Greetings, fellow game developers

This morning, as everyone else, I opened Ludum Dare website to see jam results and read blogposts regarding them. Some of these blogposts got me thinking about collecting more stats about submitted game and 5 minutes later I am making that hacky web scraping script.

Collected dataset is included at the end of this post and I invite you to look into data yourself. Please take everything written here with a grain of salt since I am nowhere near to the actual statistician.

Initially I wanted to collect info about ratings amount and grade correlation, target platforms and genres. In order to do so we have to parse grades, game links and somehow to extract the genre of the game. Since Ludum Dare doesn't allow to provide genre in game description we would have to get it from somewhere else. Fortunately, itch.io is recognized as one of the best platform for game jams and it's used by a lot of games for file hosting, and it allows to specify genre!

Dataset

So, after a bit of coding and some DOS attack-like activity in my network, I managed to collect what I needed. As Ludum Dare stats say, there were totally 3206 games submitted, out of which 2406 are in jam and 800 in compo. A minimum amount of received rating required in order to get graded is 20 and 2298 games managed to get over that threshold. Additionally, 1863 of these games provide itch.io link and 936 of them specify their genre. All statistics here based only on games which got past 20 ratings threshold.

Target platforms

In order to determine target platform I analysed both selected link tag on ludum dare page and itch.io pages (when available). And turns out that there are no surprizes here, most of the developers aim for web and windows support, and mobile games are highly underrepresented. I also expected that top-100/top-500 will be mostly represented by games playable in browser with lower proportion of other platforms, but turns out that it's not really the case. I think it's safe to say that desktop games don't really lose any points without browser support.

``` Supported platform across all entries: WEB 1825 (79.42%) Windows 1710 (74.41%) MacOS 524 (22.80%) Linux 519 (22.58%) Android 47 (2.05%) IOS 3 (0.13%)

Supported platform in TOP-50: Windows 48 (96.00%) WEB 32 (64.00%) Linux 19 (38.00%) MacOS 17 (34.00%) Android 3 (6.00%) IOS 1 (2.00%)

Supported platform in TOP-100: Windows 89 (89.00%) WEB 78 (78.00%) Linux 30 (30.00%) MacOS 29 (29.00%) Android 5 (5.00%) IOS 1 (1.00%)

Supported platform in TOP-500: WEB 428 (85.60%) Windows 392 (78.40%) MacOS 136 (27.20%) Linux 129 (25.80%) Android 12 (2.40%) IOS 1 (0.20%) ```

Genre Distribution

Across 936 entries for which we have genre data results are the following:

``` Across all entries: Puzzle 269 (28.74%) Action 212 (22.65%) Platformer 176 (18.80%) Adventure 135 (14.42%) Shooter 73 (7.80%) Simulation 55 (5.88%) Racing 40 (4.27%) Strategy 32 (3.42%) Survival 31 (3.31%) Rhythm 28 (2.99%) Interactive Fiction 26 (2.78%) Role Playing 25 (2.67%) Visual Novel 22 (2.35%) Card Game 15 (1.60%) Fighting 10 (1.07%) Educational 4 (0.43%) Sports 1 (0.11%)

In TOP-50: Puzzle 12 (32.43%) Action 9 (24.32%) Platformer 8 (21.62%) Adventure 5 (13.51%) Shooter 4 (10.81%) Simulation 4 (10.81%) Card Game 2 (5.41%) Rhythm 2 (5.41%) Strategy 2 (5.41%) Fighting 1 (2.70%) Survival 1 (2.70%) Role Playing 1 (2.70%) Interactive Fiction 1 (2.70%) Racing 1 (2.70%) ```

Puzzles and action games are the most popular choice for developers, almost half the games mention that tag. It's also nice to see that overall and top-50 distrubution is almost the same, which means that you're safe to choose any genre you want for your game and it wont affect your grades.

Rating Distribution

Now lets see what is the distribution of rating for games overall and in most popular genres. This stat is based on @AdroitConceptions suggestion.

``` Rating distribution overall In category overall: 3.44 ± 0.45 In category fun: 3.27 ± 0.53 In category innovation: 3.27 ± 0.54 In category theme: 3.51 ± 0.57 In category graphics: 3.53 ± 0.59 In category audio: 3.33 ± 0.59 In category humor: 2.92 ± 0.67 In category mood: 3.36 ± 0.52

Rating distribution for genre Action In category overall: 3.52 ± 0.45 In category fun: 3.44 ± 0.51 In category innovation: 3.27 ± 0.56 In category theme: 3.42 ± 0.60 In category graphics: 3.66 ± 0.55 In category audio: 3.42 ± 0.58 In category humor: 2.88 ± 0.63 In category mood: 3.32 ± 0.50

Rating distribution for genre Adventure In category overall: 3.40 ± 0.48 In category fun: 3.14 ± 0.56 In category innovation: 3.15 ± 0.56 In category theme: 3.40 ± 0.63 In category graphics: 3.61 ± 0.61 In category audio: 3.35 ± 0.65 In category humor: 2.97 ± 0.68 In category mood: 3.50 ± 0.54

Rating distribution for genre Puzzle In category overall: 3.55 ± 0.48 In category fun: 3.35 ± 0.55 In category innovation: 3.49 ± 0.53 In category theme: 3.67 ± 0.55 In category graphics: 3.60 ± 0.57 In category audio: 3.30 ± 0.58 In category humor: 2.83 ± 0.59 In category mood: 3.39 ± 0.51

Rating distribution for genre Platformer In category overall: 3.50 ± 0.47 In category fun: 3.38 ± 0.53 In category innovation: 3.26 ± 0.62 In category theme: 3.50 ± 0.63 In category graphics: 3.56 ± 0.58 In category audio: 3.29 ± 0.56 In category humor: 2.91 ± 0.60 In category mood: 3.34 ± 0.51

```

We can see that all categories generally rated in the same way with the exception for humor, which is usually lower than other categories. It's also interesting that Action games tend to have the higher fun grade.

