Introduction

The dataset is from Video Game Sales which was generated by a scrape of vgchartz.com. There are over 16500 records.

Because the scraping algorithm focuses on game units sale, the data is not the representation of modern gaming industry. The data also favors consoles heavily since most big PC games follows free to play platform!

Nevetheless, this analysis should give us a glimpse at the change in video game industry for the last 3 decades.

Variable Definitions:

  1. Name: Name of the game
  2. Platform: Console on which the game is running
  3. Year_of_Release: No description provided
  4. Genre: No description provided
  5. Publisher: No description provided
  6. NA_Sales: Game sales in North America (in millions of units)
  7. EU_Sales: Game sales in the European Union (in millions of units)
  8. JP_Sales: Game sales in Japan (in millions of units)
  9. Other_Sales: Game sales in the rest of the world, i.e. Africa, Asia excluding Japan, Australia, Europe excluding the E.U. and South America (in millions of units)
  10. Global_Sales: Total sales in the world (in millions of units)
  11. Critic_Score: Aggregate score compiled by Metacritic staff
  12. Critic_Count: The number of critics used in coming up with the Critic_score
  13. User_Score: Score by Metacritic’s subscribers
  14. User_Count: Number of users who gave the user_score
  15. Developer: Party responsible for creating the game
  16. Rating: The ESRB ratings (E.g. Everyone, Teen, Adults Only..etc)

Libraries and datasets

library(tidyverse)
library(ggplot2)
library(ggthemes)
library(viridis)
library(lubridate)
library(wordcloud)
library(qdap)
library(tm)
library(ngram)
library(RColorBrewer)
library(gridExtra)
library(knitr)
vgsales <- tbl_df(read.csv("vgsales.csv", stringsAsFactors = FALSE))
theme_set(theme_tufte())

Exploratory Data Analysis

We start by taking a glimpse of general characteristics about the dataset.

str(vgsales)
## Classes 'tbl_df', 'tbl' and 'data.frame':    16719 obs. of  16 variables:
##  $ Name           : chr  "Wii Sports" "Super Mario Bros." "Mario Kart Wii" "Wii Sports Resort" ...
##  $ Platform       : chr  "Wii" "NES" "Wii" "Wii" ...
##  $ Year_of_Release: chr  "2006" "1985" "2008" "2009" ...
##  $ Genre          : chr  "Sports" "Platform" "Racing" "Sports" ...
##  $ Publisher      : chr  "Nintendo" "Nintendo" "Nintendo" "Nintendo" ...
##  $ NA_Sales       : num  41.4 29.1 15.7 15.6 11.3 ...
##  $ EU_Sales       : num  28.96 3.58 12.76 10.93 8.89 ...
##  $ JP_Sales       : num  3.77 6.81 3.79 3.28 10.22 ...
##  $ Other_Sales    : num  8.45 0.77 3.29 2.95 1 0.58 2.88 2.84 2.24 0.47 ...
##  $ Global_Sales   : num  82.5 40.2 35.5 32.8 31.4 ...
##  $ Critic_Score   : int  76 NA 82 80 NA NA 89 58 87 NA ...
##  $ Critic_Count   : int  51 NA 73 73 NA NA 65 41 80 NA ...
##  $ User_Score     : chr  "8" "" "8.3" "8" ...
##  $ User_Count     : int  322 NA 709 192 NA NA 431 129 594 NA ...
##  $ Developer      : chr  "Nintendo" "" "Nintendo" "Nintendo" ...
##  $ Rating         : chr  "E" "" "E" "E" ...
colSums(is.na(vgsales))
##            Name        Platform Year_of_Release           Genre 
##               0               0               0               0 
##       Publisher        NA_Sales        EU_Sales        JP_Sales 
##               0               0               0               0 
##     Other_Sales    Global_Sales    Critic_Score    Critic_Count 
##               0               0            8582            8582 
##      User_Score      User_Count       Developer          Rating 
##               0            9129               0               0

There are quite a lot of missing values in score but they appear in Sales which means a lot of games are neglected by the rating community. It is worth our time to take a closer look at these titles in missing values analysis.

