Sources of media bias in foreign policy reporting – the case of US media

Once these initial 22 nodes were identified, their links were explored through the LittleSis database of companies, to identify board members, and their shared seats (i.e. other companies they are members of the Boards of Directors) Database of corporate ownership. Source: LittleSis. (2019, April 11). Lists. Opgehaald van LittleSis: https://littlesis.org/lists. Through cross-referencing, additional 32 boards were identified. Each Board Members' links were allocated a weight of 1. If two Board Members from one company sit on the same board of another company, the weight was adjusted to 2, and so on. This process created an Adjacency Matrix which was fed into Gephi - a Social Network Analysis (SNA) tool, for analysis and graphic representation. This tool shows a variety of metrics of centrality (a measure of how important a node is) of individual nodes in the network as a whole.

Discourse Analysis of Media - Methods and Methodology

The second hypothesis is:

Hypothesis 2: media reporting is biased towards special interests' objectives in foreign policy, by over-emphasizing foreign government abuses in states closed to US special interests, as compared to open states

The second part of the analysis will be conducted in the form of case studies of media coverage of foreign relations with two countries - Venezuela and Saudi Arabia. Media opinion texts will be selected for the period between January 20, 2015 to January 20, 2019, which represents the last two years of the Obama regime, and the first two of the Trump regime. Only opinion and editorials will be analyzed, as these represent the opinion of the media company as a legal entity, and represents content directly written or promoted by the highest positioned employees of each media company. While this may be counterintuitive at first glance, the hypothesis of this research is that the media as a whole are influenced by special interests, which, as we have seen in the SNA section, they share to a great degree. As such, if we find a large conformity of the opinion articles in media the interests of special interests, then our hypothesis will be accepted. On the other hand, the hypothesis is not accepted if the media content is independent of the concerns of special interests. In other words, opinion and editorial pieces are designed to bring about a political debate, if however, there is no substantive difference in positions, even though neutral outsiders do take substantively different positions, then the media can be seen as biased overall. The final results will then be compared with the SNA predictions, to see if they are upheld.

The media content will be based on editorials and opinion pieces, which represent the views of the editorial board and hence this will allow us to look at the systemic bias in the top tiers of media companies, allowing us to compare the results with those of the SNA, i.e. if they are actively representing the political preferences of shareholders linked to special interests. We will control for ideological leaning of media. If we find that media does in fact use narratives that are I the interest of special interests, we will accept the hypothesis.

Sources Narratives and Collection of Discursive Data

For the analysis, a discrete set of primary “discursive” sources must be selected. The first step is to assemble a set of criteria for the analysis of texts, in the form of words and phrases that indicate the leaning of media reports on the issues observed. Since we are looking at normative positions of the media towards two states, the news will be analyzed for key terms, which can show the political leanings of the paper towards the cases. Furthermore, media content will be assembled into datasets Raw data based on title, author, date, text, etc. Full database upon request.. Media is one of the forms of public information, and it is possible to access it freely and hence to establish data sets for the purpose of this study. Once the dataset is established, a word frequency analysis will be conducted to extract quantifiable data from the datasets. These will be analyzed with word analysis methods, to gain insight into the narratives presents by each of the media. The same steps will be performed on texts by the key actors identified through SNA in the previous chapter, and finally “neutral” actors. In the final step, the data will be compared across different actors, and with controls.

Case Selection - Venezuela and Saudi Arabia

The case studies will be based on two countries: Venezuela and Saudi Arabia. These two cases were selected based on the level of interest of the media on both countries, and the fact that both are oil producing countries. The first, however is a country that has moved away from international ownership of its oil production facilities, while the second has done the opposite. This is true for other sectors as well, for us the focus is on defense trade, where we see the same trend, with Venezuela decreasing purchases from the US, and increasing purchases from Russia and other competitors - needless to say, the opposite is true for S. Arabia. Finally, Saudi Arabia has invested extensively in the US, while Venezuela has done far less so. We will look at these issues in more detail.

In terms of geostrategic importance, both countries are key to US interests, primarily due to oil reserves, but also as key regional powers. Both countries have a comparable GDP, counting in the $600 to $700 billion range. Both countries have a mixed relation with US in the past, with Saudi Arabia being an uneasy ally during Arab - Israeli conflicts, and Venezuela with the 1970's nationalization of oil production, which saw international oil companies, including US ones, substituted by national ones, under state control. However, during the 1980's and the 1990's in particular, both countries allied themselves with the US more closely. Saudi Arabia agreed to Nixon's petro-dollar deal, which opened the doors for Saudis to use the excess petro-dollars to buy US debt, or invest in the country itself:

“The basic framework was strikingly simple. The U.S. would buy oil from Saudi Arabia and provide the kingdom military aid and equipment. In return, the Saudis would plow billions of their petrodollar revenue back into Treasuries and finance America's spending.” (Wong, 2016

Wong, A. (2016, May 31). Bloomberg. Opgehaald van The Untold Story Behind Saudi Arabia's 41-Year U.S. Debt Secret: https://www.bloomberg.com/news/features/2016-05-30/the-untold-story-behind-saudi-arabia-s-41-year-u-s-debt-secret)

