Citations Needed: Links to Interesting Research
05-27-2020
, 03:11 AM
05-27-2020
, 11:31 AM
Stereotypes and Politics (full text here)
Seems like I could apply this one to the conversation about belief in blank slate theories and biological determinism.
The natural experiments they use to demonstrate these distortions is pretty cool, I think:
It might be worth explaining the actual measure of exaggeration. If I'm understanding it correctly, they are using ANES responses to questions which use a 7-point likert scale. Something like this:
So, respondents are being asked, on various issues, both to choose an option that describes their own beliefs, but also to choose an option for what they think members of the other party belief. Figure 2 is then comparing (for example) responses from Republicans to responses from Democrats on what they thought Republicans would say.
Perceived partisan differences are more distorted on more salient issues
Seems like I could apply this one to the conversation about belief in blank slate theories and biological determinism.
Quote:
We start by documenting two facts, using nationally representative survey data from the American National Election Studies (ANES), which elicit political behavior (e.g. voting or making political contributions), political attitudes, and beliefs about political attitudes held by Republicans and by Democrats.
First, exaggerated beliefs about partisan differences strongly predict political engagement: individuals who perceive larger partisan differences are significantly more likely to report they voted, made political contributions, and participated in political campaigns. These patterns hold, quantitatively and qualitatively, also when we control for factors typically associated with political engagement, such as demographics, socioeconomic status, own attitudes, and strength of partisan identification. Focusing on self-identified partisans, our evidence suggests that this is primarily driven by distortions in beliefs about how distant the other party is from one's own party.
Second, while individuals exaggerate differences between Republicans and Democrats on a range of socioeconomic and political issues, such exaggerations are larger on issues that respondents consider more important and pressing for the country. In other words, individuals make larger mistakes in their perception of partisan differences on the dimensions that are more salient to them.
These facts suggest that understanding the drivers of perceived polarization is important to accounting for voting behavior. To this effect, we develop and test a model in which beliefs about political attitudes reflect stereotyping, and stereotyping is stronger for more salient issues. Stereotyping captures the tendency of probabilistic assessments to overweigh the prevalence of the types that are more likely in one group relative to a comparison group (Kahneman and Tversky, 1972; Gennaioli and Shleifer, 2010; Bordalo et al., 2016). As an example, while wealthy individuals are more prevalent among Republicans than among Democrats, only an estimated 2% of Republicans earn more than $250,000 per year. Yet, popular beliefs assume that this share is much higher — in one survey, the average stated share is above 30% (Ahler and Sood, 2018). In this approach, in line with cognitive psychology (Schneider 2004), stereotypes exaggerate true differences across groups. As a consequence, they can account for the observed exaggeration in perceived partisan differences. This mechanism has been applied to political beliefs in Bordalo et al. (2016); Gennaioli and Tabellini (2019).
First, exaggerated beliefs about partisan differences strongly predict political engagement: individuals who perceive larger partisan differences are significantly more likely to report they voted, made political contributions, and participated in political campaigns. These patterns hold, quantitatively and qualitatively, also when we control for factors typically associated with political engagement, such as demographics, socioeconomic status, own attitudes, and strength of partisan identification. Focusing on self-identified partisans, our evidence suggests that this is primarily driven by distortions in beliefs about how distant the other party is from one's own party.
Second, while individuals exaggerate differences between Republicans and Democrats on a range of socioeconomic and political issues, such exaggerations are larger on issues that respondents consider more important and pressing for the country. In other words, individuals make larger mistakes in their perception of partisan differences on the dimensions that are more salient to them.
These facts suggest that understanding the drivers of perceived polarization is important to accounting for voting behavior. To this effect, we develop and test a model in which beliefs about political attitudes reflect stereotyping, and stereotyping is stronger for more salient issues. Stereotyping captures the tendency of probabilistic assessments to overweigh the prevalence of the types that are more likely in one group relative to a comparison group (Kahneman and Tversky, 1972; Gennaioli and Shleifer, 2010; Bordalo et al., 2016). As an example, while wealthy individuals are more prevalent among Republicans than among Democrats, only an estimated 2% of Republicans earn more than $250,000 per year. Yet, popular beliefs assume that this share is much higher — in one survey, the average stated share is above 30% (Ahler and Sood, 2018). In this approach, in line with cognitive psychology (Schneider 2004), stereotypes exaggerate true differences across groups. As a consequence, they can account for the observed exaggeration in perceived partisan differences. This mechanism has been applied to political beliefs in Bordalo et al. (2016); Gennaioli and Tabellini (2019).