General Histograms

vgsales %>%
    group_by(Year_of_Release) %>%
    summarize(Sales = sum(Global_Sales, na.rm = TRUE)) %>%
    ggplot(aes(x = Year_of_Release, y = Sales)) +
    geom_col(fill = "navyblue") +
    theme(axis.text.x = element_text(angle = 90)) +
    labs(title = "Global Sales Histograms", x = "Year", y = "Sales (units)")

vgsales %>%
    group_by(Year_of_Release) %>%
    summarize(Number_of_Games = n()) %>%
    ggplot(aes(x = Year_of_Release, y = Number_of_Games)) +
    geom_col(fill = "magenta4") +
    theme(axis.text.x = element_text(angle = 90)) +
    labs(title = "Games released per Year", x = "Year", y = "Sales (units)")

Top Global Sales each year

vgsales %>%
    group_by(Year_of_Release, Publisher) %>%
    summarize(Sales = sum(Global_Sales)) %>%
    top_n(n = 1) %>%
    kable()
## Selecting by Sales
## Warning: package 'bindrcpp' was built under R version 3.4.1
Year_of_Release Publisher Sales
1980 Atari 8.36
1981 Activision 8.50
1982 Atari 19.43
1983 Nintendo 10.96
1984 Nintendo 45.56
1985 Nintendo 49.95
1986 Nintendo 16.18
1987 Nintendo 11.95
1988 Nintendo 36.44
1989 Nintendo 63.88
1990 Nintendo 35.49
1991 Nintendo 15.97
1992 Nintendo 38.11
1993 Nintendo 20.04
1994 Nintendo 24.99
1995 Sony Computer Entertainment 18.45
1996 Nintendo 73.70
1997 Sony Computer Entertainment 43.90
1998 Nintendo 48.41
1999 Nintendo 65.33
2000 Nintendo 34.05
2001 Nintendo 45.37
2002 Electronic Arts 73.01
2003 Electronic Arts 69.82
2004 Electronic Arts 67.28
2005 Nintendo 126.69
2006 Nintendo 204.42
2007 Nintendo 103.09
2008 Nintendo 90.01
2009 Nintendo 127.39
2010 Electronic Arts 80.26
2011 Electronic Arts 71.51
2012 Nintendo 56.11
2013 Nintendo 53.48
2014 Nintendo 48.88
2015 Electronic Arts 45.74
2016 Electronic Arts 26.48
2017 Sega 0.05
2020 Ubisoft 0.29
N/A Unknown 20.90

Missing Values Analysis

All missing values are in Critic Score, Critic Count and User Count. However, Developer also has an empty string for these missing values. Many of these titles are quite famous such as Super Mario Bros… thus perhaps they were released before the rating website created. A simple plot would suffice.

vgsales$Year_of_Release <- as.integer(vgsales$Year_of_Release)
vgsales %>%
    filter(is.na(Critic_Score)) %>%
    ggplot(aes(x = Year_of_Release, y = Genre)) +
    geom_jitter(alpha = 0.2, pch = 21, col = "darkslategrey") +
    labs(x = "Year of Release", title = "Missing Score by Genre and Year")

It seems our intuition is wrong; most missing scores are from 2005 and up. Nevertheless, there are a few interesting points show up in the plot:

  1. Action, Adventure, Sport, Simulation and Misc have the most missing values. This trend correlated heavily with Genre distribution so it is not a surprise.
  2. There are 2 games from 1993 that have no Genres listed. They also have no name, created by Acclaim Entertainment for GEN gaming system. Their sales combine to a total of less than 2.5 millions. These will be excluded from our dataset.
  3. Imagine: Makeup Artist is/was/will be released in 2020 which proves time travel! Joking aside, it has already presale for $ 290.000 total.
  4. Lastly, there are only 3 games from 2017 and we should also exclude them.
index <- which(vgsales$Genre == "" | vgsales$Year_of_Release %in% c(2020, 2017))
vgsales <- vgsales[-index,]
## Creating System Variable
pc <- c("PC")
console <- c("Wii", "NES", "X360", "PS3", "PS2", "SNES", "PS4", "N64", "PS", "XB", "2600",
             "XOne", "WiiU", "GC", "GEN", "DC", "SAT", "SCD", "NG", "TG16", "3DO", "GG", 
             "PCFX")
handheld <- c("GB", "DS", "GBA", "3DS", "PSP", "PSV", "WS")

