What is Cordoba Court?
Cordoba Court is an experimental social game that seeks to foster a sense of community among players by letting them share messages and small creations in a virtual space. The courtyard is modeled after the communal patios of Cordoba, Spain. Players are encouraged to cultivate it as they would a physical space, mirroring community greening and habitation practices there.
Players create a personalized totem in either a potted plant or ceramic plate, with millions of possible designs. While players can post anything they wish in the included message section of the totem, posts are not visible to others until they have been approved by moderators, ensuring a safe and respectful virtual environment for all. Overall, Cordoba Court offers a unique and engaging way for players to build a small community space online.
See additional survey information here.
Who visited?
Total Visits
1112 visits
Unique Users
865 unique users
Returning Users
113 visitors, 13% (who came back later)
A user voluntarily returning to the space is the most desirable condition as it reveals investment in the shared space. 13% of users is a higher baseline than expected, and future work will consider how to increase the amount of repeating visits.
An example of an intervention to increase returns would be special activities for second visits, like additional layers of decoration/posting, or the ability to tend the garden. Like a wiki, time spent invested in the project could enable more power and responibility over the site, much like a real shared patio space.
Dropouts
349 dropouts, 40% (consented but did not post)
Dropouts are an important performance metric in online platforms. Analytic measurements of disengagement becomes tricky for Cordoba Court, as we only collect posts and user-created signals, which departs from the surveillace/analytic infrastructures of typical social media platforms.
Future work might investigate the 'dropout' user group by coding types of dropouts (eg. technical failure, refusal to post, bounce out for irrelevance/disinterest) Subsequent design work could try to lower dropout rate for these groups. A 0% dropout rate is not necessarily ideal, as a holistic policy accounts for and accomodates disengagement where appropriate.
Total Posts
525 posts
Deleted/Overwitten Posts
11 deleted posts
When revisiting the site, users have the option to resubmit their post, overwriting the original. There was one case where a user posted a toxic message, then decided to replace it with a positive one. The number of rewrites is low given the population size, but being able to change messages or effectively delete them is important so participants have control.
Visualizing Dropouts, Posts, and Returns
This sankey diagram shows every gameplay session, where we group session events like post or did not post into buckets, to see how that session develops and where behaviors are distributed.
When did participants visit?
Visit Timeline for the Last 90 days
(dec 1,2022 -> mar 1, 2023)
What we see is a rush of engagement around the holiday season, due in part to the games release and promotion within a holiday pack of games. Posts taper off after January 1st.
How long did participants read before posting?
This lets us see the attention users spend on the existing scene before deciding to switch into a creative/productive mode. This small time period (generally between 2-3 minutes) is critical for our design, as the posts a user sees in the leadup to their post will influence their contribution. Eg. We expect posters to mirror the content they see before posting in terms of length, tone, topic. Beyond the concept of 'emotional contagion', the existing posts provide the social foundation on what a user will deem as an approrpriate contribution.
If the average user spends 50-120 seconds viewing the courtyard before posting, how might a design intervention help filter and highlight the posts which will best motivate the user to post a quality contribution?
Where did participants come from?
We do not track the location of users in this version of the app, and only have self reported data, which can be irrelevant or unreliable. This wordcloud thus represents the diversity of locations, which the size of the text being a proxy for freqency in the data. It's also worth noting there's a possibility to make a better visualization if we think about tributary accounting of location from specific to general eg. colorado -> USA -> North America so we may get a better account of demographics at a global scale. We could also account for things like typos and the use of different langauges, eg. Spain and España are different words but refer to the same country. With these limitations noted, the wordcloud is presented for what it's worth.
Neighborliness: Do posters cluster together or spread out?
We measured the distance between each newly placed object and its closest neighbor at the time of posting. We call this a neighborly proximity - or 'neighborliness'.
Posts that are 1-2 units away from an existing post are said to be neighborly.
We can think of general 'neighborliness' as a measurement of preference for either clustering, or spreading out as we co-occupy space.
For the first 200 placed objects, 82% were exactly 2 units away from an existing post.
This shows that the bulk of posters exhibit a neighborly tendancy, but avoid directly touching others, if possible.
However, as the space starts to fill, the choice to spread out disappers, which is why in both figures we see a quick decay of spread-out posting.
Open space is quickly exhausted.
From left to right we are seeing the neighborly proximities, in groups of 50. It should read like a timeline.
What's interesting here is that even in the beginning, when players had the most freedom to spread out, we still see a bias towards 2-unit distances.
Clustering seems like a general preference in co-occupied space.
If we were interested in further investigating this behavior, we might create a simulated posting environment that presents each tester with an identical option to post near or far. Would most people post in the same way with the same context?
Below is a table showing counts in sets of 100 posts.
| Distance (units) | Posts 1-100: | Posts 101-200: | Posts 201-300: | Posts 301-400: | Posts 401-500: |
|---|---|---|---|---|---|
| 1 | 26 | 52 | 84 | 89 | 92 |
| 2 | 42 | 42 | 12 | 10 | 8 |
| 3 | 14 | 6 | 2 | 1 | |
| 4 | 6 | ||||
| 5 | 6 | ||||
| 6 | 2 | ||||
| 7 | 3 |
What messages did people post?
LDA Topic Modelling
Latent Dirichlet Allocation (LDA) topic modelling shows words that are statistically associated with each other among the messages. The messages are likely too short and too few to get rich insights.
Topic 0:
fadedthanaho, stay, ho, love, life, make, dont, feliz, navidad, muchTopic 1:
hi, christmas, ive, holiday, well, hello, world, inside, play, ratsTopic 2:
please, ever, im, call, friends, courtyard, get, community, help, bushTopic 3:
happy, year, holidays, everyone, plant, new, next, wonderful, also, greatTopic 4:
love, one, best, nice, thank, game, man, cold, would, coolTopic 5:
merry, madvent, christmas, hope, happy, holidays, hey, good, see, goTopic 6:
time, may, like, dont, let, season, things, thing, good, timesCredits
-
Colter Wehmeier
Development, Direction, Art, Programming
-
Yiqi Xiao
Logo Concept, Additional Design
-
Kajetan Haas
Fractal Plate Shader and Model
-
Bill Derrah
Additional Sound
-
Cyreides
Plant and Fountain Models
-
Georgios Artopoulos
Additional Direction and Media Collection
-
Stan Ruecker
Additional Direction
-
Mohammed Rafat-Saleh
Architectural Patio Research, Additional 3D models
-
Dariah Udigish
Promotion, Application
-
Clowder Team, NCSA
Data management tool