Brazilian Twitter’s Response to Dilma’s Removal From Office

Brazilian Twitter has been busy in the last month discussing the impeachment process against Dilma, and when the Senate voted to remove her from office on May 12th, hashtags like #ByeDemocracyDay and #TchauQueridaDay were trending on Twitter both within Brazil and worldwide.

I was interested in how social media in the country would react, so I collected tweets which were sent with the hashtag #TchauQueridaDay during and after the Senate vote. I amassed more than 60,000 in total, and from this data set I was able to extract a state-specific location for approximately 20,000 tweets, using geotagging or the user’s self-reported location.

Once the data was aggregated by state, I normalized it by comparing the number of #TchauQueridaDay tweets to the total number of tweets from each state during a similar time period. This location quotient, inspired by economic base analysis, gives the state’s percentage of #TchauQueridaDay tweets relative to its total number of tweets.

A score of 1.000 means that there are relatively the same number of #TchauQueridaDay tweets as tweets in general. A number larger than 1.000 means that #TchauQueridaDay tweets are more prevalent, and that the state’s Twitter users are more pro-impeachment/anti-Dilma than the norm. Pro-impeachment states are represented by darker colors in the map below, and are led by Mato Grosso and Acre.

statesgraytable

Before continuing, there are a couple of important limitations to this data set that need to be mentioned. The first is that the location quotient really only measures pro-impeachment/anti-Dilma sentiment. In other words, we can’t infer support for Dilma simply because there was a lack of negative tweets from a specific state. Also, a state’s Twitter users are a very specific subset of the population, and their views may not be the same as the general electorate. That being said, there are still a few noteworthy results, especially when comparing the location quotients to the 2014 election, when Dilma ran against PSDB candidate Aécio Neves.

Both Mato Grosso and Acre were among the twelve states that had an Aécio majority in 2014, and of those twelve, seven states had a location quotient above 1.000 in my study. Notable amongst those states that had an Aécio majority in 2014 is Matto Grosso do Sul, which had a location quotient of just 0.659, the third lowest.

Other interesting results include Paraíba and Rio Grande do Norte, which went solidly to Dilma in 2014, but posted location quotients of 1.453 and 1.269 respectively. Bahia, traditionally a PT stronghold, and a state that gave Dilma 70% of their vote in 2014, had a location quotient of 1.031. As one would expect, Dilma’s home state of Rio Grande do Sul had a location quotient of only 0.486, which was second lowest. The full list of states and their location quotients can be found at the end of this article.

An important final point to make is that much of the data relies on the user’s self-reported location from their Twitter profile, which may not necessarily reflect the state that the tweet was sent from, or may include a non-existent location (e.g. “Land of Nothing”). However, if the state or city was not recognized as being an official municipality, the data point was discarded. Because of this nearly 40,000 tweets were not used, which resulted in states with low total numbers both in the data set and the sample. These states happened to have relatively low populations to begin with (such as Roraima, with a population of 500,000), but it does open the door to the possibility of one very active person skewing the results.

For the actual impeachment vote I plan to track both pro and anti-impeachment hashtags simultaneously in order to paint a more complete picture, and this first analysis should give a good starting point and offer a basis for comparison. Stay tuned for more.

 

State Location Quotient
MT 3.268
AC 1.792
AP 1.600
PB 1.453
RN 1.269
ES 1.242
DF 1.224
SP 1.211
SC 1.156
GO 1.108
BA 1.031
SE 1.028
MG 0.990
AL 0.961
PR 0.957
RO 0.953
RJ 0.881
CE 0.842
PE 0.841
AM 0.829
TO 0.821
MA 0.791
RR 0.778
RS 0.705
MS 0.659
PA 0.486
PI 0.337