Quote:
A central feature of multidimensional political views is that the importance attached to specific issues varies across individuals and over time. Accordingly, the natural tendency to stereotype may be diluted for issues that are not top of mind. Our assumption that issue salience modulates the strength of stereotyping resonates with the finding that respondents exaggerate partisan differences more on issues they deem more important, but its motivation runs deeper. Evidence shows that, when attention is directed to a sensorial stimulus, such as size or brightness, differences along that dimension are perceived to be larger, while differences in other, relatively neglected dimensions are judged to be smaller (Nosofsky, 1988). Crucially, it entails the testable prediction that exogenous shifts in issue salience impacts beliefs.
To attach a causal interpretation to the relationship between salience and belief distortions, going beyond the individual-level correlations described above, we exploit a major, exogenous shock to issue salience: the end of the Cold War in 1991. From the American perspective, the dissolution of the Soviet Union and the collapse of the communist regimes in Eastern Europe represented a dramatic and unexpected reduction in the salience of external threats. We present evidence from the ANES that the share of American voters who considered national defense related issues as more pressing sharply decreased right after 1991. Conversely, the share of Americans who viewed domestic issues, such as social welfare and race rations, as most pressing nearly doubled right after 1991, moving from 45 to almost 80 percent in less than 4 years.
Consistent with the model’s predictions, as external threats faded away with the end of the Cold War, exaggeration of beliefs on partisan differences dropped on external issues such as defense spending, and increased on domestic ones. Both patterns are statistically and quantitatively significant, and are driven by movement in beliefs, rather than by changes in respondents’ positions.
To attach a causal interpretation to the relationship between salience and belief distortions, going beyond the individual-level correlations described above, we exploit a major, exogenous shock to issue salience: the end of the Cold War in 1991. From the American perspective, the dissolution of the Soviet Union and the collapse of the communist regimes in Eastern Europe represented a dramatic and unexpected reduction in the salience of external threats. We present evidence from the ANES that the share of American voters who considered national defense related issues as more pressing sharply decreased right after 1991. Conversely, the share of Americans who viewed domestic issues, such as social welfare and race rations, as most pressing nearly doubled right after 1991, moving from 45 to almost 80 percent in less than 4 years.
Consistent with the model’s predictions, as external threats faded away with the end of the Cold War, exaggeration of beliefs on partisan differences dropped on external issues such as defense spending, and increased on domestic ones. Both patterns are statistically and quantitatively significant, and are driven by movement in beliefs, rather than by changes in respondents’ positions.
Quote:
So, respondents are being asked, on various issues, both to choose an option that describes their own beliefs, but also to choose an option for what they think members of the other party belief. Figure 2 is then comparing (for example) responses from Republicans to responses from Democrats on what they thought Republicans would say.
Quote:
As it appears, perceived partisan differences strongly correlate with actual ones: ANES respondents perceive partisan differences to be larger when actual differences between Republicans and Democrats are higher.
At the same time, however, beliefs about partisan differences systematically exaggerate actual ones: the constant for the regression corresponding to results plotted in Figure 2 is 0.845 (s.e. = 0.096), making all the dots fall above the 45 degree line. Taken literally, this means that in the case where actual differences were equal to zero, perceived partisan differences were exaggerated by almost 1 point (out of 7).
On average, the actual difference between Republicans and Democrats across domestic issues is approximately 1 unit (on a 7-point scale), and is around 0.7 unit for external issue (namely, defense spending). Consistent with Bordalo et al. (2016) and Westfall et al. (2015), perceived differences are almost twice as large — ranging from 1.1 (for women’s role in column 7) to 2.1 (for the broad liberal-conservative issue in column 3), with an average of 1.7.
