Introduction: Political Tweets

Politics has always been concerned with using any available method of mass media to disseminate its message, and social media is no exception. In Malta, the most popular social media sites, by share of their usage are, according to a May 2018 Misco survey:

• Facebook (87% of the population)
• Google+ (50%)
• YouTube (46%)
• Instagram (24%)
• Twitter (12%)

I’ve yet to come across anyone who actually uses Google+ regularly, so those answers are probably people who just have an account. And since YouTube and Instagram aren’t discussion oriented, you can make the case that most political discourse is on either Facebook or Twitter. Since by default most of what is said on Twitter is public, and Twitter has an API which is a dream to use in comparison to Facebook’s, this blog post will look solely at public Twitter data, obtained by querying the Twitter API, for which I used this fantastic tutorial.

Who has the largest number of followers?

Whose Tweets are the longest?

What’s the average word count of our users’s tweets?

## # A tibble: 9 x 2
##   screen_name     Average_Word_Count
##   <chr>                        <dbl>
## 1 AzzopardiJason                26.8
## 2 JosephMuscat_JM               23  
## 3 adriandeliapn                 22  
## 4 KurtFarrugia                  20.7
## 5 SimonBusuttil                 20.3
## 6 presidentMT                   18  
## 7 KonradMizzi                   17.4
## 8 chrisfearne                   12.2
## 9 ClydePuli                     11.6

Most users manage an average of around 20 words per tweet. The average Jason Azzopardi or Joseph Muscat tweet is a bit longer, the average Clyde Puli or Chris Fearne tweet a bit shorter.

Here’s what that distribution would look like as boxplots, with the thick middle line denoting the median, the coloured bars to either side of it showing the middle 50% of the data. The spread of the data shows that Joseph Muscat, Jason Azzopardi, and Adrian Delia’s word count in tweets is highly variable. For the other users, except the odd outlier (represented by the dots), not so much.

Favourites and Retweets

So far we know that Joseph Muscat and Jason Azzopardi are the most prolific twitter users when looking at volume. But how well do they engage their respective audiences?

One measure of how widely the message is disseminated is how much times a tweet has been favourited or retweeted.

Again, Joseph Muscat is the most succesful user in terms of getting his message across, with his tweets being favourited 97 times and retweeted 40 times on average. This is nearly double of what the runner up, and his direct competitor, Adrian Delia manages.

Other Twitter users who seem to succesfully get their message across include Chris Fearne, Konrad Mizzi, Kurt Farrugia and Simon Busuttil. The violin graphs below show the density of the distribution. Taking Simon Busuttil’s retweets as an example, we can see that the majority of his tweets has less than 100 retweets. A large portion of that had even less than 50, but one or two tweets had over 450 retweets.

## # A tibble: 9 x 3
##   screen_name     avg_faves avg_rts
##   <chr>               <dbl>   <dbl>
## 1 adriandeliapn      30.3    19.2  
## 2 AzzopardiJason      7.57    1.41 
## 3 chrisfearne        18.3     6.87 
## 4 ClydePuli           0.784   0.547
## 5 JosephMuscat_JM    96.8    40.7  
## 6 KonradMizzi        22.0    11.6  
## 7 KurtFarrugia       18.3    11.1  
## 8 presidentMT         7.44    3.15 
## 9 SimonBusuttil      24.4    13.9

As a side note, while creating these graphs, I realised that they looked more or less the same. And indeed, when you plot out the relationship between the number of retweets and favourites for each tweet, it’s linear. This makes sense, as to an extent, these two things are measuring the same phenomenon: popularity of the tweet.

Word Frequencies

So what are individual users more likely to tweet? Again, by using the tidytext package, which features a twitter tokenizer, we can find out. I’ve been fascinated by the simple elegance of plotting frequency on scatter plots from the moment I saw them in Julia Silge and David Robinson’s book, so this will be a shameless copy.

The principle is this: words that are equally likely to be used by both users will be near the center blue diagonal line. Words that one user prefers more than another will be off to the side. The units of the axis are percent of frequency. So a word with 1% frequency will mean that out of a hundred words in every tweet, that user is likely to use it once. That probably says a great deal about the level of information condensation in tweets. Lets start with the most logical comparison.

Sentiment of Tweets

Using a similar procedure as the one detailed in my article examining local news articles, we can quantify and visualise how much positive or negative sentiment a tweet has.

Here is what the mean sentiment score of every tweet in 2018 looks like, using the AFINN sentiment lexicon, with each dot being a tweet, and the light grey line being a perfectly neutral tweet.

A few things readily stand out:

• Chris Fearne, Joseph Muscat, Konrad Mizzi and the President have, generally, many more positive tweets than negative ones.
• Adrian Delia and Kurt Farrugia adopt a balancing act, being negative and positive in equal proportions.
• Simon Busuttil and Jason Azzopardi are more lopsided towards negative tweets than positive ones.

And here’s what the polarity scores per user look like after passing all the tweet test through the QDAP package’s polarity function. Any dot left of the vertical line passing through a polarity of 0 means that the average polarity of that user is negative, and, likewise, anything right is positive.

Now to be clear, the job of opposition figures is to criticise, while that of individuals in government is to promote their work, so the results are hardly surprising… save for two things. Adrian Delia’s relatively ‘positive’ rating, and Kurt Farrugia’s relatively ‘negative’ one. So what gives?

I think what’s happening is fascinating. From the analysis of word frequencies, we know that Kurt Farrugia’s is quick to intercept criticism and heckling in an attempt to counterbalance. This is despite the fact that most of these tweets set out to target Joseph Muscat’s account. Kurt Farrugia’s twitter presence allows Joseph Muscat’s account to not engage in these discussions himself directly. Instead, Muscat tweets about the successes and achievements of his government, giving him lopsidedly positive tweets.

Similarly, Adrian Delia’s twitter presence is the most sombre of all the PN associated accounts we’ve looked at. He to, appears, to be more than willing to allocate the trench warfare to his deputies.

Before this gets too long…