South Africa Service Delivery Protest Tracker

Sign up to learn more about our project

We aim to create an easy-to-understand source of

real-time data on service delivery protests that will

be useful to policy makers, think tanks, journalists,

academics – as well as the general public.

This project develops and refines a methodology to capture information that addresses the question: “What is the extent and nature of township protests in South Africa?” The project also seeks to evaluate the most appropriate method for consolidating information on protests from different media sources while considering reliability of sources, relevance of data, and external validity of data.

The project organized the gathered data and sources into a database beginning on February 17, which enabled us to visualize the data using a variety of platforms. It is also important to note that we updated our list of sources to include a more local and geographically disperse set of newspapers on Friday, April 18. While there was no noticeable increase in articles that mentioned service delivery protest after that date, it is important to consider this as a possible confounding variable. A full list of sources used in this research is available here.

Maps

The following maps were produced with CartoDB through both a manual coding of data and an automated coding of articles using AlchemyAPI. The maps attempt to show the variance in geographic distribution where articles have mentioned service delivery protests, as well as the how much a given location has had articles mentioning service delivery protests. The geographic distribution of mentions of service delivery protests for both methodologies report similar results, with a bulk of the mentions in urban areas near Johannesburg, Pretoria and Cape Town, and fewer articles in the surrounding less urban areas.

It is noteworthy to mention however, that the automated method reports more data points, appearing to cover a wider distribution of geographic areas. The reason for this discrepancy lies in the method of how AlchemyAPI returns key words. Any mention of a location (a city or township), is captured regardless of whether the article focuses primarily on the location or not. In contrast, in the manual methodology, the coders only coded the most relevant locations based on where incidents were happening, or where the articles’ subject matter was focused. Also, manual coders may not have been equipped to identify some locations.

Both methodologies report similar trends and proportions for the presence of violence mentioned in the articles. The most striking difference again, lies in the appearance of that the automated methodology captures a systematically greater number of geographic regions that have articles with mentions of the term “violence.”

The wider variance in geographic regions reinforces the fact that AlchemyAPI is more sensitive than manual coders and captures any mention of any location, whereas manual coders only report locations that are a primary focus in the article. The data points on the map produced by the automated methodology are thus often duplicates of an article. Another reason that a greater number of articles mentioning violence seem to be reported in the automated methodology is that AlchemyAPI relies on any query of “violence.” In contrast, the manual methodology actually verified whether the incidents or articles focus on violence, rather than recording the mere mention that the automated procedure immediately captures.

In analyzing the distribution of data points from the automated map, it is also important to keep in mind the way CartoDB maps data points. Many of the dots or data points are stacked on top of each other, thus a dark dot may actually be underneath a lighter dot or vice versa. Again, in this instance, anyone seeking to analyze the data from the map must verify exactly how many of the data points are present.

Due to such discrepancies, one cannot draw immediate assumptions that an article actually mentioned violence in one of the geographic points mapped from the automated methodology. The manual coding methodology would be much more reliable in this regard. In either situation, human verification is essential; it is crucial that any interpreter of the data could access the source or the article connected to each data point and draw her own conclusions on whether a reported location has mentioned “violent” or not. The automated map is misleading because violence tends to be mentioned more often in areas outside of major urban areas – however, the same conclusion could not be drawn by looking at map produced from manual coding.

In analyzing the distribution of data points from the automated map, it is also important to keep in mind the way CartoDB maps data points. Many of the dots or data points, are stacked on top of each other, thus a dark dot may actually be underneath a lighter dot or vice versa. Again, in this instance, anyone seeking to analyze the data from the map must verify exactly how many of the data points are present.

Charts

The following graphs visualize the number of articles mentioning service delivery protests over time and also graph the proportion of articles that mention violence.

Manual Coding: Articles Mentioning Service Delivery Protest Over Time

Learn About Tableau

Automated Coding: Articles Mentioning Service Delivery Protest Over Time

Learn About Tableau

From a brief look at both graphs, there does not appear to be a significant difference in the proportion of articles where violence was reported. The correlation of trends in the two graphs however, appears to be weaker from the dates March 23 and onward. The reasons for this may be due to the difference in definitions or levels of sensitivity of detecting “violence” between the two methodologies. As mentioned before, the automated methodology marks “violence” as present in an article if the keyword itself is returned. However, the definition of violence for a manual coder depends on entirely on subjective human interpretation.

The following bar charts graph the articles that mention service delivery protests according to the type of grievance that the protest focused on. An important point to note here is that many of the articles encompassed more than one type of grievance. Also, the manual methodology only marked a grievance if the kind of grievance (housing, electricity, employment etc.) pertained to the reasons why people were protesting. But the automated methodology may have coded the type of grievance as long as it was merely mentioned in the article as a keyword, regardless of whether the mention was related to protests.

Manual Coding: Articles Mentioning Service Delivery Protest by Category

Learn About Tableau

Automated Coding: Articles Mentioning Service Delivery Protest by Category

Learn About Tableau

From looking at both the bar charts, the top 2-4 categories of protests are reported to be different when comparing the two methodologies. The manual methodology has determined that the top grievances in articles that mention protests were “land and housing”, “electricity” and “corruption”. The automated methodology, on the other hand, reports that the top categories were “sanitation”, “elections” and “corruption”. However, none of the differences in the proportion of articles that are mentioned each category (across both methodologies) are statistically significant (according to a calculation of a standard Z-score of 1.96). This means that the difference in the proportions of categories between the two methodologies do not differ drastically beyond what would have happened if the methodologies were very similar.

