Video Game Sales Analysis

Project Overview

This project analyzes a historical dataset of video game sales (1980–2016) covering 16,715 titles across 31 platforms and 12 genres, with regional sales figures for North America, Europe, Japan, and the rest of the world, plus critic scores, user scores, and ESRB content ratings where available.

The analysis was performed to identify which platforms, genres, and regions are currently driving sales, understand how the console market has shifted generations over time, and produce evidence-based advertising recommendations for the 2017 planning period. The dataset was cleaned, explored, and modeled entirely in Python (pandas, NumPy, SciPy), with all visualizations built in Plotly.

Research Questions

The notebook addresses the following questions:

• How have video game release volumes changed from 1980 to 2016?
• How long does a gaming platform typically stay commercially active, and which platforms had already become discontinued by 2016?
• Which time period is most representative of the current console market for forecasting purposes, rather than the full 36-year history?
• Which platforms generated the most sales in that relevant period, and which are growing vs. declining?
• Does a game's critic score or user score correlate with its sales (using PS4 as a case study)?
• Which multi-platform games sell best, and how does their performance split across platforms?
• Which genres dominate the market, by both total sales and market share?
• How do platform, genre, and ESRB rating preferences differ across North America, Europe, and Japan?
• Is there a statistically significant difference in user scores between the Xbox One and PC platforms?
• Is there a statistically significant difference in user scores between the Action and Sports genres?

Dataset

• 16,715 rows × 11 raw columns; 16,444 rows after cleaning
• Core fields: name, platform, year_of_release, genre
• Regional sales: na_sales, eu_sales, jp_sales, other_sales (millions of copies), plus an engineered total_sales column
• Review fields: critic_score (0–100), user_score (0–10), rating (ESRB) — each 40–54% missing, concentrated in older titles
• Time period: 1980–2016

Full column-by-column documentation, missing-value analysis, and cleaning decisions are in dataset/README.md.

Data Cleaning and Preparation

• Standardized all column names to lowercase.
• Dropped 2 rows missing name/genre (core identifying fields).
• Converted year_of_release to a nullable integer type.
• Converted "tbd" values in user_score to NaN (not 0, to avoid implying the worst possible score for an unrated game), then cast the column to float64.
• Dropped 269 rows missing year_of_release (16,713 → 16,444 rows).
• Deliberately kept missing values in critic_score, user_score, and rating rather than imputing them — with 40–54% missingness in each column and 6,580 rows missing all three at once, the notebook judged imputation to be unreliable at that scale.
• Engineered total_sales = na_sales + eu_sales + jp_sales + other_sales.
• Confirmed no fully duplicate rows exist in the dataset.

Exploratory Data Analysis

• Temporal analysis: annual release counts, 1980–2016.
• Platform lifecycle: first/last active year and lifespan per platform, discontinued-platform classification, sales heatmap across platform × year.
• Relevant-period selection: restricting most downstream analysis to 2013–2016 based on the PS4/XOne generation launch.
• Platform performance: total sales, year-over-year growth, and a linear 2017 sales forecast per platform.
• Sales distribution: per-platform sales spread (median, mean, std), both all-time and for 2013–2016.
• Review-score impact: critic score and user score vs. sales, using PS4 as a case study.
• Cross-platform comparison: how the same multi-platform game performs across the platforms it was released on.
• Genre analysis: total sales and market share by genre.
• Regional analysis: platform, genre, and ESRB rating preferences broken out by North America, Europe, and Japan.

Statistical Analysis

Two independent-samples t-tests (Welch's t-test, unequal variances) were performed, both at α = 0.05.

Test 1 — Xbox One vs. PC user scores

• H0: No statistically significant difference between mean user_score on Xbox One and PC (μ_XOne = μ_PC).
• H1: A statistically significant difference exists (μ_XOne ≠ μ_PC).
• Test: Welch's t-test (scipy.stats.ttest_ind, equal_var=False)
• Sample sizes: Xbox One n=182 (mean=6.521, std=1.381); PC n=155 (mean=6.270, std=1.742)
• Result: t = 1.452, p = 0.1476
• Decision: p ≥ 0.05 → H0 is not rejected.
• Interpretation: No statistically significant difference was detected between Xbox One and PC user scores in this dataset.

Test 2 — Action vs. Sports user scores

• H0: No statistically significant difference between mean user_score for Action and Sports genres.
• H1: A statistically significant difference exists.
• Test: Welch's t-test (scipy.stats.ttest_ind, equal_var=False)
• Sample sizes: Action n=389 (mean=6.838, std=1.330); Sports n=160 (mean=5.238, std=1.783)
• Result: t = 10.233, p ≈ 1.45×10⁻²⁰
• Decision: p < 0.05 → H0 is strongly rejected.
• Interpretation: Action games are rated significantly higher by users than Sports games on average. The notebook offers one possible (untested) explanation — Sports titles often follow annual roster-update cycles with less year-to-year innovation than Action titles — while explicitly noting this is an interpretation, not proven causation.