At the same time, however, beliefs about partisan differences systematically exaggerate actual ones: the constant for the regression corresponding to results plotted in Figure 2 is 0.845 (s.e. = 0.096), making all the dots fall above the 45 degree line. Taken literally, this means that in the case where actual differences were equal to zero, perceived partisan differences were exaggerated by almost 1 point (out of 7).
On average, the actual difference between Republicans and Democrats across domestic issues is approximately 1 unit (on a 7-point scale), and is around 0.7 unit for external issue (namely, defense spending). Consistent with Bordalo et al. (2016) and Westfall et al. (2015), perceived differences are almost twice as large — ranging from 1.1 (for women’s role in column 7) to 2.1 (for the broad liberal-conservative issue in column 3), with an average of 1.7.
Quote:
On average, perceived partisan differences systematically exaggerate actual ones. However, there exists substantial heterogeneity in beliefs about partisan differences across issues, as well as in their exaggeration relative to actual differences (Table 3, Panel B). Such heterogeneity remains even within respondents and, as we document next, is systematically associated with the salience of the issue perceived by the respondent.
To assess the role of issue salience on beliefs, we regress the perceived partisan difference on a given issue against an indicator for whether the respondent identifies that issue as the most important problem facing the country at the time of the interview. Table 4 presents the results. The unconditional correlation between perceived partisan differences and issue salience is positive, statistically significant, and quantitatively large. According to the coefficient reported in Column 1, perceived partisan differences by an individual are 0.3 units (or, 16%) higher on the most important issues.
To assess the role of issue salience on beliefs, we regress the perceived partisan difference on a given issue against an indicator for whether the respondent identifies that issue as the most important problem facing the country at the time of the interview. Table 4 presents the results. The unconditional correlation between perceived partisan differences and issue salience is positive, statistically significant, and quantitatively large. According to the coefficient reported in Column 1, perceived partisan differences by an individual are 0.3 units (or, 16%) higher on the most important issues.
06-01-2020
, 05:16 PM
06-12-2020
, 09:22 AM
06-15-2020
, 03:57 PM
A pair of currently topical studies:
1) A large-scale analysis of racial disparities in police stops across the United States
I wasn't familiar with this method, but it's rather clever:
2) The effect of suspect race on police officers’ decisions to draw their weapons (full text)
This is similar to Fryer's Fryer's 2016 study. Fryer used data from Houston PD. I think he published an updated analysis which also uses data from Houston, but I don't have it. This author uses data from Dallas. The smaller dataset (in contrast, for example, to the one from the first study) is definitely a limitation, but it's interesting to see a similar result.
I think the discussion in this paper of prior research and limitations is useful, e.g. both
and
1) A large-scale analysis of racial disparities in police stops across the United States
Quote:
More than 20 million Americans are stopped each year for traffic violations, making this one of the most common ways in which the public interacts with the police. Due to the decentralized nature of policing in the United States—and a corresponding lack of comprehensive and standardized data—it is difficult to rigorously assess the manner and extent to which race plays a role in traffic stops. The most widely cited national statistics come from the Police–Public Contact Survey (PPCS), which is based on a nationally representative sample of approximately 50,000 people who report having been recently stopped by the police.
In addition to such survey data, some local and state agencies have released periodic reports on traffic stops in their jurisdictions, and have also made their data available to outside researchers for analysis. While useful, these datasets provide only a partial picture. For example, there is concern that the PPCS, like nearly all surveys, suffers from selection bias and recall errors. Data released directly by police departments are potentially more complete but are available only for select agencies, are typically limited in what is reported and are inconsistent across jurisdictions.
To address these challenges, we compiled and analysed a dataset detailing nearly 100 million traffic stops carried out by 21 state patrol agencies and 35 municipal police departments over almost a decade. This dataset was built through a series of public records requests filed in all 50 states. To facilitate future analysis, we have redistributed these records in a standardized form.
In addition to such survey data, some local and state agencies have released periodic reports on traffic stops in their jurisdictions, and have also made their data available to outside researchers for analysis. While useful, these datasets provide only a partial picture. For example, there is concern that the PPCS, like nearly all surveys, suffers from selection bias and recall errors. Data released directly by police departments are potentially more complete but are available only for select agencies, are typically limited in what is reported and are inconsistent across jurisdictions.
