Background Tobacco companies’ intentions to influence the WHO Framework Convention on Tobacco Control (FCTC) via the Conference of Parties (COP; the official biannual meeting where Parties review the Convention) are well documented. We aimed to analyse Twitter data to gain insights into tobacco industry tactics, arguments and allies.
Methods We retrieved 9089 tweets that included #COP8FCTC between 1 and 9 October 2018. We categorised the tweets’ content and sentiment through manual coding and machine learning. We used an investigative procedure using publicly available information to categorise the most active Twitter users and investigate tobacco industry links. Network analysis was used to visualise interactions and detect communities.
Results Most tweets were about next-generation products (NGPs) or ‘harm reduction’ (54%) and tended to argue in support of NGPs; around one-quarter were critical of tobacco control (24%). The largest proportion of most active tweeters were NGP advocates, and slightly over half of those had either links to the Philip Morris International (PMI) funded Foundation for a Smoke-Free World (FSFW) and/or to the International Network of Nicotine Consumer Organisations, a network to whom the FSFW granted US$100 300 in 2018. PMI was the most active transnational tobacco company during COP8.
Conclusions The nature of the activity on Twitter around COP8, including a substantial online presence by PMI executives and NGP advocates with links to organisations funded directly and indirectly by PMI, is highly consistent with PMI’s 2014 corporate affairs strategy, which described engaging tobacco harm reduction advocates to ‘amplify and leverage the debate on harm reduction’ around events such as the COP.
- electronic nicotine delivery devices
- harm reduction
- public policy
- tobacco industry
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
Twitter data can provide insights into public policy debates such as sugar taxation1 and standardised tobacco packaging,2 including the stakeholders involved, alliances between groups and the arguments and evidence put forward to support different positions. Such data can help public health advocates better understand and inform policy debates,1 2 and potentially also bring to light the role of self-interested parties such as corporations in shaping the discussion.
The WHO Framework Convention on Tobacco Control (FCTC) is a landmark global public health treaty that requires parties (of which there were 182 at the time of writing) to implement certain measures to reduce the demand and supply of tobacco at national level.3 Previous research has highlighted countries’ challenges in implementing the WHO FCTC4 5 and tobacco industry (TI) attempts to obstruct its implementation.6 7 Indeed historically, the TI made concerted efforts to block and weaken the development of the WHO FCTC,8–10 and since its adoption, the industry has continued to try and influence WHO FCTC negotiations via the Conference of Parties (COP).11 The COP is the governing body of the WHO FCTC, and Parties attend biannual sessions of the COP to review and promote implementation and adopt protocols, annexes and amendments to the Convention.12 The first session of the COP, referred to as COP1, took place in 2006; the most recent session (at the time of writing), COP8, was held in 2018. An example of TI interference at COP sessions includes the use of a front group, the International Tobacco Growers Association,13 to defeat the adoption of full guidelines for Articles 9 and 10 at COP4 in 2010 in Uruguay.11 Similarly, another TI front group, the International Tax and Investment Center,14 convened a meeting for finance ministers the day before COP6 where Article 6 guidelines on tobacco taxation were to be agreed, to influence tobacco tax negotiations and portray the meeting as though it was officially associated with COP6.15 Leaked internal documents from Philip Morris International (PMI) explicitly stated in 2014 that part of the company’s corporate affairs strategy was to ‘amplify and leverage the debate on harm reduction around global events (eg, COP6)’.16 A Reuters investigation also revealed that PMI set up an ‘operations room’ in a hotel near the venue of COP7 to influence delegations attending the conference.17 In the past, TI representatives have served on delegations and have reportedly accessed COP sessions by posing as members of the media and general public.18 To prevent such activities at COP8, members of the public and media were prohibited from accessing official sessions.18 Exposing examples of TI interference in tobacco control (TC) policy-making is an essential step in addressing the problem,15 and harvesting Twitter data could enhance this effort. Given that tobacco companies are active on Twitter,19 20 the lack of regulation for online media means they are able to easily promote their agendas to a widespread audience.
To date, no published studies have examined stakeholders’ social media activity during a WHO FCTC COP. Analysing Twitter activity around COP would provide new and globally relevant insights about current TC campaigns, debates and arguments, and the networks and key actors of influence within those communications. We examined the main themes and sentiment of tweets about WHO FCTC COP8, identified and classified the most active tweeters, explored how people who tweeted about COP8 engaged with one another and, lastly, explored the presence of TI links among the most active tweeters.
