Multilevel Modeling for the Intersections of Race and Gender

• In Event: Poster Session: Political Methodology

Abstract

In an increasingly data-driven world, it becomes the duty of those engaged in scholarship to ensure that data analysis does not perpetuate inequalities and erase lived experience. Using Kimberle Crenshaw’s framework of intersectionality, scholars within Political Science can combat this by studying the compounding effect these identities can have on an individual if they represent interlocking systems of oppressions (ex: experiencing both racism, sexism, and class structures). There are two goals of this work: a methodological contribution arguing an intersectional approach to Twitter data, the second is modeling the impacts of racism and sexism during the 2020 primary season on rates of angry and fearful Tweets of women and people of color. Anger and fear can be both catalysts and inhibitors to political action so it’s vital to understand their nuances. Research shows that racism and sexism exist on Twitter around 2016 election content, but the work is focused on the source of this rhetoric rather than looking at the ramifications on those targeted by this rhetoric. The goal will be to have an original Twitter dataset from the 2020 primary season that can be crossed with compiled algorithms to predict race and gender of the individual to create intersectional data. The methodological contribution compiles evidence that there are systematic differences in Twitter rhetoric at the intersections of race and gender through unsupervised machine learning cluster methods. The substantive portion deals with rates of anger and fear on Twitter during the election cycle (specifically for those who hold identities targeted by racist or sexist speech). An Interrupted Time Series Analysis is used to capture the effect of the election on the proportion of Tweets expressing political anger or fear. Overall, this work aims to ground machine learning methods of analyzing political behavior in feminist theory to produce research that both better explains the lived experience of marginalized people and emphasizes the stakes of work on race/ethnicity and gender.Political science scholars have acknowledged the importance of race/ethnicity and gender roles' in framing the way an individual views the world and thus influencing their behavior. Using the framework of intersectionality, feminist scholars, and critical race scholars, have studied the compounding effect these identities can have on an individual if they represent interlocking systems of oppression (ex: experiencing both racism and sexism). Quantitative methods in political science, however, have room for improvement in accounting for intersectionality, often studying race/ethnicity or gender solely, and not placing the influence of these intersections as the priority methodologically. It is vital for tools consistent with an intersectional framework to be readily available to political scientists. This paper attempts to acknowledge vital considerations when applying intersectional theory and identifies the potential for multilevel modeling methods for measuring intersectional effects. Multilevel models account for better relationships between groups and within groups through estimation of mixed effects and more detailed variation structures, thus making it a better tool for intersectional effects than classical regression. This piece replicates two articles that showcase the potential of multilevel models to change the way political science scholars study intersectional identities.

Author

• Melina Much, UC Irvine