To address these challenges, we compiled and analysed a dataset detailing nearly 100 million traffic stops carried out by 21 state patrol agencies and 35 municipal police departments over almost a decade. This dataset was built through a series of public records requests filed in all 50 states. To facilitate future analysis, we have redistributed these records in a standardized form.
Quote:
Our statistical analysis of these records proceeds in three steps. First, we assess potential bias in stop decisions by applying the ‘veil of darkness’ test developed by Grogger and Ridgeway. This test is based on a simple observation: because the sun sets at different times throughout the year, one can examine the racial composition of stopped drivers as a function of sunlight while controlling for time of day. In particular, we use the discontinuity created by the start (and end) of daylight-saving time (DST), comparing the racial distribution of drivers stopped in the evenings immediately before DST begins, when it is dark, to the distribution after DST begins, when it is light at the same time of day. If black drivers comprise a smaller share of stopped drivers when it is dark and accordingly difficult to determine a driver’s race, that suggests black drivers were stopped during daylight hours in part because of their race.
Quote:
To illustrate the intuition behind this method, in Fig. 2 we examine the demographic composition of drivers stopped by the Texas State Patrol at various times of day. Each panel in the plot shows stops occurring in a specific 15-min window (for example, 19:00–19:15), and the horizontal axis indicates minutes since dusk. We restrict to white and black drivers—because the ethnicity of Hispanic drivers is not always apparent, even during daylight hours—and, following Grogger and Ridgeway, we filter out stops that occurred in the ~30-min period between sunset and dusk, when it is neither ‘light’ nor ‘dark.’ For each time period, the plot shows a marked drop in the proportion of drivers stopped after dusk who are black, suggestive of discrimination in stop decisions.
Quote:
Researchers are working to identify appropriate benchmarks for exploring racial bias in the officer-involved shooting (OIS) context. Two recent studies benchmarked OIS against incidents in which officers drew weapons but did not shoot. A problem is that the decision to draw a weapon may itself be subject to bias. Using 2017 use-of-force data from the Dallas Police Department, we modeled officers’ decisions to draw their weapons as a function of suspect race and other suspect, officer, and incident character- istics. We benchmarked by limiting analyses to arrest and active aggression cases, thereby excluding interactions in which it was less likely suspects would have had weapons drawn against them. The key finding was that black suspects were no more or less likely to have weapons drawn against them than other suspects.
I think the discussion in this paper of prior research and limitations is useful, e.g. both
Quote:
Our key finding is somewhat unique relative to recent research in this area. On the one hand, recent national benchmarking studies tend to find that blacks are less likely to be shot than whites (e.g., Cesario et al., 2019; Tregle et al., 2019). The same was observed in some incident-level work (Wheeler et al., 2018; Worrall et al., 2018). On the other hand, some county-level work (Ross, 2015) has found that blacks are much more likely than whites to be shot. Our findings land somewhere in the middle. How do we reconcile these findings? The national level studies have not examined incident characteristics. The same holds with other aggregate research, whether at the county, Census block, or some other level. To our knowledge, our study is the first to attempt benchmarking at the incident level—with a focus on weapon draws.
Quote:
We have pointed out various limitations throughout the study, but some key ones should be highlighted here once again. Most importantly, our work is limited to a single department. Policies governing weapon draws obviously vary across departments, so the outcome variable in Dallas may be qualitatively different than a similar out- come in another agency. Next, while benchmarking with arrests and active aggression cases improves over several recent incident-level studies, many of which fail to attempt any benchmarking at all, it is imperfect. For example, it fails to account for the possibility that officers are biased in their decisions to confront certain classes of suspects in the first place.
There is also no clear guidance on what constitutes the appropriate list of covariates to include in a model of police weapon draws. Our predictors were selected largely out of convenience because they were available in the data. Finally, the UoF and suspect data were all reported post-incident. For example, officers could have indicated a higher level of resistance to justify the force in question.
