Research:Where in the World are Mobile Editors

Contact

Oliver Keyes

---

Research:Projects

This page documents a completed research project.

Introduction

Screenshot of the Wikimedia Mobile Web gateway.

Mobile devices are an increasingly important way for internet users to consume web content. In 2012, mobile devices were responsible for over 10 percent of web requests; in 2013, this increased to 17.4%. Many organisations, including the Wikimedia Foundation, have been investing resources in capitalising on this and developing effective mobile websites and applications. In the Wikimedia Foundation's case this comes in the form of threee efforts - one, on the part of the Mobile Web team, to design an efficient website for direct browsing (m.[wikimedia site].org), a second to design an efficient set of mobile applications for Android and iOS operating systems, and the Wikipedia Zero programme which negotiates with mobile phone carriers to grant free (zero-rated) access to Wikipedia for their users. These groups produce products in heavy usage, and the mobile teams - specifically, Web and Apps - are seeking more data on how people are using their software.

One aspect of the mobile web that is particularly interesting is where it is being heavily utilised. Historically, both landline and mobile phone ownership and usage have been most prominent in the developed world, but mobile phone ownership is increasingly prominent in lower-middle-income nations such as China and Egypt, as well as poorer nations such as India or Pakistan. While ownership was initially concentrated in the affluent, middle class elements of society, that's still a tremendous number of people, and the class breadth of ownership is expected to increase rapidly in such nations. Moreover, it is expected to touch people who have not previously had connections to the internet. The result is a substantial broadening of the cultural and geographic base of readers and potential contributors to our sites.

This broadening of cultural bases has the potential to impact on systemic bias within our projects. Tracking it has the potential to inform us as to where we should most focus our efforts - what if the developing world makes better mobile editors, or more mobile editors? What if our geographic distribution doesn't actually reflect the trends the rest of the world is seeing? What if we're not able to transform mobile readers into mobile editors? All of these questions can lead to different directions for our mobile projects.

Research questions

The research questions are:

1. How are our mobile editors distributed, and how does this compare to the distribution of desktop editors?
2. How active, on a per-country basis, are our mobile editors compared to desktop?

Dataset

To answer these questions we used the recentchanges table in MediaWiki. It contains all actions within the last 30 days, including, crucially, the IP address of the person making them, allowing for geolocation.

The entire recentchanges table for each of our "active" (non-closed, non-deleted, non-private) wikis was extracted, and filtered to exclude actions by bots, non-edit actions and anonymous edits. The IP addresses associated with these edits were then geolocated using our geolocation software, and the resulting dataset filtered to contain, for each edit:

• The revisionID;
• The pageID;
• The userID;
• The country code that user geolocated to;
• Whether the edit was tagged as a mobile edit or not ('mobile' versus 'desktop'), and;
• What project that edit came from.

RQ1: How are our mobile editors distributed, and how does this compare to the distribution of desktop editors?

An important part of understanding mobile editing is understanding where mobile editors are coming from, and whether there are geographic areas that diverge from or have commonality with desktop editors.

To test this, we take each unique editor - mobile and desktop both - and look at the geographic distribution. First, mobile and desktop are examined independent of each other, and then compared, by looking at what percentage of editors in each country has edited from mobile devices. After removing countries without representation from both mobile and desktop, and countries with fewer than 20 unique editors of any type, we ended up with 238,580 distinct users from 132 nations. These were then run through rworldmap to associate the data with geographic regions.

RQ2:

Transparency

As part of their efforts to open up Wikipedia to mobile device users, the WMF mobile team enabled mobile wikitext editing, initially in alpha form in 2012, and then as a default feature in July 2013. With any new feature, it's important to understand how it's being used and who by, and one of the ways to do this is to look at where it's being used, both in comparison to alternatives and independently. Accordingly, the Mobile team has asked Research and Analytics to look at where mobile editing is being used, and how this compares to desktop editing.

Methods

The primary dataset for both mobile and desktop edits is the recentchanges table in MediaWiki, which contains details about both individual edits and the editors who make them, including:

1. The IP address of the editor;
2. The username, as a discrete identifier of that editor;
3. Whether the editor is a bot, and;
4. Whether the edit made was through our mobile or desktop editing interfaces.

