Ideology Through the Partisan Lens: Applying Anchoring Vignettes to U.S. Survey Research

Ideology Through the Partisan Lens: Applying Anchoring Vignettes to U.S. Survey Research Abstract A common thread running through the research on the 7-point ideological scale frequently used on U.S. surveys is that the terms “liberal” and “conservative” are abstract and subject to interpretation. I contribute to this literature by using anchoring vignettes to clearly link these differences to partisanship. I show that Democratic and Republican respondents interpret and use the response categories of the ideological scale in systematically different ways such that Democrats have significantly lower thresholds for the distinctions between categories. These results not only have implications for the studies of ideology and polarization in the United States but also demonstrate the usefulness of anchoring vignettes when comparing individuals not just across countries but within them as well. When analyzing responses to closed-ended survey questions, an underlying assumption is that all respondents interpret and use the given response categories and the intervals between them in the same way. Previous research (Brady, 1985; Aldrich & Mckelvey, 1977), however, indicates that this assumption does not always hold. Particularly when asked about abstract concepts, survey respondents often do apply a variety of understandings to the terms and measurement scales used. In an attempt to address the potential for and possible problems related to these differing interpretations—a tendency known as category differential item functioning (DIF)—a more recent body of work (King, Murray, Salomon, & Tandon, 2004; King & Wand, 2007; Hopkins & King, 2010) advocates the use of vignettes to anchor conceptions of measurement scales and increase interpersonal comparability.1 Since its introduction, this method has been used in a growing number of countries and research areas. For example, the World Health Organization has used anchoring vignettes to assess cross-cultural differences in self-ratings of mobility, affect, pain, social relationships, vision, sleep, cognition, and self-care. In the realm of politics, anchoring vignettes have been used to explore state effectiveness in Eastern Europe (Grzymala-Busse, 2007), political efficacy in China and Mexico (King et al., 2004; Wand, 2013), interpretations of democracy in 19 African nations (Bratton, 2010), and expert placements of European political parties (Bakker, Jolly, Polk, & Poole, 2015). But while researchers are using anchoring vignettes to compare responses from individuals in different countries, it seems that the potential usefulness of the method for comparing responses within one single country, namely, the United States, has largely been ignored.2 Just as anchoring vignettes can account for differences in interpretation that stem from one’s country of residence, it seems they should also be helpful in exploring differences that arise from other factors that have been demonstrated to color political perceptions. In the case of the United States, perhaps the largest influencer of political interpretation is partisanship. Indeed, a wealth of previous research shows that an individual’s partisanship colors perceptions of not just subjective opinions but even the most objective of subjects (Bartels, 2002; Farwell & Weiner, 2000; Gaines, Kuklinski, Quirk, Peyton, & Verkuilen, 2007; Mitchell, Hibbing, Smith, & Hibbing, 2013; Simas, 2016). As such, I extend this line of research by using anchoring vignettes to explicitly show that partisanship is a significant driver of DIF in the use of one of the most common American survey measures: the 7-point ideological scale. I argue and show that because partisans use ideological terms in different ways, Democrats have significantly lower thresholds for the distinctions between all categories of this scale. This means that even when Democrats and Republicans place themselves at the same point, Democratic respondents should still have considerably more liberal ideologies than their Republican counterparts, and that failing to account for these perceptual differences could lead to an understatement of the ideological divergence between the two parties. I further illustrate this by focusing on engaged respondents and showing that the anchoring vignettes approach leads to predictions that are more consistent with the underlying issue preferences that the ideological scale is often intended to represent. Overall, the findings presented here have significant implications for not only the study of polarization in the U.S. electorate but also for the more general study of survey research, as they show that even within one country, there may be different understandings of the political world. Partisanship and Interpretations of Ideology The research presented here rests on two key theoretical claims. The first is that individuals interpret the points on the ideological scale in different ways. Recent research shows that even though citizens’ ability to correctly use the liberal–conservative scale has increased (Abramowitz & Saunders, 1998; Hetherington, 2001; Levendusky, 2009), there is still considerable heterogeneity in the policy preferences of individuals at each point on the ideological scale (Claassen, Tucker, & Smith, 2014; Ellis & Stimson, 2012; Feldman & Johnston, 2014; Treier & Hillygus, 2009). One explanation offered by this literature is centered on the idea that because ideology is comprised of distinct social and economic components (Carmines, Ensley, & Wagner, 2012; Inglehart, 1997; Layman & Carsey, 2002), the unidimensional scale may fail to capture the complexity of the concept. Thus, though citizens have become more proficient in using the unidimensional scale, “they may differ in the way they understand the dimension in terms of substantive policy content. While some citizens may see liberalism and conservatism as primarily about social issues, others may understand the dimension in terms of economics, while others may see both domains as relevant” (Feldman & Johnston, 2014, p. 340–341). The second major theory suggests that heterogeneity should still result because of a tendency of individuals to identify with ideological labels for symbolic rather than substantive reasons (Conover & Feldman, 1981). More specifically, there is evidence that a substantial number of individuals hold decidedly liberal policy opinions but avoid identifying themselves as such to escape the negative connotation of the label (Ellis & Stimson, 2012; Schiffer, 2000). This results in each point on the scale (and particularly those on the more conservative end) being comprised a mix of those whose policy preferences are consistent with that ideological label and those who simply prefer to identify themselves as such. A common thread running through these explanations is the notion that the terms “liberal,” “conservative,” and “moderate” are all essentially abstract and subject to different interpretations. That is, both of these explanations strongly suggest that the ideological scale is subject to DIF. Hare, Armstrong, Bakker, Carroll, and Poole (2015) explicitly articulate this proposition and develop a Bayesian Aldrich–McKelvey scaling method to model and correct for DIF in self-placements and the placements of senators and Senate candidates. While a major advancement in the literature in this field, Hare et al.’s (2015) work is more focused on the consequences of DIF rather than the causes, as their methodology does not allow for testing of how respondent characteristics affect perceptions of the cutoffs between ideological categories. Thus, my major contribution lies in the advancement and testing of my second major claim: Partisanship directly affects interpretations of the ideological scale such that there are two distinct, partisan understandings of this scale. I argue that there are at least two readily identifiable explanations for why Democrats and Republicans may understand and use ideological labels in systematically different ways. First, the two parties use the words “liberal” and “conservative” in different ways and with different connotations. From George H. W. Bush’s “‘L-word’ strategy, which reduced liberal to an unspeakable profanity” (Schiffer, 2000, p. 298) to George W. Bush’s self-branding as a “compassionate conservative,” the Republican Party has been largely successful in framing ideological terms in a manner favorable to itself such that Democratic candidates tend to prefer “progressive” to “liberal” (Ellis & Stimson, 2012; Sears & Citrin, 1985). And even when candidates themselves do not explicitly use these types of ideological labels, the media still tends to use them in a nonneutral manner (Ellis & Stimson, 2012). Overall, the partisan slant placed on ideological labels means that individuals’ receptivity of this messaging should be at least somewhat contingent on their partisanship (Zaller, 1992). Indeed, recent works show that citizens are most greatly influenced by the framing used by their own party’s leaders and media (Levendusky, 2013; Slothuus & de Vreese, 2010). Thus, assuming that each party attempts to define and use ideological terminology to its advantage, it follows that there should be at least two distinctly partisan conceptions of the ideological scale reflected in the self-placements of the public. Second, differential utilization of the ideological scale may be related to the psychological tendency of individuals to evaluate themselves not in an absolute sense but rather in comparison with others (Festinger, 1954). As Heine, Lehman, Peng, and Greenholtz (2002) explain in a clear example, calling oneself tall “depends on comparisons with appropriate targets…The same height—for example, 5 feet 9 in.—is seen as tall in some contexts (e.g., among elementary school students or Japanese women) and short in others (e.g., among professional basketball players or Dutch men)” (p. 904). How one determines the appropriate target for comparison has been the subject of much research, but the dominant conclusion appears to be that individuals most often and meaningfully make self-assessments in comparison with fellow group members or to those whom they see as most similar to themselves (Festinger, 1954; Suls, Martin, & Wheeler, 2002; Wheeler, Martin, & Suls, 1997; Zanna, Goethals, & Hill, 1975). As party identification has been shown to be a strong group identification akin to gender or race (Green, Palmquist, & Schickler, 2002; Greene, 1999), it is logical to think that when assessing oneself on political matters, individuals will use their political party as a reference group. Translating the height example to apply to ideological self-placements, then, it follows that when a Democratic individual self-categorizes his ideology as “middle of the road,” he is doing so not as an assessment of his ideology as moderate compared with some universal standard, but rather, relative to other members of the group to which he belongs and sees as most similar to himself—the Democratic Party. Assuming that a Republican respondent would similarly place himself in comparison with his party, this would offer an explanation as to why two individuals may place themselves at the same point on the scale when they hold vastly different policy positions. Method The technique of anchoring vignettes involves showing respondents a series of short texts where each provides a concrete description of the trait of interest (in this case, political ideology) of a different hypothetical individual. After reading each vignette, respondents are asked to rate that individual on the same scale used for self-placement. The value-added of these vignette placements is illustrated in Figure 1. Figure 1 View largeDownload slide The logic of anchoring vignettes. Figure 1 View largeDownload slide The logic of anchoring vignettes. Panel A of Figure 1 displays the traditional interpretation of respondent self-placements; as both Respondent A and Respondent B placed themselves at the same point on the ideological scale, they are assumed to have the same underlying ideology. As discussed above, however, this assumption is flawed if the respondents have fundamentally different definitions of the points on the ideological scale. Respondent placements of the anchoring vignettes help address this possibility. Because the vignettes describe individuals with clear policy positions, the respondents’ placements of these individuals provide insight into how the respondents define the various points on the ideological scale. That is, the individual described in each vignette has the same actual ideological rating, regardless of the party of the respondent, and so if a Democratic respondent rates the individual described as “somewhat liberal,” and a Republican respondent rates that same individual as “liberal,” this is suggestive that the two respondents have different interpretations of the package of policies the term “liberal” represents. By using these vignette placements of the same individuals as “anchors” and rescaling placements accordingly, we can then account for different perceptions of the scale and make more meaningful and accurate comparisons of self-placements based on where they are relative to these common anchors. For example, in Panel A of Figure 1, Respondents A and B placed themselves at the same point, but Respondent A placed herself between the two anchors, and Respondent B placed herself as more liberal than both of the anchors. Adjusting for this in Panel B of Figure 1, it becomes clear that the more appropriate conclusion may be that Respondent A is more conservative than Respondent B, as regardless of the labels used, Respondent A placed herself as more conservative than Anchor 2 and Respondent B did not. Vignette Texts and Data The anchoring vignettes were placed on a 1,000-respondent module of the 2012 Cooperative Congressional Election Survey (CCES; Ansolabehere, 2013)3 postelection survey. Table 1 displays the texts of the three vignettes, which were shown to respondents in random order. Table 1 Texts of Anchoring Vignettes Full sample Democrats only Republicans only Mean (SD) Mean (SD) Mean (SD) Prompt The following is a statement of political beliefs. Please read the statement and then answer the question that follows. Vignette 1 “I believe that all women should have the right to an abortion, that homosexuals should have the right to marry, and that all immigrants should be welcome. The fight over constant efforts to deny these fundamental rights keeps the government from focusing on what it really should be working on—making sure that every American has free healthcare. The rich and big businesses—especially those that harm our environment—should be forced to pay their rightful share of taxes so that our children can get a better education and our streets can be safe. Instead of continuing to pour money into unnecessary and unwinnable wars, the government should be providing for the basic needs of the people here at home.” 1.79 (1.20) 1.91 (1.12) 1.48 (1.16) Vignette 2 “I like the idea of the government doing more to make sure that all Americans have things like healthcare and a college education, but Americans are already taxed enough. I think that there needs to be a balance between what the government provides and what people can get from private vendors because the government can only do so much and needs to think about balancing the budget. I think that while we should continue to be tough on crime, we also need to consider rehabilitation rather than just incarceration for nonviolent offenders and reserve the death penalty only for cases with DNA proof. Homosexual couples should enjoy the same rights and benefits as heterosexual couples, but I believe this should be called a civil union, not a marriage.” 3.39 (1.43) 3.69 (1.37) 3.01 (1.44) Vignette 3 “I believe that the government’s first priority should be the security of this country. Abroad, we need to maintain a strong military and continue to support the efforts of our troops to lessen the threat of global terrorism. Here at home, law enforcement officers should be given the ability to identify and deport those who are in this country illegally. I also think that we need harsher punishments for criminals and broader application of the death penalty, especially for those who commit crimes against children or law enforcement officers. The government also needs to protect the values of American families by maintaining that marriage is a between a man and a woman and ensuring that every baby has the right to life. Moreover, individuals should be allowed to pray or observe religious traditions in schools if they so choose.” 5.91 (1.41) 6.00 (1.54) 5.99 (1.10) Placement question How would you describe the political ideology of this individual? (Response categories: Very Liberal [1], Liberal [2], Somewhat Liberal [3], Middle of the Road [4], Somewhat Conservative [5], Conservative [6], Very Conservative [7]) Full sample Democrats only Republicans only Mean (SD) Mean (SD) Mean (SD) Prompt The following is a statement of political beliefs. Please read the statement and then answer the question that follows. Vignette 1 “I believe that all women should have the right to an abortion, that homosexuals should have the right to marry, and that all immigrants should be welcome. The fight over constant efforts to deny these fundamental rights keeps the government from focusing on what it really should be working on—making sure that every American has free healthcare. The rich and big businesses—especially those that harm our environment—should be forced to pay their rightful share of taxes so that our children can get a better education and our streets can be safe. Instead of continuing to pour money into unnecessary and unwinnable wars, the government should be providing for the basic needs of the people here at home.” 1.79 (1.20) 1.91 (1.12) 1.48 (1.16) Vignette 2 “I like the idea of the government doing more to make sure that all Americans have things like healthcare and a college education, but Americans are already taxed enough. I think that there needs to be a balance between what the government provides and what people can get from private vendors because the government can only do so much and needs to think about balancing the budget. I think that while we should continue to be tough on crime, we also need to consider rehabilitation rather than just incarceration for nonviolent offenders and reserve the death penalty only for cases with DNA proof. Homosexual couples should enjoy the same rights and benefits as heterosexual couples, but I believe this should be called a civil union, not a marriage.” 