We should also take this opportunity to point out one of the most glaring weaknesses of this and all other OIS research: No study has yet explored the issue of racial bias in justified deadly force cases compared to those in which deadly force could have been used (i.e., was legally justified) but was not. For research in this area to move forward, it will be necessary for police departments to collect data on this important outcome.
There is also no clear guidance on what constitutes the appropriate list of covariates to include in a model of police weapon draws. Our predictors were selected largely out of convenience because they were available in the data. Finally, the UoF and suspect data were all reported post-incident. For example, officers could have indicated a higher level of resistance to justify the force in question.
We should also take this opportunity to point out one of the most glaring weaknesses of this and all other OIS research: No study has yet explored the issue of racial bias in justified deadly force cases compared to those in which deadly force could have been used (i.e., was legally justified) but was not. For research in this area to move forward, it will be necessary for police departments to collect data on this important outcome.
06-22-2020
, 11:14 AM
Pluralistic Collapse: The “Oil Spill” Model of Mass Opinion Polarization (full text)
I think the theoretical idea in this paper is really interesting, and seems important and worth thinking about. Yet the empirical data seems a little bit weak, or at least somewhat ambivalent. But I think that might just be the insufficiency of the methods. I think the idea of doing network analysis on belief correlation in survey data is reasonable, and should be useful. But it's really the kind of thing where it's also hard to find a good substitute for more qualitative research, despite the problems qualitative methods have with proving external validity. Pairing these methods with in depth interviews, or something like social media content analysis, would be fantastic, IMO.
Anecdotally, I think it's possible to observe the kind of polarization described in the article in political discussions online, including in forums like this one. Mostly in the cues that people take from other people's posts, and the inferences they draw to what might otherwise seem like unrelated topics.
Note: snipping in-text citations here for readability:
Using network analysis:
A snapshot of results (but see also Figures 6, 7):
I think the theoretical idea in this paper is really interesting, and seems important and worth thinking about. Yet the empirical data seems a little bit weak, or at least somewhat ambivalent. But I think that might just be the insufficiency of the methods. I think the idea of doing network analysis on belief correlation in survey data is reasonable, and should be useful. But it's really the kind of thing where it's also hard to find a good substitute for more qualitative research, despite the problems qualitative methods have with proving external validity. Pairing these methods with in depth interviews, or something like social media content analysis, would be fantastic, IMO.
Anecdotally, I think it's possible to observe the kind of polarization described in the article in political discussions online, including in forums like this one. Mostly in the cues that people take from other people's posts, and the inferences they draw to what might otherwise seem like unrelated topics.
Note: snipping in-text citations here for readability:
Quote:
Social and political theorists have long argued that differentiated societies are best able to integrate diverse interests when relying on a foundation of cross-cutting social and political cleavages. Unlike stultifying mono-cultures built on public consensus, cross-cutting lines of conflict and dis- agreement found in pluralistic societies embed individuals in overlapping groups. Even if people fight over differing views on taxation, they may still agree on issues of foreign policy or overlap in religious views; these areas of agreement are thought to prevent disagreement in one arena from escalating into all-out (metaphorical or literal) warfare. As long as opinion cleavages remain cross-cutting rather than all-encompassing, pluralistic disagreement should channel social conflict toward mutual tolerance, political forbearance, and other liberal ends....
Against this backdrop, recent decades have brought mounting concern among social scientists, journalists, policymakers, and the general public that the pluralistic structure of U.S. politics has turned to all-encompassing conflict between increasingly polarized camps. In their provocatively titled book How Democracies Die, Levitsky and Ziblatt (2018) frame polarization as a fundamental problem for U.S. democracy—not merely a nuisance but a potentially existential threat. Such alarmism is not uncommon in mainstream discourse. Yet despite the common belief that the United States has become dramatically more politically polarized, social-scientific support for such a pattern has been surprisingly elusive....
This raises the question: If opinions have not become markedly more polarized, why is it so widely believed—and so easy to intuit—that they have? Why, in other words, does polarization appear as such a salient social fact despite its apparent absence in the U.S. population?
