Recently I was intrigued by a post on twitter conversion rates (e.g. the likelihood that a view on your tweet results in a click on the link) by journalist Derek Thompson at the Atlantic. Derek writes that although using twitter gives him great joy, he’s not sure it results in the kinds of readership his employers would feel merits the time spent on the service. Derek found that even his most viral tweets only resulted in a conversion rate of about 3% – on par with the click-through rate of east asian display ads (i.e. quite poorly in the media world). Using the recently released twitter metrics, Derek found an average conversion of around 1.5% with the best posts hitting the 3% ceiling. Ultimately he concludes that twitter seems to be great at generating buzz within the twitter-sphere but performs poorly at translating that buzz into external influence.
This struck my curiosity, as it definitely reflected my own experience tweeting out papers and tracking the resultant clicks on the actual paper itself. However, the demands of academia are quite different than that of corporate media. In my experience ‘good’ posts do exactly result in a 2-3% conversion rate, or about 30 clicks on the DOI link for every 1000 views. A typical post I consider ‘successful’ will net about 5-8k views and thus 150-200 clicks. Below are some samples of my most ‘successful’ paper tweets this month, with screen grabs of the twitter analytics for each:
Sharing each of these papers resulted in a conversion rate of about 2%, roughly in line with Derek’s experience. These are all what I would consider ‘successful’ shares, at least for me, with > 100 engagements each. You can also see that in total, external engagement (i.e., clicking the link to the paper) is below that of ‘internal’ engagement (likes, RTs, expands etc). So it does appear that on the whole twitter shares may generate a lot of internal ‘buzz’ but not necessarily reach very far beyond twitter.
For a corporate sponsor, these conversion rates are unacceptable. But for academics, I would argue the ceiling of the actually interested audience is somewhere around 200 people, which corresponds pretty well with the average paper clicks generated by successful posts. Academics are so highly specialized that i’d wager citation success is really more about making sure your paper falls into the right hands, rather than that of people working in totally different areas. I’d suggest that even for landmark ‘high impact’ papers eventual success will still be predicted more by the adoption of your select core peer group (i.e. other scientists who study vegetarian dinosaur poop in the Himalayan range). In my anecdotal experience, I would say that I more regularly find papers that grab my interest on twitter than any other experience. Moreover, unlike ‘self finds’, it seems to me papers found on twitter are more likely to be outside my immediate wheelhouse – statistics publications, genetics, etc. This is an important, if difficult to quantify type of impact.
In general, we have to ask what exactly is a 2-3% conversion rate worth? If 200 people click my paper link, are any of them actually reading it? To probe this a bit further I used twitters new survey tool, which recently added support for multiple options, to ask my followers about how often they read papers found on twitter:
As you can see, out of 145 responses more than 50% in total said they read papers found on twitter “Occasionally” (52%) or “Frequently” (30%). If these numbers are at all representative, I think they are pretty reassuring for the academic twitter user. As many as ~45 out of ~150 respondents say they read papers on twitter “frequently” suggesting the service has become a major source for finding interesting papers, at least among its users. All together my take away is that while you shouldn’t expect to beat the general 3% curve, the ability to get your published work on the desks of as many as 50-100 members of your core audience is pretty powerful. This is a more tangible result than ‘engagements’ or conversion rates.
Finally, it’s worth noting that the picture on how this all relates to citation behavior is murky at best. A quick surf of the scientific literature on correlating citation rate and social media exposure is inconclusive at best. Two papers found by Neurocritic are examplars of my impression of this literature, with one claiming a large effect size and the other claiming none at all. In the end I suspect how useful twitter is for sharing research really depends on several factors including your field (e.g. probably more useful for machine learning than organic chemistry) and something i’d vaguely define as ‘network quality’. Ultimately I suspect the rule is quality of followers over quantity; if your end goal is to get your papers in the hands of a round 200 engaged readers (which twitter can do for you) then having a network that actually includes those people is probably worth more than being a ‘Kardashian’ of social media.