At the time of submission, none of the work presented in this thesis had yet been published. The width of the bars indicates what proportion of individuals (weighted) fell into each group.
COLLECTIVE IDENTITY IN COLLECTIVE ACTION
EVIDENCE FROM THE 2020 SUMMER BLM PROTEST
Introduction
By maintaining this sense of belonging and loyalty, working toward the group goal becomes individually rational and free riding is reduced (Chong et al., 2004; Conover, 1988). Importantly, race in America provides a source of collective identity that has motivated previous episodes of collective action (McClain et al., 2009; Sanchez & Vargas, 2016).
Collective Identity and Protest Participation
The act of protesting solidifies an individual's support for the cause, increasing the expected level of collective action signals observed during the days following the protest action compared to the non-protesting expectation. The protest action increases the expected levels of collective action signals observed during that day compared to the non-protest expectation.
Research Design
For each tweet, there is a probability measure𝜃 that represents the proportion of tweets belonging to each topic. These characterizations are sensitive when looking at the content of tweets, and these examples give us confidence in the reliability of our topic modeling.
Results
This trend continues until the day after the protest, after which the results begin to disappear. This increases appreciation for the small leaps in identity on the day of the protest.
Discussion
This contradicts expectations given the existing literature, which generally suggests that collective identity is an important driver of political participation. These results suggest a minimal impact of collective identity on protesting and protesting on collective identity.
PERSUADABLE VOTERS DECIDED THE 2022 MIDTERM
ABORTION RIGHTS AND ISSUES-BASED FRAMEWORKS FOR ELECTIONS
Introduction
The range in predictions given by academic experts for the 2022 midterm elections was considerable (Edsall, 2022). The underperformance of the Republican Party in the 2022 midterm elections is the focus of this paper.
Midterm Elections and American Politics
But the issue we argue was largely the focus of the 2022 midterm elections was abortion. In all but three midterm elections since 1916, the president's party has suffered a net loss of seats in the House of Representatives, as seen in Figure 2.1.
Data and Methods SurveysSurveys
A secondary concern is that voters may be influenced by media stories in the immediate aftermath of the election. Those who identify as Democratic would likely vote for the Democratic nominee regardless of their view of the financial and economic situation.
Results
A popular model for midterm performance, we first test the midterm-as-a-referendum model explicitly in the case of the 2022 midterm elections. The estimated average marginal effect for violent crime in the pooled model is 4.1% while in the Independent Party-based model it is 23.9%.
A REPULSIVE BOUNDED-CONFIDENCE MODEL OF OPINION DYNAMICS IN POLARIZED COMMUNITIES
Introduction
Bounded-confidence models are a class of models that assume that individuals change their opinions based on their relationships when their opinions are already close to those of their peers. We consider polarization and the notion that individuals can form their opinions by being contradictory. Similar to other bounded trust models, we maintain the idea that individuals are mostly influenced by others whose opinions are already somewhat close to our own.
We introduce the motivation for our model in Section 3.2 and define our model in Section 3.3.
Background and Motivation
Agents connected to each other will influence each other's opinions, but only if their opinions are sufficient. That is, even if two agents are connected, if their opinions are far apart, they will not consider each other as they form new opinions. That is, at time𝑡+1, we examine all neighbors of 𝑖 that are within the confidence, and then average their opinions.
Specifically, the HK model suggests that as the confidence limit increases, there is a transition between three types of steady states.
Model Statement
As the confidence boundary increases, stable states begin to show only a small number of dominant opinions (polarization). Note that the third row of 𝑀𝑖 𝑗 includes the case where two nodes have the same opinion and reject each other. Second, at repulsive edges, connected nodes that are in trust with each other may not converge into a single view.
For example, the same three-node example in Figure 3-2 converges to a state where all three nodes are still connected and within trust of each other, but do not have the same opinion.
Analytical Results
At time𝑡, assume 𝑥𝑖(𝑡) > 𝑥𝑗(𝑡) for all other nodes 𝑗, so that 𝑖 is the node with the highest opinion value at time𝑡. At time𝑡, assume 𝑥𝑖(𝑡) > 𝑥𝑗(𝑡) for all other nodes 𝑗 ∈ 𝑉, such that this𝑖 is the node with the highest value at time𝑡. At time𝑡, say𝑥𝑖(𝑡) < 𝑥𝑗(𝑡) for all other nodes 𝑗 ∈𝑉, so this𝑖 is the node with the lowest opinion value at time 𝑡.
Combining these lemmas, we can see that eventually𝑡′,𝑖𝑚 𝑎𝑥 will be within confidence of exactly one other node.
Numerical Results on Synthetic Networks
To see the numbers showing that the range of final opinions scales with the number of repulsive edges and 𝑐, see Figure 3.5 and related. In Figure 3.5, a test is shown for each set of parameters when 𝑝1 is set to 0.4. A sample of the generating process can be seen in Figure 3.7, where the blue edges represent positive edges and the black negative edges.
Again, to create simulation results, 100 trials are performed for all combinations of the parameters.
Future Work and Conclusions
INCORPORATING LATENT CLASS IDENTITIES IN QUANTITATIVE WORK
Introduction
Researchers who focus on SSS tend to understand it through the lens of identity and consciousness/placed in one of the cultural definitions of class (Jackman & . Jackman. However, the explicit maximization problem assumes that each point is a linear combination of two sets, with weights that represent grouping variables. Thus, by generating estimates for individuals using combinations of groups represented by probabilities, we generate a class scale.
