Social media is becoming more and more a part of the daily political process. Studies show that the number of people who see social media as a primary source of information is rapidly increasing each year. According to a recent Pew Research Center poll, 62 percent of Americans receive some news from social media platforms, and 18 percent very often use social media as a news source. More importantly, an increasing number of stories by traditional news outlets simply report tweets. For example, a list of funny tweets on the UK general elections, a list of tweets from world leaders condemning the horrific June 3 terror attack in London, or even an article on why Donald Trump did not tweet for 38 hours can be stand-alone news coverage in leading newspapers.
While social media is rapidly becoming an indispensable component of politics, an important aspect of social media is highly overlooked. This is the first time in modern history that political elites and ordinary citizens have an opportunity to directly communicate with each other. In fact, they frequently engage in conversations by replying to each other or by sharing, retweeting, or liking posts. As these social media users continue their conversations, they generate political text. Analysts perceive political texts as an important and valuable source of data that can be used to capture the author’s ideology. From a political science perspective, the ability to capture the ideology of elites and citizens using a common platform greatly helps in answering a very important question: Which political party represents its voters the most? In other words, which party’s ideological position is closest to that of its supporters, on a left-right ideology scale?
This question is critical because failure to recognize constituents’ preferences and modify party policies accordingly may result in problems related to democratic legitimacy and accountability. In fact, representation resides at the core of citizen-elite linkages in democracies. This linkage is shaped by a relationship in which voters delegate their rights to politicians to act in the political sphere, assuming that politicians will follow their preferences. Therefore, such a relationship entails correspondence between the ideologies of the citizens and their representatives.
In our ongoing research, we attempted to address this question by analyzing tweets by political elites and ordinary citizens in Turkey. The case of Turkey offers several advantages that do not exist in many other cases. It is one of the countries with the highest levels of social media usage. Moreover, Twitter is used by all social and political groups as a political platform, allowing for analysis of political ideologies. Similarly, when the Turkish government curtailed conventional forms of media, Twitter emerged as an alternative form of political communication in the period following the Gezi Park protests in 2013.
In order to measure the ideological positions of both political elites and citizens, we utilized a quantitative text analysis technique, Wordfish. Wordfish is a computer-based text-scaling algorithm. It is mostly used to position political texts on a left-right continuum based on the frequency of each word in each text. It has one important underlying assumption: political texts from similar ideologies tend to use similar vocabulary with similar frequency, while political texts from rival ideologies tend to use different vocabulary. For example, more liberal party leaders frequently use terms like climate, global warming, renewable energy and so on. In contrast, conservative party leaders tend to refrain from using these terms. On average, conservative politicians pay less attention to the environmental issues than the liberal ones, and instead of presenting their disagreement with environmental protection measures, they generally prefer stressing words associated with economic development such as industry, manufacturing, oil, coal and so on.
For this research, we obtained more than 4 million tweets posted between January 2012 and March 2016 by the candidates of the four major Turkish parties in the 2015 national parliament elections and the supporters of each party. We then grouped these tweets into eight groups of political texts (four texts for elites and four texts for supporters). We excluded non-political tweets (such as those about music, sports, etc.), and dropped stop words. We also removed words that were used within only one or two of the political texts. We then employed Wordfish to extract the ideological positions of the party elites and their supporters.
The scaling algorithm successfully placed the traditionally rightist parties and traditionally leftist parties on different ends of the political spectrum. As seen in Figure 1 below, the party elites and supporters of the conservative parties (the Justice and Development Party, or AKP, and the Nationalist Movement Party, or MHP) are all positioned toward one side of the scale, while the party elites and supporters of the traditionally liberal/secular parties (the People’s Democratic Party, or HDP, and the Republican People’s Party, or CHP) are positioned toward the other end. This result largely confirms the findings of previous studies on party positions in Turkey, and thus provides important support for the validity of the method.
We then calculated the differences between the elites and supporters for each party. The second figure demonstrates that HDP is the most representative party, as the difference between the ideological positions of its elites and supporters is the lowest of the four political parties. Since the corresponding difference is highest for MHP, MHP is found to be the least representative one. The data also show that the traditionally more rightist parties (AKP and MHP) are less representative of their core constituencies than the traditionally leftist parties (CHP and HDP).
These findings imply that leftist party elites in Turkey tend to use rhetoric that is more similar to that of their constituency when compared with rightist parties. This may be an indicator of a higher level of attentiveness by the leftist elites to the concerns and preferences of their support base. This suggests a higher level of legitimacy of leftist parties from a political representation perspective in Turkey.
The proposed method has also some limitations. For example we acknowledge that social media users are not a representative sample of the whole population. However, given that our primary interest is comparing representativeness across parties rather than trying to find a national level of representation, we believe this does not severely influence the validity of our findings. We also believe that as social media usage continues expanding, it will soon become more representative of the whole society. Hence, in the near future, researchers will be able to respond to even more research questions by using the texts of social media posts.
Figure 1. Ideological Placement of Turkish Party Elites and Supporters Based on Twitter Data
Figure 2. Ideological Distance Between Turkish Elites and Supporters, By Party
Abdullah Aydogan is a research scholar in the Baker Institute Center for the Middle East. Tayfun Tuna is a postdoctoral research fellow in computer science at the University of Houston.