Exploring Insights for Success in the Video Game Industry

Project Report for STAT 303-1 Fall 2025

Author

Will Harwood

Published

December 6, 2025

Abstract
This analysis examines trends in video games on the Steam platform. Success on Steam is dominated by a few archetypes, notably action-based multiplayer shooters and open world survival games. Developers aiming for high engagement should focus on Action or Simulation games, integrate multiplayer mechanics, and consider Free-to-Play models and open world features. Platform curation should go beyond raw player counts, highlighting titles that overperform relative to their genre, such as Massively Multiplayer games with high positive reviews or Casual and Indie games with above-average playtime. Overall, player counts alone do not capture success, and evaluating games in the context of genre standards is essential for good game development and effective recommendations.

1) Inspiration

As a long-time Steam user and gaming enthusiast, I chose this dataset to explore how game attributes relate to player behavior, popularity, and commercial outcomes. Steam offers a vast and diverse catalog, making it an ideal environment for uncovering trends in what players enjoy and how trends differ across genres. Because I’ve purchased and played games on the platform for years, the subject is personally interesting, and a data-driven project in this space felt both worthwhile and insightful.

2) Problem Statements

  • What kind of game should large-scale game developers make based on recent trends?

  • What features are common among the best performing games on Steam?

  • How does success vary within different genres?

3) Data

My datasets are the Steam Monthly Average Players Dataset, and the Steam Games Dataset, both from Kaggle, where the curators gathered and formatted Steam API data.

The Monthly Average Players Dataset has:

  • 612,265 observations, 7 total variables
  • 3 categorical variables
  • 4 numeric variables
  • Each observation represents a game during a specific month
  • Each variable represents the name of the game, its Steam ID, average player count, or other statistic about the game during that month.
  • In this report, 4 variables are used: month, name, avg_players, and app_id.

The Steam Games Dataset has:

  • 111,452 observations, 42 total variables
  • 27 categorical variables
  • 15 numeric variables
  • Each observation represents a unique game
  • Each variable represents the name of the game, its Steam ID, the genres it is a part of, or other fact about the game.
  • In this report, 19 variables are used: genres, positive, negative, estimated_owners, average_playtime_forever, peak_ccu, tags, required_age, price, dlc_count, windows, linux, mac, achievements, recommendations, user_score, average_playtime_2weeks, median_playtime_2weeks, and median_playtime_forever.

4) Stakeholders

The primary stakeholders are game developers and the Steam platform. Developers benefit from understanding which game features, tags, and genres are correlated with strong player engagement, since building a game in a declining or highly oversaturated genre can increase financial risk. Steam benefits because identifying which games overperform relative to their genre norms can improve how they select which games should be featured on the store and recommended to players. Insights into player count patterns across genres may help Steam highlight titles that users are more likely to enjoy, improving customer retention and visibility for new and up-and-coming games.

5) Data Cleaning

a) Cleaning - Question 1

The Monthly Average Players Dataset has no missing or obviously incorrect values. However, in the primary column used for answering this question, avg_players, 97,477 observations (of 612,265 total) sit outside of Tukey’s Fences. This is expected for a distribution of average players of different video games, since there are going to be massively popular games with tens of thousands of players as well as many poor-performing games with single-digit numbers of players. It does not make sense to remove these outliers, since the analysis would be sparse without them.

In the Steam Games Dataset, among the variables used for Question 1, only one had missing values. The genres value was set to an empty list for 6440 games in the dataset, mostly playtest games that had all numeric variables equal to 0. Because a games genres is incredibly difficult to impute without any other info besides its name, and that we do not want to consider games without any genres when analyzing trends with genres, I chose to remove observations without any genres. There were no incorrect values in the genres column.

For this question and the following, there are some columns of the Steam Games Dataset that are irrelevant. Most of the irrelevant columns contain descriptions about the game or links relating to the game. The irrelevant columns are: detailed_description, about_the_game, short_description, reviews, header_image, website, support_url, support_email, metacritic_url, screenshots, and movies. These columns were dropped from the dataset and unused for the remainder of the project.

Since this question is focusing on what kind of game a large studio should make, it makes sense to remove genres with few games in them. These genres represent small markets that do not make sense to spend a lot of time and money making a game for. I excluded genres with less than 500 games in them (only keeping genres with > ~0.5% of overall market share) from the analysis. Also, “Free to Play” had an alternate name “Free To Play”. The labels were merged to create one unified “Free to Play” genre.