We retrieved the recentchanges table for each 'active' wiki, defined as Wikimedia-hosted wikis that are not 'special' (the strategy wiki, for example), 'closed' (locked and no longer accessible for editing), 'deleted' (No longer accessible via the web) or 'private' (such as the wiki of, say, the Arbitration Committee on the English-language Wikipedia). It was limited to edits that were to existing articles, rather than the creation of new articles, and to users who were not identified as 'bots'.

After retrieving the username, IP address and mobile/desktop status of each edit between 28 January and 28 February 2014, and excluding IPv6 users, a random sample of 200,000 edits was taken, across all wikis. This was then geolocated down to the country-level, and the dataset limited to exclude users with non-geolocatable IPs, leaving 198,488 edits spread out over 191 nations. A minimum bar of 20 edits or editors (mobile + desktop) was then set for each country.

Results

Editors by country

To look at editors by country, we took each unique editor in the sample of edits, and aggregated them by country and editing interface type.

Looking solely at the raw number of mobile editors (Fig.1) we see high usage levels in the developed world, particularly North American and the United Kingdom/France, as well as Japan, but also Brazil and India. Comparing these numbers to desktop editors (Fig. 2) shows that mobile still makes up a very small chunk of our editing population - around 1% of the more prominently 'mobile' countries.

Looking at the variation between the two on a per-country basis, we generate Fig. 3. Values are between -1 and 1, where -1 would indicate total desktop dominance, 1 would indicate total mobile dominance, and 0 would indicate parity in editor numbers between the two countries. Again we see high mobile usage in North America, but also proportionately high usage in the entire Indian subcontinent (including Bangladesh), as well as the Middle East. This 'high usage', however, is still very low, with only 10.5% of editors in those proportionately high countries coming from Mobile.

• 
  Fig. 1: mobile editors, per country
• 
  Fig. 2: desktop editors, per country
• 
  Fig. 3: variation between mobile and desktop editors, per country
• 
  Fig. 4: variation between mobile and desktop editors, per country (normalised sample)

Edits by country

Edits by country simply takes the raw sample, and applies the sample filter (20 actions per country).

For mobile edits (Fig. 5) it shows the same outcome as with editors-by-country; mobile edits are most common in North America, chunks of Europe, and India, although still far less common than desktop edits (Fig. 6). Looking at the variation between mobile and desktop, we again see prominent usage in the Indian Subcontinent and the Middle East, followed by South and North America. Curiously the proportionately-high countries for mobile edits actually produce a far higher value than proportionately-high countries for mobile editors; while high-proportion countries provided a maximum of around 10.5% of editors, they can produce up to 26% of edits. Some of the proportionately-high-editor countries and regions, such as the United Kingdom, do not retain their position at the top of the table when variation in edits is factored in.

• 
  Fig. 5: mobile edits, per country
• 
  Fig. 6: desktop edits, per country
• 
  Fig. 7: variation between mobile and desktop edits, per country
• 
  Fig. 8: variation between mobile and desktop edits, per country (normalised sample)

Caveats

1. This was a sample of edits and editors from one particular slice of time - January/February 2014 - and as such is subject to some biases. In particular, we know that there is an element of seasonality to editor behaviour - the data retrieved from this sample may vary from the data retrieved from a sample in, say, August.
2. The data on editors may include a single editor multiple times - for example, a user who is active on multiple wikis could get caught. A paper by Scott A. Hale found that around 15% of active Wikipedia accounts are multilingual in nature (active over multiple projects with different language variations),

Implications

Assuming that the caveats mentioned above don't apply, it's possible to draw a couple of conclusions from this data.

The first is negative; Mobile makes up, despite its presence as an editing tool for over a year, a relatively small proportion of active editors and edits. In a numbers game, Mobile is far more effective overall as a way for readers to consume content than it is for editors to contribute to it.

The second, however, is positive. Despite the base numerical limitations on the number of contributions and the number of contributors, there are areas of the world where Mobile is punching far above its weight in terms of the edits produced by each user. Particularly in the Indian Subcontinent, a mobile editor is proportionately more productive than an equivalent desktop user.