When do people share? Comparing Instagram activity in six global cities

120,000 images from six global cities organized by average hue (distance to the center). The angle of each image is the day/time it was shared. All images use their local times (i.e. we keep offsets between the time zones). Because the temporal patterns for each city overlap, we see a uniform global image 24/7 cycle, without any separation between times of day. (This visualization and the post: Lev Manovich.)

In this post we compare patterns in Instagram activity between six cities: Bangkok, Berlin, Moscow, New York, Sao Paolo and Tokyo.

The analysis uses 120,000 images (20,000 from each city). To create this dataset, we first downloaded details of all geo-tagged Instagram images shared in the central same size area in each city during our full week (December 4-11, 2013; over 660,000 images in total). We then downloaded a random sample of 20,000 images from each city.

(This dataset was created as part of our Selficity project - see details below).

1. Numbers of Instagram images shared per hour in a 24 hour cycle

Berlin, Moscow, New York, and Sao Paolo have similar patterns: most images are shared between 1 and 11pm, with the peak around 7-8pm.

In Tokyo and Bangkok, there are two peaks: lunch time (1 or 2pm) and evening (7pm-11pm).

2. Numbers of Instagram images shared for every day of the week:

In most cities, people share most images on Saturday and Sunday. However, while in Berlin, Moscow and Tokyo and Bangkok, people appear to start their weekend already on Friday, in New York and Sao Paolo Friday is no different from previous weekdays.

(Because we are only using data for a single week, these patterns may not be typical. In particular, different Bangkok patterns maybe related to the political events in the city during that particular weeks.)

3. Number of Instagram users:

Our dataset contains twice as many users in New York than in Berlin -

4. Average number of images per user in each city:

- which means that while more people post on Instagram in NYC, on the average each user posts much fewer images (same as in Moscow and Sao Paolo)


1. Capture time versus share time.

Instagram allows users to post any image from their phones - i.e. users are not limiting to capture and immediately post images with Instagram app. Therefore, the volume of sharing does not directly tell when people take pictures, but rather than they use the app to share them.

2. Dataset details.

To create our data set, we used Gnip service to download Instagram data and images, so we were not constrained by Instagram API download limits. Both Instagram and Gnip provide only publicaly shared images. We were only downloading images with location data, which represents only a part of all shared images.

Selfiecity receives Golden Award in a data visualization competition

Our project selfiecity has received Golden Award from 2014 Information is Beautiful competition.

Selfiecity is a collaboration between the outstanding team of data visualization designers and programmers - Moritz Stefaner, Dominicus Baur and Daniel Goddemeyer - and five members of Software Studies Initiative. The collaboration was a great experience for us. Everybody worked hard. Moritz was the heart of the project designing data visualizations and the web site and making sure all pieces come together.

Amazingly, Moritz's another recent project OECD Regional Well-Being received the Silver Award in the same competition. Bravo, Moritz!

Our new animated Phototrails visualizations for Google Zeitgeist 2014 conference

Phototrails video 1 for Google Zeitgeist 2014 from Lev Manovich on Vimeo.

Phototrails video 2 for Google Zeitgeist 2014 from Lev Manovich on Vimeo.

This summer we received a commission to create new artworks to be shown during Google Zeitgeist 2014 conference. The conference is an invitation only two day event; this year it took place during September 14-16 in Paradise Valley, Arizona.

Google produced high quality video of many of the presentations. (You can also find videos of the talks from the earlier conferences at www.zeitgeistminds.com). For me personally, the highlights were the talks of Presidents Carter and Clinton, Google's own Eric Schmidt and Larry Page, and Lawrence Lessig - and also chatting with the people from Google X who were showing their amazing research.

We were asked to create animated versions of our Phototrails project. In the original project, we analyzed and visualized 2.3 million Instagram photos from 13 global cities. For the new Google Zeitgeist project, we created a number of new still visualizations using our our ImagePlot tool. We also used the animation option in ImagePlot to render a long sequence of visualization frames. The frames were rendered in 4K and then scaled to HD resolution. We used Premiere and After Effects to assemble the videos.

The two final videos which were exhibited at the conference are above. The fist video dissolves between both original and new Phototrails visualizations. The second is a slow zoom into the animated visualization of 120,000 Instagram photos from 6 cities. (Note: because of the Vimeo compression, the videos do not look as sharp as the originals).

The project was created by the original Phototrails team: Nadav Hochman, Jay Chow and Lev Manovich.

During the weeks leading to the event, we collaborated using Dropbox because each of us was in a different place: Nadav in NYC, Jay in California, and I was first in Brazil and then in Ireland. After we saw our videos playing at the site the morning of September 14th, we went back to the hotel, made some adjustments and rendered new versions. Good thing that ImagePlot (originally written by Manovich in 2010, and later expanded by Chow) kept rendering and never quit - even in Arizona's heat!