#ISA2015 as it happened on Twitter


Last week, this year’s ISA conference brought together over 5000 scholars and exhibitors from all over the world to discuss all things international, political, scholarly, hold meetings, get lunch together, and party at Mardi Gras (it was in New Orleans, after all!). Similar to last year, a lot of this discussing took also place on Twitter. Scholars-slash-tweeps rallied around the hashtag #isa2015 to talk to each other online about great (and not so great) panels, trends in IR scholarship, gender bias in academia, and (not surprisingly for an academic conference) coffee. Who was most active during ISA2015 on Twitter? What were the most hotly debated topics online? When did ISAlers tweet?

The Data

To find answers to these (and more) questions, I fired up the twitteR package for R, tweaked the code I wrote for last year’s analysis and started downloading All The Tweets. At least, that’s what I thought.

But, alas, we weren’t the only ones being attracted by the beauty of the #isa2015 hashtag. Apparently the „Independent Spirit Awards“ are a thing (and indie film thing, as it turns out). And they were awarded on the 21st of February, just when the conference was winding down on Saturday. Yes, you guessed right, film-aficionados thought that #isa2015 was a classy hashtag and jumped on to it. Boy, I would’ve loved to see their faces when they tuned into the hashtag, because they must’ve thought they were in the wrong film (I know it’s early into this post for bad puns. Sorry.). Further complications arose when I realized that there was an Unidentified Thai Thing that also used #isa2015 – this tweet showed up frequently in my #isa2015 downloads:

As far as I can tell (which is as far as Google Translates takes me) this is about a Thai girl (?) band… No idea. And I didn’t really want to investigate further because my interest in Thai girl bands is somewhat close to zero. The only problem is that it was clearly not connected to the scholarly #isa2015 and all my attempts to filter out Tweets written in Thai failed (regex’s \p{Thai} somehow didn’t work out…). So I filtered them out (half-)manually by omitting all the Thai tweeps and indie-film-twitter accounts (the latter ones were luckily mostly also tagged with #SpiritAwards, so this was kinda easy).

Since also I wanted to catch the #isa2015 coverage before the conference and because the Twitter API only allows for a search between 6-9 days back, I already started downloading patches of tweets on the 16th and did so throughout the conference up until the 25th. I then „stichted“ together the resulting data frames to create a full body of #isa2015 tweets. Of course, they don’t represent all ISA-related tweets. Tweets that didn’t include the #isa2015 hashtag don’t show up in the resulting list, simply because „ISA“ alone returns too many tweets that can’t be easily identified as #isa2015 related.


All in all, the search resulted in 1919 unique Twitter users that tagged a total of 8187 tweets with #isa2015 between 10 February 2015 and 24 February 2015. This is a substantial increase over the 2940 tweets of last year’s conference. I have two explanations for this: First, I substantially extended the search in time. Last year, the Twitter analysis was an afterthought I had during post-conference jetlag sleeplessness. Back then, the restrictions from the Twitter API allowed me to only include tweets during the conference and there also probably not all tweets, since I was pretty late to the API’s 6 – 9 days backlog window. This year I started much earlier and systematically downloaded earlier tweets which resulted in a much higher number. Second, Twitter has become more popular among researchers. A substantial number of researchers have joined Twitter since the last ISA conference, although the majority had an account before ISA 2014. In fact, we can use the downloaded data to get the precise join dates of the Twitter users of this year’s ISA. The next figure plots the quarter-yearly number of joins since the first day a #isa2015-tweep joined Twitter (an honor that goes to bobcorrigan).

Here we see that the vast majority of Twitter users that tweeted about ISA2015 joined Twitter before 2014. It is difficult to conclude from this however, whether the increase in this year’s tweets was because of additional tweeps writing about the ISA hashtag. This is because we do not know if twitter output is a function of join date (with the working hypothesis being that Twitter output increases by join date). (Note: I ran some models of Twitter activity, i.e. # of tweets per user as a function of time since join date + some controls. The models seem to suggest that the longer you’ve been on Twitter, the more frequently you tweeted about #isa2015, which would counter this working hypothesis. But those models are only quick-and-dirty and not ready for reporting. Maybe I’ll clean ‚em up and share them in a later post). So it turns out, my second explanation for the vastly higher number of this year’s tweets is somewhat shaky. Maybe the higher number really is a only consequence of the much earlier time period for which I collected (about 1600 tweets were tagged with #isa2015 prior to the conference, i.e. between 10 Feb and 18 Feb) and the more systematic downloading.

Who tweeted most?

But now to the fun part of this exercise – who was the busiest #isa2015 tweep out there? The next figure plots the 20 busiest #isa2015 twitter users:

A couple of observations: First, congrats to Raul Pacheco-Vega for winning the Award Of ISA Tweeting (not yet sponsored by SAGE, sorry) in what looks like a close race between him and Annick T.R. Wibben (who was last year’s winner). The FTGS_ISA section somewhat cheated because it is a section account that was very busy tweeting announcements for FTGS panels. But it nevertheless earned its third place by retweeting the hell out of other tweeps.

Second, the overall intensity of tweets is much higher than last year, with Raul scoring almost twice as many tweets than last year’s winner. Again, I’ve collected a substantially higher number of tweets overall, so this is increase is certainly a reflection of this. But I would be interested in the opinion of some of those high-frequency tweeters if they actually felt like they were tweeting more than last year? Feel free to share your thoughts in the comments.