Key Findings

1. Platform lifecycle: Platforms stay active for a median of 6 years (mean 7.6 years). Using a 3-year inactivity threshold, 20 platforms (PS2, XB, GBA, GC, PS, N64, SNES, and others) are classified as discontinued by 2016. The 2013 PS4/XOne launch marked a clear generation shift, which is why 2013–2016 was selected as the relevant period for the 2017 forecast.
2. Current-generation leaders (2013–2016): PS4 leads with 314.14M in sales, followed by PS3 (181.43M), XOne (159.32M), 3DS (143.25M), and X360 (136.80M). PS4 and XOne show a growing trend; PS3/X360/Wii/DS are declining as older-generation platforms.
3. Reviews (PS4 case study): critic_score has a moderate positive correlation with sales (r ≈ 0.41); user_score shows almost no correlation (r ≈ −0.03). Correlation is not causation — other factors like marketing and budget likely play a role.
4. Genre (all-time, 1980–2016): Action leads with ≈19.5% market share (1,716.52M), followed by Sports (14.9%), Shooter (11.8%), and Role-Playing (10.6%). Adventure, Strategy, and Puzzle are the smallest niches.
5. Regional differences (2013–2016):
   • Platforms: PS4 leads in both NA (24.84% share) and EU (35.97% share). Japan is dominated by 3DS (48.17% share) — a clear outlier from the Western pattern.
   • Genres: NA and EU both favor Action > Shooter > Sports. Japan favors Role-Playing (36.26% share) over Action (28.76%), with Shooter far behind (4.70%).
   • ESRB rating: M-rated games lead sales in NA (165.21M) and EU (145.32M). Japan is led by T-rated games (20.59M) instead.
6. Hypothesis tests: No significant difference in user scores between XOne and PC (p = 0.148). Highly significant difference between Action and Sports user scores (p ≈ 1.45×10⁻²⁰), with Action rated higher on average (6.84 vs. 5.24).

2017 Advertising Recommendations (from the notebook)

• Focus advertising spend on PS4 titles, and on the Action/Shooter genres in NA and EU — these show the strongest current sales and align with each region's top platform and genre preferences.
• Use a distinct strategy for Japan: prioritize Role-Playing titles on 3DS, and lean toward broader-audience (T-rated) content rather than mature-only titles, since Japan's regional profile diverges sharply from NA/EU on every dimension measured.
• Rely more on critic scores than user scores when using reviews as a marketing signal, since critic scores show a stronger (though still moderate) relationship with sales.

Visualizations

All 19 charts generated in the notebook are exported as individual PNG files in charts/, each with a full explanation of its purpose, what it shows, and its analytical interpretation in charts/README.md.

Project Structure

video-game-sales-analysis/
│
├── README.md                     # This file
├── requirements.txt               # Python dependencies
├── .gitignore
│
├── notebooks/
│   └── games_project_Ədalət_Sadıqov.ipynb   # Main analysis notebook
│
├── charts/
│   ├── README.md                  # Explanation of every chart
│   └── 01_*.png ... 19_*.png      # Exported Plotly visualizations
│
├── dataset/
│   ├── README.md                  # Full dataset documentation
│   └── games.csv                  # Source dataset (16,715 rows x 11 columns)
│
└── src/
    ├── README.md
    └── extract_charts.py          # Re-exports notebook charts to PNG via Plotly/Kaleido

Technologies Used

• Python
• pandas
• NumPy
• SciPy (scipy.stats — t-tests, linear regression)
• Plotly (plotly.express, plotly.graph_objects) — all visualizations
• Kaleido — static PNG export of Plotly figures
• ipywidgets
• Jupyter Notebook

No Matplotlib or Seaborn is used anywhere in this project.

How to Run

1. Clone the repository:

   git clone <repository-url>
   cd video-game-sales-analysis

2. Install dependencies:

   pip install -r requirements.txt

3. The dataset is already included at dataset/games.csv (see dataset/README.md for full schema documentation).
4. Open the notebook:

   jupyter notebook notebooks/games_project_Ədalət_Sadıqov.ipynb

5. Run all cells from top to bottom.

To regenerate the chart PNGs from the notebook's saved outputs without re-running the full analysis, see src/README.md.

Conclusion

Between 2013 and 2016, the video game market underwent a clear generational transition — PS4 and Xbox One displacing the previous console generation — which makes this window the most reliable basis for near-term (2017) forecasting. Within that window, sales are concentrated in a small number of platforms (led by PS4) and genres (led by Action and Shooter), critic reviews carry a moderate positive relationship with sales while user reviews do not, and regional preferences diverge sharply between Western markets and Japan across platform, genre, and content-rating dimensions alike. These patterns directly support the region-specific 2017 advertising recommendations summarized above, while the notebook consistently distinguishes statistically supported findings from interpretive explanations that were not directly tested.