This means that regardless of the size of the cross-sectional effects of race, gender, and class on outcomes, their inclusion is necessary for the models to be consistent with individuals' lived experiences.
Approaches to Class as an Identity and Predictor
For both results, we have that the prominent version of the class is slightly different. We take this approach to provide a holistic view of class that can then be combined with racial and gender identities to gather a comprehensive view of structural inequality and lived experience (Crenshaw, 1989; Yuval-Davis, 2015). . In this paper, we draw on the lessons learned from this section—class influence is susceptible to variance based on how it is operationalized, and care must be taken to properly articulate its contours.
Jackman and Jackman, 1973 showed that the boundaries of these class identities and contact patterns then led to distinct out-group views and the development of class-based identity using the work of (Tajfel, 1969).
Class as a Latent Variable
The functional form of the class group outcome is exogenous to the model description. While the prescriptions are useful in giving direction to the model, they are a key point of the mixture model. Finally, it is important that the coefficients of the class-determining variables are not identical between the mixture model and the fit of the unsupervised model (which provided the priors).
However, it is important that the mixture model does not perform worse than a non-mixture version of the same model.
Empirical Results
This indicates that there is information learned from using the response variable in the model. 1 refers to a 100% probability of having a higher class status while 0 refers to a 100% probability of belonging to the lower class. Upper-class-status Black Americans are more sympathetic than their lower-class-status counterparts (although not by a large difference), as shown by the pooled class model that splits the difference between the higher- and lower-status models.
This again shows that a researcher might underestimate support for BLM for upper class status across ages, and overestimate support for lower class individuals.
Discussion and Conclusion
For the most part, the results of the overall class model share the results of the two class extremes. This means that traditional models are likely to bias the results towards the larger class group shown in the data. By reducing the subjectivity of the class, we are able to improve the bias introduced through the researcher's intuition.
This technique opens doors to work on how classes are perceived by individuals, as well as the effects of class on outcomes.
BIBLIOGRAPHY
Data Collection
Full details of the process and the package used can be found in Kann et al., 2023. We first decided to narrow down the decision to three target cities – this allowed us to take advantage of clustering and the cities of Los Angeles, Houston, and Chicago were chosen for their large size. In the rest of this section, we discuss the choice of keywords, the works we look for to identify the protesters, and an overview of the protest data.
The keywords listed in Table A.1 fall into three different categories: (1) calls to mobilize others to actively participate in protests, (2) names of the people who were victims of injustice, and (3) phrases often sung during protests.
Topic Analysis
This means that we can imagine a prior division of topics before we have seen any documents. The most used words in each senTopic topic can be seen in Figure A.2, the size represents the number of tweets in which the word appears. A box-and-whisker plot of the percentage of BLM topics for each city for these types of tweets can be seen in Figure A.4.
In addition, in table A.5 you can see the correlation between the scores for each person and the RJST model.
Regression
In this section, we report the regression results with interaction terms to justify the clustering of sites. In addition, we include placebo tests to demonstrate the reliability of our results. These results were null for one group and in the opposite direction to the BLM results for the other.
Therefore, the significance of these results does not take away from the significance found in the paper.
THE 2022 U.S. MIDTERM ELECTION
Survey Question Wording Vote ChoiceVote Choice
Crosstabulations
Regression Result Figures
Regression Result Tables
Pooling Tests
A REPULSIVE BOUNDED-CONFIDENCE MODEL
Additional Proofs
So if 𝑥𝑖(𝑡) has the highest value mean at some time𝑡, it will always have the highest. From the definitions of𝑈𝑖(𝑡), 𝐿𝑖(𝑡), we can observe that the member of 𝑀 with the highest index will have the largest corresponding set 𝐿𝑖(𝑡) and the smallest corresponding𝑈𝑖(𝑡,𝑡), such that + 1 time member of 𝑀 will have the highest valued opinion of all members of𝑀. Suppose that there is a node 𝑗 such that𝑥𝑖(𝑡) −𝑥𝑗(𝑡) < 𝑐 and that 𝑗 has the highest valued meaning of all such nodes.
By assumption, since 𝑗 has the highest valued opinion of all nodes trusting 𝑖, 𝑊𝑖 𝑗(𝑡) is empty.
INCORPORATING LATENT CLASS IDENTITIES
ANES Details
250,000 or more individuals who responded with Rejected (-9) or Interview Discontinuance (-5) were excluded from the analysis. Individuals who responded with Rejected (-9), No Data After Selection (-7), No Interview After Selection (-6), or Interview Discontinuance (-5) were excluded from the analysis. Individuals who responded with Refused (-9), Don't Know (-8), No data after the election (-7), No interview after the election (-6) or (-5) Interview interruption were excluded from the analysis.
Individuals who responded with Refused (-9), Don't Know (-8), or Disconnected Interview (-5) were excluded from the analysis.
Stan Implementation
The table below shows the number and percentage of each type of response for each subject. In Figure D.1 the histogram for estimated group probabilities of the logits for respondents of each class from the original grouping can be seen. The estimated output for each coefficient can be seen in Table D.2 under the columns Mean and Estimated SD.
We then run the unmixed regression for each output variable, split into groups when the estimated class is less than 50% or more than 50%.
Results