Third, be reminded that this count does not capture those tweeps that tweeted frequently about the ISA, but didn’t use the #isa2015 hashtag (pro tip: if you exist, make sure to use #isa2016 if you want to be on next year’s top 20 list). I’d say this is somewhat unlikely, but it’s a cavet to keep in mind.

When do ISAlers tweet?

There is pretty shallow usage of the #isa2015 hashtag prior to the conference, although I do register some pre-conference buzz. This intensifies in the days leading up to the conference. The conference talk clearly peaks on the first day and then winds down a little over the course of the ISA. This pattern is different from last year when we saw the twee intensity climb over the course of the conference. Again, last year’s pattern may have resulted from the sketchy way I collected the tweets last year when I didn’t systematically download tweets prior to, and during the conference. I’m more willing to trust in this year’s data collection.

What do they talk about?

A fun exercise of last year’s analysis was a ranking of the most popular tweets. I summed each tweet’s favorite and retweet counts to arrive at somewhat crude proxy of a tweet’s popularity. I’ve created a list of the Top 20 #isa2015 Tweets. For readability reasons, I’ve included only the top 5 here. You find the full top 20 at the end of the post. Note that the number of favorites & retweets can change since I’ve made my analysis; the tweets are still online, after all, and can still get retweeted and/or favorited.

Again, let me make couple of observations (I’m referring to the full list of top 20 tweets here): yup, coffee is important for a conference, and the situation in New Orleans wasn’t particularly good. Next year’s organizers might want to look into this if they don’t want to risk Angry Academics (with capital “A”s, because they’re like the Avengers, just without costumes, and you don’t want to mess with coffee-deprived academics).

Also, a couple of really popular tweets addressed gender issues in the profession. I realize that some of them were made ironically or tounge-in-cheek. But their popularity reveals that they hit a nerve there. And from casually following my #isa2015 TweetDeck column, I know that there were many more tweets on that issue that didn’t receive as many favs + RTs. I think this really shows that there are still PLENTY of issues that we have to address to make not only ISA, but the entire profession (I guess on both sides of the Atlantic) a much more gender-balanced and overall less old-white-male dominated endeavor. Getting gender-balanced panels would be a start, I guess.

Other popular tweets were about presentational issues. I guess we all know that we academics can be really boring (and by “boring” I mean as in “rock”) when it comes to presenting our work—work that is actually pretty interesting but we way too often manage to make it sound like reading out loud safety instructions of a lawn mower (I explicitly include myself here). So I think we all should work harder at making better presentations, be a little better prepared, and maybe try to finally get rid of those tables on our slides if we do quant work. (To be fair, I saw great presentations at ISA as well—but Twitter being the snarky medium it is, is much faster at pointing fingers that giving praise)

Anything else?

Well… since I’m on the train right now and have another hour to kill, I might as well report some of the stuff that I’ve trying out with these tweets. 8000+ tweets are a lot of text to work with and naturally I was curious about what we could do with all this data (naturally for a social scientist, I guess. My sister only looked puzzled at me when I tried to explain to her what I was trying to do. Geekness galore!)

Following Jay’s forays into natural language processing, I was fiddling around with the code he used to analyze National Security Strategy texts. I know ISA tweets are way less cool stuff to play around with, but I did it anyway. What did I find out? Well, me being a total NLP-n00b, not much. I don’t want to bore you with word clouds, since they are apparently a really bad way to convey textual information. But I was intrigued by Jay’s R function that plots correlational associations of words in the text body (i.e. within tweets in my case).

Take the following plot for example.

We can observe that the word “coffee” (which has been established as an important conference theme) is most closely associated with the word “get” in tweets. So, the conference tweeters preoccupation with coffee has likely not been due to its abundance at the conference venue. It also reveals the impact many retweets can have on such correlational procedures: recall that the phrase “where did you get that coffee” has been retweeted 39 times. This considerably increases the instances in which “get” and “coffee” appear in the same tweet (retweets are collected as regular tweets by the twitter package). Nevertheless the overall correlation isn’t that high. It is .29, to be precise, much lower than the correlation of “new” with “Orleans”, which obviously correlates highly in the text body at .72.

One could play around with that function endlessly, but I’ll leave it at that. If you scroll down you find a link to the data + R code to tinker with it yourself.


Speaking of data + R code, let me close this exercise on a somewhat critical note on doing analyses like this one (and, even more relevant doing „serious“ research) with Twitter data. A major problem of the Twitters API’s time restrictions is that it limits the time you can go back to download tweet. As I publish this post, unless someone else happened to also download all the #isa2015 tweets, there is no way to independently reproduce the data collection process, even if I give the R code, simply because it wouldn’t return as many tweets (or none at all). You’d just have to believe me that I captured all the #isa2015 tweets and didn’t screw anything major when cleaning the data from Thai girl bands and indie filmmakers’ tweets. And you shouldn’t. For something as simple as this analysis, this probably isn’t a major problem. But there is more serious research being done using Twitter data. Unless Twitter doesn’t open up its databases for researchers (which it probably won’t because they can make way too much money with their database. And, yes, I’m aware of the Twitter research grants, but they don’t solve the general accessibility and replicability problem), this poses serious constraints on reproducing data collection Twitter research.

Still, I hope you enjoyed this little piece about #isa2015 as it happened on Twitter. Make sure to tune in next year and leave ideas, suggestions or rants in the comments or tweet me @felixhaass.


Top 20 Tweets