Written evidence from the University of Sheffield (DFF0025)

We first started research on online abuse in tweets sent to UK politicians in the one month periods in the run-up to the 2015 and 2017 UK general elections [Gorrell et al, 2018]. Most recently, we studied 19 million tweets by and to MPs sent between January and June 2019.

The analysis focuses on tweets using obscenenouns (“cunt”, “twat”, etc), racist or otherwise bigoted language, and milder insults (“you idiot”, “coward”). In this way, we define “abuse”broadly; “hate speech”, where religious, racial or gender groups are denigrated, would be included in this, but we do not limit our analysis to hate speech. We include all manner of personal attacks and threats. Obscene language more generally (e.g. “fucking”, “bloody”) was not counted as abusiveas it was less likely to be targeted at the politician personally.

These data allowed us to explore the following questions:

1. What influences the amount of abuse a politician receives?
2. What are the topics of concern to those who send abuse?
3. What differences do we see across the three time periods studied?

This written submission reports some of our latest research findings. Further details on the methodology and earlier research can be found in [Gorrell et al, 2018; Greenwood et al, 2019].

Information about Our Datasets

In 2019, we analysed tweets by 573 MPs and retweets of and replies to these MPs, posted in the period of 1 Jan 2019 to 31 Jun 2019. Gender, ethnicity and party distribution are shown below:

In that period:

●       The MPs posted just over 500,000 tweets

●       Were retweeted over 11 million times

●       Received over 7.5 million replies

Table 1 shows a breakdown of the volumes of tweets sent by MPs, the total number of replies, and the abusive replies they have received. Abuse towards MPs was studied in the one month periods immediately preceding the 2015 and 2017 general elections, and most recently, in the first six months of 2019. This affords a comparison of abuse levels over time.

Table 1: Statistics of dataset analyzed

As can be seen above, abusive replies to MPs in June 2019 reached unprecedentedly high levels (60,225 abusive tweets) compared to earlier months in 2019, as well as in the two months prior to the 2015 and 2017 general elections. As can be seen from these statistics, general level of engagement with MPs (i.e. replies to MP tweets) has increased significantly between 2015 and 2019. 

As shown in the graphic below, month on month between January and June 2019, abusive reply tweets to MPs climbed up, with statistically significant increases in all but one case (April to May).

Who received the most abuse?

In 2019, the individuals receiving the most abuse in absolute terms are David Lammy (27,000), Jeremy Corbyn (26,000), Boris Johnson (16,000), Theresa May (13,000) and Jeremy Hunt (9,000). This is a significant change from the most abused politicians in 2015 and 2017. Details for these period appear in https://arxiv.org/abs/1804.01498.

David Lammy being top targeted MP with abusive replies is new as compared to 2015 and 2017 – he overtakes the most prominent politicians such as the recent PM (Theresa May) and the leader of the opposition (Corbyn). Lammy is also the main target for racist abuse.

The abuse received by individuals split by party (inner ring) and gender (middle ring, pink for female and blue for male) is expressed in the sunburst graph below, where a wider segment indicates more abuse in absolute terms, and a darker segment indicates more abuse as a percentage of replies received. It is evident for example that Boris Johnson received a high proportion of abusive replies (it is 7.5%) whereas Jeremy Corbyn received a greater volume, but as a proportion of replies received, he received less than Boris. The graph is available in interactive form at the link below, where all MPs can be viewed, and the graph can be rescaled to a fixed segment width per person to show demographic effects more clearly.

http://demos.gate.ac.uk/politics/buzzfeed2019/sunburst.html?file=sunburst.json

Overall, Conservative MPs, on average, attract significantly more abuse than Labour ones.

Also, on average, male MPs attract significantly more abuse than female ones.

Abuse levels are spiky in nature, fluctuating up and down triggered by particular events or opinions stated in tweets, e.g. up to 15% - 18% for Boris Johnson in some periods, while his average over the 6 month period is 7.5%. To illustrate this, a two-month timeline for the five individuals receiving the greatest volume of abuse is given. The drop around May 12th is due to data collection error.

Note that abuse as a percentage of replies on that day is plotted here, so a peak could actually result from quite a low volume of abuse if a low volume of replies was also received.

May and June 2019

http://demos.gate.ac.uk/politics/buzzfeed2019/abuse-timeline.html?file=most-abuse-may-jun.tsv

The Boris Johnson peak in May arises from this tweet, criticising Jeremy Corbyn in the context of mental health.

Sexism

Sexist abuse is the most prevalent, as compared to homophobia and racism.

Women receive more sexist abuse, and the individuals receiving the most sexist abuse in absolute terms are Theresa May, Jess Phillips, Anna Soubry and Diane Abbott, all of whom received significantly more sexist abuse than typical (p<0.001). In terms of general abuse, which includes a broad range of uncivil address, men receive significantly more however (p<0.001).

Interestingly, women received more religious abuse than men. Directed homophobic abuse is rare.