3.39 (1.43) 3.69 (1.37) 3.01 (1.44) Vignette 3 “I believe that the government’s first priority should be the security of this country. Abroad, we need to maintain a strong military and continue to support the efforts of our troops to lessen the threat of global terrorism. Here at home, law enforcement officers should be given the ability to identify and deport those who are in this country illegally. I also think that we need harsher punishments for criminals and broader application of the death penalty, especially for those who commit crimes against children or law enforcement officers. The government also needs to protect the values of American families by maintaining that marriage is a between a man and a woman and ensuring that every baby has the right to life. Moreover, individuals should be allowed to pray or observe religious traditions in schools if they so choose.” 5.91 (1.41) 6.00 (1.54) 5.99 (1.10) Placement question How would you describe the political ideology of this individual? (Response categories: Very Liberal [1], Liberal [2], Somewhat Liberal [3], Middle of the Road [4], Somewhat Conservative [5], Conservative [6], Very Conservative [7]) Table 1 Texts of Anchoring Vignettes Full sample Democrats only Republicans only Mean (SD) Mean (SD) Mean (SD) Prompt The following is a statement of political beliefs. Please read the statement and then answer the question that follows. Vignette 1 “I believe that all women should have the right to an abortion, that homosexuals should have the right to marry, and that all immigrants should be welcome. The fight over constant efforts to deny these fundamental rights keeps the government from focusing on what it really should be working on—making sure that every American has free healthcare. The rich and big businesses—especially those that harm our environment—should be forced to pay their rightful share of taxes so that our children can get a better education and our streets can be safe. Instead of continuing to pour money into unnecessary and unwinnable wars, the government should be providing for the basic needs of the people here at home.” 1.79 (1.20) 1.91 (1.12) 1.48 (1.16) Vignette 2 “I like the idea of the government doing more to make sure that all Americans have things like healthcare and a college education, but Americans are already taxed enough. I think that there needs to be a balance between what the government provides and what people can get from private vendors because the government can only do so much and needs to think about balancing the budget. I think that while we should continue to be tough on crime, we also need to consider rehabilitation rather than just incarceration for nonviolent offenders and reserve the death penalty only for cases with DNA proof. Homosexual couples should enjoy the same rights and benefits as heterosexual couples, but I believe this should be called a civil union, not a marriage.” 3.39 (1.43) 3.69 (1.37) 3.01 (1.44) Vignette 3 “I believe that the government’s first priority should be the security of this country. Abroad, we need to maintain a strong military and continue to support the efforts of our troops to lessen the threat of global terrorism. Here at home, law enforcement officers should be given the ability to identify and deport those who are in this country illegally. I also think that we need harsher punishments for criminals and broader application of the death penalty, especially for those who commit crimes against children or law enforcement officers. The government also needs to protect the values of American families by maintaining that marriage is a between a man and a woman and ensuring that every baby has the right to life. Moreover, individuals should be allowed to pray or observe religious traditions in schools if they so choose.” 5.91 (1.41) 6.00 (1.54) 5.99 (1.10) Placement question How would you describe the political ideology of this individual? (Response categories: Very Liberal [1], Liberal [2], Somewhat Liberal [3], Middle of the Road [4], Somewhat Conservative [5], Conservative [6], Very Conservative [7]) Full sample Democrats only Republicans only Mean (SD) Mean (SD) Mean (SD) Prompt The following is a statement of political beliefs. Please read the statement and then answer the question that follows. Vignette 1 “I believe that all women should have the right to an abortion, that homosexuals should have the right to marry, and that all immigrants should be welcome. The fight over constant efforts to deny these fundamental rights keeps the government from focusing on what it really should be working on—making sure that every American has free healthcare. The rich and big businesses—especially those that harm our environment—should be forced to pay their rightful share of taxes so that our children can get a better education and our streets can be safe. Instead of continuing to pour money into unnecessary and unwinnable wars, the government should be providing for the basic needs of the people here at home.” 1.79 (1.20) 1.91 (1.12) 1.48 (1.16) Vignette 2 “I like the idea of the government doing more to make sure that all Americans have things like healthcare and a college education, but Americans are already taxed enough. I think that there needs to be a balance between what the government provides and what people can get from private vendors because the government can only do so much and needs to think about balancing the budget. I think that while we should continue to be tough on crime, we also need to consider rehabilitation rather than just incarceration for nonviolent offenders and reserve the death penalty only for cases with DNA proof. Homosexual couples should enjoy the same rights and benefits as heterosexual couples, but I believe this should be called a civil union, not a marriage.” 3.39 (1.43) 3.69 (1.37) 3.01 (1.44) Vignette 3 “I believe that the government’s first priority should be the security of this country. Abroad, we need to maintain a strong military and continue to support the efforts of our troops to lessen the threat of global terrorism. Here at home, law enforcement officers should be given the ability to identify and deport those who are in this country illegally. I also think that we need harsher punishments for criminals and broader application of the death penalty, especially for those who commit crimes against children or law enforcement officers. The government also needs to protect the values of American families by maintaining that marriage is a between a man and a woman and ensuring that every baby has the right to life. Moreover, individuals should be allowed to pray or observe religious traditions in schools if they so choose.” 5.91 (1.41) 6.00 (1.54) 5.99 (1.10) Placement question How would you describe the political ideology of this individual? (Response categories: Very Liberal [1], Liberal [2], Somewhat Liberal [3], Middle of the Road [4], Somewhat Conservative [5], Conservative [6], Very Conservative [7]) As Table 1 shows, the vignette texts were kept free of any demographic information (gender, race, age, etc.) that may potentially influence placements and to encourage as much response consistency as possible. Though the same exact issues are not discussed in each vignette, all three reference both social issues and issues related to the role of government, allowing for a broad interpretation that is not focused solely on one policy dimension. And while the ideological scale has seven points, only three vignettes were used, as King and Wand (2007) assert that it is not necessary that the number of vignettes used equal the number of points on the scale in question. Rather, vignettes should be chosen to maximize their joint ability to differentiate between respondents. In other words, a vignette is only informative to the extent that respondents could logically place themselves below, equal to, and above it. For example, if one was measuring self-assessments of personal wealth in the general population, then a vignette that described an individual with an economic situation similar to that of Bill Gates would not be useful, as it should be expected that all respondents would rate themselves as lower than this vignette. Additionally, vignettes that are too similar to one another do not provide much additional leverage. The logic behind this statement is highlighted by the example presented in Figure 1. If the two respondents had failed to see a meaningful difference between Anchors 1 and 2, then it would not have been apparent that Respondent A saw herself between the two anchors, and the differences between the respondents would have been masked. With these two goals of avoiding extreme or indiscriminate vignettes and the time and cost constraints of the survey administration, I ended up with three vignettes that aimed to represent the three major ideological categories on the scale: “liberal,” “middle of the road,” and “conservative.”4 In total, 16% of respondents failed to place any of the vignettes.5 Dropping those who did not place themselves (4.8% of those placing vignettes) and those who gave line item responses (10.1% of those placing vignettes) reduces the analytical sample to about 71.4% of the total.6 The rightmost columns of Table 1 show the mean placements for each vignette for the full sample and for each group of partisans.7 These means show that the intended rank ordering of the vignettes was not far off. Looking at the partisan differences in the placements, there is preliminary evidence of different interpretations of the scale. Republican respondents placed all three vignettes at more liberal positions on the scale. The difference in partisans’ placements of Vignette 3 is not significant, however. This suggests that (a) differences in interpretation are limited to just the liberal points on the scale; and/or (b) an additional vignette on the more conservative end of the scale may have been helpful. Still, there are at least some partisan differences that need to be explored, and I can proceed with overall confidence in the ability of the vignette ratings to usefully discriminate between respondents. Assumptions Use of anchoring vignettes involves two primary measurement assumptions: response consistency and vignette equivalence (King et al., 2004). Response consistency is the assumption that respondents use the categories of the scale the same way when placing both themselves and the individuals featured in the vignettes. If this assumption holds, it implies that if two respondents have the same underlying level of ideology, a more conservative self-rater should also be a more conservative vignette rater. Though there is no theoretical reason to believe that this assumption would be violated, I test for response consistency by regressing each respondent’s average vignette ratings on self-placement while controlling for underlying issue preferences with scores from two different issue indices featured in the CCES Common Content.8 The ordered probit coefficient for the vignette placements is positive and significant (p = .01), indicating that more conservative self-raters were on average, more conservative vignette raters and suggesting that response consistency can be safely assumed. The second assumption, vignette equivalence, is the assumption that each vignette is in fact measurable on the same scale. This is perhaps the more important of the two assumptions to check, as the unidimensionality of the ideological scale is often debated. Yet despite potential concerns, the means presented in Table 1 suggest that this assumption, too, has been met. While there was variation in the placements of the vignettes, the intended ordinality of vignettes is largely preserved. Indeed, 85.9% of the analytical sample gave ratings consistent with the intended rank ordering. Analytical Model Ordinal scales with relatively few categories are typically analyzed with ordered probit models. The central assumption of these types of ordinal models is that underlying the observed j responses is a continuous, latent scale on which there are j-1 points that serve as thresholds (τ) differentiating between the observed categories. That is, for the 7-point ideological scale, the ordered probit model estimates six thresholds that represent the latent values at which an individual crosses over from being in one ideological category to the next. The thresholds estimated by an ordered probit model, however, are assumed to be the same across all individuals. This assumption is at odds with the concept of DIF and the hypothesis being explored here, as I expect these thresholds to vary depending on the party of the individual respondent. Using a more concrete example, I expect that the point a Democratic respondent sees as being the divide between someone who is “middle of the road” versus someone who is “somewhat conservative” should be significantly different from the dividing point of a Republican. While a basic ordered probit model does not allow for these potential difference, the joint compound hierarchical ordered probit (CHOPIT) model does. For a full specification, see Rabe-Hesketh and Skrondal (2002) and King et al. (2004), but to summarize, the CHOPIT model explicitly tests and accounts for DIF by incorporating the ratings of the vignettes and estimating a set of thresholds for each independent variable in the model. This allows me to test for the effects of partisan-driven DIF while controlling for other possible factors. It is this ability to model thresholds and specifically link DIF to my independent variables that provides new leverage in comparison with other methods such as the nonparametric approach advocated by Wand (2013) or the Bayesian Aldrich–McKelvey scaling advanced by Hare et al. (2015). Thus, I rely on a CHOPIT model where the dependent variable is the respondent's ideological self-placement. Threshold estimates from the independent variable Democratic partisanship will reveal if there are significant differences between these respondents and Republican respondents, who are the omitted baseline category.9 Results Table 2 presents results from both a standard ordered probit model and the CHOPIT model. Table 2 Ordered Probit and Compound Hierarchical Ordered Probit Regression of Ideological Self-Rating Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Democratic respondent −2.33*** (.17) −2.53*** (.19) −0.75*** (.15) −0.40** (.13) −0.26 (.14) −0.03 (.13) −0.35** (.13) −0.52** (.16) Independent respondent −0.99*** (.26) −1.11*** (.31) −0.30 (.28) −0.24 (.23) −0.29 (.23) 0.12 (.24) −0.22 (.25) −0.19 (.29) Age 0.01*** (.00) 0.01*** (.00) 0.00 (.00) 0.01* (.00) 0.01 (.00) −0.00 (.00) −0.00 (.00) −0.01 (.00) Female 0.03 (.12) 0.08 (.15) 0.11 (.14) 0.07 (.13) 0.09 (.13) −0.03 (.12) 0.07 (.12) 0.06 (.15) Education −0.01 (.04) −0.06 (.05) −0.05 (.05) −0.08 (.04) −0.02 (.04) −0.07 (.04) −0.05 (.04) −0.05 (.05) Political sophistication −0.38 (.20) −0.46 (.27) −0.03 (.26) −0.05 (.23) −0.08 (.23) −0.17 (.21) −0.08 (.23) −0.06 (.26) Non-White 0.39** (.19) 0.61* (.24) 0.36 (.23) 0.34 (.19) 0.16 (.19) 0.14 (.18) 0.23 (.19) −0.11 (.23) Income −0.01 (.02) −0.02 (.03) −0.05* (.02) −0.02 (.02) −0.03 (.02) −0.01 (.02) 0.01 (.02) −0.01 (.02) Church attendance −0.04 (.03) −0.03 (.04) −0.00 (.03) 0.02 (.03) 0.03 (.03) −0.01 (.03) −0.02 (.03) 0.01 (.04) South 0.03 (.12) 0.07 (.27) 0.15 (.14) −0.00 (.12) 0.05 (.13) 0.00 (.21) 0.09 (.12) 0.03 (.15) Constant a −2.57*** (.48) −2.40*** (.39) −1.92*** (.38) −0.35 (.39) 0.01 (.39) 1.18*** (44) Anchor 1 −3.53*** (.48) Anchor 2 −1.84*** (.42) Anchor 3 0.17 (.43) Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Democratic respondent −2.33*** (.17) −2.53*** (.19) −0.75*** (.15) −0.40** (.13) −0.26 (.14) −0.03 (.13) −0.35** (.13) −0.52** (.16) Independent respondent −0.99*** (.26) −1.11*** (.31) −0.30 (.28) −0.24 (.23) −0.29 (.23) 0.12 (.24) −0.22 (.25) −0.19 (.29) Age 0.01*** (.00) 0.01*** (.00) 0.00 (.00) 0.01* (.00) 0.01 (.00) −0.00 (.00) −0.00 (.00) −0.01 (.00) Female 0.03 (.12) 0.08 (.15) 0.11 (.14) 0.07 (.13) 0.09 (.13) −0.03 (.12) 0.07 (.12) 0.06 (.15) Education −0.01 (.04) −0.06 (.05) −0.05 (.05) −0.08 (.04) −0.02 (.04) −0.07 (.04) −0.05 (.04) −0.05 (.05) Political sophistication −0.38 (.20) −0.46 (.27) −0.03 (.26) −0.05 (.23) −0.08 (.23) −0.17 (.21) −0.08 (.23) −0.06 (.26) Non-White 0.39** (.19) 0.61* (.24) 0.36 (.23) 0.34 (.19) 0.16 (.19) 0.14 (.18) 0.23 (.19) −0.11 (.23) Income −0.01 (.02) −0.02 (.03) −0.05* (.02) −0.02 (.02) −0.03 (.02) −0.01 (.02) 0.01 (.02) −0.01 (.02) Church attendance −0.04 (.03) −0.03 (.04) −0.00 (.03) 0.02 (.03) 0.03 (.03) −0.01 (.03) −0.02 (.03) 0.01 (.04) South 0.03 (.12) 0.07 (.27) 0.15 (.14) −0.00 (.12) 0.05 (.13) 0.00 (.21) 0.09 (.12) 0.03 (.15) Constant a −2.57*** (.48) −2.40*** (.39) −1.92*** (.38) −0.35 (.39) 0.01 (.39) 1.18*** (44) Anchor 1 −3.53*** (.48) Anchor 2 −1.84*** (.42) Anchor 3 0.17 (.43) Note. N = 611; Cell entries are estimated coefficients with robust standard errors in