The words and language used every day reveal who we are and what we want. They translate our internal thoughts and emotions to others as means of communication. In our previous posts, we analyzed candidates’ Facebook data to understand topics being discussed and how candidates interacted with the public. In this post we will discuss the evolution of the leading Republican and Democratic candidates’ campaign on their Facebook pages since they announced their candidacy. We have deepened our analysis to include linguistic cues to understand the personal and psychological attributes of commenters.
To do so, we utilized topic modeling to discover topics of candidates’ posts, and the Linguistic Inquiry and Word Count tool to analyze the linguistic and psychological indices of comments. The algorithm compared each comment to a dictionary file containing words in psychology-relevant categories, searching posts and comments for relevant words, and then assigned them to one of the categories. Next, the indices were scored from 0 to 100 based on the percentage of all words in the comment. For example, if a comment has 100 words and three of them express a sad tone, then it has a three percent score on the sadness index. Our dataset is comprised of 18,726 posts and 18,500,013 comments spanning a 17-month period. For each candidate’s commenters, we computed the following indices over the following periods:
For the analytic thinking index, commenters across the three periods are presented in Figure 1. In this figure and subsequent figures, the blue bars represent period 1, orange for period 2, and green for period 3. Higher scores suggest more formal, logical, and hierarchical thinking patterns. This index is important for revealing a person’s educational attainment. According to one study, “When Small Words Foretell Academic Success: The Case of College Admissions Essays,” people with higher analytic thinking index scores tend to earn higher grades in college.
Across the three periods, the commenters of Democratic candidates showed higher scores compared to those of Republicans commenters, with Clinton commenters scoring the highest on average. A noticeable phenomenon in this index is the improvement of averages as we move from period 1 to 3. As candidates drop out, there is more attention available for getting one’s message out.
Comparing Trump and Clinton commenters across the three periods, we find Trump commenters are typically below the average score, while Clinton commenters are again significantly higher than the average. The scores of Trump’s commenters have improved in period 2 and 3 which could be justified by the migration of dropped-out Republican candidates’ supporters to him. Clinton’s commenters have scored high scores that are above the average across the three periods.
Figure 1: Analytic index of candidates’ commenters
The clout index reveals the confidence and leadership that people show in their writing. The averages of first and second periods are similar, while the last period has a lower average. In the last period, both Trump and Clinton have been attacking each other which resulted in people being less confident about their candidate. For example, viewers became less confident when Trump uses the email controversy to attack Clinton, and Clinton uses the tax returns issue to mock Trump. Defaming your opponent is a well-known psychological manipulation strategy that is common in the presidential campaigns.
Generally, commenters of Republican candidates did well in this index compared to their Democratic counterparts. Among Democratic candidates and across all periods, commenters of Sanders always scored lower scores compared to Clinton’s commenters.
Figure 2: Clout index of candidates’ commenters
In the authenticity index, when people reveal themselves in an authentic or honest way, they use language that is more personal, humble, and vulnerable. Higher scores suggest honesty while lower scores suggest deception. When comparing Clinton and Trump commenters, Clinton commenters scored the lowest scores across the three periods while Trump’s commenters showed around average scores.
Across the three periods, commenters of Sanders scored higher scores compared to Clinton’s commenters. The averages of this index are decreasing as we move across the periods which suggest less honest and more deceptive statements.
Figure 3: Authenticity index of candidates’ commenters
Analyzing commenters’ emotions of anger, anxiety, and sadness revealed interesting findings. Across all periods, Republican candidates’ commenters showed higher scores in anger and sadness indices. In the anxiety index, Clinton commenters scored higher in the last two periods compared to Trump’s commenters. One possible justification is the fact that Trump has been attacking Clinton more frequently once his Republican counterparts exited. Those attacks likely raised this index for Clinton’s commenters. Offline events trigger people’s emotions, and this is noticeable in the second period where the high average score and higher scores from Republican commenters are due to the Brussels attacks on March 22, 2016. Similarly, Sanders’ commenters scored the highest level of sadness in the last period due to violence at the Nevada Democratic Convention.
Figure 4: Anger index of candidates’ commenters
Figure 5: Anxiety index of candidates’ commenters
Figure 6: Sadness index of candidates’ commenters
Reactions, linguistic, and psychological indices of candidates’ commenters reveal interesting findings. Democrats’ commenters scored the highest scores in analytical and anxiety indices. Republicans’ commenters scored the highest scores on clout, authenticity, anger, and sadness indices. The value in understanding these commenter attributes is to shed light on the type of supporters each candidate attracts, campaign strategies that affect supporters of opposition (i.e. email controversy and tax returns) and lastly, which events trigger the commenters’ tone. This gives us a clue of what to expect going into the final period of the 2016 elections.
Vikash Bajaj and Kendra L. Smith of Arizona State University contributed research to this post.
Facebook is a donor to the Brookings Institution. The findings, interpretations, and conclusions posted in this piece are solely those of the authors and not influenced by any donation.