The kept genres were: Indie, Casual, Action, Adventure, Simulation, Strategy, RPG, Early Access, Free to Play, Sports, Racing, Massively Multiplayer, Utilities, Design & Illustration, Violent.

The final cleaned Steam Games data for question 1 analysis contained 104,631 games out of the 111,452 initially present in the dataset.

b) Cleaning - Question 2

The Monthly Average Players Dataset was not used for this question.

In the Steam Games Dataset, there are some missing values. In the discount column, there are 97,410 missing values. Because this represents almost 90% of all games in the dataset, this means that the discount column is not very useful for any analysis. It was considered that discount being NA could mean a game has never been on discount, however, there are some games with a discount value of 0. For these reasons, I chose to drop discount values from the dataset, and not use it for the project.

Another column in the Steam Games Dataset, peak_ccu, representing the Peak Concurrent Playercount for a game, had some subtle issues despite not missing any data. Take the game Alien Breed 3: Descent, which has a peak_ccu value of 3, despite having an estimated owner count somewhere between 200,000 and 500,000, and hundreds of reviews. Games with a peak_ccu below 0.5% of the lower bound of their estimated_owners were flagged as potential outliers, since such low peak concurrent counts appear unusually small relative to their owner base. By doing this I found 18,185 games with probably incorrect peak playercount values. In order to fix the incorrect peak playercount values, I chose to find the variable most correlated with peak_ccu among games with correct values, and then create a linear function to impute the incorrect values. I figured that this method of imputing values was more likely to give accurate peak_ccu values than using the mean of the column or means of various categories, because there were a few numeric variables in the dataset that are statistically related to peak concurrent playercount. The most correlated variable ended up being the upper bound of estimated owners. The fixed values were combined with the correct ones into a new dataset for analysis.

In the final cleaned peak_ccu data, there were 24,136 outliers. Similar to last question, player counts tend to have a few very high values and a lot values close to 0, so this is expected, and nothing was done about the outliers.

Most games do not have a metacritic_score. It is set to 0 for 96% of the games in dataset. For this reason, I chose to drop metacritic score from the data.

There were no other incorrect or missing values in the Steam Games Dataset.

Among the other numeric variables, the number of detected outliers was as follows:

  • required_age: 1,747
  • price: 5,102
  • dlc_count: 13,784
  • achievements: 8,193
  • recommendations: 15,157
  • user_score: 44
  • positive: 18,765
  • negative: 18,750
  • average_playtime_forever: 15,229
  • average_playtime_2weeks: 2,333
  • median_playtime_forever: 15,229
  • median_playtime_2weeks: 2,333

No cleaning was necessary for the outliers present. Game data is often skewed, with popular games amassing thousands of reviews, recommendations, hours of average playtime, etc., and removing them would take away important parts of the dataset.

Besides the given numeric column, I also wanted to correlate the tags of a game to see if they had any correlation with a game’s maximum number of players. To do this, I created a one-hot-encoded dataframe with each game’s top 5 tags. If a game did not have 5 tags, I kept as many as they had. Each game came with a list of tags and the number of players that voted for that tag, so the top 5 tags should be an accurate representation of what very specific categories the game fits in to. If tags were not limited, any tag that any player voted for would be present, which introduces a lot of noise and bad tags. There were 444 different tags that appeared in the top 5 tags of any game in the dataset. 37,423 games were untagged and removed from the final data.

windows, linux, and mac are boolean categories representing if a game supported the platform or not. I kept all categories to be used to correlate with success statistics.

c) Cleaning - Question 3

To analyze how success varies by genre, it is necessary to analyze success metrics (like peak_ccu) as well as genres. Fortunately, both of those have already been cleaned up for the previous questions. To create a dataset with both values cleaned, I found all indices of the first question’s dataset (which only contained games that had a genre with more than 500 games in it) that were also in the second question’s dataset (which had imputed peak playercount data). Lastly, I filtered the games in this combined data to also be among that games that had enough tags to be in the one-hot encoded tags data created in question 2. All together, this created a dataset with 73,674 observations, all with good genre, peak ccu, and tag data.

This dataset contained no missing or incorrect values.

The numeric data used for this question was peak_ccu, average_playtime_forever, positive, and negative. There were 7166 outliers in peak_ccu, 14899 in average playtime, 12333 in the number of positive reviews, and 11816 in the negative reviews. I kept all outliers to avoid removing the most successful games from my analysis.

