Game Characteristics

What do the other game attributes look like? Specifically I was interested in a game’s expansions and mechanics, which were each stored in a single column, with multiple expansions and mechanics separated by commas. I used splitstackshape::concat.split to create list columns inside the tibble with those values, then counted the number of items in the list for each game to get a count of expansions and mechanics. Minimum number of players also had a long tail, with values above 5 infrequent, so I lumped those into one category.

board_games <- board_games %>% 
  splitstackshape::concat.split("expansion", sep = ",", structure = "list") %>% 
  splitstackshape::concat.split("mechanic", sep = ",", structure = "list") %>% 
  mutate(expansion_n = map_dbl(expansion_list, function(x) sum(!is.na(x))), 
        mechanic_n = map_dbl(mechanic_list, function(x) sum(!is.na(x)))) %>% 
  mutate(min_players_fct = as.factor(case_when(min_players == 0 ~ NA_character_,
                                                  min_players == 1 ~ "1",
                                                  min_players == 2 ~ "2",
                                                  min_players == 3 ~ "3",
                                                  min_players == 4 ~ "4", 
                                                  min_players >= 5 ~ "5+",
                                                  TRUE ~ NA_character_)))

p1 <- ggplot(board_games, aes(x = min_players_fct)) + 
  geom_bar(fill = pal[1], stat = "count") +
  labs(x = "Minimum number of players", 
       y = "Number of games") +
  bbplot::bbc_style() + 
  theme(axis.text=element_text(size=10),
        axis.title=element_text(size=12,face="bold"))

p2 <- filter(board_games, playing_time <= 600) %>% 
  ggplot(aes(x = playing_time)) + 
  geom_histogram(fill = pal[1], bins = 15) +
  labs(x = "Play time (min.)", 
       y = "Number of games") +
  bbplot::bbc_style() + 
  theme(axis.text=element_text(size=10),
        axis.title=element_text(size=12,face="bold"))

p3 <- ggplot(board_games, aes(x = mechanic_n)) + 
  geom_histogram(fill = pal[1], bins = 20) +
  labs(x = "Number of mechanics", 
       y = "Number of games") +
  bbplot::bbc_style() + 
  theme(axis.text=element_text(size=10),
        axis.title=element_text(size=12,face="bold"))

p4 <- filter(board_games, expansion_n < 100) %>% 
  ggplot(aes(x = expansion_n)) + 
  geom_histogram(fill = pal[1], bins = 50) +
  labs(x = "Number of expansions", 
       y = "Number of games") +
  bbplot::bbc_style() + 
  theme(axis.text=element_text(size=10),
        axis.title=element_text(size=12,face="bold"))

p <- cowplot::plot_grid(p1, p2, p3, p4, nrow = 2)
title <- cowplot::ggdraw() + 
  cowplot::draw_label("Distributions of game characteristics", 
                      fontface='bold', size = 16)
cowplot::plot_grid(title, p, ncol = 1, rel_heights=c(0.1, 1))

The cowplot package helped me get all four plots into one figure. Most games require a minimum of two players, and the median play time is 45 minutes. The plot only shows games <=10 hours, but there were 56 board games with play time over 10 hours. These are likely a mix of games played over multiple sessions and erroneous data. Most games employ several game mechanics, and most games don’t have any expansions, but there were 9 games with more than 100 expansions that I removed from the plot, including one game with 420 expansions!

Game Mechanics

I was most interested in further exploring the game mechanics, and since each game can have multiple mechanics, I used tidyr::separate_rows to get each mechanic on its own line.

board_games_mech <- board_games %>% 
  tidyr::separate_rows(mechanic, sep = ",") %>% 
  filter(!is.na(mechanic)) %>% 
  group_by(mechanic) %>% 
  add_tally() %>% 
  ungroup() %>% 
  mutate(mechanic = fct_reorder(mechanic, n))

p5 <- board_games_mech %>% 
  ggplot(aes(x = mechanic)) + 
  geom_bar(stat = "count", fill = pal[5]) +
  coord_flip() + 
  labs(x = "Game mechanic",
      y = "Number of games") +
  bbplot::bbc_style() + 
  theme(axis.text=element_text(size=8),
        axis.title=element_text(size=12,face="bold"))

p6 <- board_games_mech %>% 
  ggplot(aes(x = mechanic, y = average_rating)) +
  geom_boxplot() +
  coord_flip() +  
    theme(axis.text.y = element_blank(),
        axis.title.y = element_blank()) +
  labs(y = "Average rating") +
  bbplot::bbc_style() + 
  theme(axis.text=element_text(size=8),
        axis.title=element_text(size=12,face="bold"))