Against this backdrop, recent decades have brought mounting concern among social scientists, journalists, policymakers, and the general public that the pluralistic structure of U.S. politics has turned to all-encompassing conflict between increasingly polarized camps. In their provocatively titled book How Democracies Die, Levitsky and Ziblatt (2018) frame polarization as a fundamental problem for U.S. democracy—not merely a nuisance but a potentially existential threat. Such alarmism is not uncommon in mainstream discourse. Yet despite the common belief that the United States has become dramatically more politically polarized, social-scientific support for such a pattern has been surprisingly elusive....
This raises the question: If opinions have not become markedly more polarized, why is it so widely believed—and so easy to intuit—that they have? Why, in other words, does polarization appear as such a salient social fact despite its apparent absence in the U.S. population?
Quote:
The common empirical strategy in prior research is to select and analyze a limited set of survey questions on political issues, focusing on questions the researcher identifies as most politically relevant (e.g., opinions on abortion, climate change, taxation, or foreign policy). By selecting on an issue’s already-established political relevance, however, the researcher implicitly assumes that polarization occurs via heightened alignment along existing lines of political debate, akin to a fence the researcher can observe getting taller over time.
But what if polarization is less like a fence getting taller over time and more like an oil spill that spreads from its source to gradually taint more and more previously “apolitical” attitudes, opinions, and preferences? DellaPosta and colleagues (2015) show that even many initially apolitical lifestyle characteristics, from musical taste to belief in astrology, can become politicized as signals for deeper beliefs and preferences—a tendency most saliently captured in the popular image of the “latte liberal.”
This suggests a different answer to the puzzle posed at the outset: rather than heightened alignment across already-politicized opinion dimensions, the crux of contemporary polarization might lie in the increased breadth of opinions and preferences that have come to be associated with political identities and beliefs. This broadening of opinion alignment to encompass areas typically thought of as nonpolitical would not be picked up in studies that only consider polarization along existing lines of political debate. However, the widespread impression that polarization has dramatically increased may reflect real politicization of an increasingly diverse span of beliefs and preferences that once cut across (or lay apart from) ideological divides.
But what if polarization is less like a fence getting taller over time and more like an oil spill that spreads from its source to gradually taint more and more previously “apolitical” attitudes, opinions, and preferences? DellaPosta and colleagues (2015) show that even many initially apolitical lifestyle characteristics, from musical taste to belief in astrology, can become politicized as signals for deeper beliefs and preferences—a tendency most saliently captured in the popular image of the “latte liberal.”
This suggests a different answer to the puzzle posed at the outset: rather than heightened alignment across already-politicized opinion dimensions, the crux of contemporary polarization might lie in the increased breadth of opinions and preferences that have come to be associated with political identities and beliefs. This broadening of opinion alignment to encompass areas typically thought of as nonpolitical would not be picked up in studies that only consider polarization along existing lines of political debate. However, the widespread impression that polarization has dramatically increased may reflect real politicization of an increasingly diverse span of beliefs and preferences that once cut across (or lay apart from) ideological divides.
Quote:
Rather than focusing on a selected subset of political opinion items, I broaden the scope of analysis to include any opinion question ever presented in the General Social Survey (GSS) between 1972 and 2016. Building on Boutyline and Vaisey (2017), I represent cor- relations between opinions in a representative sample of the U.S. population as relational ties in a network of interconnected beliefs. Rather than a traditional social network analysis where nodes are human or organizational actors and the ties or edges between nodes represent social connections (Wasserman and Faust 1994), the belief network conceives of each attitude, opinion, or belief as a node; the strength of the edges or ties between beliefs reflects the extent to which two beliefs are correlated in survey data....
To understand what polarization looks like in a structure of associated beliefs, we can compare it to its opposite: cross-cutting opinion pluralism. In a pluralistic belief network, attitudes may be correlated with one another, but these correlations are cross-cutting. For example, perhaps attitudes toward gun control are correlated with attitudes toward abortion. But attitudes toward gun control may also be correlated with attitudes toward racial justice, which may themselves not be correlated with abortion attitudes; abortion attitudes, in turn, may be correlated with attitudes toward LGBT rights, despite the latter remaining uncorrelated with gun control attitudes. As a whole, this hypothetical belief network consists of relationships that cut across and offset each other without cohering into large cohesive clusters of densely connected attitudes. Substantively, this lack of coherence means two people who disagree on one or even most things will still likely be able to find some issue on which they agree, creating an opportunity to bridge the ideological distance between them.