vgsales <- vgsales %>%
    mutate(System = ifelse(Platform %in% pc, "pc",
                           ifelse(Platform %in% console, "console", "handheld")))
max_mis <- vgsales %>% filter(is.na(Critic_Score)) %>% 
    group_by(System) %>% filter(Global_Sales == max(Global_Sales))
vgsales %>%
    filter(is.na(Critic_Score)) %>%
    ggplot(aes(x = Platform, y = Global_Sales, col = System)) +
    geom_point(alpha = 0.5, pch = 21) +
    geom_text(data = max_mis, label = max_mis$Name, vjust = "inward", hjust = "inward") +
    coord_flip() +
    labs(title = "Missing Score sorted by Global Sales and Platform")

Even with a missing score, these games still generate a healthy amount of revenue with Super Mario Bros earned over 40 millions in sales. The data shows no trend that any system got an edge in getting scored.

To conclude the missing values portion of this report, lets take a look at the wordcloud.

clean <- function(x){
    x <- replace_contraction(x)
    x <- replace_ordinal(x)
    x <- removePunctuation(x)
    x <- tolower(x)
    x <- removeWords(x, c(stopwords("en"), "game"))
}
missing <- filter(vgsales, is.na(Critic_Score))
titles <- concatenate(missing$Name)
titles <- clean(titles)
unigram <- ngram(titles, n = 1)
freq <- tbl_df(get.phrasetable(unigram))

## Creating wordcloud
wordcloud(freq$ngrams, freq$freq, random.order = FALSE, rot.per = 0.35,
          color=brewer.pal(8, "Dark2"), max.words = 200)

There are a lot of sequels among these missing scores. Other words are quite typical such as pro, world, soccer, portable, battle, gundam, edition, championship…

Sales vs Devs, Publisher and Platform

First, lets pay tribute to the people who make and distribute our games.

vgsales %>%
    select(Publisher, Developer) %>%
    sapply(function(x) {length(unique(x))})
## Publisher Developer 
##       582      1697
top20 <- head(names(sort(table(vgsales$Publisher), decreasing = TRUE)),20)
p1 <- vgsales %>%
    filter(Publisher %in% top20) %>%
    mutate(Publisher = factor(Publisher, levels = rev(top20))) %>%
    group_by(Publisher) %>%
    summarize(Number_of_Devs = n_distinct(Developer)) %>%
    top_n(20, wt = Number_of_Devs) %>%
    ggplot(aes(x = Publisher, y = Number_of_Devs)) +
    geom_col(fill = "darkslategrey", alpha = 0.5) +
    coord_flip()
p2 <- vgsales %>%
    filter(Publisher %in% top20) %>%
    mutate(Publisher = factor(Publisher, levels = rev(top20))) %>%
    group_by(Publisher) %>%
    summarize(Number_of_Games = n()) %>%
    top_n(20, wt = Number_of_Games) %>%
    ggplot(aes(x = Publisher, y = Number_of_Games)) +
    geom_col(fill = "firebrick4", alpha = 0.5) +
    coord_flip() +
    theme(axis.title.y = element_blank(), axis.text.y = element_blank())
grid.arrange(p1,p2, ncol = 2, top = "Distribution of games and devs for top 20 Publishers")

There are a lot more Developers than Publishers. In fact, most big Publishers have over 100 Developers working for them. Notice the Unknown Publisher is in the top 20 also. Unknown here means the collection of unrecorded publisher and not the name “Unknown”.

The more interesting point is the massive amount of titles big Publishers have compare to the rest. Even with only the top 20 we can clearly see the big difference in distributive power. How about the sale?

toppub <- vgsales %>%
    filter(Publisher %in% c("Electronic Arts", "Activision", "Ubisoft", "Namco Bandai Games", "Nintendo")) %>%
    group_by(Publisher, Year_of_Release) %>%
    summarize(total = sum(Global_Sales))

vgsales %>%
    group_by(Publisher, Year_of_Release) %>%
    summarize(total = sum(Global_Sales)) %>%
    ggplot(aes(x = Year_of_Release, y = total)) +
    geom_point(alpha = 0.5, pch = 21) +
    geom_point(data = toppub, aes(col = Publisher), size = 1.5) +
    geom_line(data = toppub, aes(col = Publisher), size = 1.3) +
    scale_color_viridis(discrete = TRUE) +
    theme(legend.position = c(0.2, 0.7)) +
    labs(title = "Time series of Global Sales by Publishers", y = "Sales (in million units)")

The sales distribution is amazing! The top 5 Publishers take home multiple times the sales of the rest combined. While it is expected given the distribution of game titles and developers, the plot shows how hard it is to compete with these big buys in the industry.

