.svg)
Exploring engagement, toxicity, and civic conversations across online and offline spaces in the Toronto by-election.
April 9th, 2024
The Samara Centre for Democracy monitored activity on Twitter* during the 2023 Toronto Mayoral By-election as part of our SAMbot project, a multi-year machine learning initiative that measures abusive content received by Canadian political candidates online. We found that online abuse was prevalent during this election period (May 13 to June 26, 2023).
We monitored a total of 124,730 tweets during the election period.
3,988 of these tweets were identified as abusive.
90% of abusive tweets targeted top 9 candidates
30 Twitter users accounted for 10% of abusive tweets
Our findings emphasize that online spaces and offline spaces are not separate worlds. What happens in the online realm is intertwined with the physical world, and a part of one shared civic dialogue. As we grapple with how technology is influencing our democracy, we must consider how the working conditions for candidates on the local digital campaign trail are shaping who runs for, and who wins, municipal office.
*The social media platform formerly known as “Twitter” has rebranded as “X” since we monitored this election. The platform was still called “Twitter” when we collected this data, so we refer to the platform as “Twitter” throughout this report.
**Candidate images are sourced from candidates’ by-election campaign materials.
John Tory was elected mayor of Toronto in the fall of 2022, in an election where only 29.7% of eligible voters cast their ballots. Tory resigned from the mayoral chair in February 2023 and a mayoral by-election was called for Toronto in the spring of 2023. This election saw an increase in turnout from the 2022 election — the by-election had a voter turnout of 38.5%.
The 2023 Toronto mayoral by-election remarkably had 102 candidates of diverse personal and professional backgrounds. Some had formerly held significant political positions, some were long-time local advocates, and others were largely unknown candidates who did not provide much public information about their candidacies.
For 45 days, from the end of the candidate nomination period to the end of election day (May 13 to June 26, 2023), we monitored 53 mayoral candidates on Twitter. The remaining candidates did not have public or active Twitter accounts as of the end of the election’s nomination period.
Olivia Chow is a former school trustee and Toronto city councillor. She served as the Member of Parliament (MP) for Trinity-Spadina from 2006 to 2014. She placed third in the 2014 Toronto mayoral race, and was ultimately elected mayor of Toronto in the 2023 mayoral by-election.
Ana Bailão served three terms on the Toronto city council from 2010-2022, two as a councillor and one as deputy mayor. She retired from council in 2022 and did not run in that year’s municipal election. She was endorsed by outgoing Mayor of Toronto John Tory during her 2023 mayoral campaign.
Mark Saunders formerly served as the chief of police for the Toronto Police Service. He ran as the Progressive Conservative candidate for Member of Provincial Parliament (MPP) in Don Valley West during the 2022 Ontario election. He was endorsed by Premier of Ontario Doug Ford during his 2023 mayoral campaign.
Anthony Furey is a former Toronto Sun columnist, talk radio host, and news media executive.
Josh Matlow is a Toronto city councillor for Ward 12, and has been on city council since 2010. While still a councillor, Matlow ran in the mayoral by-election. He remains a councillor as of publication.
Mitzie Hunter served as the MPP for Scarborough—Guildwood from 2013-2023 and Ontario’s minister of education from 2016-2018. She resigned from the Ontario legislature to participate in the 2023 Toronto mayoral by-election.
Chloe Brown is a policy analyst, activist, and former Toronto mayoral candidate. She placed third in the 2022 Toronto mayoral race.
Brad Bradford is a Toronto city councillor for Ward 19, and has served in that role since 2018. While still a councillor, Bradford ran in the mayoral by-election. He remains a councillor today.
Chris Saccoccia, commonly known as “Chris Sky,” is a Canadian property developer. He has been active in advocating against public health measures since the outbreak of COVID-19.
We have analyzed the online engagement and abuse received by mayoral candidates in order to share insights about where online abuse manifested and compare the online engagement different candidates received on Twitter during the election period. While we focus on particular candidates in the following pieces, their experiences are used as a lens to better understand the process and experience of running for municipal office in Canada.
The eventual mayor and a fringe candidate had similar engagement and levels of abuse — but very different experiences online.
Chloe Brown and Anthony Furey were among the most successful mayoral candidates, but as political outsiders, they faced barriers breaking into media opportunities.