In the table above, Fisher’s exact test was used to establish if a word is significantly more often used in a tweet to a female or amale MP, as a proportion of all abuse terms found in tweets to that group. (This may occasionally mean that a word ap-pears on the female list because it is used unexpectedly frequently given that women receive less abuse, even though it may be similarly or more frequently used against maleMPs on a per-individual basis.) Only abuse terms that appear more than 500 times across all the years/genders combined are included in the table. Stronger words, selected according to the Ofcom guide or in keeping with it, are indicated in bold.

The impression given by the table reinforces the above finding that men tend to receive more abuse, with the suggestion of a richer vocabulary of insults being targeted to a greater extent at men. Even terms that need not technically be considered gendered, such as “bastard” and “git”, still suggest a male target through routine usage, making them hard to classify. It is notable however that women are morelikely to be addressed with “fuck off”, or “fuck you” (as aproportion of all abuse words received), phrases which areperhaps unified by being expressions of aggression, and bythe words “scum” and “turd”, which might be thought of as“dirt” words.

The bar chart gives an indication in quantitative terms of which abuse words are being used in tweets to male and female MPs (the table above gives significant differences but not an idea of the extent of word use or how male-female experience differs in absolute terms). The y-axis gives counts of this term being used per person, averaged over the time periods, for the 20 most used abuse terms. So, for example, on average a woman MP might expect to be addressed with “idiot” six times in any one timeperiod (though note that abuse does not spread itself in a normal distribution, so “average” is not a usual case, most MPs receiving less than that, and a few much more).

A note about gender: Quantitative findings such as those in our work do not address the complex culture surrounding what it means to be female, male or otherwise in our society. Work from the Pew Research Center (2014), as well as expanding on similar findings about female vs.male online abuse levels in a random sample, notes that women were “more likely to be upset” by abuse received. Indeed, the meaning of threats and abuse for any particular demographic varies a great deal depending on factors such as history of oppression, physical vulnerability and the security of their social standing. As Chemaly comments in her piece for Time,“women take online harassment more seriously not because we are hysterics, but because we reasonably have to” (Chemaly 2014). This work presents, therefore, only part of the story. Pew also found that younger women, in the 18-24 age range, are disproportionately targeted; a group not represented in our sample of politicians.

Racism

Overall, MPs from ethnic minorities receive more racist abuse.

In particular, the individuals receiving the most racist abuse in our sample are David Lammy, Diane Abbott and Sajid Javid (p<0.001). The ethnic minority individuals receiving the most general abuse are David Lammy, James Cleverly and Diane Abbott, receiving significantly more general abuse than is typical (p<0.001). The timeline below graphs the abuse of all types (not just racist abuse) received by these three ethnic minority MPs that received the most general abuse:

Most abused ethnic minority MPs

http://demos.gate.ac.uk/politics/buzzfeed2019/abuse-timeline.html?file=ethnic-minority-abuse.tsv

The person targeted by most racist abuse is David Lammy. Overall, he received 594 racist replies. Amongst these, 169 replies containing the n word; 37 of these have now been removed by Twitter, deleted, or the author’s account has been suspended.

There were very few cases of religious abuse directed at the MPs that we could detect reliably.

What topics attract abusive replies in 2019?

The diagram below shows what topics discussed by most-abused MPs triggered the most abusive replies. For each of the MPs, we show the 5 topics that attracted the most abuse, as well as the “other” category, which gives the total percent for tweets on a topic other than these five. The five topics attracting the most abusive replies vary from one MP to the next, hence in the chart there are more than 6 categories.

Most notably, tweets about Brexit by these MPs tended to attract significant amount of abuse for all MPs except Diane Abbott (see examples and discussion below). James Cleverly also particularly attracted abuse for his comments on certain topics, and examples are given in the final section below, though in his case the dominant subject on which his tweets were abusively responded to is Brexit, as for other prominent MPs.

22/08/2019

References

[Chemaly 2014] Chemaly, S. 2014. There’s no comparing male and female harassment online. http://time.com/3305466/male-female-harassment-online/. Accessed: 2019-03-27

[Gorrell et al. 2018] Gorrell, G.; Greenwood, M. A.; Roberts, I.;Maynard, D.; and Bontcheva, K. 2018. Twits, twats and twaddle:Trends in online abuse towards uk politicians. In Twelfth International AAAI Conference on Web and Social Media.

[Greenwood et al. 2019] Greenwood, M. A.; Bakir, M.E.; Gorrell, G.; Song, X.; Roberts, I.;  and Bontcheva, K. Online Abuse of UK MPs from 2015 to 2019: Working Paper. https://arxiv.org/abs/1904.11230

[Pew 2014] Pew. 2014. Online harassment. https://www.pewresearch.org/wp-content/uploads/sites/9/2014/10/PI_OnlineHarassment_72815.pdf. Accessed: 2019-03-27