least used vin14 and vin15 (3.2%). The most commonly used background songs were bossavibez and springtheme (12.9%) and the least used was thorfinn1. Videos more than two standard deviations were also excluded to prevetn outliers from skewing the results.

---

Results

Overall Model Fit

• R-squared: 0.294 → About 29% of the variation in likesCount is explained by the model.
• Adjusted R-squared: 0.045 → Adjusted for model complexity, relatively low explanatory power.
• F-statistic p-value: 0.291 → The overall model is not statistically significant at conventional levels.

While individual variables may show interesting patterns, the model as a whole is exploratory and not predictive.

Song Effects

Top 3 Most Positive Song Coefficients:

1️⃣ song_name_thorfinn4.mp3: +4.018 likes (p = 0.047, statistically significant)

2️⃣ song_name_tavernahoy.wav: +3.611 likes (p = 0.094)

3️⃣ song_name_springtheme.wav: +3.394 likes (p = 0.084)

Top 3 Most Negative Song Coefficients:

1️⃣ song_name_bossavibez.wav: +0.273 likes (very small, p = 0.884)

2️⃣ song_name_bossatamb.wav: +1.877 likes (p = 0.217)

3️⃣ song_name_bossalowed.wav: +1.245 likes (p = 0.552)

Insights:

The standout result is song_name_thorfinn4.mp3, which shows a positive and significant effect on likes. This suggests it may be an engaging track for this audience.

Other songs show non-significant coefficients, so their effects should be interpreted cautiously.

Background Effects

Top 3 Most Positive Background Coefficients:

1️⃣ background_vin15.mp4: +3.151 likes (p = 0.237)

2️⃣ background_vin9.mp4: +2.468 likes (p = 0.406)

3️⃣ background_vin11.mp4: -5.460 likes (least negative among negatives) (p = 0.048)

Top 3 Most Negative Background Coefficients:

1️⃣ background_vin13.mp4: -7.448 likes (p = 0.010, statistically significant)

2️⃣ background_vin12.mp4: -6.336 likes (p = 0.031, statistically significant)

3️⃣ background_vin6.mp4: -5.947 likes (p = 0.025, statistically significant)

Insights:

• Several backgrounds have statistically significant negative effects on likes, particularly background_vin13.mp4, background_vin12.mp4, and background_vin6.mp4.
• This suggests some background choices may detract from audience engagement. Notably from the same regressio but ran on videoviewCount instead vin12 actually had the strongest positive effect on views. Puzzling results...

Font Effect

Roboto-Regular.ttf: +0.071 likes (p = 0.942)

→ Essentially neutral effect, not statistically significant.

Summary of Significance

• song_name_thorfinn4.mp3: statistically significant positive effect.
• Backgrounds vin13, vin12, and vin6: statistically significant negative effects.
• Other variables are exploratory and not statistically distinguishable from zero.

---

Conclusion

Key Takeaways:

✅ song_name_thorfinn4.mp3: Using this song correlates with significantly more likes. More emphasis will be placed on this in future batches.

🚫 background_vin13.mp4, vin12.mp4, vin6.mp4: Associated with lower likes and statistically significant, and hence less emphasis should be placed on these.

Cautions:

• The model’s overall explanatory power is weak and not significant.
• Most effects are not statistically significant and should be interpreted as exploratory.
• Larger sample sizes and interaction terms (song × background) could improve understanding.
• A causal relationship is arguable given statistical significance of some variables, though omitted variable bias may still play a role.