Since its inception social media, and particularly Twitter, has become a data source of choice for computationally minded researchers into social systems. A cursory query of papers with “Twitter” or “microblog” in the title on Google Scholar returns 925 results from the first 6 months of 2015 alone. The allure of Twitter as a data source is clear; reasonable sample sizes are easily obtained and the data are rich, containing not only the notoriously restrictive 140 character content itself but also myriad meta-data such as time, an estimate of location and engagement (favourites and retweets).
More recently these data sources have shown great success in development and humanitarian fields, notably in the field of crisis mapping. However, there is still much that remains to be understood regarding how platforms such as Twitter are used and by whom (there’s a great review of this problem space by Zeynep Tufecki).
In low-income country contexts the question of biases due to differing levels of access is ever present. Since Twitter was conceived as a mobile platform (the 140 character limit echoing the length of a single SMS) one could argue that it is theoretically open to anyone with a phone of any kind. That includes the Motorola and Nokia handsets that many of my generation will cringingly recall, and that barrier to entry is becoming extremely low. In reality of course, the adoption of platforms such as these is more complex and relies heavily on preferential attachment; in other words, if you already have many users, it is easy to attract more. On the other hand, with few active users there is little to attract more users to join.
Of course Twitter is the first to admit that people use social media in different ways, often determined by how they access the platform; from the hyper-connected multi-device addicts to the occasional desktop user passively listening in a few times a month.
Since the device meta-data seems to have been somewhat neglected in development oriented studies, we decided to analyse the statistics of different device usage as part of a recent collaboration with Bill and Melinda Gates Foundation. Specifically, we considered the vaccine debate in India, Pakistan, Kenya and Nigeria throughout 2014 via Twitter, Facebook (accessed via DataSift) and mainstream media. The device data (twitter.source in DataSift terminology) is described in more detail here.
The benefits of such a meta-study are clear; firstly the type of device clearly signals the users’ economic well-being allowing us to understand if we are tuned in to the most vulnerable populations. Secondly the utility of microblogging in a sudden onset disaster is very different when all users rely on a low power mobile device compared to a website accessed via desktops. Finally, the device signals how the medium is used; messages that are created by widgets on news or blogging websites suggest more systematic syndication behaviour rather than spontaneous personal expression.
Firstly we group together the devices that are observed in our dataset into 4 categories determined ex-post; (i) Android devices (ii) automated/semi automated services (iii) Apple devices and (iv) feature phone devices.
Firstly we see that the generally lower cost Android handsets are more widely used than high end Apple devices in all countries which makes sense since none of these countries are high-income. Conversely, Kenyan users made significantly more use of lower end feature phones (i.e. sending tweets via SMS or rudimentary mobile web technologies) than any other country which also coincides with our intuition on the relative adoption of technologies in each of these countries. Finally, services which connected Twitter to the general internet ecosystem through widgets are keenly adopted in Nigeria more than any other country under study.
Finally, we present the top 15 sources used in each country to demonstrate the variety of ways in which Twitter is used.
Such a study begins to help us to better understand who is using social media and how. Some caveats inevitably apply; for one, we have considered the subset of users engaged in the vaccination debate who are likely not representative of the entire user base. Secondly, Twitter is just one platform of many possible platforms. In many cases, other alternative services are prefered as they have a larger community within the country, focus on topics that are of greater interest to the population of that country or maybe better support a non-Latin script.
Top image: A collection of mobile handset devices. (Wind.com.NY via Creative Commons)