parentheses; *** p<.001, ** p<.01, * p<.05. CHOPIT = compound hierarchical ordered probit. aEstimated τs for the ordered probit model are −3.16,−2.28,−1.79, −0.55, 0.05, and 0.91. Table 2 Ordered Probit and Compound Hierarchical Ordered Probit Regression of Ideological Self-Rating Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Democratic respondent −2.33*** (.17) −2.53*** (.19) −0.75*** (.15) −0.40** (.13) −0.26 (.14) −0.03 (.13) −0.35** (.13) −0.52** (.16) Independent respondent −0.99*** (.26) −1.11*** (.31) −0.30 (.28) −0.24 (.23) −0.29 (.23) 0.12 (.24) −0.22 (.25) −0.19 (.29) Age 0.01*** (.00) 0.01*** (.00) 0.00 (.00) 0.01* (.00) 0.01 (.00) −0.00 (.00) −0.00 (.00) −0.01 (.00) Female 0.03 (.12) 0.08 (.15) 0.11 (.14) 0.07 (.13) 0.09 (.13) −0.03 (.12) 0.07 (.12) 0.06 (.15) Education −0.01 (.04) −0.06 (.05) −0.05 (.05) −0.08 (.04) −0.02 (.04) −0.07 (.04) −0.05 (.04) −0.05 (.05) Political sophistication −0.38 (.20) −0.46 (.27) −0.03 (.26) −0.05 (.23) −0.08 (.23) −0.17 (.21) −0.08 (.23) −0.06 (.26) Non-White 0.39** (.19) 0.61* (.24) 0.36 (.23) 0.34 (.19) 0.16 (.19) 0.14 (.18) 0.23 (.19) −0.11 (.23) Income −0.01 (.02) −0.02 (.03) −0.05* (.02) −0.02 (.02) −0.03 (.02) −0.01 (.02) 0.01 (.02) −0.01 (.02) Church attendance −0.04 (.03) −0.03 (.04) −0.00 (.03) 0.02 (.03) 0.03 (.03) −0.01 (.03) −0.02 (.03) 0.01 (.04) South 0.03 (.12) 0.07 (.27) 0.15 (.14) −0.00 (.12) 0.05 (.13) 0.00 (.21) 0.09 (.12) 0.03 (.15) Constant a −2.57*** (.48) −2.40*** (.39) −1.92*** (.38) −0.35 (.39) 0.01 (.39) 1.18*** (44) Anchor 1 −3.53*** (.48) Anchor 2 −1.84*** (.42) Anchor 3 0.17 (.43) Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Democratic respondent −2.33*** (.17) −2.53*** (.19) −0.75*** (.15) −0.40** (.13) −0.26 (.14) −0.03 (.13) −0.35** (.13) −0.52** (.16) Independent respondent −0.99*** (.26) −1.11*** (.31) −0.30 (.28) −0.24 (.23) −0.29 (.23) 0.12 (.24) −0.22 (.25) −0.19 (.29) Age 0.01*** (.00) 0.01*** (.00) 0.00 (.00) 0.01* (.00) 0.01 (.00) −0.00 (.00) −0.00 (.00) −0.01 (.00) Female 0.03 (.12) 0.08 (.15) 0.11 (.14) 0.07 (.13) 0.09 (.13) −0.03 (.12) 0.07 (.12) 0.06 (.15) Education −0.01 (.04) −0.06 (.05) −0.05 (.05) −0.08 (.04) −0.02 (.04) −0.07 (.04) −0.05 (.04) −0.05 (.05) Political sophistication −0.38 (.20) −0.46 (.27) −0.03 (.26) −0.05 (.23) −0.08 (.23) −0.17 (.21) −0.08 (.23) −0.06 (.26) Non-White 0.39** (.19) 0.61* (.24) 0.36 (.23) 0.34 (.19) 0.16 (.19) 0.14 (.18) 0.23 (.19) −0.11 (.23) Income −0.01 (.02) −0.02 (.03) −0.05* (.02) −0.02 (.02) −0.03 (.02) −0.01 (.02) 0.01 (.02) −0.01 (.02) Church attendance −0.04 (.03) −0.03 (.04) −0.00 (.03) 0.02 (.03) 0.03 (.03) −0.01 (.03) −0.02 (.03) 0.01 (.04) South 0.03 (.12) 0.07 (.27) 0.15 (.14) −0.00 (.12) 0.05 (.13) 0.00 (.21) 0.09 (.12) 0.03 (.15) Constant a −2.57*** (.48) −2.40*** (.39) −1.92*** (.38) −0.35 (.39) 0.01 (.39) 1.18*** (44) Anchor 1 −3.53*** (.48) Anchor 2 −1.84*** (.42) Anchor 3 0.17 (.43) Note. N = 611; Cell entries are estimated coefficients with robust standard errors in parentheses; *** p<.001, ** p<.01, * p<.05. CHOPIT = compound hierarchical ordered probit. aEstimated τs for the ordered probit model are −3.16,−2.28,−1.79, −0.55, 0.05, and 0.91. Respondent age, gender, race, education, political sophistication, income, church attendance, and region of residence are also controlled, and analyses are weighted for probability of selection.10 The first two columns show the coefficient estimates for the two models. The bottom of the second column displays the estimates of the vignette locations. These estimates are monotonically increasing. This fact supports my earlier claims of vignette equivalence and the appropriateness of the unidimensional scale, as they again show a consistency in the ordering of the vignettes. The remaining columns display the other unique aspect of the CHOPIT model: estimated thresholds (τ) for every independent variable in the model. The large majority of these estimates are not statistically significant. Particularly notable are the null results for both education and political sophistication. This means that if two individuals identify with the same party but have different levels of education or sophistication, their interpretations of the ideological scale should not differ. However, the negative (and significant in four of the six cases) threshold estimates for the Democratic respondent variable, indicate that partisanship does affect impressions of the distinctions between ideological categories. That is, if two individuals are of the same level of education or sophistication but one is a Democrat and the other a Republican, then they should have significantly different perceptions of the ideological scale. Thus, the key difference exposed by the CHOPIT model is related to partisanship, and not knowledge. The significant, negative coefficient for τ1 means that when placing a given vignette, Democrats are more likely than Republicans to choose “liberal” rather than “extremely liberal. Because the parameterization of the CHOPIT model makes interpretation of higher-order thresholds dependent on those prior, I follow Grol-Prokopczyk, Freese, and Hauser (2011) and more clearly present the results by visually illustrating the mean estimates for each category of respondents in the left-hand panel of Figure 2.11 Figure 2 View largeDownload slide Estimated mean thresholds for the ideological scale Figure 2 View largeDownload slide Estimated mean thresholds for the ideological scale As this figure shows (and difference of means tests confirm), Democratic and Republican respondents have significantly different thresholds for distinguishing between each set of responses on the 7-point ideological scale. So if two respondents hold the same issue positions but are from different parties, the Democratic respondent is more likely to classify his issue positions as more conservative than his Republican counterpart, as his perceptions of where the thresholds between categories lie are all lower or more liberal. Consistent with theories of the negative affect associated with the term “liberal,” these differences in perceptions are largest at the left end of the scale. For example, the mean Republican rating for τ1 is 3.01. This means that for a Republican, any latent value below this should be classified as “extremely liberal.” For a Democrat, however, 3.01 falls between the mean estimates for τ3 and τ4. This means that what is defined as “extremely liberal” by a Republican could be defined as anywhere from “extremely liberal” to “middle of the road” by a Democrat. At the opposite end, the similar estimates for Democratic τ5 and Republican τ4 suggest that what is “conservative” to a Democrat may only be “somewhat conservative” to a Republican. Note that this contrasts with Table 1’s suggestion of agreement over the meaning of “conservative,” as these threshold results show that differences in interpretation do in fact exist at all points. As such, it is clear that within each party, there are distinct definitions of the labels associated with the entire ideological scale. Implications for Ideological Self-Identification So far, I have shown that Democrats and Republicans interpret the ideological scale in significantly different ways. How, then, do these different conceptions of the scale points and thresholds between them impact our interpretations of ideological self-identification? The importance of accounting for these perceptual differences can most simply be seen by comparing the ordered probit and CHOPIT coefficients in Table 2. While both models produce significant, negative estimates for the Democratic respondent variable, the CHOPIT model estimates a larger effect of partisanship. This suggests that failing to account for differences in the perceptions of the ideological scale may lead to an understatement of the ideological polarization between members of the two parties. Yet, the polarization literature (Abramowitz, 2011, 2013) shows that the type of partisan differentiation exposed here is most prevalent among and perhaps limited to those who take interest and participate in the political process. In addition, engaged citizens are the most likely to be exposed to consistently partisan messages (Iyengar & Hahn, 2009) and the most motivated to respond to those messages (Slothuus & de Vreese, 2010). It is likely, then, that the type of distortion that I am claiming exists is strongest among them.12 Because my analyses are necessarily limited to only those respondents who placed themselves and the vignettes, it is possible that those who are willing and/or able to assess ideology on this scale do so exactly because they are engaged citizens, and thus, my method may be driving my results. To assess this possibility, I compare willingness/ability to place the vignettes with characteristics identified as predictors of political engagement. The full comparison is available in Supplemental Appendix D, but to summarize, the probit analysis reveals that there are no statistically significant differences due to of education or intensity of either partisanship or ideology. As Abramowitz (2011) finds these to be three of the strongest predictors of political engagement, I reference this as evidence that the results presented here are not simply artifacts of my restricted sample. Still, I further explore the possibility that partisan distortion may be stronger among a specific type of individual by rerunning the CHOPIT model among two subsets of my sample: those who report taking part in one or more political act beyond voting (i.e., the politically engaged; 56% of my analytical sample) and those who do not (44% of my analytical sample). Table 3 displays the key coefficients of interest for both subsamples.13 Table 3 Ordered Probit and Compound Hierarchical Ordered Probit Regression of Ideological Self-Rating by Political Engagement Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Politically engaged respondents N=341 Democratic respondent −3.11*** (.28) −3.57*** (.32) −1.38*** (.19) −0.71*** (.18) −0.57*** (.21) −0.38* (.18) −0.72*** (.19) −0.52* (.22) Anchor 1 −3.59*** (.70) Anchor 2 −1.60* (.62) Anchor 3 1.18 (.63) Unengaged respondents N=270 Democratic respondent −1.93*** (.22) −1.92*** (.26) −0.33 (.23) −0.22 (.21) −0.07 (.21) 0.25 (.20) −0.07 (.20) −0.56* (.25) Anchor 1 −3.18*** (.69) Anchor 2 −1.55*** (.63) Anchor 3 0.13 (.64) Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Politically engaged respondents N=341 Democratic respondent −3.11*** (.28) −3.57*** (.32) −1.38*** (.19) −0.71*** (.18) −0.57*** (.21) −0.38* (.18) −0.72*** (.19) −0.52* (.22) Anchor 1 −3.59*** (.70) Anchor 2 −1.60* (.62) Anchor 3 1.18 (.63) Unengaged respondents N=270 Democratic respondent −1.93*** (.22) −1.92*** (.26) −0.33 (.23) −0.22 (.21) −0.07 (.21) 0.25 (.20) −0.07 (.20) −0.56* (.25) Anchor 1 −3.18*** (.69) Anchor 2 −1.55*** (.63) Anchor 3 0.13 (.64) Note. Cell entries are estimated coefficients with robust standard errors in parentheses; *** p<.001, ** p<.01, * p<.05. CHOPIT = compound hierarchical ordered probit. Full results are available in Supplementary Appendix E. Table 3 Ordered Probit and Compound Hierarchical Ordered Probit Regression of Ideological Self-Rating by Political Engagement Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Politically engaged respondents N=341 Democratic respondent −3.11*** (.28) −3.57*** (.32) −1.38*** (.19) −0.71*** (.18) −0.57*** (.21) −0.38* (.18) −0.72*** (.19) −0.52* (.22) Anchor 1 −3.59*** (.70) Anchor 2 −1.60* (.62) Anchor 3 1.18 (.63) Unengaged respondents N=270 Democratic respondent −1.93*** (.22) −1.92*** (.26) −0.33 (.23) −0.22 (.21) −0.07 (.21) 0.25 (.20) −0.07 (.20) −0.56* (.25) Anchor 1 −3.18*** (.69) Anchor 2 −1.55*** (.63) Anchor 3 0.13 (.64) Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Politically engaged respondents N=341 Democratic respondent −3.11*** (.28) −3.57*** (.32) −1.38*** (.19) −0.71*** (.18) −0.57*** (.21) −0.38* (.18) −0.72*** (.19) −0.52* (.22) Anchor 1 −3.59*** (.70) Anchor 2 −1.60* (.62) Anchor 3 1.18 (.63) Unengaged respondents N=270 Democratic respondent −1.93*** (.22) −1.92*** (.26) −0.33 (.23) −0.22 (.21) −0.07 (.21) 0.25 (.20) −0.07 (.20) −0.56* (.25) Anchor 1 −3.18*** (.69) Anchor 2 −1.55*** (.63) Anchor 3 0.13 (.64) Note. Cell entries are estimated coefficients with robust standard errors in parentheses; *** p<.001, ** p<.01, * p<.05. CHOPIT = compound hierarchical ordered probit. Full results are available in Supplementary Appendix E. Looking first as those who are not engaged, there is little evidence of partisan distortion, as only one of the threshold estimates is statistically significant. However, this should not be construed as evidence that those who are not engaged do not think ideologically. To the contrary, the estimated anchors for this subset of the sample are still monotonically increasing and consistent with a unidimensional interpretation of the scale. In addition, the coefficient for the partisanship variable is still negative and significant, indicating a predictable relationship between partisan and ideological self-identifications. Rather, the differences between the two models just show that those who are not engaged are not as polarized as their more interested and involved counterparts. This is consistent with works on motivated reasoning and the argument that while partisanship determines the direction of bias, “citizens’ engagement with politics should influence the strength of bias” ( Slothuus & de Vreese, 2010, p. 633). And indeed, restricting the analysis to just the more interested and involved respondents reveals greater bias and significant polarization. The right-hand panel of Figure 2 shows that the partisan differences in the threshold estimates are even greater than those uncovered when analyzing all respondents. The mean Republican estimate for τ1 (−2.63) now falls between the Democratic estimates for τ4 and τ5. This means that there is even less overlap between partisan perceptions than previously estimated. What each party defines as “middle of the road” (4) is seen as fully ideological by the opposite party. That is, a Democratic 4 is equivalent to a Republican 1, and a Republican 4 is equivalent to a Democratic 6. To further illustrate this, Table 4 maps this lack of overlap back onto actual policy preferences. I construct a 0–1 scale from preferences on Congressional bills regarding the budget, taxes, the birth control exemption, the Affordable Healthcare Act, the Keystone Pipeline, and ending “Don’t Ask, Don’t Tell” (α = .79).14 Though these bills span a wide range of policy areas, they load onto a single factor,15 suggesting a fit with the unidimensional nature of the ideology measure. Table 4 Comparison of the Engaged Respondents’ Policy Preferences Average roll-call battery placement Analytical sample Full CCES Mean (SD) Mean (SD) Democrats Republicans Democrats Republicans Extremely liberal (1) .09 .08 .36 (.16) — (.14) (.28) n = 33 n = 2,153 n = 27 Liberal (2) .11 .50 .11 .41 (.13) n = 1 (.14) (.23) n = 69 n = 3,362 n = 47 Somewhat liberal (3) .14 .50 .13 .34 (.12) n = 1 (.15) (.24) n = 32 n = 2,315 n = 91 Middle of the road (4) .16 .42 .20 .48 (.19) (.19) (.19) (.25) n = 18 n = 9 n = 1,942 n = 710 Somewhat conservative (5) .23 .63 .28 .61 (.17) (.28) (.23) (.25) n = 6 n = 33 n = 360 n = 1,612 Conservative (6) .07 .74 .37 .72 (.12) (.22) (.26) (.22) n = 3 n = 63 n = 226 n = 4,189 Extremely conservative (7) .81 .45 .79 — (.18) (.26) (.19) n=54 n =133 n =3,068 Average roll-call battery placement Analytical sample Full CCES Mean (SD) Mean (SD) Democrats Republicans Democrats Republicans Extremely liberal (1) .09 .08 .36 (.16) — (.14) (.28) n = 33 n = 2,153 n = 27 Liberal (2) .11 .50 .11 .41 (.13) n = 1 (.14) (.23) n = 69 n = 3,362 n = 47 Somewhat liberal (3) .14 .50 .13 .34 (.12) n = 1 (.15) (.24) n = 32 n = 2,315 n = 91 Middle of the road (4) .16 .42 .20 .48 (.19) (.19) (.19) (.25) n = 18 n = 9 n = 1,942 n = 710 Somewhat conservative (5) .23 .63 .28 .61 (.17) (.28) (.23) (.25) n = 6 n = 33 n = 360 n = 1,612 Conservative (6) .07 .74 .37 .72 (.12) (.22) (.26) (.22) n = 3 n = 63 n = 226 n = 4,189 Extremely conservative (7) .81 .45 .79 — (.18) (.26) (.19) n=54 n =133 n =3,068 Note. Details on the construction of the roll-call battery can be found in Supplementary Appendix B. CCES = Cooperative Congressional Election Survey. Table 4 Comparison of the Engaged Respondents’ Policy Preferences Average roll-call battery placement Analytical sample Full CCES Mean (SD) Mean (SD) Democrats Republicans Democrats Republicans Extremely liberal (1) .09 .08 .36 (.16) — (.14) (.28) n = 33 n = 2,153 n = 27 Liberal (2) .11 .50 .11 .41 (.13) n = 1 (.14) (.23) n = 69 n = 3,362 n = 47 Somewhat liberal (3) .14 .50 .13 .34 (.12) n = 1 (.15) (.24) n = 32 n = 2,315 n = 91 Middle of the road (4) .16 .42 .20 .48 (.19) (.19) (.19) (.25) n = 18 n = 9 n = 1,942 n = 710 Somewhat conservative (5) .23 .63 .28 .61 (.17) (.28) (.23) (.25) n = 6 n = 33 n = 360 n = 1,612 Conservative (6) .07 .74 .37 .72 (.12) (.22) (.26) (.22) n = 3 n = 63 n = 226 n = 4,189 Extremely conservative (7) .81 .45 .79 — (.18) (.26) (.19) n=54 n =133 n =3,068 Average roll-call battery placement Analytical sample Full CCES Mean (SD) Mean (SD) Democrats Republicans Democrats Republicans Extremely liberal (1) .09 .08 .36 (.16) — (.14) (.28) n = 33 n = 2,153 n = 27 Liberal (2) .11 .50 .11 .41 (.13) n = 1 (.14) (.23) n = 69 n = 3,362 n = 47 Somewhat liberal (3) .14 .50 .13 .34 (.12) n = 1 (.15) (.24) n = 32 n = 2,315 n = 91 Middle of the road (4) .16 .42 .20 .48 (.19) (.19) (.19) (.25) n = 18 n = 9 n = 1,942 n = 710 Somewhat conservative (5) .23 .63 .28 .61 (.17) (.28) (.23) (.25) n = 6 n = 33 n = 360 n = 1,612 Conservative (6) .07 .74 .37 .72 (.12) (.22) (.26) (.22) n = 3 n = 63 n = 226 n = 4,189 Extremely conservative (7) .81 .45 .79 — (.18) (.26) (.19) n=54 n =133 n =3,068 Note. Details on the construction of