The categories analyzed in this question were genres and tags. The genres used were: Indie, Casual, Action, Adventure, Simulation, Strategy, RPG, Early Access, Free to Play, Sports, Racing, Massively Multiplayer, Utilities, Design & Illustration, Violent. Similarly to above, I only wanted to keep genres with a significant number of games present. The data had 444 different tags present.

6) Data Analysis

a) Analysis 1

b) Analysis 2

What features are common among the best performing games on Steam?

Methodology

My first step was to check if any features in particular correlate well with a games success. Initially, I only looked at numeric variables present in the Games Dataset. I correlated each numeric variable to 3 chosen “success” stats: peak_ccu (peak concurrent players), average_playtime_forever, and median_playtime_forever. The correlation heatmap is plotted in Figure 3.

However, I also wanted to see if any tags of a game would correlate to success. To do this, I merged the cleaned playercount data with the one-hot encoded Game Tags data prepared in 5b. I dropped every column except peak_ccu and all of the tags. Peak Concurrent Players is the best way to measure the success of a game since a lot of niches games have dedicated player bases driving up playtime. Afterwards I correlated every variable in the dataset (400+ different tags) with peak playercount. I took the 20 most positively correlated and 20 most negatively correlated tags, to avoid crowding the plot, and displayed them in Figure 4.

To get a better look at only the best-performing Steam Games, I sorted and filtered the clean dataset to only include the top 1000 games by peak playercount. I chose top 1000 because that represents roughly the top 1% of games in the dataset, which is a good goal for a game to reach. Then I grabbed each games top 5 most voted-for tags by the Steam community and aggregated all 1000 games tags into a list before counting the unique occurences of each tag. Since tags are often more specific about a game’s content than its genre, this should help us make narrower deductions about what kind of games are popular on Steam than genre alone gave us. The 30 most commonly appearing tags in the top games are displayed in Figure 5.

Lastly, I wanted to find out what kind of tags appear with each other the most among successful games. If developers can identify common successful tag pairs, it would help narrow down features they should put into their games. To do this, I borrowed the dataset made for Figure 3 with peak playercount and one-hot encoded tag information. I sorted and filtered to only have the top 1000 games by max players, and then removed the playercount column so that only the tag information for the top games was left. To find the number of tag co-occurrences I multiplied this matrix by its transpose, creating a 449x449 matrix of total tag co-occurrences in the top 1000 games. For appealing visualization, I filtered out any tags with less than 30 appearances. The top tags were plotted in a heatmap in Figure 6. Underneath the heatmap is a table containing the 15 most co-occurring tags in the games selected, with repeats (tag A and tag B vs tag B and tag A) removed.

Figure 3

peak_ccu average_playtime_forever median_playtime_forever
required_age 0.073453 0.051255 0.020551
price 0.056024 0.057227 0.038195
dlc_count 0.021401 0.036153 0.015961
windows 0.000839 0.001179 0.000780
mac 0.022686 0.031282 0.017946
linux 0.022171 0.021071 0.011096
achievements 0.015393 0.015543 0.007444
recommendations 0.439511 0.182814 0.040074
user_score -0.000899 0.000473 0.000595
positive 0.509238 0.204040 0.035680
negative 0.408061 0.195260 0.036887
average_playtime_2weeks 0.135902 0.164185 0.080127
median_playtime_2weeks 0.097327 0.148620 0.076562

recommendations is the number of recommendations a game has, positive the number of positive reviews, etc. windows, mac, and linux refer to if a game supports that operating system.

Unsurprisingly, a game’s required age or windows support do not say much about how many peak concurrent players it is likely to have. The only metrics that correlate well with success metrics are other metrics of success.

Worthy of note is the mild positive correlation between success and price. This is likely because of the abundance of poorly made Free to Play games on Steam.

Figure 4

Positive Negative
Genre Corr Genre Corr
0 Multiplayer 0.114464 Indie -0.034362
1 Open World 0.078566 Adventure -0.019557
2 Co-op 0.051658 Puzzle -0.016628
3 FPS 0.047952 Arcade -0.014305
4 Free to Play 0.047323 2D -0.013125
5 Hunting 0.044917 2D Platformer -0.011854
6 Online Co-Op 0.043284 VR -0.011815
7 Survival 0.039269 Visual Novel -0.011328
8 Mythology 0.034798 Hidden Object -0.010161
9 Open World Survival Craft 0.034306 3D -0.009694
10 MMORPG 0.034282 Exploration -0.009693
11 Souls-like 0.033993 Action-Adventure -0.009671
12 Shooter 0.026959 Simulation -0.009612
13 Competitive 0.026925 Platformer -0.009144
14 Action RPG 0.024048 Shoot 'Em Up -0.007478
15 MOBA 0.023334 Bullet Hell -0.007391
16 Looter Shooter 0.022294 Side Scroller -0.007230
17 Hero Shooter 0.021828 Cute -0.007160
18 Creature Collector 0.020661 Interactive Fiction -0.007152
19 Battle Royale 0.017832 Retro -0.007036