Twitter data collection
We used NCapture to download a dataset of tweets that included the hashtag associated with the eighth session of the WHO FCTC COP, #COP8FCTC. Tweets were downloaded on 16 October 2018 and content spanned an 8-day period from 1 to 9 October 2018 (COP8 took place 1–6 October 2018). This dataset comprised 9089 tweets, of which 1453 were original tweets from 386 unique users, and the remaining 7636 posts comprised retweets. NCapture does not necessarily collect all tweets with a particular hashtag; the number is determined by Twitter and depends on the number of Tweets available and the amount of traffic on Twitter. Foreign language tweets were included and were translated by members of the wider research team (these included 198 in Spanish, 51 tweets in French, 18 in Portuguese, 9 in German, 7 in Italian and 6 in other European languages). We excluded 90 tweets that did not have sufficient textual information to code (ie, they included only an image, a meme or an extremely brief tweet comprising only one or two words). Thus, the final sample of original tweets was 1363 (though for network analyses, we used the full dataset of 9089 tweets and retweets).
Coding tweet content and sentiment
We adapted a coding framework used in earlier research on standardised packaging2 and coded tweets in terms of content and sentiment. For content, we coded whether each tweet related to: (1) a specific TC policy such as tobacco taxation (hereafter this variable is referred to as ‘TC-related’); (2) next-generation nicotine products (NGPs) (ie, vaping, e-cigarettes, other ‘reduced-risk’ nicotine products or ‘tobacco harm reduction’, hereafter labelled ‘NGP-related’) and (3) the WHO FCTC or TC more generally, that is, WHO FCTC ratification, governance or TC as a concept/sector (‘FCTC-related’). In terms of sentiment, we coded tweets as either: (1) informative/neutral, (2) argument or (3) critical (ie, derogatory or abusive). Full details of the coding definitions are provided in online supplemental file 1. A randomly selected subsample of tweets (n=200) was independently coded by LR and KE-R with inter-rater reliability (Cohen’s kappa coefficient) ranging from 0.90 to 0.98. Discrepant results were resolved through comparison and discussion, and the codebook was subsequently refined.
Using the 200 coded tweets, we trained machine learning classifiers to predict the content and sentiment of the remaining 1163 original tweets. To use Twitter text data as input data for machine learning models, they must be converted into a numerical format. This is done using a combination of data processing techniques from machine learning (Tf-Idf allowing for one-word, two-word and three-word combinations) and natural language processing (stop word removal and lemmatisation). We also incorporated numerical data about the tweets and the users (number of retweets, followers, following, mentions, hashtags, URLs, tweets by user in dataset, tweets by user overall, length of original tweet and length of processed tweet) and along with the ‘numerically encoded’ tweets, these data formed the input for machine learning models. The final machine learning model that performed best on the testing dataset of each tweet category (ie, tweet content and sentiment) was selected to predict the classes for the unseen data (see online supplemental file 2 for details on model metrics). Lastly, we used descriptive statistics to examine and present data on tweet sentiment and content (including other hashtags and Twitter accounts mentioned) and the extent of retweets.
Categorising the most active Twitter users and investigating TI links
For each Twitter account that posted two or more original tweets using #COP8FCTC (n=152 Twitter users), we used a manual stepwise investigative procedure using publicly available information (eg, mainly sourced from the Twitter user’s landing page, LinkedIn and via Google) to categorise each Twitter account and investigate any links with the TI. We categorised and investigated only Twitter users who had written two or more tweets using #COP8FCTC because we conceived those users as having attempted to influence TC policy (as opposed to, eg, individuals who wrote only a single tweet or who only retweeted others’ content). We developed the following categories based on modified definitions used in other research2 21: (1) TI actor, (2) NGP industry actor, (3) NGP advocate, (4) public health advocate, (5) member of public and (6) other (see table 3 for definitions). We recorded whether the Twitter account was anonymous or not; we defined an account as anonymous if it had a non-identifiable username, or if it had a seemingly identifiable username but otherwise no identifiable information on the account landing page (as an example, an account with the name ‘Tom Smith’ that provided no further personal information, account description, affiliations, occupational information, location or photographs was coded as anonymous). Five members of the research team conducted the investigative procedure; to ensure consistency and to pretest the protocol, we each independently coded the same five Twitter accounts, before discussing discrepancies and refining the definitions. Team members then each independently coded approximately 30 Twitter accounts. We subsequently double-coded 10 Twitter accounts from another team member’s subset of 30 accounts, and met to compare and check coding, and ensure there were no discrepancies.