the roll-call battery can be found in Supplementary Appendix B. CCES = Cooperative Congressional Election Survey. The values in the table are the mean roll-call battery scores for engaged Democrats and Republicans who place themselves at each point on the ideological scale. For comparison and to show the generalizability of my results, I calculate the means for both the engaged respondents in my analytical sample and for the engaged respondents in the full CCES. Despite the fact that self-placements show that partisans often do place themselves at the same point, the issue preferences are more consistent with the threshold estimates from the CHOPIT model. Only the most conservative Democrats and the most liberal Republicans appear to share the same preferences, and even then, they place themselves at vastly different points on the ideological scale. The preferences in Table 4 also reflect the lack of differentiation among the liberal points on the scale that is suggested by the threshold estimates. In my sample, the estimated difference between a Democrat who places at “extremely liberal” and a Democrat who places at “somewhat liberal” is only 0.05. In contrast, the difference between a Republican who places at “extremely conservative” and a Republican who places at “somewhat conservative” is 0.18. This suggests that “liberal” does indeed fit a narrower range of preferences than “conservative.” In total, it appears that the anchoring vignettes approach offers a view of the ideological scale that more accurately reflects the true divide in respondents’ policy preferences and suggests that polarization among the engaged public may be even greater than is assumed when partisan-driven DIF is not taken into account. Discussion In the age of increased partisan polarization, it seems that even political terminology carries a partisan charge. Using anchoring vignettes, I advance the literature by going beyond just showing that individuals interpret ideological labels differently, but by also arguing that these differences can be predictably linked to partisan affiliation, as Democratic respondents have consistently lower thresholds for the distinctions between categories. Particularly among the politically engaged, what is labeled as “middle of the road” by a member of one party would be defined as distinctly ideological by a member of the opposite party, and, once these differences in interpretation are taken into account, the resulting estimates of the distinctions between categories are more reflective of the polarization in respondents’ policy preferences. Future work should probe further into this finding, as the question of how to best measure ideology is subject of much debate and any method that may help minimize the variation of issue preferences connected to each point of the ideological scale would help improve its predictive powers. Altogether, a substantial proportion of Democratic and Republican citizens do have different perceptions of the ideological scale. An interesting extension of this work would be to explore whether there are also differences within these two partisan groups. For example, it may be that among Republicans, those who identify with the Tea Party may have different perceptions of what it means to be “conservative” than those who do not. Moreover, it is unlikely that the DIF uncovered here is limited to just the ideological self-placement questions. The findings presented here suggest that some of the differences in where respondents place political figures and candidates that are typically attributed to projection may be because of different interpretations of the ideological scale. While anchoring vignettes have most often been used to increase the cross-cultural comparability of multinational surveys, my findings also make an important contribution to the study of public opinion by highlighting the potential usefulness of the technique in studies of single countries. The finding of partisan interpretations of the ideological scale presented here suggests that elite polarization may be creating what is the equivalent of two partisan “subcultures” within the United States. Given the broad geographic and ethnic diversity of the United States, it also seems likely that the ideological scale and other commonly used survey questions may be interpreted differently depending on factors such as the region of the country one lives in or the length of time that one’s family has been in the United States. Beyond the United States, countries wherein there are major divides based on factors such as religion or ethnicity may also benefit from more careful consideration of whether individuals are in fact giving comparable survey responses. Thus, my work has broader implications for the future direction of all survey research, as it demonstrates the potential usefulness of anchoring vignettes and the overall need for construction and analyses of survey questions measuring subjective concepts that acknowledge that even in areas that may share common borders, there still may be multiple, distinct understandings of the political world. Funding Funding for the research was provided by the Department of Research at the University of Houston. A previous version of this article was presented at the 2013 Annual Meeting of the American Political Science Association. Footnotes 1See also http://gking.harvard.edu/vign 2Wand (2013) uses data from the 2004 ANES to present a nonparametric analysis of self-reported preferences for government services but uses respondents’ placements of the two major-party presidential candidates as anchors rather than vignettes. 3For details on the survey methodology, see http://projects.iq.harvard.edu/cces/home 4In Fall 2011, I piloted five vignettes on a small sample (N = 33) of undergraduate students. Approximately 43% of those respondents placed at least two of the vignettes at the same point on the ideological scale, suggesting that reducing the number of vignettes shown to respondents would not necessarily sacrifice additional explanatory leverage. 5The large majority of respondents took an “all-or-nothing” approach. Of the 174 respondents who did not place all three vignettes, 14 placed just two and the remaining 160 placed none. 6A discussion of the differences between those who did and did not place the vignettes will be taken up later in the article. 7Plots of the full distributions of the vignette ratings are available in Supplemental Appendix A. 8Question text and more information about the construction of these indices can be found in Supplemental Appendix B. 9Partisan leaners are coded as partisans. 10Responses used to create Ideological Self-Placement and all independent variables come from the preelection wave of the CCES. Age is a continuous variable derived from the source variable “birthyr.” Female is a recoding of the variable “gender.” Education, income, and church attendance are categorical variables derived directly from the CCES source variables “educ,” “faminc,” and “pew-churchatd.” Sophistication is the number of correct answers given when asked to recall the party of the respondent’s governor, senior senator, junior senator, and House member (cc310a-d). Non-White is a dichotomous recoding of “race.” All responses other than “White” are coded as 1. South is coded 1 if the variable “inputstate” indicates the respondent is a resident of AL, AR, DE, FL, GA, KY, LA, MD, MS, NC, OK, SC, TN, TX, VA, or WV. 11See Supplemental Appendix C for computational details. 12I thank two anonymous reviewers for suggesting this line of discussion and analyses. 13Full model results are available in Supplemental Appendix E. 14See Supplementary Appendix B for full details about the questions used to construct this measure. 15The rotated factor loadings are all >.38. Elizabeth Simas is an Assistant Professor of Political Science at the University of Houston. Her research examines the behavior of both candidates and voters in U.S. elections. Acknowledgements The author is grateful for helpful comments from Jason Casellas. References Abramowitz A. I. ( 2011 ). The disappearing center: Engaged citizens, polarization, and American democracy . New Haven : Yale University Press . Abramowitz A. I. ( 2013 ). The polarized public? Why American government is so disfunctional . Boston : Pearson . Abramowitz A. I. , Saunders K. L. , Abramowitz A. ( 1998 ). Ideological realignment in the U.S. electorate . The Journal of Politics , 60 , 634 – 652 . https://doi.org/10.2307/2647642 Google Scholar CrossRef Search ADS Aldrich J. H. , Mckelvey R. D. ( 1977 ). A method of scaling with applications to the 1968 and 1972 presidential elections . The American Political Science Review , 71 ( 1 ), 111 – 130 . https://doi.org/10.2307/1956957 Google Scholar CrossRef Search ADS Ansolabehere S. ( 2013 ) Cooperative congressional election study, 2012: Common content . Release 1: April 15, 2013. Cambridge, MA : Harvard University . Bakker R. , Jolly S. , Polk J. , Poole K. ( 2015 ). The European common space: Extending the use of anchoring vignettes. http://dx.doi.org/10.1017/S0022381614000449. https://doi.org/10.1017/S0022381614000449 Bartels L. M. ( 2002 ). Beyond the running tally: Partisan bias in political perceptions . Political Behavior . https://doi.org/10.1023/A:1021226224601 Brady H. ( 1985 ). The perils of survey research: Inter-personally incomparable survey responses . Political Methodology , 11 , 269 – 291 . Bratton M. ( 2010 ). Anchoring the "D-Word" in Africa . Journal of Democracy , 21 , 106 – 113 . Google Scholar CrossRef Search ADS Carmines E. G. , Ensley M. J. , Wagner M. W. ( 2012 ). Who fits the left-right divide? Partisan polarization in the American electorate . American Behavioral Scientist , 56 , 1631 – 1653 . https://doi.org/10.1177/0002764212463353 Google Scholar CrossRef Search ADS Claassen C. , Tucker P. , Smith S. S. ( 2014 ). Ideological labels in America . Political Behavior , 37 , 253 – 278 . https://doi.org/10.1007/s11109-014-9272-6 Google Scholar CrossRef Search ADS Conover P. J. , Feldman S. ( 1981 ). The origins and meaning of liberal/conservative self-identifications . American Journal of Political Science , 25 , 617 – 645 . https://doi.org/10.1007/s11109-010-9145-6 Google Scholar CrossRef Search ADS Ellis C. , Stimson J. ( 2012 ). Ideology in America . New York, NY : Cambridge University Press . Farwell L. , Weiner B. ( 2000 ). Bleeding hearts and the heartless: popular perceptions of liberal and conservative ideologies . Personality and Social Psychology Bulletin , 26 , 845 – 852 . https://doi.org/10.1177/0146167200269009 Google Scholar CrossRef Search ADS Feldman S. , Johnston C. ( 2014 ). Understanding the determinants of political ideology: Implications of structural complexity . Political Psychology , 35 , 337 – 358 . https://doi.org/10.1111/pops.12055 Google Scholar CrossRef Search ADS Festinger L. ( 1954 ). A theory of social comparison processes . Human Relations , 7 , 117 – 140 . https://doi.org/10.1177/001872675400700202 Google Scholar CrossRef Search ADS Gaines B. J. , Kuklinski J. H. , Quirk P. J. , Peyton B. , Verkuilen J. ( 2007 ). Same facts, different interpretations: Partisan motivation and opinion on Iraq . Journal of Politics , 69 , 957 – 974 . https://doi.org/10.1111/j.1468-2508.2007.00601.x Google Scholar CrossRef Search ADS Green D. , Palmquist B. , Schickler E. ( 2002 ). Partisan hearts and minds: Political parties and social identities of voters . New Haven : Yale University Press . Greene S. ( 1999 ). Understanding party identification: A social identity approach . Political Psychology , 20 , 393 – 403 . https://doi.org/10.1111/0162-895X.00150 Google Scholar CrossRef Search ADS Grol-Prokopczyk H. , Freese J. , Hauser R. M. ( 2011 ). Using anchoring vignettes to assess group differences in general self-rated health . Journal of Health and Social Behavior , 52 , 246 – 261 . https://doi.org/10.1177/0022146510396713 Google Scholar CrossRef Search ADS PubMed Grzymala-Busse A. ( 2007 ). Rebuilding leviathan: Party competition and state exploitation in post-communist democracies . New York, NY : Cambridge University Press . Hare C. , Armstrong D. A. , Bakker R. , Carroll R. , Poole K. T. ( 2015 ). Using Bayesian Aldrich-McKelvey scaling to study citizens’ ideological preferences and perceptions . American Journal of Political Science , 59 , 759 – 774 . https://doi.org/10.1111/ajps.12151 Google Scholar CrossRef Search ADS Heine S. J. , Lehman D. R. , Peng K. , Greenholtz J. ( 2002 ). What’s wrong with cross-cultural comparisons of subjective Likert scales? The reference-group effect . Journal of Personality and Social Psychology , 82 , 903 – 918 . https://doi.org/10.1037/0022-3514.82.6.903 Google Scholar CrossRef Search ADS PubMed Hetherington M. J. ( 2001 ). Resurgent mass partisanship: The role of elite polarization . American Political Science Review , 95 , 619 – 631 . https://doi.org/10.1017/S0003055401003045 Google Scholar CrossRef Search ADS Hopkins D. J. , King G. ( 2010 ). Improving anchoring vignettes: Designing surveys to correct interpersonal incomparability . Public Opinion Quarterly , 74 , 201 – 222 . https://doi.org/10.1093/poq/nfq011 Google Scholar CrossRef Search ADS Inglehart R. ( 1997 ). Modernization and postmodernization: Cultural, economic, and political change in 43 societies . Princeton : Princeton University Press . Iyengar S. , Hahn K. S. ( 2009 ). Red media, blue media: Evidence of ideological selectivity in media use . Journal of Communication , 59 ( 1 ), 19 – 39 . https://doi.org/10.1111/j.1460-2466.2008.01402.x Google Scholar CrossRef Search ADS King G. , Murray C. J. L. , Salomon J. A , Tandon A. ( 2004 ). Enhancing the validity and cross-cultural comparability of measurement in survey research . American Political Science Review , 98 ( 1 ), 191 – 207 . https://doi.org/10.1017/S000305540400108X Google Scholar CrossRef Search ADS King G. , Wand J. ( 2007 ). Comparing incomparable survey responses: New tools for anchoring vignettes . Political Analysis , 15 ( 1 ), 46 – 66 . Google Scholar CrossRef Search ADS Layman G. C. , Carsey T. M. ( 2002 ). Party polarization and "conflict extension" in the American electorate . American Journal of Political Science , 46 , 786 – 802 . https://doi.org/10.2307/3088434 Google Scholar CrossRef Search ADS Levendusky M. S. ( 2009 ). The partisan sort: How liberals became democrats and conservatives became republicans . Chicago : Chicago University Press . Levendusky M. S. ( 2013 ). Why do partisan media polarize viewers? American Journal of Political Science , 57 , 611 – 623 . https://doi.org/10.1111/ajps.12008 Google Scholar CrossRef Search ADS Mitchell D.-G. , Hibbing M. V. , Smith K. B. , Hibbing J. R. ( 2013 ). Side by side, worlds apart: desired policy change as a function of preferences AND perceptions . American Politics Research , 42 , 338 – 363 . https://doi.org/10.1177/1532673X13498619 Google Scholar CrossRef Search ADS Rabe-Hesketh S. , Skrondal A. . 2002 . Estimating CHOPIT models in GLLAMM: Political efficacy example from King et al. 2002. Retrieved from : http://www.gllamm.org/examples.html Retrieved August 12, 2013 Schiffer A. ( 2000 ). I’m not that liberal: Explaining conservative democratic identification . Political Behavior , 22 , 293 – 310 . https://doi.org/10.1023/A:1010626029987 Google Scholar CrossRef Search ADS Sears D. , Citrin J. ( 1985 ). Tax revolt: Something for nothing in California . Cambridge : Harvard University Press . Simas E. N. ( 2016 ). Perceptions of the heterogeneity of party elites in the United States . Party Politics . Advance online publication. https://doi.org/10.1177/1354068816668676 Slothuus R. , de Vreese C. H. ( 2010 ). Political parties, motivated reasoning, and issue framing effects . The Journal of Politics , 72 , 630 – 645 . https://doi.org/10.1017/S002238161000006X Google Scholar CrossRef Search ADS Suls J. , Martin R. , Wheeler L. ( 2002 ). Social comparison: Why, with whom, and with what effect? Current Directions in Psychological Science , 11 , 159 – 163 . https://doi.org/10.1111/1467-8721.00191 Google Scholar CrossRef Search ADS Treier S. , Hillygus D. S. ( 2009 ). The nature of political ideology in the contemporary electorate . Public Opinion Quarterly , 73 , 679 – 703 . https://doi.org/10.1093/poq/nfp067 Google Scholar CrossRef Search ADS Wand J. ( 2013 ). Credible comparisons using interpersonally incomparable data: Nonparametric scales with anchoring vignettes . American Journal of Political Science , 57 ( 1 ), 249 – 262 . https://doi.org/10.1111/j.1540-5907.2012.00597.x Google Scholar CrossRef Search ADS Wheeler L. , Martin R. , Suls J. ( 1997 ). The proxy model of social comparison for self-assessment of ability . Personality and Social Psychology Review , 1 ( 1 ), 54 – 61 . https://doi.org/10.1207/s15327957pspr0101_4 Google Scholar CrossRef Search ADS PubMed Zaller J. ( 1992 ). The nature and origin of mass opinions . New York, NY : Cambridge University Press . Zanna M. P. , Goethals G. R. , Hill J. F. ( 1975 ). Evaluating a sex-related ability: Social comparison with similar others and standard setters . Journal of Experimental Social Psychology , 11 ( 1 ), 86 – 93 . https://doi.org/10.1016/S0022-1031(75)80013-8 Google Scholar CrossRef Search ADS © The Author 2017. Published by Oxford University Press on behalf of The World Association for Public Opinion Research. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png International Journal of Public Opinion Research Oxford University Press

Ideology Through the Partisan Lens: Applying Anchoring Vignettes to U.S. Survey Research

Loading next page...
 
/lp/ou_press/ideology-through-the-partisan-lens-applying-anchoring-vignettes-to-u-s-bm1rJA0Xsp
Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press on behalf of The World Association for Public Opinion Research. All rights reserved.
