Twitter, Public Opinion, and the 2011 Nigerian Presidential Election

This paper analyzes a corpus of Nigerian Tweets collected during the run-up to the 2011 Nigerian Presidential election and compares it with official election returns and polling data. We found that counts of the mentions on Twitter of the two major candidates correlated strongly with polling and election results when compared across the country's geopolitical regions, though the same data overrepresented two other candidates who did not fare as well in the polls or at the ballot box. Sentiment extracted from Twitter was less accurate in capturing mean levels of support for the two major candidates and, in particular, showed a strong negativity bias against the incumbent president. Twitter sentiment did mirror regional trends in the polling and election results though not as strongly as seen for mention counts. We demonstrate methodologically how to sample these data, extract sentiment, and compare this sentiment with ground-truth polling data and election results. Although social media clearly capture opinion about contentious issues, our results suggest that the opinion represented there may not always accurately reflect true public opinion.