One of the arenas in which Nigeria’s election is being contested is on social media. Thanks to the increasing penetration of smartphones, traditional barriers of access to broadband Internet have been lifted and an enthusiastic population is eager to have its views heard on social media regarding issues as diverse as fuel subsidy removal and the purchase of presidential jets. The election is no exception.
This year’s election cycle is possibly Nigeria’s most expensive so far with both parties being significantly well resourced despite ongoing economic woes. A trip down the streets of Lagos or Kaduna will reveal miles of campaign posters and billboards. Nigerian TV, radio and the Internet are suffused with campaign jingles, infomercials and documentaries.
Whereas traditional media is largely paid for, social media is a mix of raw citizen participation and pecuniary interests – a scenario that reflects the actual elections in Nigeria. Just as many people are paid to vote for one party or the other with cash or bags of rice, many are also paid to post on social media for their candidates. Nonetheless, many more on social media are not paid to post or vote. Therefore, analyzing posts and the sentiment they present could be a good proxy for polling data if the people posting are a fair representation of the voting public.
The analyses presented below attempts to do this using Twitter data. Tweets related to Nigeria’s elections have been pulled from Datasift for the period between December 1, 2014 through March 24, 2015, using the hashtags and handles represented in the table below. A total of 26,504 tweets were analyzed – 3,705 in December, 10,933 in January, 6,570 in February and 5,296 in March. This represents a random sample of 1 percent of the total 2.6 million tweets actually pulled by Datasift.
|Mohammadu Buhari (Challenger)||Goodluck Ebele Jonathan (Incumbent)|
|#Febuhari, #March4Buhari, #ThisisBuhari, @thisisbuhari, @ProfOsinbajo, @buhariosinbajo, #Buhari, #osinbajo, #GMB15, #IChooseGMB, “Yemi Osinbajo”, #Iamready, #iwillvoteapc|
#change (geo tagged), #ihavedecided
|#forwardnigeria, #gejvictory2015, #WeTriumphStill, #GEJWinsIt, #gejwins, @ourgej/#ourgej, @ArcNamadiSambo, #gej4nigeria, #GOODLUCKNIGERIA, #GEJVictory2015, #GEJ2015, #IChooseGEJ, “Arc Namadi Sambo”, #continuity, @presgoodluck, @presidentGEJ, @GEJ_Nigeria|
|“Bipartisan” Tags||#NigeriaDecides, “March 28 Nigeria”, “Election Nigeria”, “Nigeria Decides”|
Word Count: Who is More popular?
The analysis included a raw count of words that most frequently occur in all the tweets posted in the entire period reviewed, which is graphically represented in the word cloud below. This reveals that Buhari’s handle @thisisbuhari has the most absolute mentions over the period and GEJ’s #WeTriumphStill is a close second. It also explores the evolution of words and tags on a monthly basis.
As figure 2 shows, GEJ clearly gains momentum in number of absolute mentions in February and March. The All Progress Congress vice-presidential candidate Yemi Osinbajo seems to have been very popular in December after his announcement as running mate but plateaus over time. Words and phrases like @inecnigeria, postponement, military, Boko Haram are popular in February but fade out in other months. The ascendancy of #wetriumphstill, which was used by the GEJ team to celebrate gains against Boko Haram as well as the GEJ campaign, shows that GEJ’s increasing popularity may be linked to these perceived gains against the insurgency.
Sentiment Analysis: Who has the Most Support?
In addition, an analysis was conducted using standard natural language processing tools to determine the sentiment of tweets. The analysis measures the emotional tone of the tweet to analyze whether it expresses a sentiment. Next, it scores the sentiment on a scale of +1 to -1, where +1 represents a very positive sentiment and -1 a negative sentiment. It identifies the structure of the tweet to identify sentiment-bearing phrases or clauses, usually adjective-noun expressions such as “evil dictator” – the software is trained to know by association that when humans use the word “evil” before a noun, it usually is not a positive sentiment and “dictator” is more likely to refer to Buhari. It also identifies named entities in the tweet and is able to link that to the sentiment. For example, a tweet says “RT @daily_trust: Osun monarchs reject Jonathan endorsement by Ooni #NigeriaDecides”. The software picks the word “reject “and assigns a negative score to the tweet for Jonathan. The reader can learn more here.
The results of the sentiment analysis for all the tweets are visualized in the chart below.
The chart shows Buhari leading Jonathan by a factor of two to one in December with his lead peaking in February. The February dip for Jonathan may have been due to popular disapproval of the election postponement. However, pro-Jonathan sentiments seem to have picked up in March, possibly due to gains being made against Boko Haram, dwindling opposition resources or more aggressive campaigning by his team on and off social media.
Hashtag Count: Who Has the Most Focused Twitter Operation?
The raw count of hashtags over time shows GEJ consistently ahead of Buhari. As of March, GEJ had five of the top 10 most popular hashtags to Buhari’s three. This is probably indicative a more focused Twitter presence, a sign of organized participation as opposed to Buhari’s largely organic citizen-led social media movement. The timing of hashtag appearances also tells a story with #ihavedecided, a movement of Buhari supporters becoming popular in January right after its launching and #gejwinsit showing up in March.
Representativeness: Who is Tweeting and How do They Compare to the Population?
Twitter presence in Nigeria is at best a skewed sample of the population, being more linked to the elite than the general public. This is supported by the data, which shows that the highest percentage of tweets are linked to individuals who are reported as senior managers, writers, consultants, engineers, entrepreneurs and scientists – all elite professions in Nigeria. This shows that the opinions revealed may indeed only represent a section of the electorate. It may also reflect the aspirational nature of Nigerians in that the titular identifications used on Twitter may not necessarily reflect actual jobs or status of the tweeter.
A more accurate representation of Nigeria’s larger social media population might be mined by analyzing posts in more grassroots websites such as Naira Land, Eskimi, Linda Ikeji’s blog or Facebook. Future analysis could include data from these platforms.
It is possible given the nature of the social structure in Nigeria that the opinions of the elite will cascade down the economic ladder, but this is at best conjecture. At any rate, even if we were to disagree with the filter-down theory, it is at least interesting to know the outcome of the political campaigning conducted via Twitter for all enthusiasm and vitriol generated on the platform this election season.
Tobi Oluwatola is a doctoral fellow at the Pardee RAND Graduate School. He writes from Los Angeles. He can be reached on twitter @tobioluwatola and email firstname.lastname@example.org. The author thanks Andrew Cady and Chris Skeels for their data support and Angela O’Mahony and Osoba Osonde for their helpful comments on an earlier draft. Mistakes are mine.