Over the past year I have been learning and adapting methods for studying and summarising social media activity around health conferences and awareness campaigns.

This blog ties up that work, bringing the key pieces of work together in one place. My hope is that other people and organisations can use these techniques to plan, monitor and evaluate their own social media activities.

I am taking a sabbatical over the next few months, returning to the front line of clinical work in a care of the elderly unit in a large teaching hospital. I will not be able to extract and analyse tweets between April and July (inclusive). I will continue to monitor selected campaigns that relate to the clinical attachment (eg #EndPJParalysis and #EndPJParalysis which recently announced a 70 day campaign, beginning 17 April 2018).

This blog lists the main learning points from my activities over the past year, with links to the relevant blogs.

1. You can’t rush social media work. You have to plan weeks or months ahead and  should incorporate a single, simple and specific hashtag into your campaign materials. Even with careful planning, people will mishear, mistype, or invent their own hashtags (let’s call it “hashtag drift”). Check and share your planned hashtag and materials with others, both in the real world and on social media. Use analytical tools and the Twitter search function to see whether other people have used the same hashtag for other purposes. Read my blog on planning, tweeting and summarising a health campaign or conference.

Figure: An illustration of “hashtag drift”: The official hashtag for the Royal College of Paediatrics and Child Health conference (13-15 March) is #RCPCH18. However, checking at the end of the first morning of the conference there is also some (minimal) activity using #RCPCH2018. Containing “hashtag drift” is challenging once it takes off, and means that both hashtags should be included in social network analysis during and after the event. (For tweets, though, keep using #RCPCH18).

Source: Followthehashtag (click on links above)

2. Reserve plenty of time for analysis during and after the campaign (and up to 10 days after the main activities end). I have written up a “how to” guide about running social network analysis using my prefered method (NodeXL) (*), with supporting files providing further advice and an Excel template based on a previous analysis. (I have also produced a “how to” guide on extracting retweets from NodeXL files and a “how to” guide using TAGS for times when NodeXL is not able to obtain a full extract from Twitter. More recently I have worked on approaches for identifying the most “mentioned” tweeters, both in original tweets and replies, to work with both NodeXL and other tools including TAGS). Social network analysis is useful because of the interactions and influencers it will demonstrate, but also because it gives you an extract of tweets using your selected search terms that is archived and can be analysed in the future. As explained in the guide, you may need to go back and attempt extracts in blocks (eg by day or by hashtag). You should look beyond the main hashtag to see what other people are tweeting around the campaign. Think of it a bit like a literature review: if you set out to study immunisation in a review of the published literature you would miss a lot of useful references if you did not factor in different spellings (US and UK English which use “z” and “s” respectively) and words (eg vaccination). The same applies to social network analysis. Almost every campaign and conference that I have studied has “hashtag drift” as described above: eg people attending the American College of Cardiology last weekend tweeted using #ACC18 (official hashtag), but also using #ACC2018. In the colorectal cancer awareness campaign featured in the more advanced sections of the “how to” guide there was tweeting using a plethora of hashtags and terms. Tweeting, though digital in format, is little different from the game of “Chinese whispers” (or should I say “telephone“). Even once you have identified the terms you would like to search for, you may need to pause until Twitter catches up with all the activity. It can take days for Twitter and NodeXL to catch up, and sometimes a full extract is never produced (NodeXL can only look back 9-10 days). You have to be persistent but also realistic. I was unable to extract more than a few minutes of World AIDS Day last December, because of the sheer volume of tweeting. I did, however, manage to extract a reasonably complete extract of UK “World AIDS Day OR #WorldAIDSDay” tweets (figure below, source NodeXL) which I have yet to summarise (any volunteers, please take a look at the “how to” guide and have a go, though perhaps after attempting a smaller analysis first!) Having said, however, that you can’t rush planning and analysis of a campaign, when you have a NodeXL extract available the analysis can be very quick if you apply simple rules. This is all described in the “how to” guides above.
3. Be sceptical. Question everything. Check before you share. There are lots of potential pitfalls, biases and confounding factors in social network analysis. There will be trouble makers and attention seekers in any large campaign, and others like to crash conferences in the virtual domain. Sometimes these accounts will appear as “influencers” in social network analysis. If you share the automated outputs from social network analysis you may end up appearing to legitimise these accounts. At the very least check the accounts listed in these automated summaries before you tag them in a tweet. Ideally, however, you should look at individual tweets, and identify the people who tweet the highest quality content to gain a fuller picture. Often, but not always, the most shared tweets will also be the most informative, which is what makes the approach laid out in step 2 so powerful.
4. Keep tweeting and curating, but do it with purpose. There is no doubt that there is a lot to learn from Twitter. Lots of academic content is launched on social media weeks or months ahead of print, and you can learn a lot from colleagues and tweeters from across the world. You can gain clinical tips and maximise the opportunities to learn from other people’s work (eg with EndPJParalysis campaign), which is why many in the quality improvement movement were early adopters of social media. However, social media is also designed to be addictive and distracting, so we need to be careful not to overuse. My blog on the value of tweeting for healthcare workers summarises some of the key points. There are a few simple rules that will help increase the impact of your own tweeting as summarised in a JPH article and this accompanying blog: keep the message short and clear; include a URL for further information; add a relevant hashtag; mention others that might be interested or you would like to credit; add a picture that provides more information.

These four observations should help individuals and organisations plan, collate and evaluate their Twitter activity (both tweeting and observing). However, currently very few organisations, conferences or campaigns, take this systematic approach. The purpose of this blog is to allow anybody to have a go.

The ScotPublicHealth blog will now return to more Public Health and clinical posts. Job done for the moment on social network analysis, but watch for future outputs from this work at conferences and in publications.

Dr Graham Mackenzie, Consultant in Public Health, 13 March 2018

(*) NB NodeXL has changed the format of its extract (sometime in September 2018). As a result the method described above needs to change a little. Instead of searching for tweets by looking for posts that do not start with “RT” it is necessary to look in a column called “Retweet ID”. When this field is blank, the post is a tweet. This is in fact a more accurate way of identifying tweets, and had been available as an option before this format change. Use this information to populate column N in the example template. I will update this blog, NodeXL instructions and example template when time.