Social network analysis
Social network analysis of the tweets was performed using network-x, and visualisation was done using Gephi. We built two kinds of networks (mentions-in and re-tweets) using the original dataset of 9089 tweets for this analysis. Re-tweet networks and mentions-in networks are directed weighted graphs where the nodes are the Twitter users and the edges are the retweet and mention relationship, respectively, where the direction of the relationship is shown by the direction of the edge. Both re-tweets and mentions-in show means of interaction in a Twitter social network, with re-tweets showing information-spreading to followers, while the mentions-in network represents actions to directly include or ‘tag’ another user in tweets.
The study received favourable ethical approval from the REACH Committee at the University of Bath (ref: EP17/18 237) and complies with Twitter’s terms of service.22
Tweet content and sentiment
The largest proportion of tweets (n=622, 54%) were about NGPs or ‘tobacco harm reduction’ and most of these put forward an argument in favour of NGPs (n=387, 62%) or were critical of people and organisations perceived as opposing NGPs (n=220, 35%) (table 1). Examples of tweets coded for sentiment are shown in table 2.
Only a small proportion of tweets (10%) were about TC measures outlined in the WHO FCTC (eg, taxation or standardised packaging). Of the 418 tweets that were FCTC related, the majority were critical towards the WHO FCTC or TC in general (n=274), and around half of those critical FCTC-related tweets were posted by NGP advocates (49%, n=133). Tweets that were NGP related were retweeted on average six times, compared with an average of five retweets for tweets that were FCTC related, and only three retweets for TC-related tweets.
Characteristics of the ‘most active’ Twitter users
Advocates of NGPs comprised the largest number of Twitter users who had posted twice or more with the hashtag #COP8FCTC (33%), and they also posted the largest volume of tweets (38%) (see table 3). There were almost as many public health advocates in the group of most active Twitter users as there were NGP advocates (n=43 vs n=50), though the number of tweets posted by public health advocates was considerably lower than the number for NGP advocates (n=285 vs n=467). Almost one-fifth of the original tweets with #COP8FCTC were from TI actors (see table 3). Of the 27 TI actors, three were the official accounts for transnational tobacco companies (@JTI_global, @PMIScience and @InsidePMI), 11 were individual accounts from PMI executives and 3 accounts belonged to an individual employee of either British American Tobacco, Imperial Brands or Japan Tobacco International. Ten of the TI actors’ accounts were from a foundation, think-tank or similar organisation that receives direct funding from a transnational tobacco company (or employees of such an organisation); examples included the PMI-funded Foundation for a Smoke-Free World,23 the Consumer Choice Center (has received PMI funding and JTI (Japan Tobacco International) funding at time of writing)24 25 and Factasia.org (received PMI funding at time of writing).26
None of the tobacco or NGP industry actors or public health advocates posted using an anonymous account, whereas around half of the 41 NGP advocates who tweeted in an individual capacity (rather than an organisational capacity) had anonymous accounts.
Investigating TI links among most active users
Since by definition, the public health advocates in our sample of Twitter users excluded individuals and organisations with TI links (table 3), the results of the investigative procedure focus on NGP advocates, members of the public and ‘other’ users. For the latter two categories, we found no evidence of any link between users and the TI; similarly, for 23 of the 50 NGP advocates, we found no publicly available evidence of any link with the TI.
Of the remaining 27 NGP advocates, six had financial links with the Foundation for a Smoke-Free World (FSFW), in that they were either an employee, board member (or other member of governance team) or director of an organisation that receives funding from FSFW. Those organisations included Knowledge-Action-Change (which was granted US$1 051 364 from FSFW in 2018),27 the International Network of Nicotine Consumer Organisations (INNCO; received a FSFW grant of US$100 300 in 2018),28 the Centre of Research Excellence on Indigenous Sovereignty granted US$978 500 by FSFW in 2018)29 and the Twitter account for the ‘Nicotine Science and Policy’ website (produced by Knowledge-Action-Change).30
Slightly over half (n=27) of the 50 NGP advocates were affiliated with INNCO (this number includes the official INNCO Twitter account); Twitter users/accounts’ affiliations to INNCO comprised national vaping consumer organisations that were members of INNCO at the time of COP8 (eg, New Nicotine Alliance Australia and ProVapeo Mexico)28 (n=6), individuals affiliated with those national member organisations (n=14), individuals who have been part of INNCO’s governance or advisory groups (n=4) and two individuals who, in 2016, appear to have played an active role in helping to establish INNCO.28 We found no evidence that the individuals affiliated with INNCO or its member organisations were themselves funded by FSFW or by the TI directly. Tweets by INNCO and its affiliates as described above constituted 63% (n=296) of the tweets posted by the NGP advocates.
