Racialized Ideology and Structure in Online Rental Advertisements: Using Natural Language Processing to Uncover Racialized Discourse

Abstract

This dissertation analyzes how the discursive representation of racist ideology in rental ads link up with the racialized social system in the US, a system I define, following Bonilla Silva (1997) as a “societ[y] in which economic, political, social, and ideological levels are partially structured by the placement of actors in racial categories or race.” Throughout this dissertation, I develop a connection between discourse and housing outcomes as likely implicated in processes that reproduce segregation and gentrification from three viewpoints. First, in Chapter 1 I examine a particular aspect of the advertisements: how they describe neighborhood quality. I use text analysis and regression techniques to see how those descriptions vary in neighborhoods with different racial compositions and different socioeconomic status. I find that advertisements in White and Asian neighborhoods describe those places more positively than advertisements in Black or Latinx neighborhoods. I also investigate the ways that text writers in some Black neighborhoods might resist the trend of devaluing Black neighborhoods. In Chapter 2 I look at a second, somewhat broader area of discourse: exclusionary language. I investigate how it varies across race and socioeconomic class, and add consideration of gentrifying areas. I find that some kinds of discursive exclusion, like background and credit checks, are more common in Black neighborhoods, while others, like strong tenant expectation, are more common in White neighborhoods. Finally, in Chapter 3, I return to the discursive themes from the first two chapters, but consider their variation across the sixteen metropolitan areas in my study. I find that the racialization of discourse about neighborhood quality is consistent across the whole study area. I also find that, in general, some cities have more exclusionary discourse than others, and that that variation can be partially explained by metropolitan-level variables. For instance, cities with lower levels of segregation tend to have more language about background and credit checks.