It is clear that no tag is a guarantee of success based on these correlations. However, there are a lot of Multiplayer-adjacent tags in the top 20 most correlated, such as Multiplayer, Co-op, Online Co-op, and MMORPG. This suggests that adding multiplayer features to a game might give it the slightest edge in garnering popularity.

We also see that the theme of gun-based combat games dominating repeats itself here. There are 4 shooter tags in the top 20 most correlated: FPS, Shooter, Looter Shooter, and Hero Shooter.

One potential insight that did not appear in the first analysis is the popularity of Survival games, with both Survival and Open World Survival Craft appearing in the top 20.

On the side of negative correlations, most are very weak. It makes sense that Indie is the lowest, however, since Indie games do not have the budget and support of games built by large studios, and therefore are less likely to garner a lot of players.

Figure 5

count
tags
Action 370
Multiplayer 351
Free to Play 176
Open World 168
FPS 166
RPG 143
Strategy 137
Singleplayer 132
Adventure 116
Shooter 110
Simulation 104
Co-op 88
Survival 81
Horror 64
Indie 64
Story Rich 64
Sci-fi 60
First-Person 56
Massively Multiplayer 47
Online Co-Op 45
Zombies 40
Turn-Based Strategy 39
Hack and Slash 37
RTS 37
Classic 36
Fantasy 36
Atmospheric 35
Sports 34
Casual 34
Early Access 34

The top 30 tags back up some of the correlations seen in Figure 4. Multiplayer, Shooter, and Survival games are among the most popular type of game in Steam’s top 1000 games by playercount.

The line plots from Figure 1 revealed that RPG and Strategy were among the more stagnant genres, but here we can see they still command sizeable percentages of Steam’s top 1000 games, at 14.3% and 13.7%, respectively.

Because of the sheer volume of Indie games available, it makes sense that there are 64 in the top 1000.

Interestingly, Simulation games have almost as many appearances as Shooter games. Based on the growth seen in Figure 1, it might be the case that Simulation climbs into the top 5 of this list within the next few years.

Figure 6
Code 
sorted = ccuAndTags.sort_values("peak_ccu", ascending=False)
top50 =sorted.iloc[:1000, 1:].astype(int)
occur = top50.T.dot(top50)
maxOccurs = occur.max()
smalltags = list(maxOccurs.loc[maxOccurs<30].index)
data = occur.drop(index=smalltags, columns=smalltags)
plt.figure(figsize=(13,7))
a = sns.heatmap(data, vmin=0, vmax=50, cmap="coolwarm", square=True, linewidths=.01)
gradient = a.collections[0].colorbar
gradient.set_ticks([10,20,30,40,50])
gradient.set_ticklabels([10,20,30,40,">50"])
plt.xticks(ticks=np.arange(len(data.columns))+0.5, labels=data.columns, fontsize=7)
plt.yticks(ticks=np.arange(len(data.index))+0.5, labels=data.index, fontsize=7)
plt.title("Only among tags with >30 occurrences",fontsize=9)
plt.suptitle("Heatmap of Tag Co-occurrence in Top 1000 Steam Games", ha="center", x=0.55, y=.95)
plt.show()
sorted = data.reset_index().melt(id_vars="index").sort_values("value", ascending=False)
sorted.loc[(sorted["index"] != sorted["variable"])].head(30).iloc[::2].rename(columns={"index":"Tag 1", "variable":"Tag 2", "value":"Occurrences"})

Tag 1 Tag 2 Occurrences
20 Multiplayer Action 148
10 FPS Action 114
514 Multiplayer Free to Play 110
400 Multiplayer FPS 96
1036 FPS Shooter 86
787 Shooter Multiplayer 81
1026 Action Shooter 79
22 Open World Action 72
29 Singleplayer Action 67
76 Action Adventure 61
494 Action Free to Play 57
856 Multiplayer Open World 56
874 Action RPG 47
767 Co-op Multiplayer 46
859 RPG Open World 44

Co-occurrence counts how often both tags appear in the same game, not how similar the tags are.