EA and Ubisoft joined late but started to dominate the market very quickly. Note that EA overtaked most of the competition by 2010. However, their practice of buying out their competition is not very popular with the fans.

EA acquisitions list

EA reddit breakdown

toppub <- vgsales %>%
    filter(Publisher %in% c("Electronic Arts", "Activision", "Ubisoft", "Namco Bandai Games", "Nintendo")) %>%
    gather("Region", "Sales", ends_with("Sales")) %>%
    group_by(Publisher, Year_of_Release, Region) %>%
    filter(Region != "Global_Sales") %>%
    summarize(total = sum(Sales))
vgsales %>%
    gather("Region", "Sales", ends_with("Sales")) %>%
    group_by(Publisher, Year_of_Release, Region) %>%
    filter(Region != "Global_Sales") %>%
    summarize(total = sum(Sales)) %>%
    ggplot(aes(x = Year_of_Release, y = total)) +
    geom_point(alpha = 0.5, pch = 21) + 
    facet_wrap(~Region, scales = "free_y") +
    geom_point(data = toppub, aes(col = Publisher), size = 0.5) +
    geom_line(data = toppub, aes(col = Publisher), size = 1.1) +
    scale_color_viridis(discrete = TRUE) +
    theme(legend.position = "top", axis.title.x = element_blank()) +
    labs(title = "Time series of Global Sales by Publishers and Region", y = "Sales (in million units)")

When sorted through regions, sales distribution is the same for NA and US. However, Nintendo and Manco Bandai Games are the only big players in Japan market.

Moreover, NA an EU markets generates much more sales than Japan and others - population matters here.

Notice there is a downward trend after 2010 in sales across the industry.

Reproducing the above decline using different dataset is reccomended to reach any conclusion.

It is not a gaming analysis without system war. Three contenders are: console, PC and handheld.

vgsales %>%
    group_by(System, Year_of_Release) %>%
    summarize(total = sum(Global_Sales, na.rm = TRUE)) %>%
    ggplot(aes(x = Year_of_Release, y = total, fill = System)) +
    geom_col(position = "stack") +
    labs(y = "Sales (in million $)", title = "Global Sales by System and time") +
    scale_fill_viridis(discrete = TRUE)

This is a dangerous statistic. While the data clearly shows PC games sales is practically nothing compared to the other two, remember most revenue of PC games come from MOBA, RPG, … which are free-to-play!

PC generates more revenue

Lets take a closer look at the console markets where there are three big players: xbox, playstation, nintendo and others. The classification only include the big names and gloat over many. However, it should still give us a fair observation

xbox <- c("X360", "XB", "XOne")
nintendo <- c("Wii", "WiiU", "N64", "GC", "NES") 
playstation <- c("PS", "PS2", "PS3", "PS4")

vgsales %>%
    filter(System  == "console") %>%
    mutate(console_type = ifelse(Platform %in% xbox, "xbox",
                                 ifelse(Platform %in% nintendo, "nintendo", 
                                        ifelse(Platform %in% playstation, "playstation",
                                               "others")))) %>%
    group_by(console_type, Year_of_Release) %>%
    summarize(total = sum(Global_Sales)) %>%
    ggplot(aes(x = Year_of_Release, fill = console_type)) +
    geom_density(position = "fill") +
    labs(title = "Distribution of game titles sorted by console type") +
    scale_fill_viridis(discrete = TRUE)

vgsales %>%
    filter(System  == "console", Year_of_Release <= 2010) %>%
    mutate(console_type = ifelse(Platform %in% xbox, "xbox",
                                 ifelse(Platform %in% nintendo, "nintendo", 
                                        ifelse(Platform %in% playstation, "playstation",
                                               "others")))) %>%
    group_by(console_type, Year_of_Release) %>%
    summarize(total = sum(Global_Sales)) %>%
    ggplot(aes(x = Year_of_Release, y = total, col = console_type)) +
    geom_point(alpha = 0.2) + 
    geom_line(alpha = 0.2) +
    stat_smooth(method = "lm", se = FALSE) +
    theme(legend.position = c(0.2, 0.7)) +
    labs(title = "Console sales before 2010 across console platform", y = "Sales (in units)") +
    scale_color_viridis(discrete = TRUE)