pp <- cowplot::plot_grid(p5, p6, nrow = 1)
title <- cowplot::ggdraw() + cowplot::draw_label("Number of ratings and average rating of games by mechanic", fontface='bold', size = 16)
cowplot::plot_grid(title, pp, ncol = 1, rel_heights=c(0.1, 1))

board_games_mech %>% 
  group_by(mechanic) %>% 
  summarize(n = n(), average_rating = mean(average_rating)) %>% 
  ggplot(aes(x = n, y = average_rating)) + 
  geom_point(color = pal[5]) + 
    labs(x = "Number of games",
       y = "Average rating", 
       title = "Number of games and average rating of games by mechanic") +
  bbplot::bbc_style() + 
  theme(axis.text=element_text(size=8),
        axis.title=element_text(size=12,face="bold"),
        plot.title=element_text(size=16,face="bold"))

There are 51 different mechanics, with Dice Rolling and Hand Management being the most frequent. There is some variability in rating across mechanics, and the ratings of the games in a mechanic do not seem related to the number of games in that mechanic. Notably, Roll/Spin and Move stands out as having comparably lower ratings.

I’m not familiar with what all of these mechanics actually mean, so I wanted to see the highest rated and most rated game in the most popular mechanics to get an idea of the type of games that use those mechanics.

board_games_mech %>% 
  group_by(mechanic) %>% 
  summarize(top_rated = first(name, order_by = desc(average_rating)), most_rated = first(name, order_by = desc(users_rated))) %>% 
  arrange(desc(mechanic)) %>% 
  head(10) %>% 
  kable()

Board Game Mechanic Association Rules

Since games frequently have multiple mechanics, are there game mechanics that commonly co-occur? To answer this question, I used association rules to identify commonly co-occurring game mechanics. In order to use the arules package for association rule mining, the data has to be in “transactions” format. In order to get it in this format, I first restructured the data so that each mechanic was on its own line. From there, I transformed the data into a list, where each item in the list represents a single game and contains a factor array of that game’s mechanic(s). This can then be converted into “transactions” format.

library("arules")
library("arulesViz")

board_games_mech <- board_games %>% 
  select(game_id, mechanic) %>% 
  tidyr::separate_rows(mechanic, sep = ",") %>% 
  filter(!is.na(mechanic)) %>% 
  group_by(game_id) %>% 
  # filter(n() > 1) %>% # removing games with only one mechanic
  ungroup() %>% 
  mutate_all(as.factor)

mechanics_list <-  split(board_games_mech$mechanic, board_games_mech$game_id)
mechanics_transactions <- as(mechanics_list, "transactions")
summary(mechanics_transactions)

> transactions as itemMatrix in sparse format with
>  9582 rows (elements/itemsets/transactions) and
>  51 columns (items) and a density of 0.04900938 
> 
> most frequent items:
>           Dice Rolling        Hand Management         Set Collection 
>                   2438                   2176                   1347 
>        Hex-and-Counter Variable Player Powers                (Other) 
>                   1244                   1223                  15522 
> 
> element (itemset/transaction) length distribution:
> sizes
>    1    2    3    4    5    6    7    8    9   10   11   12   15   18 
> 2852 2716 1987 1113  524  217  111   30   23    4    2    1    1    1 
> 
>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
>   1.000   1.000   2.000   2.499   3.000  18.000 
> 
> includes extended item information - examples:
>                          labels
> 1                        Acting
> 2 Action / Movement Programming
> 3 Action Point Allowance System
> 
> includes extended transaction information - examples:
>   transactionID
> 1             1
> 2             2
> 3             3

The board game data is now stored as a matrix. The density in a sparse matrix indicates the proportion of non-empty cells: 0.05. The most frequent items and itemset/transaction length distribution match the distributions above, showing that Dice Rolling is the most frequent mechanic, and 2852 games only have one mechanic. If I had not already done this exploratory analysis, I would usually use itemFrequencyPlot to visualize the most frequent mechanics.