One organisation belonging to INNCO, ProVapeo Mexico, attempted to mobilise its members—and presumably other NGP-using members of the public—to advocate for vaping using the #COP8FCTC hashtag. An appeal on their website stated: ‘Please log in to Twitter and send a message explaining how vaping has helped you and also tell COP8 that they should support vaping as an alternative for smokers who are trying to quit tobacco consumption. Hashtag #COP8FCTC send tweet to @fctcofficial @vera_dacosta copy us @provapeomexico’ (see online supplemental file 3). We identified 128 tweets that mentioned @provapeomexico, consistent with ProVapeo Mexico’s instructions (the vast majority of these individuals tweeted only once).
Social network analysis
Figure 1 shows the results of the mentions-in network. The size of the nodes are proportional to the number of times that user has received mentions, and the colours of the nodes represent belonging to the same community, as identified by Gephi. The close-up of this network diagram shows several organisations as prominent nodes: Campaign for Tobacco-Free Kids, Framework Convention Alliance, Food and Drug Administration Tobacco, WHO FCTC Official (along with Tedros Adhanom, Director General of WHO and Vera DaCosta, then Head of WHO FCTC Secretariat), INNCO, New Nicotine Alliance and PMI Science. These organisations received mentions and in turn mentioned other users most frequently.
The results for the re-tweets network are shown in figure 2. The size of the nodes in this figure are proportional to the number of posts retweeted by the user and nodes belonging to the same communities identified by Gephi are represented in the same colour. Figure 2 shows the nodes in the orange community contain mostly public health individuals who tend to retweet each other’s tweets and that this community is relatively isolated in the COP8 dataset. The pale green community, which is also relatively self-contained, features ProVapeo Mexico and the tweets and retweets generated by its campaign to mobilise members of the public. The other denser clusters indicate that users in those communities very frequently retweeted each other’s tweets. TI actors appear both within the purple and blue communities of NGP advocates; the purple community includes INNCO and INNCO members such as Vape India, the Tobacco Harm Reduction Association of Canada and New Nicotine Alliance Australia, while the blue community comprises the FSFW, the ‘Nicotine Science and Policy’ account (a website produced by FSFW grantee Knowledge-Action-Change30) and the TI-funded Consumer Choice Center. However, there was a high level of retweeting and interaction across these two (purple and blue) communities.
Most of the Twitter content regarding and during WHO FCTC COP8 advocated for NGPs or ‘tobacco harm reduction’ and was critical of TC. The majority of tweets were composed either by TI or NGP industry representatives or by NGP advocates; NGP advocates made up the largest group of Twitter users in our sample. They were responsible for posting the largest number of original tweets, and their tweets were retweeted to a greater extent than those written by other users. In other words, the activity generated by NGP advocates around COP8 was far greater than that generated by public health advocates.
The extensive activity by NGP advocates with links to organisations funded directly and indirectly by PMI (including factasia.org and Foundation for a Smoke-Free World,23 26 and Knowledge-Action-Change and INNCO,27 28 respectively), as well as a substantial online presence by PMI executives themselves, suggests a strategic approach by PMI to influence COP8 debates. Such an approach is highly consistent with the listed actions in the company’s leaked 2014 corporate affairs strategy, which included, to: ‘Establish the concept of harm reduction as legitimate public policy in tobacco regulation’, ‘Identify and engage non-traditional 3rd party stakeholders/allies (e-cigarette manufacturers and retailers, adult consumers of RRP products, tobacco harm reduction advocates…’ and ‘Amplify and leverage the debate on harm reduction around global events (eg, COP6)’.16
The TI has a long history of using front groups and intermediaries to remove its ‘fingerprints’ from information and use others’ voices to make its arguments appear more credible and widely supported than they actually are.15 31 Over many decades, the TI has repeatedly used smokers’ rights groups to influence policy, groups that were organised and predominantly funded by tobacco companies themselves.32 Hence, it is perhaps unsurprising that new ‘vaping consumer advocacy’ groups that receive TI funding are emerging and attempting to influence the WHO FCTC.