vgsales %>%
    filter(System  == "console", Year_of_Release > 2010) %>%
    mutate(console_type = ifelse(Platform %in% xbox, "xbox",
                                 ifelse(Platform %in% nintendo, "nintendo", 
                                        ifelse(Platform %in% playstation, "playstation",
                                               "others")))) %>%
    group_by(console_type, Year_of_Release) %>%
    summarize(total = sum(Global_Sales)) %>%
    ggplot(aes(x = Year_of_Release, y = total, col = console_type)) +
    geom_point(alpha = 0.2) + 
    geom_line(alpha = 0.2) +
    stat_smooth(method = "lm", se = FALSE) +
    labs(title = "Console sales after 2010 across console platform", y = "Sales (in units)") +
    scale_color_viridis(discrete = TRUE)

Few observations:

  1. Nintendo was the only big player before 1995 and thus they had a complete control over the market
  2. After 1995, Xbox and playstation gained the majority of market shares
  3. There are only 3 big players in console in the last 20 years, all others are going extinct.
  4. Again we see a sharp decline in sales after 2010

Sales vs Genre and Rating

If you are not familiar with games rating, here is the short version. Note that we have an empty string which means missing value for one of the rating category

AO - Adult only 18+ M - Mature Only 17+ T - Teen E10+ - Everyone 10+ E - Everyone EC - Early Childhood RP - Rating Pending

vgsales %>%
    ggplot(aes(x = Rating, y = Genre, col = Genre)) +
    geom_jitter(alpha = 0.4, pch = 21) +
    theme(legend.position = "none") +
    scale_color_viridis(discrete = TRUE)

There are many games without rating. In contrast, most Publisher avoids K-A, EC and AO rating since the playerbase is too specific. Genre is not affected by time as much.

vgsales %>%
    group_by(Genre, Name) %>%
    summarise(Sales = sum(Global_Sales)) %>%
    top_n(n = 1) %>%
    kable()
## Selecting by Sales
Genre Name Sales
Action Grand Theft Auto V 56.57
Adventure Assassin’s Creed 11.27
Fighting Super Smash Bros. Brawl 12.84
Misc Wii Play 28.92
Platform Super Mario Bros. 45.31
Puzzle Tetris 35.84
Racing Mario Kart Wii 35.52
Role-Playing Pokemon Red/Pokemon Blue 31.37
Shooter Call of Duty: Black Ops 30.82
Simulation Nintendogs 24.67
Sports Wii Sports 82.53
Strategy Pokemon Stadium 5.45

Lastly, lets take a look at how different regions favor different genre.

vgsales %>%
    gather("Region", "Value", c("NA_Sales", "EU_Sales", "JP_Sales", "Other_Sales")) %>%
    group_by(Region, Genre) %>%
    summarize(Sales = sum(Value)) %>%
    top_n(n = 3) %>%
    ggplot(aes(x = Region, y = Sales, group = Region, fill = Genre)) +
    geom_col(position = "stack") +
    scale_fill_viridis(discrete = TRUE) +
    labs(title = "Top Genre by Sales per Region")
## Selecting by Sales

Lastly, lets take a look at the Name wordcloud

names <- concatenate(vgsales$Name)
names <- clean(names)

unigram <- ngram(names, n = 1)
freqs <- get.phrasetable(unigram)
wordcloud(freqs$ngrams, freqs$freq, random.order = FALSE, rot.per = 0.35,
          color=brewer.pal(8, "Dark2"), max.words = 100)

Final Notes

  1. Game sales peaked in 2008 and has been decreasing ever since. However, the data is not a representation of overall gaming scene. Modern games focus on free to play platform, dominated by PC games.
  2. The data is flawed since it only takes into account game sales where most revenue from games comes from microtransactions. Again, we did not have the data to show PC games on the rise.
  3. EA, Nintendo, Sony are the last few big companies can manage to survive
  4. Most bad games die out instantly in the last few years which signify the rise in standard of games.
  5. The analysis focuses on consoles due to the data structure.