Once again, Action dominates, appearing in 8 of the top 15 tag co-occurrences. These co-occurrences indicate that the most successful games share a common cluster of tags: Action, Multiplayer, Free-to-Play, FPS, Shooter games. These 5 tags describe the most popular games on Steam’s platform, such as Counter-Strike and PUBG.

Looking at some of the less popular tags, we can identify some interesting trends from the heatmap. Simulation games often are also Multiplayer or Strategy games, but not Action or FPS (First Person Shooter).

Popular Survival games typically have Multiplayer and Open World features.

Horror games do not have a lot of co-occurrences with any other popular tag. This suggests that the Horror genre is less reliant on broad tags like Action or Shooter, and has more diversity of game type.

Answer

Across all analyses, the most common tags in the top 1000 games, the strongest correlations with peak playercount, and the top tag co-occurrences, a clear pattern emerges. The dominant profile of a highly popular Steam game centers on Action, Multiplayer, Free to Play, and FPS/Shooter elements. These tags appear both frequently (Action = 370/1000 games, Multiplayer = 351/1000) and with strong positive correlations. This combination describes many of Steam’s most successful games, such as Counter-Strike and PUBG.

A second set of popular features involves Open World, Survival, and Adventure. These appear in the top correlated tags (Open World, Survival, Open World Survival Craft) and also rank highly in total frequency among successful games. This trend reflects the growing appeal of large, explorable environments.

Some other popular features that do not fit into any set archetype of game are Co-op, Horror, and Story-Rich.

c) Analysis 3

How does success vary within different genres?

Methodology

To start, I used the same genres from Analysis 1, only the ones with enough games within them to give meaningful insights, and the cleaned data prepared in Cleaning 3. For each genre, I found each game that had that genre, and then extracted four statistics: average_playtime_forever, peak_ccu, positive (total number of positive reviews), and a new variable, the percent of reviews that were positive, taken by dividing the positive column by the sum of positive and negative.

After all the data was gathered for each genre, I created a 2x2 grid of boxplots with each genre having their own entries in each plot. Each boxplot represented one of the statistics extracted from the process above. The plot is in Figure 7, with mean and quartile stats in tabular form underneath.

To get a closer look at what kinds of features are popular within in each genre, I grouped the games dataset by genre and sorted it by descending peak_ccu. For each genre, I took the top 300 games by peak playercount and found how many times each tag appears in the most popular titles. I chose the top 300 in each genre to make sure that there was enough data for significant results as well as get a look at a distribution of successful games, not just the top few. In Figure 8, the top 15 most commonly appearing tags within the most popular games of each genre are presented, with some of the results in a table below. The idea behind this is to look at which tags, and therefore which features or aspects of a game, tend to do well within each genre. This allows us to make very particular claims about what is popular in a given genre.

Figure 7

Average Playtime Peak Concurrent Players Percent Positive Reviews Number of Positive Reviews
mean 25% 50% 75% mean 25% 50% 75% mean 25% 50% 75% mean 25% 50% 75%
genre
Action 513.2 83.0 212.0 394.0 3110.74 5.00 297.92 1073.66 75.58 63.04 82.36 97.46 1880.42 4.0 15.0 84.0
Adventure 523.16 93.0 221.0 389.0 1952.16 3.00 297.92 556.50 76.52 65.62 83.33 96.66 1075.42 5.0 17.0 91.0
Casual 422.93 55.0 187.0 298.0 755.74 2.00 297.92 556.50 77.23 65.94 85.5 100.0 352.15 3.0 12.0 45.0
Design & Illustration 3720.15 16.75 127.0 1499.75 1560.47 2.00 7.00 297.92 75.29 65.38 80.0 94.74 1389.57 4.0 13.0 52.25
Early Access 672.5 57.0 210.0 522.5 1712.58 3.00 85.00 556.50 75.35 62.34 83.15 100.0 589.33 3.0 12.0 52.0
Free To Play 942.62 13.0 73.0 377.0 4833.21 297.92 556.50 2625.12 72.65 60.77 75.0 86.36 5185.35 21.0 84.0 462.0
Indie 395.71 71.0 197.0 322.0 1121.33 3.00 297.92 556.50 76.88 66.25 83.77 98.51 634.87 4.0 15.0 67.0
Massively Multiplayer 1169.95 61.0 271.0 1045.0 6266.79 297.92 556.50 2625.12 64.08 50.41 66.28 79.88 4465.12 10.0 59.0 492.0
RPG 774.09 114.0 274.0 673.0 2690.88 7.00 297.92 597.00 75.44 64.1 81.25 94.38 1607.98 5.0 24.0 160.0
Racing 560.44 55.5 186.0 429.5 1226.84 3.50 297.92 556.50 72.45 57.18 79.55 95.24 853.8 4.0 15.0 76.0
Simulation 738.37 75.0 227.5 563.25 1530.61 3.00 297.92 556.50 72.24 58.58 78.36 92.61 1082.98 5.0 20.0 114.0
Sports 915.33 49.0 182.0 539.0 1772.63 3.00 297.92 556.50 72.27 59.68 78.51 93.69 876.98 4.0 15.0 72.0
Strategy 707.88 81.0 242.0 564.5 1416.84 5.00 297.92 556.50 74.66 62.5 80.49 95.0 944.21 4.0 19.0 110.0
Utilities 4797.67 21.0 228.0 1715.0 1279.52 2.00 11.00 556.50 72.65 60.0 77.07 93.68 957.19 4.0 12.0 52.0
Violent 233.61 21.25 166.5 264.75 857.00 297.92 297.92 556.50 58.41 42.13 58.64 78.21 121.25 4.0 13.0 42.0