Received ratings and Grade correlation

One of the common topics discussed in posts regarding results was about how high amount of ratings leads to lower average grade. For some people it may sound logical and I wanted to check if such statement has any truth behind it it. For measuring correlation we'll be using something called Pearson's correlation coefficient, which probably fimiliar to anyone who came across statistics course at some point.

Turns out that correlation coefficient between amount of ratings received and grade in overall top is 0.22233

For other categories results are the following:

text Ratings received/Grade correlation across all entries Ratings received/Grade correlation for overall: 0.22233126091016753 Ratings received/Grade correlation for fun: 0.19439265109626402 Ratings received/Grade correlation for innovation: 0.13478324542490097 Ratings received/Grade correlation for theme: 0.13059698797880057 Ratings received/Grade correlation for graphics: 0.2075506952357996 Ratings received/Grade correlation for audio: 0.2045200770843993 Ratings received/Grade correlation for humor: 0.2031874502415293 Ratings received/Grade correlation for mood: 0.19442989187360174

Correlation coefficient between 0 and 0.3 generally regarded as a very weak correlation - so that means that while higher amount of ratings can lead to higher grades such influence is not really significant. Besides, I think now we can say that our initial hypothesis is incorrect.

Across games which managed to get into top 100 correlation coefficient looks like that: text Ratings received/Grade correlation for overall: 0.28650349819411464 Ratings received/Grade correlation for fun: 0.13743718059002868 Ratings received/Grade correlation for innovation: 0.12956207793693833 Ratings received/Grade correlation for theme: -0.003790955798309322 Ratings received/Grade correlation for graphics: -0.047956136459542414 Ratings received/Grade correlation for audio: -0.053283386265215796 Ratings received/Grade correlation for humor: 0.08796826193357084 Ratings received/Grade correlation for mood: 0.032409899490679184

So, this is actually really intersting. First of all there is noticeable difference in correlation for games in top-100 and across all entries. While coefficient for all games is somewhere around 0.2 almost everywhere, in top-100 it drops almost to zero for all categories except innovation, fun and overall (in next section we'll see more about fun and overall correlation). We can conclude that low amount of ratings doesn't prevent your game from successfully reaching top-100. I don't feel confident to state any other meaning of such result but it's certainly worth remembering.

The section Received ratings and Grade correlation in the initial version of this post looked totally different and was calculated absolutely wrong. The reason for that is because I actually did a correlation between amount of received ratings and position in category top. And that was simply incorrect for two reason: first of all - we're calculating correlation both in jam and compo, and compo has significantly less entries which yields different positions. The second reason is caused by the fact that position is actually decreasing while average grade rises (1st position is better than 2nd position, but grade of 5 is better than grade of 4) - which results in negative correlation and can be confusing for some readers. I apologize for such misinformation.

Correlation between different categories

Another interesting insight can be spotted if we take a look at how different categories correlate with each other. It can show which categories are rated in similiar way and vice verse.

Since there are a lot of categories full correlation output is a bit long and can be seen on pastebin. A sneak peak from full output: ```text Overall/Fun Grade correlation: 0.8935648248913685 Overall/Innovation Grade correlation: 0.7239394376255062 Overall/Theme Grade correlation: 0.5898030292263634 Overall/Graphics Grade correlation: 0.704619779209486 Overall/Audio Grade correlation: 0.6761667301174727 Overall/Humor Grade correlation: 0.4370432044059216 Overall/Mood Grade correlation: 0.7307469349058053

Highest grade correlation between overall and fun: 0.8935648248913685 Lowest grade correlation between theme and humor: 0.26277385324255415 ```

So, what have we got. First of all, we can see that correlation between fun and overall rating is pretty close to 1. That means that people usually tend to give high overall rating for fun games and vice verse. Also, as expected, mood has pretty high correlation with graphics and audio.

Besides, humor seems to have lowest correlation with almost each of the categories. I am not going to draw any conclusion from that fact but it's quite a surprising result.

How can you look into dataset yourself

Collected data used for this research available on Google Spreadsheets: https://docs.google.com/spreadsheets/d/1nkWSSBI9JDTxYfKzh3ZX8XPEOHNCivJVpXGMlmJ8XbU/edit?usp=sharing

I would love to share with you source code of data scraping script but I don't want to cause unintented ddos on ldjam.com and itch.io so I am not going to post it here. If you really want the source code feel free to DM me on twitter

To sum up

Whoah, that was a fun morning. I believe these stats can be interesting to fellow game jammers because they were definitely interesting for me. Overall, that was an awesome Ludum Dare. If you managed to finish your game you should be damn proud of yourself! See all of you in April!

P.S. If this post was interesting and you value time put into that little research please share link to this page with your friends - I'll appreciate it.

P.P.S. In case you feel like some more interesting stats can be derived please write about it in the comments and I'll do my best to include it in the post.