Manual Coding: Articles Mentioning Service Delivery Protest by Type of Mention

Learn About Tableau

One key dimension that the manual team coded for, that the automation team did not, was the nature of how service delivery protests was discussed for each article. We placed each article in one of the following categories: 1) the article broadly focuses on service delivery protests as a topic of discussion, 2) the article reports a single service delivery protest incident, 3) the article focuses on another topic besides service delivery protests but mentions “service delivery protest” at least once.

According to the manual coding methodology and an analysis of the breakdown of articles, approximately one-third of articles sourced focus on reporting a service delivery protest as an incident, one-third focus on service delivery protests as a broader topic while the remaining third cover another topic with a brief mention of service delivery protests.

Manual Coding: Articles Mentioning Service Delivery Protest Over Time (Police vs. No Police)

Learn About Tableau

Automated Coding: Articles Mentioning Service Delivery Protest Over Time (Police vs. No Police)

Learn About Tableau

Sources

The project monitored the following sources for articles that mention service delivery protest:

  1. Beeld Suid-Afrika "http://www.beeld.com/"
  2. Meander Chronicle "http://meanderchronicle.co.za"
  3. Rekord East "http://rekordeast.co.za"
  4. Die Kwêvoël "http://www.kwevoel.co.za"
  5. AllAfrica News: South Africa "http://allafrica.com/southafrica/"
  6. Business Day Live "http://www.bdlive.co.za"
  7. St Francis Chronicle "http://stfrancischronicle.com"
  8. Somerset Budget & Pearston Advocate... "http://www.somersetbudget.co.za"
  9. News24 South Africa "http://www.news24.com/"
  10. Mail & Guardian News Feed - National "http://mg.co.za/rss/national"
  11. Latest from the Institute "http://www.sairr.org.za"
  12. DIE HOORNDIE HOORN "http://www.diehoorn.co.za"
  13. 48hours.co.za "http://48hours.co.za"
  14. CityPress South Africa "http://www.citypress.co.za/"
  15. EMM Alert: SouthAfrica "http://emm.newsbrief.eu/NewsBrief/alertedition/all/SouthAfrica.html"
  16. The Mercury - IOL "http://www.iol.co.za:80/cmlink/1.1538217"
  17. Politicsweb -  Daily news headlines "http://www.politicsweb.co.za"
  18. Cape Business News "http://www.cbn.co.za/"
  19. kathugazette.co.za "http://www.kathugazette.co.za"
  20. Tabloid Newspapers "http://tabloidmedia.co.za"
  21. Google Alerts - South Africa "http://www.google.com/alerts/feeds/18374825296320795236/11309043107303853878"
  22. NGO Pulse "http://www.ngopulse.org/not_used/383"
  23. TimesLIVE "http://www.timeslive.co.za"
  24. Al-Qalam "http://alqalam.co.za"
  25. Sowetan LIVE "http://www.sowetanlive.co.za"
  26. Daily Maverick "http://www.dailymaverick.co.za/"
  27. Kormorant "http://www.kormorant.co.za"
  28. Ons Kontrei "http://www.onskontrei.co.za"
  29. Community "http://www.grocotts.co.za/taxonomy/term/58/all"
  30. Isolezwe/The Star - News South Africa Extended "http://www.iol.co.za/news-south-africa-extended-1.679178"
  31. TheStar News RSS Feed "http://www.iol.co.za:80/cmlink/1.1073924"
  32. sportseye.co.za/web "http://sportseye.co.za/web"
  33. DispatchLIVE "http://www.dispatchlive.co.za"
  34. THISABILITY - The Power Of The Soul "http://www.thisability.co.za"
  35. News Mpumalanga Category RSS "http://www.iol.co.za:80/cmlink/1.661"
  36. HeraldLIVE "http://www.heraldlive.co.za"
  37. Rapport Suid-Afrika "http://www.rapport.co.za/"
  38. Zeerust News "http://www.zeerustnews.co.za"
  39. ZithetheleZithethele "http://www.zithethele.com"
  40. IRIN | Africa | Southern Africa | South Africa "http://www.IRINnews.org"
  41. Township Times "http://www.townshiptimes.co.za"
  42. www.Witness.co.za Feed "http://www.witness.co.za"
  43. Mail & Guardian News Feed "http://mg.co.za/rss/"
  44. Local News "http://www.mnkystudio.com/news"

About

About the Service Delivery Tracker

The South Africa Service Delivery Protest Tracker aims to aggregate news articles and other online media sources and visualize service delivery protests both geographically and over time. Building upon the methodology the Council on Foreign Relations used to build the Nigeria Security Tracker, we aim to develop an online platform that automates the process of tracking and aggregating information on service delivery protests (including geographic location and nature of grievance) pulled from media reports, and other news sources. The hope is that this tool will be source of real-time data on protests that will be useful for policy makers, think tanks, journalists, academics as well as the general public.


Meet the team

Le Chen

methodology researcher

Janice Dean

methodology researcher

Jesper Frant

project manager

Rachana Kumar

technology lead

Dean Zambrano

technology advisor

Anne Nelson

faculty advisor

Want to know more? | a little about us

We are Master of Public Administration students at Columbia University's School of International Public Affairs, working with Ambassador John Campbell, Africa Fellow at the Council on Foreign Relation, to conduct a media monitoring project to determine the scale and nature of service delivery protests in South Africa.


Contact