Massively Multiplayer and RPG games lead in Average Playtime. On the other hand, it makes sense for Free to Play games to have the lowest median Average Playtime, since gamers are less likely to feel the need to finish or spend time on a game they did not pay for.

Examining Peak Concurrent Players, Massively Multiplayer maintains the highest median, now accompanied by Free to Play. Games in the Free to Play category benefit from the absence of a price barrier, increasing accessibility and attracting larger audiences. In contrast, Design & Illustration and Utilities exhibit substantially lower median peak concurrent players. This aligns with expectations, as most applications in these genres, such as illustration tools or wallpaper apps, are not multiplayer, and therefore users do not engage simultaneously with others.

Looking at the number of Positive Reviews, Free To Play and Massively Multiplayer games dominate, with medians of 462 and 492 positive reviews, respectively. These genres tend to attract larger player bases, which naturally results in more reviews. Smaller or niche genres such as Violent and Design & Illustration have far fewer positive reviews, consistent with their smaller audiences.

Lastly, in Percent Positive Reviews, most genres exhibit high median approval, generally above 70%. Adventure, Action, Indie, and Casual games show particularly strong reception, with medians around 80%. In contrast, Massively Multiplayer games have a lower median of roughly 66%, likely reflecting the huge variance in player experiences in games within the genre.

Figure 8

Casual Indie Sports Action Adventure ... Early Access Racing Utilities Violent Design & Illustration
tags count tags count tags count tags count tags count ... tags count tags count tags count tags count tags count
0 Casual 129 Multiplayer 98 Sports 222 Action 153 Action 111 ... Early Access 143 Racing 240 Utilities 288 Violent 231 Design & Illustration 277
1 Free to Play 89 Action 57 Simulation 133 Multiplayer 143 Open World 94 ... Multiplayer 83 Multiplayer 108 Design & Illustration 103 Indie 203 Animation & Modeling 176
2 Indie 87 Indie 55 Multiplayer 109 FPS 86 Adventure 90 ... Action 67 Action 93 Software 84 Action 202 Utilities 156
3 Multiplayer 81 Free to Play 52 Racing 86 Open World 65 Multiplayer 77 ... Free to Play 54 Simulation 91 Animation & Modeling 79 Adventure 132 Game Development 89
4 Action 66 Survival 42 Action 65 Free to Play 60 RPG 54 ... Survival 53 Automobile Sim 79 Video Production 61 Gore 123 Photo Editing 54
5 Simulation 61 Simulation 35 Free to Play 62 Shooter 59 Survival 40 ... Simulation 53 Indie 75 VR 57 Casual 60 Education 50
6 Adventure 52 Co-op 34 Indie 59 Co-op 42 Singleplayer 39 ... Open World 51 Sports 73 Early Access 52 Early Access 48 Software 49
7 Strategy 37 Open World 33 Casual 43 RPG 42 Free to Play 37 ... Strategy 50 Driving 55 Game Development 52 RPG 46 Video Production 49
8 RPG 29 Strategy 31 Automobile Sim 40 Singleplayer 39 Horror 35 ... Indie 41 Casual 41 Audio Production 37 VR 37 Early Access 44
9 Singleplayer 27 RPG 28 Strategy 28 Adventure 35 Co-op 34 ... RPG 40 Free to Play 38 Education 37 Simulation 36 Software Training 29
10 Puzzle 27 Adventure 28 Singleplayer 24 Survival 28 Story Rich 34 ... FPS 39 Arcade 31 Software Training 33 Strategy 36 Audio Production 28
11 Sandbox 22 FPS 27 Management 24 Horror 26 First-Person 24 ... Adventure 29 Open World 29 Indie 33 Horror 31 VR 26
12 Anime 22 Horror 27 Driving 23 First-Person 24 FPS 24 ... Base-Building 25 Adventure 25 Photo Editing 26 Free to Play 26 Simulation 23
13 Funny 19 First-Person 26 PvP 23 Online Co-Op 24 Open World Survival Craft 22 ... Sandbox 25 Singleplayer 21 Casual 24 Nudity 21 Casual 21
14 Massively Multiplayer 18 Singleplayer 23 Soccer 22 Zombies 20 Indie 21 ... Casual 23 Early Access 17 Simulation 23 Sexual Content 17 Indie 19

