The Influence of Partisan Motivated Reasoning on Public Opinion

Our work also adds to studies of how party endorsements in general affect public opinion (e.g., Arceneaux 2008; Bullock 2011; Nicholson 2012). Consider, for instance, Bullock’s (2011) recent paper, which tests the effects of a partisan endorsement on support for a policy by varying the availability of a source endorsement. He concludes (2011, p. 512), “party cues are influential, but partisans… are generally affected at least as much—and sometimes much more—by exposure to substantial amounts of policy information.” What Bullock does not probe deeply, however, is the conditions under which partisan endorsements are likely to slant evaluations.

Note that motivated reasoning encompasses a range of distinct goals, including defending prior opinions, impression motivation, and behavioral motivation (see Kunda 1999), but here we follow political science work to date focusing on directional and accuracy goals.

These various moderators somewhat contradict Taber and Lodge’s (2006, p. 767) conclusion that: “despite our best efforts to promote the even-handed treatment of policy arguments in our studies, we find consistent evidence of directional partisan bias…Our participants may have tried to be evenhanded, but they found it impossible to be fair-minded.” Of course even Taber and Lodge themselves find moderating effects of opinion strength and sophistication (also see Druckman 2012).

This can be accomplished in a variety of other ways, with the underlying rationale being to increase, “the stakes involved in making a wrong judgment or in drawing the wrong conclusion, without increasing the attractiveness of any particular opinion” (Kunda 1990, p. 481). One approach is to inform respondents that their decision is important, will be judged by peers, will have to be justified, will be made public, or will affect someone else (also see, e.g., Tetlock 1983; Tetlock et al. 1989; Lerner and Tetlock 1999; Tetlock 1986, all of whom do not explicitly look at social expectations but use it as a clear implicit component of their treatments).

As will become clear, we follow this approach (i.e., inducing participants to believe they will have to justify their responses). This approach differs from the one taken by Taber and Lodge (2006, p. 759), who ask respondents to, “view information in an evenhanded way so [as to] explain the issue to other students.” The potential problem with not asking explicitly for general justification is highlighted by Lord et al. (1984) who find that inducing people to form accurate preferences requires not only encouraging them to be unbiased, but also inducing them to justify their opinion. Taber and Lodge’s manipulation asks respondents to put their prior opinions aside and requires them to “explain the issue” to others. However, individuals may have understood this to mean that they need to present some facts to others; they may not have been induced to consider alternative viewpoints or justify their opinions. This is why we follow this other experimental work by asking respondents to justify their specific opinions (e.g., Redlawsk 2002; Tetlock 1983). Indeed, Houston and Fazio (1989, p. 65) explain that removing attitudinal slant requires “directing people to focus on the nature of the judgmental process” (also see Creyer et al. 1990; Lerner and Tetlock 1999).

We are careful here because such an endorsement could work to generate something akin to an accuracy goal given that conflict can generate elaboration (e.g., Chong and Druckman 2007) which is what we posit; however, it also is possible that the endorsement just leads to a moderation of opinions. We thank an anonymous reviewer for this point.

We thank Laurel Harbridge for pointing out the important distinction between unanimity and cross-partisan situations.

We contracted with a survey research company (Bovitz Inc.) to collect the data. The sample was drawn from a panel of respondents who have opted into complete online surveys. The panel was originally developed based on a random-digit-dial (RDD) telephone survey, where to enter the panel a respondent needed to have access to the Internet. (In this sense, it is a non-probability sample in the same way as those taken by firms such as YouGov are non-probability samples.) The panel has continued to grow based on ongoing RDD recruiting and referrals. From the panel, which has ~1 million members, a given sample is drawn using a matching algorithm to ensure that those screened to qualify for the survey constitute a sample that demographically represents the United States.

