YouTube Statistics Analysis

This repository presents a structured data analysis of global YouTube channel statistics using R. The project explores trends across content categories and countries using both descriptive and inferential statistics. Key performance indicators such as subscriber count, total views, and upload volume are analyzed. The final output includes visualizations, simulation-based inference, and clear interpretations.

Report: View HTML Report
Dataset: Kaggle – Global YouTube Statistics 2023

---

Objectives

• Explore patterns in YouTube performance metrics across countries and content types
• Identify key factors influencing subscriber count and video views
• Compare performance between categories using statistical testing and bootstrap simulation

---

Dataset Overview

The dataset contains information on top-ranking YouTube channels globally, sourced from Kaggle. It includes various performance metrics and metadata describing the channels.

Data Dictionary (Used Variables)

Variable	Description
Youtuber	Name of the YouTube channel
Rank	Global ranking based on subscriber count
Grade	Letter grade assigned by third-party evaluators
subscribers	Total number of subscribers
video.views	Total accumulated video views
video.uploads	Number of videos uploaded
category	Content category (e.g., Entertainment, Music)
title	Channel description or tagline
Country	Country of origin
Abbreviation	Country abbreviation

---

Analytical Workflow

1. Data Preparation

• Imported the dataset and cleaned column names
• Converted data types and removed duplicates or incomplete records

2. Descriptive Analysis

• Computed summary statistics for subscribers, views, and uploads
• Aggregated metrics by category and country
• Visualized trends using bar charts

3. Data Visualization

• Developed an interactive bubble chart (subscribers vs video views)
• Used highcharter for bubble chart and 3D column chart

4. Inferential Analysis - Normality and Non-Parametric Tests

• Performed Shapiro-Wilk normality tests on "uploads" and "subscribers" to assess their distribution.
  • Key finding: Both variables were found to be non-normally distributed (p-value < 0.05).
• Conducted a Wilcoxon signed-rank test to compare the median values of "uploads" and "subscribers."
  • Key finding: There is a significant difference in the median values between "uploads" and "subscribers" (p-value < 0.05).
• Applied Spearman's rank correlation to assess the monotonic relationship between "uploads" and "subscribers."
  • Key finding: There is no statistically significant correlation between "uploads" and "subscribers" (p-value > 0.05), despite a very weak positive correlation (rho = 0.052).

5. Inferential Analysis - Bootstrap Simulation for Regression

• Performed a bootstrap simulation (100 resamples) to estimate the relationship between "uploads" and "subscribers" using a linear regression model.
  • Calculated the mean slope, mean intercept, and mean R-squared from the simulated models.
  • Critical finding: The standard deviations for slope, intercept, and R-squared were all zero, indicating a potential issue with the simulation or data (e.g., lack of variability in resamples or an error in the bootstrapping process). This makes the current bootstrap results unreliable.
  • The mean R-squared was very low (0.0063 or 0.63%), suggesting that "uploads" is a poor predictor of "subscribers" in this linear model, even if the simulation issue is resolved.

---

Key Insights

• Entertainment and Music are the leading content categories by subscriber and view counts.
• Upload volume alone does not significantly predict higher subscriber numbers (supported by low Spearman's correlation and very low R-squared from bootstrap).
• Countries like the United States and India dominate in the number of top channels.
• There is statistically significant evidence of subscriber count differences across content categories, and also a significant difference in median values between uploads and subscribers (Wilcoxon test).
• The relationship between uploads and subscribers is complex: while their central tendencies differ, there's no strong or significant monotonic correlation, and linear models show very little predictive power. Further investigation is needed into the bootstrap simulation's zero standard deviations to confirm its findings.

---

Tools Used

• Programming Language: R
• Libraries: tidyverse, psych, highcharter, stringr
• Documentation: R Markdown
• Version Control: Git and GitHub

---

Project Structure

global-youtube-statistics/
├── data/
│   └── Global YouTube Statistics.csv
├── youtube_stats_analysis.Rmd       # R Markdown source file
├── youtube_stats_analysis.html      # Rendered HTML report
├── .gitignore
└── README.md                        # Project documentation

---

How to Run

1. Clone the repository:

   git clone https://github.com/ameerahrazali/global-youtube-statistics.git

2. Open the .Rmd file in RStudio.
3. Run or knit the document to generate the HTML report.

---

Future Improvements

While the current analysis provides valuable initial insights, especially concerning the non-normal distribution of key variables and the weak relationship between uploads and subscribers, the dataset contains richer information that can be utilized in future enhancements. These improvements aim to deepen our understanding and address some of the limitations encountered:

• Address Bootstrap Simulation Issue:

  • Prioritize resolving the zero standard deviation issue in the bootstrap simulation. This is crucial for obtaining reliable estimates and confidence intervals for regression coefficients.
  • Once resolved, use the bootstrap to properly estimate the sampling distribution of regression coefficients and R-squared, providing more robust insights into the relationship between uploads and subscribers.

• Expand Predictive Modeling:

  • Incorporate channel_type, video_views_rank, and country_rank to analyze performance by content specialization and channel standing.
  • Analyze recent performance trends using video_views_for_the_last_30_days and subscribers_for_last_30_days to understand short-term growth dynamics.
  • Add a financial perspective using:
    • lowest_monthly_earnings
    • highest_monthly_earnings
    • lowest_yearly_earnings
    • highest_yearly_earnings
  • Explore how channel age (using created_year, created_month, and created_date) affects growth trajectories.

• Integrate External Factors:

  • Integrate country-level socioeconomic indicators, such as:
    • Gross tertiary education enrollment (%)
    • Population
    • Unemployment rate
    • Urban_population
  • Visualize channel locations using Latitude and Longitude for geospatial analysis to identify regional trends or popular hubs.

These enhancements would enable more comprehensive insights into YouTube success factors, allow for multivariate modeling to identify stronger predictors, and provide a more robust statistical foundation by fully leveraging the bootstrap methodology.