15 rows × 30 columns

The top tags of Action and Free to Play reveal the Multiplayer Shooter archetype of game uncovered in previous analysis.

One insight this plot has that previous ones do not is the intersection of popular tags in the Simulation and Strategy genres. Both have Simulation, Multiplayer, and Strategy as their top 3 tags, and City Builder and Sandbox tags lower down their lists. This suggests another popular archetype of popular game that aligns with well known titles such as The Sims and Roller Coaster Tycoon.

The heavy overlap between the popular tags of Utilities and Design & Illustration suggest that they share a lot of games. This makes sense, as games such as Blender, which is a multi-purpose animation and modeling app, fit into both genres.

There is also a lot of tag-sharing between the Massively Multiplayer and RPG genres. A lot of popular titles fit into both genres, such as Final Fantasy and EVE Online.

Answer

Each genre has vastly different standards for success depending on which metrics are being explored.

Successful Massively Multiplayer games tend to be Open World RPGs, and they tend to have far more average playtime and peak players than other games. On the other hand, they often struggle to garner more than 80% positive reviews.

Casual and Indie games do not tend to draw players in for a long time nor do they have large playerbases, but they have among the highest percentages of positive reviews in the industry. In addition, they have a more even spread of tags across their popular titles, suggesting that consumers looking for Casual and Indie games are less likely to look for a specific archetype of game.

The most popular Utilities and Design & Illustration apps focus on Animation, Modeling, and Photo/Video Development. They do not tend to have the eye-popping playercounts that other genres do, but can have dedicated fans who put more hours into their games than fans of other genres.

7) Recommendations to Stakeholders

a) Recommendations 1

For studios looking to develop the next hit game, I would focus on making an Action or Simulation game. Action games have dominated Steam’s player counts as far back as the data goes, with the most popular games on the platform consistently being Action-based PvP Shooters. The biggest games in the Action genre are household titles like Counter-Strike and PUBG, and any studio looking to capture some of the massive Action market should follow their lead. Simulation games have been on a recent tear, with games like Farming Simulator, EA FC, and Football Manager garnering hundreds of thousands of players. There seems to be a growth in the number of players who enjoy playing management-style games where they get to put themselves in a unique and enjoyable environment. Otherwise, my recommendation is to stay away from RPG and Simulation games. They have had stagnant playercounts and represent a shrinking share of Steam’s total playercount, so releasing a game in either of those genres may be more difficult.

The primary limitation of this analysis is that the success of a genre is driven by the success of its top few games. A few recent releases of huge budget Action games, like Counter-Strike or Elden Ring, may inflate overall player counts and make it look like Action as a whole is doing well without giving us a look at the typical Action game. There is no guarantee that releasing a similar game will capture any of the market that the most popular games have.