First, I’m using the frequent itemsets target to find combinations of game mechanics that commonly occur together.

itemsets <- apriori(mechanics_transactions, 
                    parameter = list(supp=0.01, minlen = 2, maxlen=5, 
                                     target = "frequent itemsets"))

summary(itemsets)

> set of 73 itemsets
> 
> most frequent items:
>                  Dice Rolling               Hand Management 
>                            26                            19 
>        Variable Player Powers                Set Collection 
>                            16                            10 
> Area Control / Area Influence                       (Other) 
>                             9                            72 
> 
> element (itemset/transaction) length distribution:sizes
>  2  3 
> 67  6 
> 
>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
>   2.000   2.000   2.000   2.082   2.000   3.000 
> 
> summary of quality measures:
>     support            count      
>  Min.   :0.01012   Min.   : 97.0  
>  1st Qu.:0.01158   1st Qu.:111.0  
>  Median :0.01472   Median :141.0  
>  Mean   :0.01810   Mean   :173.5  
>  3rd Qu.:0.01993   3rd Qu.:191.0  
>  Max.   :0.05469   Max.   :524.0  
> 
> includes transaction ID lists: FALSE 
> 
> mining info:
>                    data ntransactions support confidence
>  mechanics_transactions          9582    0.01          1

There are 73 combinations of mechanics that occur in at least 1% of the data: 67 two mechanic combinations, and 6 three mechanic combinations.

One important metric for frequent itemsets is support, which indicates how frequently the itemset appears in the data set, or the proportion of games that contain the combination of mechanics.

Below are the 10 mechanic combinations with the highest support. The most frequent combination of mechanics is Dice Rolling and Variable Player Powers, which occurred in about 5.5% of all games. This combination makes sense, because those two mechanics occurred with high frequency (they were the 1st and 4th highest, respectively). Interestingly, the combination of the two most frequent mechanics, Dice Rolling and Hand Management, only occurs 6th most frequently as a pair.

DATAFRAME(itemsets) %>% 
  arrange(desc(support)) %>% 
  head(10) %>% 
  rename("mechanics" = items) %>% 
  kable(row.names = F, digits = 3)

Moving on to association rule mining, which gives output in the form of if this then that rules. For example, if a game employs mechanic X, it is likely to also employ mechanic Y.

association_rules <- apriori(mechanics_transactions, 
                             parameter = list(supp=0.001, conf=0.8, maxlen=5))

summary(association_rules)

> set of 51 rules
> 
> rule length distribution (lhs + rhs):sizes
>  3  4  5 
>  9 37  5 
> 
>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
>   3.000   4.000   4.000   3.922   4.000   5.000 
> 
> summary of quality measures:
>     support           confidence          lift            count      
>  Min.   :0.001044   Min.   :0.8000   Min.   : 3.144   Min.   :10.00  
>  1st Qu.:0.001148   1st Qu.:0.8363   1st Qu.: 3.983   1st Qu.:11.00  
>  Median :0.001357   Median :0.8667   Median : 7.123   Median :13.00  
>  Mean   :0.002089   Mean   :0.8826   Mean   :10.141   Mean   :20.02  
>  3rd Qu.:0.001513   3rd Qu.:0.9253   3rd Qu.:11.109   3rd Qu.:14.50  
>  Max.   :0.007201   Max.   :1.0000   Max.   :30.419   Max.   :69.00  
> 
> mining info:
>                    data ntransactions support confidence
>  mechanics_transactions          9582   0.001        0.8

There are 51 association rules that occur in at least .1% of the data. The support threshold is very low for association rules because there are so many games with only one mechanic that don’t fit any association rules.

Association rule mining employs an additional metric, confidence, which indicates how often the rule is true, or the proportion of games that have the left hand side (LHS) that also contain the right hand side (RHS). Crayon Rail System only appears in <1% of games, but 86% of the time a game is labeled as both Crayon Rail System and Route/Network Building, it is also labeled as Pick-up and Deliver, with this association rule occurring in 12 games.

DATAFRAME(association_rules) %>% 
  head(11) %>% 
  kable(row.names = F, digits = 3)

Below are the 10 association rules with the highest confidence. For the first 5, in 100% of the games that the combination of LHS mechanics occur, the RHS mechanic occurs as well. So 100% of the games labeled as Auction/Bidding, Stock Holding, and Tile Placement are also labeled as Route/Network Building, and this association occurs in 23 games.

DATAFRAME(association_rules) %>% 
  arrange(desc(confidence)) %>% 
  head(10) %>% 
  kable(row.names = F, digits = 3)

Network graph of game mechanics

This shows constellations of mechanics that frequently occur together.

require(widyr)
require(igraph)
require(ggraph)
require(ggforce)

board_games %>% 
  select(name, mechanic) %>% 
  tidyr::separate_rows(mechanic, sep = ",") %>% 
  filter(!is.na(mechanic)) %>% 
  mutate_all(as.factor) %>% 
  pairwise_cor(mechanic, name, sort = T) %>% 
  filter(correlation > .15) %>% 
  graph_from_data_frame() %>% 
  ggraph() +
  geom_edge_link() +
  geom_node_point() +
  geom_node_text(aes(label = name), repel = T) + 
  theme_void() + 
  labs(title = "Network of game mechanics that commonly occur together")