ISSN
0954-2892
eISSN
1471-6909
D.O.I.
10.1093/ijpor/edx006
Publisher site
See Article on Publisher Site

Abstract

Abstract A common thread running through the research on the 7-point ideological scale frequently used on U.S. surveys is that the terms “liberal” and “conservative” are abstract and subject to interpretation. I contribute to this literature by using anchoring vignettes to clearly link these differences to partisanship. I show that Democratic and Republican respondents interpret and use the response categories of the ideological scale in systematically different ways such that Democrats have significantly lower thresholds for the distinctions between categories. These results not only have implications for the studies of ideology and polarization in the United States but also demonstrate the usefulness of anchoring vignettes when comparing individuals not just across countries but within them as well. When analyzing responses to closed-ended survey questions, an underlying assumption is that all respondents interpret and use the given response categories and the intervals between them in the same way. Previous research (Brady, 1985; Aldrich & Mckelvey, 1977), however, indicates that this assumption does not always hold. Particularly when asked about abstract concepts, survey respondents often do apply a variety of understandings to the terms and measurement scales used. In an attempt to address the potential for and possible problems related to these differing interpretations—a tendency known as category differential item functioning (DIF)—a more recent body of work (King, Murray, Salomon, & Tandon, 2004; King & Wand, 2007; Hopkins & King, 2010) advocates the use of vignettes to anchor conceptions of measurement scales and increase interpersonal comparability.1 Since its introduction, this method has been used in a growing number of countries and research areas. For example, the World Health Organization has used anchoring vignettes to assess cross-cultural differences in self-ratings of mobility, affect, pain, social relationships, vision, sleep, cognition, and self-care. In the realm of politics, anchoring vignettes have been used to explore state effectiveness in Eastern Europe (Grzymala-Busse, 2007), political efficacy in China and Mexico (King et al., 2004; Wand, 2013), interpretations of democracy in 19 African nations (Bratton, 2010), and expert placements of European political parties (Bakker, Jolly, Polk, & Poole, 2015). But while researchers are using anchoring vignettes to compare responses from individuals in different countries, it seems that the potential usefulness of the method for comparing responses within one single country, namely, the United States, has largely been ignored.2 Just as anchoring vignettes can account for differences in interpretation that stem from one’s country of residence, it seems they should also be helpful in exploring differences that arise from other factors that have been demonstrated to color political perceptions. In the case of the United States, perhaps the largest influencer of political interpretation is partisanship. Indeed, a wealth of previous research shows that an individual’s partisanship colors perceptions of not just subjective opinions but even the most objective of subjects (Bartels, 2002; Farwell & Weiner, 2000; Gaines, Kuklinski, Quirk, Peyton, & Verkuilen, 2007; Mitchell, Hibbing, Smith, & Hibbing, 2013; Simas, 2016). As such, I extend this line of research by using anchoring vignettes to explicitly show that partisanship is a significant driver of DIF in the use of one of the most common American survey measures: the 7-point ideological scale. I argue and show that because partisans use ideological terms in different ways, Democrats have significantly lower thresholds for the distinctions between all categories of this scale. This means that even when Democrats and Republicans place themselves at the same point, Democratic respondents should still have considerably more liberal ideologies than their Republican counterparts, and that failing to account for these perceptual differences could lead to an understatement of the ideological divergence between the two parties. I further illustrate this by focusing on engaged respondents and showing that the anchoring vignettes approach leads to predictions that are more consistent with the underlying issue preferences that the ideological scale is often intended to represent. Overall, the findings presented here have significant implications for not only the study of polarization in the U.S. electorate but also for the more general study of survey research, as they show that even within one country, there may be different understandings of the political world. Partisanship and Interpretations of Ideology The research presented here rests on two key theoretical claims. The first is that individuals interpret the points on the ideological scale in different ways. Recent research shows that even though citizens’ ability to correctly use the liberal–conservative scale has increased (Abramowitz & Saunders, 1998; Hetherington, 2001; Levendusky, 2009), there is still considerable heterogeneity in the policy preferences of individuals at each point on the ideological scale (Claassen, Tucker, & Smith, 2014; Ellis & Stimson, 2012; Feldman & Johnston, 2014; Treier & Hillygus, 2009). One explanation offered by this literature is centered on the idea that because ideology is comprised of distinct social and economic components (Carmines, Ensley, & Wagner, 2012; Inglehart, 1997; Layman & Carsey, 2002), the unidimensional scale may fail to capture the complexity of the concept. Thus, though citizens have become more proficient in using the unidimensional scale, “they may differ in the way they understand the dimension in terms of substantive policy content. While some citizens may see liberalism and conservatism as primarily about social issues, others may understand the dimension in terms of economics, while others may see both domains as relevant” (Feldman & Johnston, 2014, p. 340–341). The second major theory suggests that heterogeneity should still result because of a tendency of individuals to identify with ideological labels for symbolic rather than substantive reasons (Conover & Feldman, 1981). More specifically, there is evidence that a substantial number of individuals hold decidedly liberal policy opinions but avoid identifying themselves as such to escape the negative connotation of the label (Ellis & Stimson, 2012; Schiffer, 2000). This results in each point on the scale (and particularly those on the more conservative end) being comprised a mix of those whose policy preferences are consistent with that ideological label and those who simply prefer to identify themselves as such. A common thread running through these explanations is the notion that the terms “liberal,” “conservative,” and “moderate” are all essentially abstract and subject to different interpretations. That is, both of these explanations strongly suggest that the ideological scale is subject to DIF. Hare, Armstrong, Bakker, Carroll, and Poole (2015) explicitly articulate this proposition and develop a Bayesian Aldrich–McKelvey scaling method to model and correct for DIF in self-placements and the placements of senators and Senate candidates. While a major advancement in the literature in this field, Hare et al.’s (2015) work is more focused on the consequences of DIF rather than the causes, as their methodology does not allow for testing of how respondent characteristics affect perceptions of the cutoffs between ideological categories. Thus, my major contribution lies in the advancement and testing of my second major claim: Partisanship directly affects interpretations of the ideological scale such that there are two distinct, partisan understandings of this scale. I argue that there are at least two readily identifiable explanations for why Democrats and Republicans may understand and use ideological labels in systematically different ways. First, the two parties use the words “liberal” and “conservative” in different ways and with different connotations. From George H. W. Bush’s “‘L-word’ strategy, which reduced liberal to an unspeakable profanity” (Schiffer, 2000, p. 298) to George W. Bush’s self-branding as a “compassionate conservative,” the Republican Party has been largely successful in framing ideological terms in a manner favorable to itself such that Democratic candidates tend to prefer “progressive” to “liberal” (Ellis & Stimson, 2012; Sears & Citrin, 1985). And even when candidates themselves do not explicitly use these types of ideological labels, the media still tends to use them in a nonneutral manner (Ellis & Stimson, 2012). Overall, the partisan slant placed on ideological labels means that individuals’ receptivity of this messaging should be at least somewhat contingent on their partisanship (Zaller, 1992). Indeed, recent works show that citizens are most greatly influenced by the framing used by their own party’s leaders and media (Levendusky, 2013; Slothuus & de Vreese, 2010). Thus, assuming that each party attempts to define and use ideological terminology to its advantage, it follows that there should be at least two distinctly partisan conceptions of the ideological scale reflected in the self-placements of the public. Second, differential utilization of the ideological scale may be related to the psychological tendency of individuals to evaluate themselves not in an absolute sense but rather in comparison with others (Festinger, 1954). As Heine, Lehman, Peng, and Greenholtz (2002) explain in a clear example, calling oneself tall “depends on comparisons with appropriate targets…The same height—for example, 5 feet 9 in.—is seen as tall in some contexts (e.g., among elementary school students or Japanese women) and short in others (e.g., among professional basketball players or Dutch men)” (p. 904). How one determines the appropriate target for comparison has been the subject of much research, but the dominant conclusion appears to be that individuals most often and meaningfully make self-assessments in comparison with fellow group members or to those whom they see as most similar to themselves (Festinger, 1954; Suls, Martin, & Wheeler, 2002; Wheeler, Martin, & Suls, 1997; Zanna, Goethals, & Hill, 1975). As party identification has been shown to be a strong group identification akin to gender or race (Green, Palmquist, & Schickler, 2002; Greene, 1999), it is logical to think that when assessing oneself on political matters, individuals will use their political party as a reference group. Translating the height example to apply to ideological self-placements, then, it follows that when a Democratic individual self-categorizes his ideology as “middle of the road,” he is doing so not as an assessment of his ideology as moderate compared with some universal standard, but rather, relative to other members of the group to which he belongs and sees as most similar to himself—the Democratic Party. Assuming that a Republican respondent would similarly place himself in comparison with his party, this would offer an explanation as to why two individuals may place themselves at the same point on the scale when they hold vastly different policy positions. Method The technique of anchoring vignettes involves showing respondents a series of short texts where each provides a concrete description of the trait of interest (in this case, political ideology) of a different hypothetical individual. After reading each vignette, respondents are asked to rate that individual on the same scale used for self-placement. The value-added of these vignette placements is illustrated in Figure 1. Figure 1 View largeDownload slide The logic of anchoring vignettes. Figure 1 View largeDownload slide The logic of anchoring vignettes. Panel A of Figure 1 displays the traditional interpretation of respondent self-placements; as both Respondent A and Respondent B placed themselves at the same point on the ideological scale, they are assumed to have the same underlying ideology. As discussed above, however, this assumption is flawed if the respondents have fundamentally different definitions of the points on the ideological scale. Respondent placements of the anchoring vignettes help address this possibility. Because the vignettes describe individuals with clear policy positions, the respondents’ placements of these individuals provide insight into how the respondents define the various points on the ideological scale. That is, the individual described in each vignette has the same actual ideological rating, regardless of the party of the respondent, and so if a Democratic respondent rates the individual described as “somewhat liberal,” and a Republican respondent rates that same individual as “liberal,” this is suggestive that the two respondents have different interpretations of the package of policies the term “liberal” represents. By using these vignette placements of the same individuals as “anchors” and rescaling placements accordingly, we can then account for different perceptions of the scale and make more meaningful and accurate comparisons of self-placements based on where they are relative to these common anchors. For example, in Panel A of Figure 1, Respondents A and B placed themselves at the same point, but Respondent A placed herself between the two anchors, and Respondent B placed herself as more liberal than both of the anchors. Adjusting for this in Panel B of Figure 1, it becomes clear that the more appropriate conclusion may be that Respondent A is more conservative than Respondent B, as regardless of the labels used, Respondent A placed herself as more conservative than Anchor 2 and Respondent B did not. Vignette Texts and Data The anchoring vignettes were placed on a 1,000-respondent module of the 2012 Cooperative Congressional Election Survey (CCES; Ansolabehere, 2013)3 postelection survey. Table 1 displays the texts of the three vignettes, which were shown to respondents in random order. Table 1 Texts of Anchoring Vignettes Full sample Democrats only Republicans only Mean (SD) Mean (SD) Mean (SD) Prompt The following is a statement of political beliefs. Please read the statement and then answer the question that follows. Vignette 1 “I believe that all women should have the right to an abortion, that homosexuals should have the right to marry, and that all immigrants should be welcome. The fight over constant efforts to deny these fundamental rights keeps the government from focusing on what it really should be working on—making sure that every American has free healthcare. The rich and big businesses—especially those that harm our environment—should be forced to pay their rightful share of taxes so that our children can get a better education and our streets can be safe. Instead of continuing to pour money into unnecessary and unwinnable wars, the government should be providing for the basic needs of the people here at home.” 1.79 (1.20) 1.91 (1.12) 1.48 (1.16) Vignette 2 “I like the idea of the government doing more to make sure that all Americans have things like healthcare and a college education, but Americans are already taxed enough. I think that there needs to be a balance between what the government provides and what people can get from private vendors because the government can only do so much and needs to think about balancing the budget. I think that while we should continue to be tough on crime, we also need to consider rehabilitation rather than just incarceration for nonviolent offenders and reserve the death penalty only for cases with DNA proof. Homosexual couples should enjoy the same rights and benefits as heterosexual couples, but I believe this should be called a civil union, not a marriage.” 3.39 (1.43) 3.69 (1.37) 3.01 (1.44) Vignette 3 “I believe that the government’s first priority should be the security of this country. Abroad, we need to maintain a strong military and continue to support the efforts of our troops to lessen the threat of global terrorism. Here at home, law enforcement officers should be given the ability to identify and deport those who are in this country illegally. I also think that we need harsher punishments for criminals and broader application of the death penalty, especially for those who commit crimes against children or law enforcement officers. The government also needs to protect the values of American families by maintaining that marriage is a between a man and a woman and ensuring that every baby has the right to life. Moreover, individuals should be allowed to pray or observe religious traditions in schools if they so choose.” 5.91 (1.41) 6.00 (1.54) 5.99 (1.10) Placement question How would you describe the political ideology of this individual? (Response categories: Very Liberal [1], Liberal [2], Somewhat Liberal [3], Middle of the Road [4], Somewhat Conservative [5], Conservative [6], Very Conservative [7]) Full sample Democrats only Republicans only Mean (SD) Mean (SD) Mean (SD) Prompt The following is a statement of political beliefs. Please read the statement and then answer the question that follows. Vignette 1 “I believe that all women should have the right to an abortion, that homosexuals should have the right to marry, and that all immigrants should be welcome. The fight over constant efforts to deny these fundamental rights keeps the government from focusing on what it really should be working on—making sure that every American has free healthcare. The rich and big businesses—especially those that harm our environment—should be forced to pay their rightful share of taxes so that our children can get a better education and our streets can be safe. Instead of continuing to pour money into unnecessary and unwinnable wars, the government should be providing for the basic needs of the people here at home.” 1.79 (1.20) 1.91 (1.12) 1.48 (1.16) Vignette 2 “I like the idea of the government doing more to make sure that all Americans have things like healthcare and a college education, but Americans are already taxed enough. I think that there needs to be a balance between what the government provides and what people can get from private vendors because the government can only do so much and needs to think about balancing the budget. I think that while we should continue to be tough on crime, we also need to consider rehabilitation rather than just incarceration for nonviolent offenders and reserve the death penalty only for cases with DNA proof. Homosexual couples should enjoy the same rights and benefits as heterosexual couples, but I believe this should be called a civil union, not a marriage.” 