In this conceptualization, opinion pluralism collapses into polarization via the consolidation of previously cross-cutting alignments into increasingly broad and encompassing ones.... If knowing someone’s attitude on issue X allows us to predict attitudes on issues Y and Z, then X, Y, and Z make up a constrained set of beliefs. Applying this logic to the polarization debate, researchers examine correlations or alignments between issues rather than distributions of attitudes within issues. When strong correlations across multiple dimensions of opinion indicate high levels of constraint—for example, when knowing someone’s attitude on abortion helps predict their attitude on gun control—the population can be said to be more polarized.
To understand what polarization looks like in a structure of associated beliefs, we can compare it to its opposite: cross-cutting opinion pluralism. In a pluralistic belief network, attitudes may be correlated with one another, but these correlations are cross-cutting. For example, perhaps attitudes toward gun control are correlated with attitudes toward abortion. But attitudes toward gun control may also be correlated with attitudes toward racial justice, which may themselves not be correlated with abortion attitudes; abortion attitudes, in turn, may be correlated with attitudes toward LGBT rights, despite the latter remaining uncorrelated with gun control attitudes. As a whole, this hypothetical belief network consists of relationships that cut across and offset each other without cohering into large cohesive clusters of densely connected attitudes. Substantively, this lack of coherence means two people who disagree on one or even most things will still likely be able to find some issue on which they agree, creating an opportunity to bridge the ideological distance between them.
In this conceptualization, opinion pluralism collapses into polarization via the consolidation of previously cross-cutting alignments into increasingly broad and encompassing ones.... If knowing someone’s attitude on issue X allows us to predict attitudes on issues Y and Z, then X, Y, and Z make up a constrained set of beliefs. Applying this logic to the polarization debate, researchers examine correlations or alignments between issues rather than distributions of attitudes within issues. When strong correlations across multiple dimensions of opinion indicate high levels of constraint—for example, when knowing someone’s attitude on abortion helps predict their attitude on gun control—the population can be said to be more polarized.
Quote:
06-23-2020
, 01:06 AM
Quote:
Pluralistic Collapse: The “Oil Spill” Model of Mass Opinion Polarization (full text)
I think the theoretical idea in this paper is really interesting, and seems important and worth thinking about. Yet the empirical data seems a little bit weak, or at least somewhat ambivalent. But I think that might just be the insufficiency of the methods. I think the idea of doing network analysis on belief correlation in survey data is reasonable, and should be useful. But it's really the kind of thing where it's also hard to find a good substitute for more qualitative research, despite the problems qualitative methods have with proving external validity. Pairing these methods with in depth interviews, or something like social media content analysis, would be fantastic, IMO.
Anecdotally, I think it's possible to observe the kind of polarization described in the article in political discussions online, including in forums like this one. Mostly in the cues that people take from other people's posts, and the inferences they draw to what might otherwise seem like unrelated topics.
I think the theoretical idea in this paper is really interesting, and seems important and worth thinking about. Yet the empirical data seems a little bit weak, or at least somewhat ambivalent. But I think that might just be the insufficiency of the methods. I think the idea of doing network analysis on belief correlation in survey data is reasonable, and should be useful. But it's really the kind of thing where it's also hard to find a good substitute for more qualitative research, despite the problems qualitative methods have with proving external validity. Pairing these methods with in depth interviews, or something like social media content analysis, would be fantastic, IMO.
Anecdotally, I think it's possible to observe the kind of polarization described in the article in political discussions online, including in forums like this one. Mostly in the cues that people take from other people's posts, and the inferences they draw to what might otherwise seem like unrelated topics.
As far as the U.S. becoming more polarized, I agree we've become more vocal in our differences but I don't think that means that we have more or greater differences today than in the past. I don't think things have changed much in the latter regard throughout human history. What's changed is social media and our ability to become more vocal.
06-23-2020
, 09:38 AM
06-23-2020
, 01:15 PM
06-23-2020
, 03:36 PM