To explore the possibility of extra-ordinary pre-treatment effects, we content analyzed news articles from The New York Times and The USA Today from June 2008 to approximately June 2009 that included one of the following terms in the headline or lead paragraph: “energy policy,” “energy crisis,” “energy shortage,” or “energy plan.” From these, we selected articles that met specific criteria to ensure they are about the U.S. energy situation. This resulted in a total of 67 articles (28 from the USA Today and 39 from the NYT). We found that 39 % mentioned some type of partisan content (from one party) and 6 % mentioned some sort of bipartisanship. These results suggest nothing out of the norm a la pre-treatment and that partisanship plays a role in these discussions.

We asked pre-test respondents whether they thought the Act was sponsored by Democrats or Republicans, and we found no significant differences in presumed attributions.

Personal communication, Charles Taber 12/28/09, and personal communication Milton Lodge 12/31/09. The closest example we could find was Boiney et al. (1997, p. 8) who ask respondents to decide whether to introduce a new product for a company with a directional manipulation telling them that the product is profitable and that past proposals have been turned down too quickly. We build on this general approach. Redlawsk (2002) manipulates motivation in a study of motivated reasoning, but focuses on on-line versus memory-based processing; he assumes on-line is the default, and then manipulates memory-based processing by telling people they will have to list everything they can remember and justify their choice. This latter aspect will likely prompt more accuracy processing, which is what Redlawsk (2002) wants to show—i.e., that memory-based processing moderates motivated reasoning.

We thank Charles Taber for suggesting this specific approach; personal communication, 1/4/10.

We use one-tailed tests throughout as is conventional given clear directional predictions; see Blalock 1979; hence our 90 % confidence intervals.

The question wording and distribution of each response for all control variables is reported in Table 6.

Note that moving in the opposite direction of an out-party endorsement is consistent with others who find a similar backlash effect (Cohen 2003; Redlawsk 2002).

Note that the directional processing motivation Conditions (2, 5, 8, 11, and 14) significantly exceeded the no manipulation processing Conditions (1, 4, 7, 10, and 13) in only one of five cases. The one case is the same party endorsement, no motivation relative to same party endorsement, directional motivation conditions (Conditions 4 and 5, p < 0.05). The no endorsement conditions with no processing manipulation (1) and a directional processing inducement (2), perhaps surprisingly, register significant increases in support for the policy. Interestingly, the increase in support in these conditions stems entirely from movement among Democrats (evidence on this is available upon request from the authors). In short, in the absence of any processing inducement, Democrats seem to engage in motivated reasoning to a greater extent than Republicans when they are induced to think about and justify their views. This presumably reflects that energy is an issue owned by Democrats (see Druckman et al. 2009a).

There also is the question about whether partisan motivated reasoning leads to polarization. Elite partisan polarization itself appears to increase motivated reasoning (e.g., Druckman et al. 2013; Levendusky 2010; Slothuus and de Vreese 2010). In other words, partisan reasoning will be most likely to occur on issues where the parties conflict or are most dissimilar. We cannot directly examine this because we look at a single case at a single point in time, and, thus, there is no objective variation in polarization. We also do not manipulate polarization (perceptions) as Levendusky (2010) and Druckman et al. (2013) do. We did, however, measure perceptions of partisan similarity. Specifically, we asked “In general, to what extent do you think Democrats and Republicans take similar or dissimilar policy positions?” on a 1–7 scale with higher scores indicating greater similarity. Although we do not report the results from this analysis here (these are available upon request from the authors), we find clear evidence that partisan motivated reasoning occurs to a greater extent among those who view the parties as most different on this issue. This provides further evidence suggesting that partisan motivated reasoning can exacerbate polarization.

We employ a median split for this measure which allows us to focus on what are more likely to be qualitatively distinct groups (as do Druckman and Nelson 2003; Miller and Krosnick 2000, p. 305). We find consistent results, albeit slightly weaker, using a continuous measure rather than using a median split. Of note, we find our party trust measure does not moderate support for the Act in conditions where a party endorsement was not provided, as one would expect given trust should only moderate support for the Act in cases where partisan motivated reasoning occurs.