To overcome this, it may be more helpful to study the distribution of average players in each genre, and how it has shifted over time. This way, it is possible to see what percentage of a genre’s player base is made up of its top few games. A genre that is growing more “organically”, that is, not relying on a few popular games to make up its player count, may show increased median average monthly player counts across games in the genre. It may also be helpful to find a better system of identifying a game’s genres than Steam’s built in labels. Often, games with any aspect of a genre, even if it is a small part of the game, get labeled as being in that genre. This makes Steam’s genres not the most accurate way of determining the type and style of a game.

b) Recommendations 2

Developers who are looking to add features into their game that increases the chance of success should be trying to add as many Multiplayer aspects as possible. Multiplayer consistently shows up as the most correlated tag to peak playercount, the second most common tag in the top 1000 games, and makes up 3 of the 4 most popular tag pairs among popular games. It is clear that gamers want to be able to spend time with other players in game, and developers should lean into this as much as possible.

I also recommend that developers of Action Shooter games should be making their games Free-to-Play. The most popular shooters are overwhelmingly free, and trying to convince consumers to pay for a game in a market saturated by successful free games is a tough proposition. These games are able to make money through in-game sales after the player has already downloaded.

My last suggestion for developers is to incorporate more Open World maps into their games. Similarly to Multiplayer features, Open World appears as one of the most correlated to peak playercount, and is a very common tag amongst the most successful games on Steam.

The biggest constraint to using game tags to analyze successful features of games is that tags are often not specific enough to analyze the features of a game. It might be the case that Open World games are becoming more popular because of increased adoption of teleport mechanics to traverse the map, but there is no “teleportation” tag that would identify that. Similarly, multiplayer games come in many forms. Should players work together as a team or solo against each other? Should they face off against other players or bots? Figuring out which type of multiplayer game tends to be more successful is difficult going off of tag data alone.

Looking deeper into the tags category might help gain insight into Multiplayer games, as PvP and Teamwork are both tags on Steam. But in order to get answers about specific features like teleportation, nothing short of a deep dive into each successful game within a category of interest and noting their features will give insight into the attributes that make games popular.

c) Recommendations 3

Steam should carefully consider which games to feature on their home page by analyzing how that game is performing relative to its genre. If you only display games based on the number of players they have, the home page will be dominated by Free to Play and Action games, which ignores consumers who prefer niche genres. Break down home page displays by category, and use different metrics than player count to help steer players towards games that could become successful. For example, in the Massively Multiplayer category, which typically has low percentages of positive reviews, games with large shares of positive reviews should be featured. The low median percentage of positive reviews in the genre might imply players are unsatisfied with the current selection of games, and by recommending games with higher percent positive reviews could help grow the genre. Similarly, Casual and Indie games tend to have low average playtime, so by featuring games in those genres with high playtime, Steam may be able to capture more customers.

For developers, careful consideration should be given to the most popular tags of the genre they are developing a game in. Typical Utilities and Design & Illustration titles focus on Animation, Modeling, and Illustration. Trying to develop a Utilities app for Steam that does not fit into one of these subcategories might not be the best idea. If someone is interested in building a Strategy game, they should consider making it as a Sci-fi, City Builder, or Historical game as well, since those tags have done well and already have established player bases who may try their game out.

Some limitations of these conclusions are the lack of good success metrics in the Steam Games dataset. Peak Concurrent Players is a good way to measure a games’ popularity, but companies care about making money, not having a lot of consumers. While there is price data and rough estimates of the number of people who own each game, the range of potential owners each game could have is too big to make claims from. Additionally, games on Steam go on sale very often, so simply multiplying price by owners does not give anywhere near an accurate amount of money made by each game.

To better quantify success on Steam, better data is needed on game prices, discounts, and spending habits. If a game studio had access to this they would be able to find very specific trends in spending based on tag, genre, and time period, and would be able to optimally price and release their game to maximize success.

8) Conclusion

Across all analyses, Steam is shaped by a handful of dominant archetypes, such as action-based multiplayer and open world exploration, yet every genre operates under its own definition of success. Developers and Executives at gaming companies and Steam need to understand that the gaming industry is vast, and the strategies used for one particular kind of game are not universally applicable. Action and multiplayer shooters dominate raw playercounts, but that dominance is heavily skewed by a few blockbuster titles and masks strong performers in smaller categories. Massively Multiplayer games hinge on long playtime and huge populations but struggle with reviews, whereas Casual and Indie games attract smaller but highly satisfied audiences, and Utilities and creative tools maintain tiny market share but unusually dedicated users. These patterns, and many others present in the data above, show that playercounts alone cannot evaluate a game or guide platform recommendations. Steam should judge games relative to the norms of their genre, and developers should examine the specific mechanics behind popular tags rather than relying on the tags themselves, especially since financial success is difficult to infer from publicly available data.

Appendix

A helpful website for verifying and finding Steam data is SteamDB