3 Methodology
3.3 Qualitative analysis
The study will also include a qualitative textual analysis of some of the articles coded during the content analysis. In a textual analysis, the researcher attempts to discern the likely interpretations that could be made from that text (Lester‐Roushanzamir & Raman, 1999; A. McKee, 2003). Researchers use textual analysis to consider not just the words used, but the motivations behind the use of those words and how the words might be interpreted (McKee, 2003). Textual analysis offers a chance to examine the ways in which ideas or events are given importance as well as how readers might understand and respond to the text (Shah, 1999). The textual analysis should provide an additional layer of understanding regarding the use of
unnamed sources and the ethical justifications for that use.
Since the researcher is conducting the main body of coding, he identified unnamed sourced newspaper articles that would prove helpful in discussion. After the content analysis was completed, the articles were examined to help expand and comment on the quantitative findings. The author also examined the specific use of unnamed sources within the framework of the ethical theory of utilitarianism, for which the textual analysis will provide support. The textual analysis is offered in the discussion section.
3.4 Summary
The research design outlined in this section should provide a rich set of reliable and valid data regarding the use of unnamed sources over time. With data analysis—both descriptive and statistical—the research questions should be easily answered. The data allows for discussion over whether the frequency of unnamed source use has changed over time. It also allows us to
see whether journalists in earlier decades took different approaches toward identifying
sources, independent verification, and explanations for anonymity. The findings will add to the discussion regarding the balance between the reporter’s mission to report the news and the need for transparency to establish credibility.
4 Results
With the literature review and methodology set, the researcher embarked upon the coding and analyzing. The process went smoothly except for a couple of impediments which will be discussed in the following paragraphs.
4.1 Intercoder reliability
A second coder was trained on the coding protocol and given specific instructions regarding how to make the best judgments in specific situations. The coder examined a random sample of 127 articles to test for intercoder reliability. Four of the articles were excluded from the sample because of errors with the story ID code. The remaining 123 articles represent roughly 10 percent of the total sample. The sample achieved an acceptable overall simple agreement rate of 85.4 percent. Table 4.1 shows the results for the individual variables, both as simple agreement and Scott’s Pi.
Table 4.1: Intercoder Reliability
Variable Percent Agreement Scott’s PI
Newspaper 100 1 Story ID 100 1 Story Type 57.3 0.440 Unnamed Sources 94.2 0.882 Attempt at ID 82.7 0.624 Independent Verification 82.7 0.601
Explanation for Anonymity 98.3 0.791
Number of Sources 68.03 0.487
In reviewing the data set in Table 4.1, two variables lowered the reliability ratings tremendously—story type and number of sources. Despite copious instructions, the coders often disagreed when categorizing each story. For instance, a new drug approval could be coded as “government” or “health,” or a story about an Army general talking to a foreign official could be coded as “military” or “foreign affairs.” The simple agreement for this category achieved just 57.3 percent. Another troublesome variable was Number of Sources. The coders often disagreed on just how many unnamed sources were included in an article. Often, long articles would cite several different unnamed sources and refer to them in different ways, leaving the coder guessing at the exact number. The number of sources category achieved a simple agreement of just 68 percent. Fortunately, neither category represents an important part of the findings of the dissertation. When those variables are removed from the overall simple agreement calculation, the level of agreement rises to 93 percent.
As can be seen in the table, only two of the Scott’s Pi calculations reached acceptable levels—Anonymous Sources and Explanation for Anonymity. The other Scott’s Pi numbers were far lower, below levels considered acceptable.6 The low Scott’s Pi numbers are surprising given the corresponding simple agreement. For example, Attempt at ID and Independent Verification both achieved 82.7 percent simple agreement, yet their Scott’s Pi numbers were in the low 60s.
Several researchers consider Scott’s Pi and other chance‐correcting coefficients as an inadequate method to test intercoder reliability in certain circumstances. Content analysis author Neundorf notes that percent agreement is actually the most appropriate measurement in instances “wherein each pair of coded measures is either a hit or a miss” (2002, p. 149).
6
Several researchers in peer‐reviewed journals have opted to not release chance‐corrected coefficients such as Scott’s Pi or Cohen’s Kappa because they are mathematically biased against binary variables (Bachen, Raphael, Lynn, K. McKee, & Philippi, 2008; Xenos & Foot, 2005). In effect, these authors report that the calculations over‐correct for chance agreement for categories with only two possible responses. Given the work of these other scholars, this researcher feels the Attempt at ID and Independent Verification variables are sufficiently reliable given the 82.3 percent simple agreement. Furthermore, the lowest reliability rating reported by Martin‐Kratzer and Thornton (2007) was .83, and those authors never said whether the figure was simple agreement or a chance‐corrected coefficient.