The downing of Malaysian Airlines Flight 17 on July 17, 2014 over the Crimean Peninsula was an international tragedy. It is believed that the plane was downed by a surface-to-air missile, killing all 283 passengers and 15 crew on board. With the debris field within rebel occupied territory, it wasn’t until July 23, 2014 –nearly a week later–before the victims could be identified and returned to their families.
During that week, the downing of Malaysian Airlines Flight 17 (MH17) became an international political brouhaha. Aid groups and independent rescue workers struggled for access to the crash site, while political leaders took sides. At the center of the debate was Russia’s President Vladimir Putin. Headlines asked if Putin was somehow culpable for the flight downing, speculating whether his government had provided the actual surface-to-air missile. Some organizations even called for Putin’s head, suggesting that the international community should step in.
It would be a stretch to say that Putin was popular before the tragedy. Still, the anti-Putin rhetoric increased a few decibels in the week following the event. For me, it became difficult to differentiate between saber-rattling from those with very specific agendas, and the true voice of the people. Following the downing of MH17, did world opinion move against the Russian president? To measure this, I turned to the “pulse of the planet”: Twitter.
Correlation between MH17 and Putin
To see if MH17 had an impact on Putin’s popularity, I first needed to establish a correlation between the crash and conversations mentioning Putin. I created a collection of 12,459 randomly sampled Twitter accounts that had tweeted at least once between July 1, 2014 and July 28, 2014. These dates were chosen to give roughly two weeks of tweets on either side of the crash date (July 17). This gave me 3,096,921 tweets from before and after the crash.
I then measured the Share of Voice (SoV) for tweets mentioning Vladimir Putin and tweets mentioning MH17. Share of Voice is the estimated fraction of the global Twitter population mentioning either of these terms in a single day. It is typically represented in Parts Per Mil (ppm) which is the number of people out of 1,000 making mention. So if the Share of Voice for Vladimir Putin is 10 ppm, that means roughly 10 people out of 1,000 referenced Putin on that particular day. Extrapolating to the global Twitter population of 271million active users (as of June 30), this would be approximately 2.71 million people that day.
Figure 1 shows the daily Share of Voice for Putin and MH17 before and after the crash. Inspecting the graph, Tweets mentioning Putin are ~3 times more likely between July 17, 2014 (the date of the crash) and July 23, 2014 (when the remains of the victims were returned home). During this period, mentions of MH17 are also high, dropping considerably after the victims were repatriated. A Pearson’s Correlation test between mentions of MH17 and Putin confirm the correlation (p-value = 1.4×10-7). The data confirms that mention of Putin is highly correlated with the crash of MH17. It is also evident, given there was no conversation about MH17 prior to the tragedy, that the crash caused the increase in mentions of Putin.
Figure 1: Daily Share of Voice for Mentions of MH17 and Putin
Having established that the global mentions of Putin dramatically increased after the crash, I then looked at the content of these increased mentions. On Twitter, a common method for analyzing large numbers of tweets is through sentiment analysis, which assigns a score between -1 (very negative sentiment) to +1 (very positive sentiment) to each tweet. Aggregating these scores together can provide insight as to how positively or negatively people in our sample see Vladimir Putin. By tracking the sentiment day-by-day, I can tell if the MH17 crash caused people to feel more or less positively towards Putin.
I trained the sentiment model from the 3,096,921 tweets by looking for positive and negative emoticons. Then I used the sentiment classifier to assign an average daily score to Putin mentions, and mentions of MH17. Results are shown in Figure 2. The average sentiment has quite a bit of variation from day to day. To help visualize trends, I plotted a best fit curve against the data.
Figure 2: Daily Sentiment for Putin and MH17
Surprisingly, there is no evidence that the crash affected Twitter sentiment for Putin! From the graph we can see that once the repatriation of the victims’ remains finally got underway, the sentiment around MH17 saw a sharp uptick, but towards Putin it is essentially flat during the period under study. In fact, it looks like Putin’s global sentiment began to dip about a week before the actual crash.
Segmentation by Location
Having seen that globally the sentiment for Putin did not change following the crash, I wondered how the sentiment would vary based on location of the tweets. The tragedy was very much Europe-focused. Other locations were less directly affected by the events. So I segmented the tweets based on their continent of origin, and looked at the change in sentiment both before and after the crash (Figure 3).
Figure 3: Sentiment by Continent Before and After the Crash
In all continents except Europe, the average sentiment actually increased after the crash. Also, the variation in sentiment is much larger. For example, in North America (prior to the crash), most of the Tweets had a sentiment near -1 (with a handful of outliers). In North America the opinion of Putin was mostly uniformly negative across those sampled. After the crash, the average sentiment was still close to -1, but there is far more variation in how people feel. A large variation in sentiment indicates that people were divided in their opinion of Putin, with some condemning the Russian leader while others taking a more neutral stance. This pattern repeats itself for South America, Asia and Australia.
Europe, on the other hand, is quite different. Before the crash of MH17, Europe had the most positive sentiment towards Putin and the largest variation. The opinion of people in Europe towards Putin was quite divided, with some being against Putin and others being for Putin. However, after the crash, there is not only a drop in the average sentiment, but the variation in sentiment also decreases. The data shows that the tragedy of MH17 has begun to solidify opinion in Europe against Putin.
So did the downing of Malaysian Airlines Flight 17 change world opinion of Russian President Vladimir Putin? It has had an effect, but not what I was expecting. Within Europe, the crash has galvanized opinion against Putin, while in the rest of the world it has opened debate. The net affect is that global public opinion did not change but the debate about Putin shifted from the European to the international stage.