3.39 (1.43) 3.69 (1.37) 3.01 (1.44) Vignette 3 “I believe that the government’s first priority should be the security of this country. Abroad, we need to maintain a strong military and continue to support the efforts of our troops to lessen the threat of global terrorism. Here at home, law enforcement officers should be given the ability to identify and deport those who are in this country illegally. I also think that we need harsher punishments for criminals and broader application of the death penalty, especially for those who commit crimes against children or law enforcement officers. The government also needs to protect the values of American families by maintaining that marriage is a between a man and a woman and ensuring that every baby has the right to life. Moreover, individuals should be allowed to pray or observe religious traditions in schools if they so choose.” 5.91 (1.41) 6.00 (1.54) 5.99 (1.10) Placement question How would you describe the political ideology of this individual? (Response categories: Very Liberal [1], Liberal [2], Somewhat Liberal [3], Middle of the Road [4], Somewhat Conservative [5], Conservative [6], Very Conservative [7]) Table 1 Texts of Anchoring Vignettes Full sample Democrats only Republicans only Mean (SD) Mean (SD) Mean (SD) Prompt The following is a statement of political beliefs. Please read the statement and then answer the question that follows. Vignette 1 “I believe that all women should have the right to an abortion, that homosexuals should have the right to marry, and that all immigrants should be welcome. The fight over constant efforts to deny these fundamental rights keeps the government from focusing on what it really should be working on—making sure that every American has free healthcare. The rich and big businesses—especially those that harm our environment—should be forced to pay their rightful share of taxes so that our children can get a better education and our streets can be safe. Instead of continuing to pour money into unnecessary and unwinnable wars, the government should be providing for the basic needs of the people here at home.” 1.79 (1.20) 1.91 (1.12) 1.48 (1.16) Vignette 2 “I like the idea of the government doing more to make sure that all Americans have things like healthcare and a college education, but Americans are already taxed enough. I think that there needs to be a balance between what the government provides and what people can get from private vendors because the government can only do so much and needs to think about balancing the budget. I think that while we should continue to be tough on crime, we also need to consider rehabilitation rather than just incarceration for nonviolent offenders and reserve the death penalty only for cases with DNA proof. Homosexual couples should enjoy the same rights and benefits as heterosexual couples, but I believe this should be called a civil union, not a marriage.” 3.39 (1.43) 3.69 (1.37) 3.01 (1.44) Vignette 3 “I believe that the government’s first priority should be the security of this country. Abroad, we need to maintain a strong military and continue to support the efforts of our troops to lessen the threat of global terrorism. Here at home, law enforcement officers should be given the ability to identify and deport those who are in this country illegally. I also think that we need harsher punishments for criminals and broader application of the death penalty, especially for those who commit crimes against children or law enforcement officers. The government also needs to protect the values of American families by maintaining that marriage is a between a man and a woman and ensuring that every baby has the right to life. Moreover, individuals should be allowed to pray or observe religious traditions in schools if they so choose.” 5.91 (1.41) 6.00 (1.54) 5.99 (1.10) Placement question How would you describe the political ideology of this individual? (Response categories: Very Liberal [1], Liberal [2], Somewhat Liberal [3], Middle of the Road [4], Somewhat Conservative [5], Conservative [6], Very Conservative [7]) Full sample Democrats only Republicans only Mean (SD) Mean (SD) Mean (SD) Prompt The following is a statement of political beliefs. Please read the statement and then answer the question that follows. Vignette 1 “I believe that all women should have the right to an abortion, that homosexuals should have the right to marry, and that all immigrants should be welcome. The fight over constant efforts to deny these fundamental rights keeps the government from focusing on what it really should be working on—making sure that every American has free healthcare. The rich and big businesses—especially those that harm our environment—should be forced to pay their rightful share of taxes so that our children can get a better education and our streets can be safe. Instead of continuing to pour money into unnecessary and unwinnable wars, the government should be providing for the basic needs of the people here at home.” 1.79 (1.20) 1.91 (1.12) 1.48 (1.16) Vignette 2 “I like the idea of the government doing more to make sure that all Americans have things like healthcare and a college education, but Americans are already taxed enough. I think that there needs to be a balance between what the government provides and what people can get from private vendors because the government can only do so much and needs to think about balancing the budget. I think that while we should continue to be tough on crime, we also need to consider rehabilitation rather than just incarceration for nonviolent offenders and reserve the death penalty only for cases with DNA proof. Homosexual couples should enjoy the same rights and benefits as heterosexual couples, but I believe this should be called a civil union, not a marriage.” 3.39 (1.43) 3.69 (1.37) 3.01 (1.44) Vignette 3 “I believe that the government’s first priority should be the security of this country. Abroad, we need to maintain a strong military and continue to support the efforts of our troops to lessen the threat of global terrorism. Here at home, law enforcement officers should be given the ability to identify and deport those who are in this country illegally. I also think that we need harsher punishments for criminals and broader application of the death penalty, especially for those who commit crimes against children or law enforcement officers. The government also needs to protect the values of American families by maintaining that marriage is a between a man and a woman and ensuring that every baby has the right to life. Moreover, individuals should be allowed to pray or observe religious traditions in schools if they so choose.” 5.91 (1.41) 6.00 (1.54) 5.99 (1.10) Placement question How would you describe the political ideology of this individual? (Response categories: Very Liberal [1], Liberal [2], Somewhat Liberal [3], Middle of the Road [4], Somewhat Conservative [5], Conservative [6], Very Conservative [7]) As Table 1 shows, the vignette texts were kept free of any demographic information (gender, race, age, etc.) that may potentially influence placements and to encourage as much response consistency as possible. Though the same exact issues are not discussed in each vignette, all three reference both social issues and issues related to the role of government, allowing for a broad interpretation that is not focused solely on one policy dimension. And while the ideological scale has seven points, only three vignettes were used, as King and Wand (2007) assert that it is not necessary that the number of vignettes used equal the number of points on the scale in question. Rather, vignettes should be chosen to maximize their joint ability to differentiate between respondents. In other words, a vignette is only informative to the extent that respondents could logically place themselves below, equal to, and above it. For example, if one was measuring self-assessments of personal wealth in the general population, then a vignette that described an individual with an economic situation similar to that of Bill Gates would not be useful, as it should be expected that all respondents would rate themselves as lower than this vignette. Additionally, vignettes that are too similar to one another do not provide much additional leverage. The logic behind this statement is highlighted by the example presented in Figure 1. If the two respondents had failed to see a meaningful difference between Anchors 1 and 2, then it would not have been apparent that Respondent A saw herself between the two anchors, and the differences between the respondents would have been masked. With these two goals of avoiding extreme or indiscriminate vignettes and the time and cost constraints of the survey administration, I ended up with three vignettes that aimed to represent the three major ideological categories on the scale: “liberal,” “middle of the road,” and “conservative.”4 In total, 16% of respondents failed to place any of the vignettes.5 Dropping those who did not place themselves (4.8% of those placing vignettes) and those who gave line item responses (10.1% of those placing vignettes) reduces the analytical sample to about 71.4% of the total.6 The rightmost columns of Table 1 show the mean placements for each vignette for the full sample and for each group of partisans.7 These means show that the intended rank ordering of the vignettes was not far off. Looking at the partisan differences in the placements, there is preliminary evidence of different interpretations of the scale. Republican respondents placed all three vignettes at more liberal positions on the scale. The difference in partisans’ placements of Vignette 3 is not significant, however. This suggests that (a) differences in interpretation are limited to just the liberal points on the scale; and/or (b) an additional vignette on the more conservative end of the scale may have been helpful. Still, there are at least some partisan differences that need to be explored, and I can proceed with overall confidence in the ability of the vignette ratings to usefully discriminate between respondents. Assumptions Use of anchoring vignettes involves two primary measurement assumptions: response consistency and vignette equivalence (King et al., 2004). Response consistency is the assumption that respondents use the categories of the scale the same way when placing both themselves and the individuals featured in the vignettes. If this assumption holds, it implies that if two respondents have the same underlying level of ideology, a more conservative self-rater should also be a more conservative vignette rater. Though there is no theoretical reason to believe that this assumption would be violated, I test for response consistency by regressing each respondent’s average vignette ratings on self-placement while controlling for underlying issue preferences with scores from two different issue indices featured in the CCES Common Content.8 The ordered probit coefficient for the vignette placements is positive and significant (p = .01), indicating that more conservative self-raters were on average, more conservative vignette raters and suggesting that response consistency can be safely assumed. The second assumption, vignette equivalence, is the assumption that each vignette is in fact measurable on the same scale. This is perhaps the more important of the two assumptions to check, as the unidimensionality of the ideological scale is often debated. Yet despite potential concerns, the means presented in Table 1 suggest that this assumption, too, has been met. While there was variation in the placements of the vignettes, the intended ordinality of vignettes is largely preserved. Indeed, 85.9% of the analytical sample gave ratings consistent with the intended rank ordering. Analytical Model Ordinal scales with relatively few categories are typically analyzed with ordered probit models. The central assumption of these types of ordinal models is that underlying the observed j responses is a continuous, latent scale on which there are j-1 points that serve as thresholds (τ) differentiating between the observed categories. That is, for the 7-point ideological scale, the ordered probit model estimates six thresholds that represent the latent values at which an individual crosses over from being in one ideological category to the next. The thresholds estimated by an ordered probit model, however, are assumed to be the same across all individuals. This assumption is at odds with the concept of DIF and the hypothesis being explored here, as I expect these thresholds to vary depending on the party of the individual respondent. Using a more concrete example, I expect that the point a Democratic respondent sees as being the divide between someone who is “middle of the road” versus someone who is “somewhat conservative” should be significantly different from the dividing point of a Republican. While a basic ordered probit model does not allow for these potential difference, the joint compound hierarchical ordered probit (CHOPIT) model does. For a full specification, see Rabe-Hesketh and Skrondal (2002) and King et al. (2004), but to summarize, the CHOPIT model explicitly tests and accounts for DIF by incorporating the ratings of the vignettes and estimating a set of thresholds for each independent variable in the model. This allows me to test for the effects of partisan-driven DIF while controlling for other possible factors. It is this ability to model thresholds and specifically link DIF to my independent variables that provides new leverage in comparison with other methods such as the nonparametric approach advocated by Wand (2013) or the Bayesian Aldrich–McKelvey scaling advanced by Hare et al. (2015). Thus, I rely on a CHOPIT model where the dependent variable is the respondent's ideological self-placement. Threshold estimates from the independent variable Democratic partisanship will reveal if there are significant differences between these respondents and Republican respondents, who are the omitted baseline category.9 Results Table 2 presents results from both a standard ordered probit model and the CHOPIT model. Table 2 Ordered Probit and Compound Hierarchical Ordered Probit Regression of Ideological Self-Rating Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Democratic respondent −2.33*** (.17) −2.53*** (.19) −0.75*** (.15) −0.40** (.13) −0.26 (.14) −0.03 (.13) −0.35** (.13) −0.52** (.16) Independent respondent −0.99*** (.26) −1.11*** (.31) −0.30 (.28) −0.24 (.23) −0.29 (.23) 0.12 (.24) −0.22 (.25) −0.19 (.29) Age 0.01*** (.00) 0.01*** (.00) 0.00 (.00) 0.01* (.00) 0.01 (.00) −0.00 (.00) −0.00 (.00) −0.01 (.00) Female 0.03 (.12) 0.08 (.15) 0.11 (.14) 0.07 (.13) 0.09 (.13) −0.03 (.12) 0.07 (.12) 0.06 (.15) Education −0.01 (.04) −0.06 (.05) −0.05 (.05) −0.08 (.04) −0.02 (.04) −0.07 (.04) −0.05 (.04) −0.05 (.05) Political sophistication −0.38 (.20) −0.46 (.27) −0.03 (.26) −0.05 (.23) −0.08 (.23) −0.17 (.21) −0.08 (.23) −0.06 (.26) Non-White 0.39** (.19) 0.61* (.24) 0.36 (.23) 0.34 (.19) 0.16 (.19) 0.14 (.18) 0.23 (.19) −0.11 (.23) Income −0.01 (.02) −0.02 (.03) −0.05* (.02) −0.02 (.02) −0.03 (.02) −0.01 (.02) 0.01 (.02) −0.01 (.02) Church attendance −0.04 (.03) −0.03 (.04) −0.00 (.03) 0.02 (.03) 0.03 (.03) −0.01 (.03) −0.02 (.03) 0.01 (.04) South 0.03 (.12) 0.07 (.27) 0.15 (.14) −0.00 (.12) 0.05 (.13) 0.00 (.21) 0.09 (.12) 0.03 (.15) Constant a −2.57*** (.48) −2.40*** (.39) −1.92*** (.38) −0.35 (.39) 0.01 (.39) 1.18*** (44) Anchor 1 −3.53*** (.48) Anchor 2 −1.84*** (.42) Anchor 3 0.17 (.43) Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Democratic respondent −2.33*** (.17) −2.53*** (.19) −0.75*** (.15) −0.40** (.13) −0.26 (.14) −0.03 (.13) −0.35** (.13) −0.52** (.16) Independent respondent −0.99*** (.26) −1.11*** (.31) −0.30 (.28) −0.24 (.23) −0.29 (.23) 0.12 (.24) −0.22 (.25) −0.19 (.29) Age 0.01*** (.00) 0.01*** (.00) 0.00 (.00) 0.01* (.00) 0.01 (.00) −0.00 (.00) −0.00 (.00) −0.01 (.00) Female 0.03 (.12) 0.08 (.15) 0.11 (.14) 0.07 (.13) 0.09 (.13) −0.03 (.12) 0.07 (.12) 0.06 (.15) Education −0.01 (.04) −0.06 (.05) −0.05 (.05) −0.08 (.04) −0.02 (.04) −0.07 (.04) −0.05 (.04) −0.05 (.05) Political sophistication −0.38 (.20) −0.46 (.27) −0.03 (.26) −0.05 (.23) −0.08 (.23) −0.17 (.21) −0.08 (.23) −0.06 (.26) Non-White 0.39** (.19) 0.61* (.24) 0.36 (.23) 0.34 (.19) 0.16 (.19) 0.14 (.18) 0.23 (.19) −0.11 (.23) Income −0.01 (.02) −0.02 (.03) −0.05* (.02) −0.02 (.02) −0.03 (.02) −0.01 (.02) 0.01 (.02) −0.01 (.02) Church attendance −0.04 (.03) −0.03 (.04) −0.00 (.03) 0.02 (.03) 0.03 (.03) −0.01 (.03) −0.02 (.03) 0.01 (.04) South 0.03 (.12) 0.07 (.27) 0.15 (.14) −0.00 (.12) 0.05 (.13) 0.00 (.21) 0.09 (.12) 0.03 (.15) Constant a −2.57*** (.48) −2.40*** (.39) −1.92*** (.38) −0.35 (.39) 0.01 (.39) 1.18*** (44) Anchor 1 −3.53*** (.48) Anchor 2 −1.84*** (.42) Anchor 3 0.17 (.43) Note. N = 