Indeed, INNCO appears to have been set up in 2016 with assistance from members of Knowledge-Action-Change for the very purpose of influencing the WHO FCTC.28 Those involved in establishing INNCO acknowledged that ‘most consumers do not belong, or wish to belong, to formal organisations’ yet argued the need for a new international nicotine consumer organisation as a way of ‘tackling issues at an international level, in particular with the UN [United Nations] system and WHO’.28 They identified the WHO FCTC COP meetings as a ‘big ticket’ measure that INNCO should target.28 However, Parties rejected INNCO’s 2018 application for WHO FCTC COP observer status.33
Our findings have several implications. First, TC advocates and policy-makers should be aware that a key strategy of the TI appears to be engaging with and funding tobacco harm reduction consumer groups, presumably to augment pro-NGP messages and the appearance of a groundswell of independent consumer advocates. The TI’s use of Twitter during COP is likely indicative of a more comprehensive strategy to influence policy for NGPs, which past evidence suggests may also involve direct lobbying, corporate social responsibility activities, and creating and disseminating misleading evidence.31 Our findings highlight the need for ongoing monitoring, investigation and exposure of links between tobacco companies and seemingly independent entities that promote an industry agenda,15 particularly since such links have often been deliberately obfuscated15 32 and are therefore not immediately apparent to policy-makers and journalists. At a national level, advocates could monitor policy submissions, reports and media articles written by tobacco harm reduction groups, investigate any links with tobacco companies and their known allies and disseminate material to expose such links via the media and to key stakeholders. In terms of Twitter activity specifically, public health advocates could consider increasing Tweet volume and communicating the need for evidence-based TC measures during WHO FCTC COP and other significant policy windows. This could help ensure messages supporting TC policies are not overshadowed by those promoting NGPs and criticising the WHO FCTC, though evaluations should be carried out to examine the effectiveness of this approach.
Our study has limitations. First, since data were collected using Twitter’s public API, our dataset may not contain every tweet containing #COP8FCTC and did not include tweets from private Twitter accounts. Nonetheless, it seems very unlikely our sample of tweets differs systematically from those not captured, as the use of NCapture likely resulted in a random, rather than a biased, sample of tweets. The predictions of our machine learning models did not use any information from ‘url’ links or media embedded in the tweets, which means we may have lost some richness of data in terms of tweet content and sentiment. However, due to a relatively small data set, training complex models to incorporate ‘images’ along with text and numerical features were not feasible. Lastly, our study does not provide a nuanced analysis of the arguments for and against various NGPs, nor does it examine qualitative differences in the tweets other than their sentiment. Future research could draw on additional data sources, such as blogs, media articles and relevant reports, to describe TI and NGP advocates’ arguments and activities during COP in greater detail.
Notwithstanding these limitations, our study provides insights as to the Twitter activity during COP8 and the extent of TI involvement in shaping and amplifying harm reduction debates, both directly and indirectly through organisations it funds.
What this paper adds
What is already known on this subject
Philip Morris International’s (PMI) leaked 2014 corporate affairs strategy indicated the company was planning to influence the proceedings of the WHO Framework Convention on Tobacco Control (FCTC) Conference of Parties (COP) and planned to work with and amplify the voices of harm reduction advocates to do so.
What important gaps in knowledge exist on this topic
To date, no published studies have explored the Twitter debates around WHO FCTC COP, nor investigated the Twitter activity of tobacco companies and their third-party allies around the WHO FCTC COP.
What this paper adds
Most of the Twitter activity during COP8 related to ‘tobacco harm reduction’, yet a significant proportion of this content was created and propagated by tobacco industry actors (particularly PMI executives) and harm reduction advocates who were linked to organisations funded directly and indirectly by PMI.
We wish to thank Mercedes Maria Carballo, Britta K. Matthes, Kathrin Lauber and Arsenios Tselengidis of the Tobacco Control Research Group at the University of Bath for help with translating tweets into English, and Karin Silver for advice on investigative procedure.
Contributors LR and KE-R conceptualised the study; all authors contributed to the refinement of the study design and methods. LR collected the data. LR and KE-R coded a subset of tweets, and AJ performed machine learning and social network analysis, with assistance from GW. LR, TL, GW, KR and KE-R conducted the investigative procedure. LR drafted the initial manuscript; all authors reviewed and contributed to subsequent drafts.
Funding The University of Bath is the Research Partner of the STOP initiative, which is funded by Bloomberg Philanthropies (www.Bloomberg.org). This work is also partly funded by CRUK grant CRUK A25745.
Competing interests None declared.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement Data may be shared on reasonable request.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.