Expanding the Scope of Selective Exposure: An Objective Approach to Measurement of Media Ideology

Abstract

The literature on selective exposure has made inestimable advances in our understanding of media and its effect on politics and the general public. However, much of the research on this topic has relied on potentially inaccurate assumptions. In our paper we apply an open-source, publicly available, high-dimensional measurement of meaning through word co-occurrence context (Shaoul & Westbury, 2010), which has historically been applied to questions of semantic relationships between words. This method allows scholars to avoid pitfalls from previous assumptions and determine previously unknown ideological positions of previously unknown sources. We demonstrate the validity and range of this method and provide a series of best practices for scholars who wish to employ this tool in their own research. Our method will ultimately expand the variety of research questions available as well as improving inferences, opening up new lines of research for scholars studying media consumption and political behavior.

Notes

¹There is also an extensive literature that argues dependence on government sources introduces bias into media (e.g., Covert & Wasburn, 2007; Dickson, 1994; Smith, McCarthy, McPhail, & Augustyn, 2001).

²One of the seminal works on this measure, Congress: A Political-Economic History of Roll Call Voting (Poole & Rosenthal 2000), has been cited more than 2,200 times according to a Google Scholar search conducted on February 15, 2013.

³We choose to associate larger numbers with the Republican party so that it matches the coding of DW-NOMINATE scores, which use positive numbers for the Republican party and negative numbers for the Democratic party. This coding is arbitrary, but visually fits the “left-right spectrum” we usually think of when we think about ideology in American politics.

⁴This finding, while potentially not what the reader may expect, fits with the time period studied here. Testing this is beyond the scope of this paper, but because of the emergence of the Tea Party and reshaping of the Republican party during the time period we study and a relatively stability of the Democratic platform once they took over majority status in Congress and won the presidency, we are not surprised that Republicans are considered the more “changeable” of the two parties.

FIGURE 4 Correlation between DW-NOMINATE score and the stable/changeable distance ratio (derived from a corpus of UseNet postings), for 115 Democrat and 87 Republican members of Congress.

⁵We reverse the numerator and denominator for immigrants to parallel our interest group rating. While most of the interest groups advocate for the target group, the U.S. Border Control supports restricting immigration which means higher numbers represent a less positive valence toward immigrants. We have calibrated our anchors to reflect this for ease of interpretation, but it does not substantively change our results in any way.

⁶Euclidean Distance is defined as the square root of the summed squared differences between all elements in vectors of any length. It is normalized here by computing all distances, subtracting their average, and dividing by their standard deviation. This normalization makes the units in Table 2 and 3 directly comparable.

⁷“Directional” and “proximity” used in this context have specific meanings to the literature on voting behavior. “Directional” means that voters cast their ballots purely based on partisan cues (a Republican votes for a Republican, a Democrat votes for a Democrat) (Rabinowitz & MacDonald, 1989), while “proximity” means that voters cast their ballot for the candidate who is closer to their ideal point in a dimensional space (Hinich & Enelow, 1984).