611; Cell entries are estimated coefficients with robust standard errors in parentheses; *** p<.001, ** p<.01, * p<.05. CHOPIT = compound hierarchical ordered probit. aEstimated τs for the ordered probit model are −3.16,−2.28,−1.79, −0.55, 0.05, and 0.91. Table 2 Ordered Probit and Compound Hierarchical Ordered Probit Regression of Ideological Self-Rating Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Democratic respondent −2.33*** (.17) −2.53*** (.19) −0.75*** (.15) −0.40** (.13) −0.26 (.14) −0.03 (.13) −0.35** (.13) −0.52** (.16) Independent respondent −0.99*** (.26) −1.11*** (.31) −0.30 (.28) −0.24 (.23) −0.29 (.23) 0.12 (.24) −0.22 (.25) −0.19 (.29) Age 0.01*** (.00) 0.01*** (.00) 0.00 (.00) 0.01* (.00) 0.01 (.00) −0.00 (.00) −0.00 (.00) −0.01 (.00) Female 0.03 (.12) 0.08 (.15) 0.11 (.14) 0.07 (.13) 0.09 (.13) −0.03 (.12) 0.07 (.12) 0.06 (.15) Education −0.01 (.04) −0.06 (.05) −0.05 (.05) −0.08 (.04) −0.02 (.04) −0.07 (.04) −0.05 (.04) −0.05 (.05) Political sophistication −0.38 (.20) −0.46 (.27) −0.03 (.26) −0.05 (.23) −0.08 (.23) −0.17 (.21) −0.08 (.23) −0.06 (.26) Non-White 0.39** (.19) 0.61* (.24) 0.36 (.23) 0.34 (.19) 0.16 (.19) 0.14 (.18) 0.23 (.19) −0.11 (.23) Income −0.01 (.02) −0.02 (.03) −0.05* (.02) −0.02 (.02) −0.03 (.02) −0.01 (.02) 0.01 (.02) −0.01 (.02) Church attendance −0.04 (.03) −0.03 (.04) −0.00 (.03) 0.02 (.03) 0.03 (.03) −0.01 (.03) −0.02 (.03) 0.01 (.04) South 0.03 (.12) 0.07 (.27) 0.15 (.14) −0.00 (.12) 0.05 (.13) 0.00 (.21) 0.09 (.12) 0.03 (.15) Constant a −2.57*** (.48) −2.40*** (.39) −1.92*** (.38) −0.35 (.39) 0.01 (.39) 1.18*** (44) Anchor 1 −3.53*** (.48) Anchor 2 −1.84*** (.42) Anchor 3 0.17 (.43) Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Democratic respondent −2.33*** (.17) −2.53*** (.19) −0.75*** (.15) −0.40** (.13) −0.26 (.14) −0.03 (.13) −0.35** (.13) −0.52** (.16) Independent respondent −0.99*** (.26) −1.11*** (.31) −0.30 (.28) −0.24 (.23) −0.29 (.23) 0.12 (.24) −0.22 (.25) −0.19 (.29) Age 0.01*** (.00) 0.01*** (.00) 0.00 (.00) 0.01* (.00) 0.01 (.00) −0.00 (.00) −0.00 (.00) −0.01 (.00) Female 0.03 (.12) 0.08 (.15) 0.11 (.14) 0.07 (.13) 0.09 (.13) −0.03 (.12) 0.07 (.12) 0.06 (.15) Education −0.01 (.04) −0.06 (.05) −0.05 (.05) −0.08 (.04) −0.02 (.04) −0.07 (.04) −0.05 (.04) −0.05 (.05) Political sophistication −0.38 (.20) −0.46 (.27) −0.03 (.26) −0.05 (.23) −0.08 (.23) −0.17 (.21) −0.08 (.23) −0.06 (.26) Non-White 0.39** (.19) 0.61* (.24) 0.36 (.23) 0.34 (.19) 0.16 (.19) 0.14 (.18) 0.23 (.19) −0.11 (.23) Income −0.01 (.02) −0.02 (.03) −0.05* (.02) −0.02 (.02) −0.03 (.02) −0.01 (.02) 0.01 (.02) −0.01 (.02) Church attendance −0.04 (.03) −0.03 (.04) −0.00 (.03) 0.02 (.03) 0.03 (.03) −0.01 (.03) −0.02 (.03) 0.01 (.04) South 0.03 (.12) 0.07 (.27) 0.15 (.14) −0.00 (.12) 0.05 (.13) 0.00 (.21) 0.09 (.12) 0.03 (.15) Constant a −2.57*** (.48) −2.40*** (.39) −1.92*** (.38) −0.35 (.39) 0.01 (.39) 1.18*** (44) Anchor 1 −3.53*** (.48) Anchor 2 −1.84*** (.42) Anchor 3 0.17 (.43) Note. N = 611; Cell entries are estimated coefficients with robust standard errors in parentheses; *** p<.001, ** p<.01, * p<.05. CHOPIT = compound hierarchical ordered probit. aEstimated τs for the ordered probit model are −3.16,−2.28,−1.79, −0.55, 0.05, and 0.91. Respondent age, gender, race, education, political sophistication, income, church attendance, and region of residence are also controlled, and analyses are weighted for probability of selection.10 The first two columns show the coefficient estimates for the two models. The bottom of the second column displays the estimates of the vignette locations. These estimates are monotonically increasing. This fact supports my earlier claims of vignette equivalence and the appropriateness of the unidimensional scale, as they again show a consistency in the ordering of the vignettes. The remaining columns display the other unique aspect of the CHOPIT model: estimated thresholds (τ) for every independent variable in the model. The large majority of these estimates are not statistically significant. Particularly notable are the null results for both education and political sophistication. This means that if two individuals identify with the same party but have different levels of education or sophistication, their interpretations of the ideological scale should not differ. However, the negative (and significant in four of the six cases) threshold estimates for the Democratic respondent variable, indicate that partisanship does affect impressions of the distinctions between ideological categories. That is, if two individuals are of the same level of education or sophistication but one is a Democrat and the other a Republican, then they should have significantly different perceptions of the ideological scale. Thus, the key difference exposed by the CHOPIT model is related to partisanship, and not knowledge. The significant, negative coefficient for τ1 means that when placing a given vignette, Democrats are more likely than Republicans to choose “liberal” rather than “extremely liberal. Because the parameterization of the CHOPIT model makes interpretation of higher-order thresholds dependent on those prior, I follow Grol-Prokopczyk, Freese, and Hauser (2011) and more clearly present the results by visually illustrating the mean estimates for each category of respondents in the left-hand panel of Figure 2.11 Figure 2 View largeDownload slide Estimated mean thresholds for the ideological scale Figure 2 View largeDownload slide Estimated mean thresholds for the ideological scale As this figure shows (and difference of means tests confirm), Democratic and Republican respondents have significantly different thresholds for distinguishing between each set of responses on the 7-point ideological scale. So if two respondents hold the same issue positions but are from different parties, the Democratic respondent is more likely to classify his issue positions as more conservative than his Republican counterpart, as his perceptions of where the thresholds between categories lie are all lower or more liberal. Consistent with theories of the negative affect associated with the term “liberal,” these differences in perceptions are largest at the left end of the scale. For example, the mean Republican rating for τ1 is 3.01. This means that for a Republican, any latent value below this should be classified as “extremely liberal.” For a Democrat, however, 3.01 falls between the mean estimates for τ3 and τ4. This means that what is defined as “extremely liberal” by a Republican could be defined as anywhere from “extremely liberal” to “middle of the road” by a Democrat. At the opposite end, the similar estimates for Democratic τ5 and Republican τ4 suggest that what is “conservative” to a Democrat may only be “somewhat conservative” to a Republican. Note that this contrasts with Table 1’s suggestion of agreement over the meaning of “conservative,” as these threshold results show that differences in interpretation do in fact exist at all points. As such, it is clear that within each party, there are distinct definitions of the labels associated with the entire ideological scale. Implications for Ideological Self-Identification So far, I have shown that Democrats and Republicans interpret the ideological scale in significantly different ways. How, then, do these different conceptions of the scale points and thresholds between them impact our interpretations of ideological self-identification? The importance of accounting for these perceptual differences can most simply be seen by comparing the ordered probit and CHOPIT coefficients in Table 2. While both models produce significant, negative estimates for the Democratic respondent variable, the CHOPIT model estimates a larger effect of partisanship. This suggests that failing to account for differences in the perceptions of the ideological scale may lead to an understatement of the ideological polarization between members of the two parties. Yet, the polarization literature (Abramowitz, 2011, 2013) shows that the type of partisan differentiation exposed here is most prevalent among and perhaps limited to those who take interest and participate in the political process. In addition, engaged citizens are the most likely to be exposed to consistently partisan messages (Iyengar & Hahn, 2009) and the most motivated to respond to those messages (Slothuus & de Vreese, 2010). It is likely, then, that the type of distortion that I am claiming exists is strongest among them.12 Because my analyses are necessarily limited to only those respondents who placed themselves and the vignettes, it is possible that those who are willing and/or able to assess ideology on this scale do so exactly because they are engaged citizens, and thus, my method may be driving my results. To assess this possibility, I compare willingness/ability to place the vignettes with characteristics identified as predictors of political engagement. The full comparison is available in Supplemental Appendix D, but to summarize, the probit analysis reveals that there are no statistically significant differences due to of education or intensity of either partisanship or ideology. As Abramowitz (2011) finds these to be three of the strongest predictors of political engagement, I reference this as evidence that the results presented here are not simply artifacts of my restricted sample. Still, I further explore the possibility that partisan distortion may be stronger among a specific type of individual by rerunning the CHOPIT model among two subsets of my sample: those who report taking part in one or more political act beyond voting (i.e., the politically engaged; 56% of my analytical sample) and those who do not (44% of my analytical sample). Table 3 displays the key coefficients of interest for both subsamples.13 Table 3 Ordered Probit and Compound Hierarchical Ordered Probit Regression of Ideological Self-Rating by Political Engagement Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Politically engaged respondents N=341 Democratic respondent −3.11*** (.28) −3.57*** (.32) −1.38*** (.19) −0.71*** (.18) −0.57*** (.21) −0.38* (.18) −0.72*** (.19) −0.52* (.22) Anchor 1 −3.59*** (.70) Anchor 2 −1.60* (.62) Anchor 3 1.18 (.63) Unengaged respondents N=270 Democratic respondent −1.93*** (.22) −1.92*** (.26) −0.33 (.23) −0.22 (.21) −0.07 (.21) 0.25 (.20) −0.07 (.20) −0.56* (.25) Anchor 1 −3.18*** (.69) Anchor 2 −1.55*** (.63) Anchor 3 0.13 (.64) Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Politically engaged respondents N=341 Democratic respondent −3.11*** (.28) −3.57*** (.32) −1.38*** (.19) −0.71*** (.18) −0.57*** (.21) −0.38* (.18) −0.72*** (.19) −0.52* (.22) Anchor 1 −3.59*** (.70) Anchor 2 −1.60* (.62) Anchor 3 1.18 (.63) Unengaged respondents N=270 Democratic respondent −1.93*** (.22) −1.92*** (.26) −0.33 (.23) −0.22 (.21) −0.07 (.21) 0.25 (.20) −0.07 (.20) −0.56* (.25) Anchor 1 −3.18*** (.69) Anchor 2 −1.55*** (.63) Anchor 3 0.13 (.64) Note. Cell entries are estimated coefficients with robust standard errors in parentheses; *** p<.001, ** p<.01, * p<.05. CHOPIT = compound hierarchical ordered probit. Full results are available in Supplementary Appendix E. Table 3 Ordered Probit and Compound Hierarchical Ordered Probit Regression of Ideological Self-Rating by Political Engagement Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Politically engaged respondents N=341 Democratic respondent −3.11*** (.28) −3.57*** (.32) −1.38*** (.19) −0.71*** (.18) −0.57*** (.21) −0.38* (.18) −0.72*** (.19) −0.52* (.22) Anchor 1 −3.59*** (.70) Anchor 2 −1.60* (.62) Anchor 3 1.18 (.63) Unengaged respondents N=270 Democratic respondent −1.93*** (.22) −1.92*** (.26) −0.33 (.23) −0.22 (.21) −0.07 (.21) 0.25 (.20) −0.07 (.20) −0.56* (.25) Anchor 1 −3.18*** (.69) Anchor 2 −1.55*** (.63) Anchor 3 0.13 (.64) Ordered probit coefficient CHOPIT coefficient CHOPIT τ1 CHOPIT τ2 CHOPIT τ3 CHOPIT τ4 CHOPIT τ5 CHOPIT τ6 Politically engaged respondents N=341 Democratic respondent −3.11*** (.28) −3.57*** (.32) −1.38*** (.19) −0.71*** (.18) −0.57*** (.21) −0.38* (.18) −0.72*** (.19) −0.52* (.22) Anchor 1 −3.59*** (.70) Anchor 2 −1.60* (.62) Anchor 3 1.18 (.63) Unengaged respondents N=270 Democratic respondent −1.93*** (.22) −1.92*** (.26) −0.33 (.23) −0.22 (.21) −0.07 (.21) 0.25 (.20) −0.07 (.20) −0.56* (.25) Anchor 1 −3.18*** (.69) Anchor 2 −1.55*** (.63) Anchor 3 0.13 (.64) Note. Cell entries are estimated coefficients with robust standard errors in parentheses; *** p<.001, ** p<.01, * p<.05. CHOPIT = compound hierarchical ordered probit. Full results are available in Supplementary Appendix E. Looking first as those who are not engaged, there is little evidence of partisan distortion, as only one of the threshold estimates is statistically significant. However, this should not be construed as evidence that those who are not engaged do not think ideologically. To the contrary, the estimated anchors for this subset of the sample are still monotonically increasing and consistent with a unidimensional interpretation of the scale. In addition, the coefficient for the partisanship variable is still negative and significant, indicating a predictable relationship between partisan and ideological self-identifications. Rather, the differences between the two models just show that those who are not engaged are not as polarized as their more interested and involved counterparts. This is consistent with works on motivated reasoning and the argument that while partisanship determines the direction of bias, “citizens’ engagement with politics should influence the strength of bias” ( Slothuus & de Vreese, 2010, p. 633). And indeed, restricting the analysis to just the more interested and involved respondents reveals greater bias and significant polarization. The right-hand panel of Figure 2 shows that the partisan differences in the threshold estimates are even greater than those uncovered when analyzing all respondents. The mean Republican estimate for τ1 (−2.63) now falls between the Democratic estimates for τ4 and τ5. This means that there is even less overlap between partisan perceptions than previously estimated. What each party defines as “middle of the road” (4) is seen as fully ideological by the opposite party. That is, a Democratic 4 is equivalent to a Republican 1, and a Republican 4 is equivalent to a Democratic 6. To further illustrate this, Table 4 maps this lack of overlap back onto actual policy preferences. I construct a 0–1 scale from preferences on Congressional bills regarding the budget, taxes, the birth control exemption, the Affordable Healthcare Act, the Keystone Pipeline, and ending “Don’t Ask, Don’t Tell” (α = .79).14 Though these bills span a wide range of policy areas, they load onto a single factor,15 suggesting a fit with the unidimensional nature of the ideology measure. Table 4 Comparison of the Engaged Respondents’ Policy Preferences Average roll-call battery placement Analytical sample Full CCES Mean (SD) Mean (SD) Democrats Republicans Democrats Republicans Extremely liberal (1) .09 .08 .36 (.16) — (.14) (.28) n = 33 n = 2,153 n = 27 Liberal (2) .11 .50 .11 .41 (.13) n = 1 (.14) (.23) n = 69 n = 3,362 n = 47 Somewhat liberal (3) .14 .50 .13 .34 (.12) n = 1 (.15) (.24) n = 32 n = 2,315 n = 91 Middle of the road (4) .16 .42 .20 .48 (.19) (.19) (.19) (.25) n = 18 n = 9 n = 1,942 n = 710 Somewhat conservative (5) .23 .63 .28 .61 (.17) (.28) (.23) (.25) n = 6 n = 33 n = 360 n = 1,612 Conservative (6) .07 .74 .37 .72 (.12) (.22) (.26) (.22) n = 3 n = 63 n = 226 n = 4,189 Extremely conservative (7) .81 .45 .79 — (.18) (.26) (.19) n=54 n =133 n =3,068 Average roll-call battery placement Analytical sample Full CCES Mean (SD) Mean (SD) Democrats Republicans Democrats Republicans Extremely liberal (1) .09 .08 .36 (.16) — (.14) (.28) n = 33 n = 2,153 n = 27 Liberal (2) .11 .50 .11 .41 (.13) n = 1 (.14) (.23) n = 69 n = 3,362 n = 47 Somewhat liberal (3) .14 .50 .13 .34 (.12) n = 1 (.15) (.24) n = 32 n = 2,315 n = 91 Middle of the road (4) .16 .42 .20 .48 (.19) (.19) (.19) (.25) n = 18 n = 9 n = 1,942 n = 710 Somewhat conservative (5) .23 .63 .28 .61 (.17) (.28) (.23) (.25) n = 6 n = 33 n = 360 n = 1,612 Conservative (6) .07 .74 .37 .72 (.12) (.22) (.26) (.22) n = 3 n = 63 n = 226 n = 4,189 Extremely conservative (7) .81 .45 .79 — (.18) (.26) (.19) n=54 n =133 n =3,068 Note. Details on the construction of the roll-call battery can be found in Supplementary Appendix B. CCES = Cooperative Congressional Election Survey. Table 4 Comparison of the Engaged Respondents’ Policy Preferences Average roll-call battery placement Analytical sample Full CCES Mean (SD) Mean (SD) Democrats Republicans Democrats Republicans Extremely liberal (1) .09 .08 .36 (.16) — (.14) (.28) n = 33 n = 2,153 n = 27 Liberal (2) .11 .50 .11 .41 (.13) n = 1 (.14) (.23) n = 69 n = 3,362 n = 47 Somewhat liberal (3) .14 .50 .13 .34 (.12) n = 1 (.15) (.24) n = 32 n = 2,315 n = 91 Middle of the road (4) .16 .42 .20 .48 (.19) (.19) (.19) (.25) n = 18 n = 9 n = 1,942 n = 710 Somewhat conservative (5) .23 .63 .28 .61 (.17) (.28) (.23) (.25) n = 6 n = 33 n = 360 n = 1,612 Conservative (6) .07 .74 .37 .72 (.12) (.22) (.26) (.22) n = 3 n = 63 n = 226 n = 4,189 Extremely conservative (7) .81 .45 .79 — (.18) (.26) (.19) n=54 n =133 n =3,068 Average roll-call battery placement Analytical sample Full CCES Mean (SD) Mean (SD) Democrats Republicans Democrats Republicans Extremely liberal (1) .09 .08 .36 (.16) — (.14) (.28) n = 33 n = 2,153 n = 27 Liberal (2) .11 .50 .11 .41 (.13) n = 1 (.14) (.23) n = 69 n = 3,362 n = 47 Somewhat liberal (3) .14 .50 .13 .34 (.12) n = 1 (.15) (.24) n = 32 n = 2,315 n = 91 Middle of the road (4) .16 .42 .20 .48 (.19) (.19) (.19) (.25) n = 18 n = 9 n = 1,942 n = 710 Somewhat conservative (5) .23 .63 .28 .61 (.17) (.28) (.23) (.25) n = 6 n = 33 n = 360 n = 