However, the Saudis were also concerned about dealing with the US in such a way, particularly due to the rising salafi far-right voices, which were opposed to the US on principle. Part of this group would later fund Al-Qaeda activities, and possibly terrorist groups linked to the 9/11 attacks. (Benjamin, 2016

Benjamin, M. (2016). Kingdom of the Unjust. New York: OR Books. and Wong, 2016

Wong, A. (2016, May 31). Bloomberg. Opgehaald van The Untold Story Behind Saudi Arabia's 41-Year U.S. Debt Secret: https://www.bloomberg.com/news/features/2016-05-30/the-untold-story-behind-saudi-arabia-s-41-year-u-s-debt-secret)

Similarly, Venezuelan governments in the 1980's and 1990's were investing much of the petro-dollars in the US, while at the same time, they opened up the nationalized oil reserves (under the holding company PDVSA) to foreign shared rights to extraction. This meant that a portion of the income on oil was transferred to US companies directly Lander, L., & Lуpez-Maya, M. (2007, September 25). Venezuela's Oil Reform and Chavismo . Opgehaald van NACLA: https://nacla.org/article/venezuelas-oil-reform-and-chavismo . During this period, PDVSA directors had developed a strong relation with the US government. With the election of Hugo Chavez, the deal with the US companies was renegotiated, particularly a number of reforms to the sector.

“The reform puts the brakes on the trend towards privatization of PDVSA that had been developing in the previous years, without blocking private investment in the sector” (Lander & Lуpez-Maya, 2007

Lander, L., & Lуpez-Maya, M. (2007, Sep 25). Venezuela's Oil Reform and Chavismo. Opgehaald van NACLA: https://nacla.org/article/venezuelas-oil-reform-and-chavismo)

These reforms were unpopular in the US governments, particularly due to the limits imposed on privatization, but also due to a longstanding policy of getting Venezuela out of OPEC, which Chavez reversed, committing strongly to the organization. This is evidenced by the attempted coup in 2002, siding with the military and civilian coup-leaders Reeker, P. T. (2002). Venezuela: Change of Government. Washington, D.C.: US State Department, among which were pro-privatization ex-heads of PDVSA. (Lander & Lуpez-Maya, 2007 Ibid.). Furthermore, Venezuela started increasingly trading oil with China, a move seen as trying to gain market independence from the US AssociatedPress. (2005, August 21). Venezuela opens oil office in China. Opgehaald van ChinaDaily: http://www.chinadaily.com.cn/english/doc/2005-08/21/content_470826.htm . Here the main divergence over time with Saudi Arabia was established.

Another key sector is the arms imports from the US, since it is a key indicator of the level of alliance between two countries. In our cases, we will look at the level of trade from the US and its competitors over a period of 40 years, to see the long term trends. In the bellow chart, we can see that Saudi Arabia has variably purchased arms from the US, making up to 96% of its total arms purchases, but averaging around 59%. While the trend varies a lot, it is clear that the US is the most important arms trade partner to S. Arabia, as on average, it accounts for more than half of the Saudi arms imports. Saudi Arabia has not imported arms from Russia in the given period, according to the database. In regards to Venezuela, however, the graph shows a much different trend, with a 17% average (or 25% average before the coup attempt). At the same time, Venezuela starts to trade with Russia in arms in the 1990's, but from 2004, Russia becomes one of the major arms trade partners, with an average of 42% of all arms trade conducted by Venezuela.

Figure 2 - Arms purchases by Venezuela and S. Arabia

We can see that they are very comparable, with Venezuela leading slightly in terms of press freedom. As such, is we assume that the media is not bias if issues are treated by the authors with an overall average frequency of political terms outlined above. If, however, the open state of S. Arabia is less held up to scrutiny on the four categories of issues described above, then we can assume that the media is positively bias towards them. Finally, if Venezuela is held to lover scrutiny on the issues, the hypothesis will also be rejected.

The text is processed in a way that allows the RapidMiner software to understand the text better. The text processing involves tokenization, where the text is split into a sequence of tokens, in this case individual words. Subsequently, the tokens were filtered for stopwords, i.e. removing prepositions, conjunctions, determiners, and other words that clutter the text, without being descriptive of notions held by the narrator of the text. Along with other minor adjustments, the final operation is to create n-grams, which are sets of two tokens. This last step allows us to see phrases of two tokens, such as “human rights”. The resulting dataset consists of all single and two tokens, with a numerical value attached to each token representing the keyword density in which they appear in each text, and an average of frequencies across all texts of a dataset. There are two datasets, representing all opinion articles written in the month subsequent to the 10th of January in Venezuela since (i.e. the start of the presidential crisis), and since 2nd of October in the case of Saudi Arabia (murder of Khashoggi). The codding words and phrases were then extracted from the broader dataset, creating the final dataset, which can be seen in Appendix I. Here is a few examples from each of the five categories, where zero represents no mentions, and the higher the keyword density, the more often the token was used, on average in the country based dataset:

Table 3 - Example results table