Mayoral runner-up Ana Bailão proved that offline success didn’t necessarily require high online engagement.
Xiao Hua Gong spent a lot on advertising, and though those expenses didn’t translate into success at the polls, it showcased how big campaign budgets alone can reach into communities — sometimes in unexpected ways.
We monitored the 2023 Toronto mayoral by-election for 45 days, from May 13, 2023 at 00:00 ET to June 26, 2023 at 23:59 ET.
We tracked the Twitter mentions of 54 of 102 total mayoral candidates. Candidates that were not tracked did not possess a public or active Twitter account as of the end of the election’s nomination deadline.
Our SAMbot project has tracked abuse in federal, provincial, and municipal races since 2021. However, since data from each election is collected during different time periods, with different lengths, and with different totals of tracked candidates, it is not useful, nor advisable, to compare SAMbot data across elections.
We do not collect data on retweets, as counting the same tweet more than once can distort the analysis. We only evaluate text within a tweet; content such as images, audio, or videos that may spread abuse cannot be evaluated by the machine learning tools that we use.
In our SAMbot project, we use machine learning models to assess abusive language. These models are ever-evolving, which means that during each deployment, our data is more accurate and informed.
Please note that compared to our previous SAMbot deployments, we have evaluated abusive tweets significantly differently in this election.
Our machine learning model makes a confidence prediction to assess whether a tweet should be considered “abusive.” When measuring abuse, our model gives each tweet a score from 0% to 100% for each category, based on how confident it is that the tweet is abusive in nature.
Previously with SAMbot deployments, we used a 51% confidence prediction to evaluate abuse; we have changed to now use a 70% confidence prediction. This change means that our analysis will be more accurate, and that cumulative results will paint a better picture of how abuse is distributed across the entire election and across all candidates.
Simultaneously, this change also means that some nuanced and subtle forms of abusive language may be missed by our machine learning model, and will make it appear at first glance as if there is comparatively less abusive content present, which is not necessarily the case. Machine learning models will never be able to monitor all abusive language across an election, as the subjective nature of what constitutes “abuse” does not permit the possibility of 100% accuracy. This methodological change allows us to more accurately represent how abuse is distributed overall.
We have made this change as part of our ever-evolving intention to strive for more accurate and ethical methodological practices within the field of social media and machine learning research. Using confidence intervals in this way is in line with recommendations for social science research.
This change makes abuse volumes look considerably lower than in previous elections we have tracked (however, SAMbot data should never be compared across elections regardless). Please consider these significant methodological changes while interpreting the data in this report.
SAMbot deployments use machine learning tools — software applications that run automated tasks. Using machine learning allows us to analyze tweets at a massive scale. Through our SAMbot project, we can evaluate millions of tweets for how likely abusive they are. We track all English and French tweets sent to candidates. Each tweet tracked, whether a reply, quote tweet or mention, was analyzed against five abuse categories using a machine learning tool called Perspective API:
Perspective API provides us with a confidence prediction to assess whether a tweet meets an abuse category. When a tweet is evaluated, it’s given a score from 0% to 100% for each category, based on how certain the machine learning model is that the tweet meets that abuse category. If the tweet is assessed as >=70% likely to meet an abuse category, we determine that the tweet has met the criteria. If a tweet meets at least one of the five abuse categories at the >=70% confidence interval, it is counted as an abusive tweet. The abusive tweet category serves to aggregate all tweets that meet at least one abuse category.
The Samara Centre for Democracy, Engagement and Abuse on Toronto's Digital Campaign Trail: The 2023 Toronto Mayoral By-election Report (Toronto: The Samara Centre for Democracy, 2024), https://www.samaracentre.ca/engagement-and-abuse-on-torontos-digital-campaign-trail-the-2023-toronto-mayoral-by-election-report.
For access to the data included in this report, including Twitter engagement data, abuse data, candidate debate attendance, polling results, and more, please refer to the SAMbot 2023 Toronto Mayoral By-election Data Release.
If you have any other questions about our data or the SAMbot project, please reach out to us at hello@samaracentre.ca.
Exploring engagement, toxicity, and civic conversations across online and offline spaces in the Toronto by-election.