1,612 Conservative (6) .07 .74 .37 .72 (.12) (.22) (.26) (.22) n = 3 n = 63 n = 226 n = 4,189 Extremely conservative (7) .81 .45 .79 — (.18) (.26) (.19) n=54 n =133 n =3,068 Note. Details on the construction of the roll-call battery can be found in Supplementary Appendix B. CCES = Cooperative Congressional Election Survey. The values in the table are the mean roll-call battery scores for engaged Democrats and Republicans who place themselves at each point on the ideological scale. For comparison and to show the generalizability of my results, I calculate the means for both the engaged respondents in my analytical sample and for the engaged respondents in the full CCES. Despite the fact that self-placements show that partisans often do place themselves at the same point, the issue preferences are more consistent with the threshold estimates from the CHOPIT model. Only the most conservative Democrats and the most liberal Republicans appear to share the same preferences, and even then, they place themselves at vastly different points on the ideological scale. The preferences in Table 4 also reflect the lack of differentiation among the liberal points on the scale that is suggested by the threshold estimates. In my sample, the estimated difference between a Democrat who places at “extremely liberal” and a Democrat who places at “somewhat liberal” is only 0.05. In contrast, the difference between a Republican who places at “extremely conservative” and a Republican who places at “somewhat conservative” is 0.18. This suggests that “liberal” does indeed fit a narrower range of preferences than “conservative.” In total, it appears that the anchoring vignettes approach offers a view of the ideological scale that more accurately reflects the true divide in respondents’ policy preferences and suggests that polarization among the engaged public may be even greater than is assumed when partisan-driven DIF is not taken into account. Discussion In the age of increased partisan polarization, it seems that even political terminology carries a partisan charge. Using anchoring vignettes, I advance the literature by going beyond just showing that individuals interpret ideological labels differently, but by also arguing that these differences can be predictably linked to partisan affiliation, as Democratic respondents have consistently lower thresholds for the distinctions between categories. Particularly among the politically engaged, what is labeled as “middle of the road” by a member of one party would be defined as distinctly ideological by a member of the opposite party, and, once these differences in interpretation are taken into account, the resulting estimates of the distinctions between categories are more reflective of the polarization in respondents’ policy preferences. Future work should probe further into this finding, as the question of how to best measure ideology is subject of much debate and any method that may help minimize the variation of issue preferences connected to each point of the ideological scale would help improve its predictive powers. Altogether, a substantial proportion of Democratic and Republican citizens do have different perceptions of the ideological scale. An interesting extension of this work would be to explore whether there are also differences within these two partisan groups. For example, it may be that among Republicans, those who identify with the Tea Party may have different perceptions of what it means to be “conservative” than those who do not. Moreover, it is unlikely that the DIF uncovered here is limited to just the ideological self-placement questions. The findings presented here suggest that some of the differences in where respondents place political figures and candidates that are typically attributed to projection may be because of different interpretations of the ideological scale. While anchoring vignettes have most often been used to increase the cross-cultural comparability of multinational surveys, my findings also make an important contribution to the study of public opinion by highlighting the potential usefulness of the technique in studies of single countries. The finding of partisan interpretations of the ideological scale presented here suggests that elite polarization may be creating what is the equivalent of two partisan “subcultures” within the United States. Given the broad geographic and ethnic diversity of the United States, it also seems likely that the ideological scale and other commonly used survey questions may be interpreted differently depending on factors such as the region of the country one lives in or the length of time that one’s family has been in the United States. Beyond the United States, countries wherein there are major divides based on factors such as religion or ethnicity may also benefit from more careful consideration of whether individuals are in fact giving comparable survey responses. Thus, my work has broader implications for the future direction of all survey research, as it demonstrates the potential usefulness of anchoring vignettes and the overall need for construction and analyses of survey questions measuring subjective concepts that acknowledge that even in areas that may share common borders, there still may be multiple, distinct understandings of the political world. Funding Funding for the research was provided by the Department of Research at the University of Houston. A previous version of this article was presented at the 2013 Annual Meeting of the American Political Science Association. Footnotes 1See also http://gking.harvard.edu/vign 2Wand (2013) uses data from the 2004 ANES to present a nonparametric analysis of self-reported preferences for government services but uses respondents’ placements of the two major-party presidential candidates as anchors rather than vignettes. 3For details on the survey methodology, see http://projects.iq.harvard.edu/cces/home 4In Fall 2011, I piloted five vignettes on a small sample (N = 33) of undergraduate students. Approximately 43% of those respondents placed at least two of the vignettes at the same point on the ideological scale, suggesting that reducing the number of vignettes shown to respondents would not necessarily sacrifice additional explanatory leverage. 5The large majority of respondents took an “all-or-nothing” approach. Of the 174 respondents who did not place all three vignettes, 14 placed just two and the remaining 160 placed none. 6A discussion of the differences between those who did and did not place the vignettes will be taken up later in the article. 7Plots of the full distributions of the vignette ratings are available in Supplemental Appendix A. 8Question text and more information about the construction of these indices can be found in Supplemental Appendix B. 9Partisan leaners are coded as partisans. 10Responses used to create Ideological Self-Placement and all independent variables come from the preelection wave of the CCES. Age is a continuous variable derived from the source variable “birthyr.” Female is a recoding of the variable “gender.” Education, income, and church attendance are categorical variables derived directly from the CCES source variables “educ,” “faminc,” and “pew-churchatd.” Sophistication is the number of correct answers given when asked to recall the party of the respondent’s governor, senior senator, junior senator, and House member (cc310a-d). Non-White is a dichotomous recoding of “race.” All responses other than “White” are coded as 1. South is coded 1 if the variable “inputstate” indicates the respondent is a resident of AL, AR, DE, FL, GA, KY, LA, MD, MS, NC, OK, SC, TN, TX, VA, or WV. 11See Supplemental Appendix C for computational details. 12I thank two anonymous reviewers for suggesting this line of discussion and analyses. 13Full model results are available in Supplemental Appendix E. 14See Supplementary Appendix B for full details about the questions used to construct this measure. 15The rotated factor loadings are all >.38. Elizabeth Simas is an Assistant Professor of Political Science at the University of Houston. Her research examines the behavior of both candidates and voters in U.S. elections. Acknowledgements The author is grateful for helpful comments from Jason Casellas. References Abramowitz A. I. ( 2011 ). The disappearing center: Engaged citizens, polarization, and American democracy . New Haven : Yale University Press . Abramowitz A. I. ( 2013 ). The polarized public? Why American government is so disfunctional . Boston : Pearson . Abramowitz A. I. , Saunders K. L. , Abramowitz A. ( 1998 ). Ideological realignment in the U.S. electorate . The Journal of Politics , 60 , 634 – 652 . https://doi.org/10.2307/2647642 Google Scholar CrossRef Search ADS Aldrich J. H. , Mckelvey R. D. ( 1977 ). A method of scaling with applications to the 1968 and 1972 presidential elections . The American Political Science Review , 71 ( 1 ), 111 – 130 . https://doi.org/10.2307/1956957 Google Scholar CrossRef Search ADS Ansolabehere S. ( 2013 ) Cooperative congressional election study, 2012: Common content . Release 1: April 15, 2013. Cambridge, MA : Harvard University . Bakker R. , Jolly S. , Polk J. , Poole K. ( 2015 ). The European common space: Extending the use of anchoring vignettes. http://dx.doi.org/10.1017/S0022381614000449. https://doi.org/10.1017/S0022381614000449 Bartels L. M. ( 2002 ). Beyond the running tally: Partisan bias in political perceptions . Political Behavior . https://doi.org/10.1023/A:1021226224601 Brady H. ( 1985 ). The perils of survey research: Inter-personally incomparable survey responses . Political Methodology , 11 , 269 – 291 . Bratton M. ( 2010 ). Anchoring the "D-Word" in Africa . Journal of Democracy , 21 , 106 – 113 . Google Scholar CrossRef Search ADS Carmines E. G. , Ensley M. J. , Wagner M. W. ( 2012 ). Who fits the left-right divide? Partisan polarization in the American electorate . American Behavioral Scientist , 56 , 1631 – 1653 . https://doi.org/10.1177/0002764212463353 Google Scholar CrossRef Search ADS Claassen C. , Tucker P. , Smith S. S. ( 2014 ). Ideological labels in America . Political Behavior , 37 , 253 – 278 . https://doi.org/10.1007/s11109-014-9272-6 Google Scholar CrossRef Search ADS Conover P. J. , Feldman S. ( 1981 ). The origins and meaning of liberal/conservative self-identifications . American Journal of Political Science , 25 , 617 – 645 . https://doi.org/10.1007/s11109-010-9145-6 Google Scholar CrossRef Search ADS Ellis C. , Stimson J. ( 2012 ). Ideology in America . New York, NY : Cambridge University Press . Farwell L. , Weiner B. ( 2000 ). Bleeding hearts and the heartless: popular perceptions of liberal and conservative ideologies . Personality and Social Psychology Bulletin , 26 , 845 – 852 . https://doi.org/10.1177/0146167200269009 Google Scholar CrossRef Search ADS Feldman S. , Johnston C. ( 2014 ). Understanding the determinants of political ideology: Implications of structural complexity . Political Psychology , 35 , 337 – 358 . https://doi.org/10.1111/pops.12055 Google Scholar CrossRef Search ADS Festinger L. ( 1954 ). A theory of social comparison processes . Human Relations , 7 , 117 – 140 . https://doi.org/10.1177/001872675400700202 Google Scholar CrossRef Search ADS Gaines B. J. , Kuklinski J. H. , Quirk P. J. , Peyton B. , Verkuilen J. ( 2007 ). Same facts, different interpretations: Partisan motivation and opinion on Iraq . Journal of Politics , 69 , 957 – 974 . https://doi.org/10.1111/j.1468-2508.2007.00601.x Google Scholar CrossRef Search ADS Green D. , Palmquist B. , Schickler E. ( 2002 ). Partisan hearts and minds: Political parties and social identities of voters . New Haven : Yale University Press . Greene S. ( 1999 ). Understanding party identification: A social identity approach . Political Psychology , 20 , 393 – 403 . https://doi.org/10.1111/0162-895X.00150 Google Scholar CrossRef Search ADS Grol-Prokopczyk H. , Freese J. , Hauser R. M. ( 2011 ). Using anchoring vignettes to assess group differences in general self-rated health . Journal of Health and Social Behavior , 52 , 246 – 261 . https://doi.org/10.1177/0022146510396713 Google Scholar CrossRef Search ADS PubMed Grzymala-Busse A. ( 2007 ). Rebuilding leviathan: Party competition and state exploitation in post-communist democracies . New York, NY : Cambridge University Press . Hare C. , Armstrong D. A. , Bakker R. , Carroll R. , Poole K. T. ( 2015 ). Using Bayesian Aldrich-McKelvey scaling to study citizens’ ideological preferences and perceptions . American Journal of Political Science , 59 , 759 – 774 . https://doi.org/10.1111/ajps.12151 Google Scholar CrossRef Search ADS Heine S. J. , Lehman D. R. , Peng K. , Greenholtz J. ( 2002 ). What’s wrong with cross-cultural comparisons of subjective Likert scales? The reference-group effect . Journal of Personality and Social Psychology , 82 , 903 – 918 . https://doi.org/10.1037/0022-3514.82.6.903 Google Scholar CrossRef Search ADS PubMed Hetherington M. J. ( 2001 ). Resurgent mass partisanship: The role of elite polarization . American Political Science Review , 95 , 619 – 631 . https://doi.org/10.1017/S0003055401003045 Google Scholar CrossRef Search ADS Hopkins D. J. , King G. ( 2010 ). Improving anchoring vignettes: Designing surveys to correct interpersonal incomparability . Public Opinion Quarterly , 74 , 201 – 222 . https://doi.org/10.1093/poq/nfq011 Google Scholar CrossRef Search ADS Inglehart R. ( 1997 ). Modernization and postmodernization: Cultural, economic, and political change in 43 societies . Princeton : Princeton University Press . Iyengar S. , Hahn K. S. ( 2009 ). Red media, blue media: Evidence of ideological selectivity in media use . Journal of Communication , 59 ( 1 ), 19 – 39 . https://doi.org/10.1111/j.1460-2466.2008.01402.x Google Scholar CrossRef Search ADS King G. , Murray C. J. L. , Salomon J. A , Tandon A. ( 2004 ). Enhancing the validity and cross-cultural comparability of measurement in survey research . American Political Science Review , 98 ( 1 ), 191 – 207 . https://doi.org/10.1017/S000305540400108X Google Scholar CrossRef Search ADS King G. , Wand J. ( 2007 ). Comparing incomparable survey responses: New tools for anchoring vignettes . Political Analysis , 15 ( 1 ), 46 – 66 . Google Scholar CrossRef Search ADS Layman G. C. , Carsey T. M. ( 2002 ). Party polarization and "conflict extension" in the American electorate . American Journal of Political Science , 46 , 786 – 802 . https://doi.org/10.2307/3088434 Google Scholar CrossRef Search ADS Levendusky M. S. ( 2009 ). The partisan sort: How liberals became democrats and conservatives became republicans . Chicago : Chicago University Press . Levendusky M. S. ( 2013 ). Why do partisan media polarize viewers? American Journal of Political Science , 57 , 611 – 623 . https://doi.org/10.1111/ajps.12008 Google Scholar CrossRef Search ADS Mitchell D.-G. , Hibbing M. V. , Smith K. B. , Hibbing J. R. ( 2013 ). Side by side, worlds apart: desired policy change as a function of preferences AND perceptions . American Politics Research , 42 , 338 – 363 . https://doi.org/10.1177/1532673X13498619 Google Scholar CrossRef Search ADS Rabe-Hesketh S. , Skrondal A. . 2002 . Estimating CHOPIT models in GLLAMM: Political efficacy example from King et al. 2002. Retrieved from : http://www.gllamm.org/examples.html Retrieved August 12, 2013 Schiffer A. ( 2000 ). I’m not that liberal: Explaining conservative democratic identification . Political Behavior , 22 , 293 – 310 . https://doi.org/10.1023/A:1010626029987 Google Scholar CrossRef Search ADS Sears D. , Citrin J. ( 1985 ). Tax revolt: Something for nothing in California . Cambridge : Harvard University Press . Simas E. N. ( 2016 ). Perceptions of the heterogeneity of party elites in the United States . Party Politics . Advance online publication. https://doi.org/10.1177/1354068816668676 Slothuus R. , de Vreese C. H. ( 2010 ). Political parties, motivated reasoning, and issue framing effects . The Journal of Politics , 72 , 630 – 645 . https://doi.org/10.1017/S002238161000006X Google Scholar CrossRef Search ADS Suls J. , Martin R. , Wheeler L. ( 2002 ). Social comparison: Why, with whom, and with what effect? Current Directions in Psychological Science , 11 , 159 – 163 . https://doi.org/10.1111/1467-8721.00191 Google Scholar CrossRef Search ADS Treier S. , Hillygus D. S. ( 2009 ). The nature of political ideology in the contemporary electorate . Public Opinion Quarterly , 73 , 679 – 703 . https://doi.org/10.1093/poq/nfp067 Google Scholar CrossRef Search ADS Wand J. ( 2013 ). Credible comparisons using interpersonally incomparable data: Nonparametric scales with anchoring vignettes . American Journal of Political Science , 57 ( 1 ), 249 – 262 . https://doi.org/10.1111/j.1540-5907.2012.00597.x Google Scholar CrossRef Search ADS Wheeler L. , Martin R. , Suls J. ( 1997 ). The proxy model of social comparison for self-assessment of ability . Personality and Social Psychology Review , 1 ( 1 ), 54 – 61 . https://doi.org/10.1207/s15327957pspr0101_4 Google Scholar CrossRef Search ADS PubMed Zaller J. ( 1992 ). The nature and origin of mass opinions . New York, NY : Cambridge University Press . Zanna M. P. , Goethals G. R. , Hill J. F. ( 1975 ). Evaluating a sex-related ability: Social comparison with similar others and standard setters . Journal of Experimental Social Psychology , 11 ( 1 ), 86 – 93 . https://doi.org/10.1016/S0022-1031(75)80013-8 Google Scholar CrossRef Search ADS © The Author 2017. Published by Oxford University Press on behalf of The World Association for Public Opinion Research. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

International Journal of Public Opinion ResearchOxford University Press

Published: Oct 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off