We present a field experiment that uses geo-referenced smartphone advertisements to measure ethnic preferences at a highly disaggregated level. Different types of banners advertising a vote matching tool are randomly displayed to mobile Internet users in Bosnia and Herzegovina, while recording their spatial coordinates. Differences in the response (click) rate to different ethnic cues on these banners are used to measure temporal and spatial variation in ethnic preferences among the population of Bosnia and Herzegovina. Our study lays out the theoretical and practical underpinnings of this technology and discusses its potential for future applications, but also highlights limitations of this approach.

Existing research has shown that online networks are often segregated along identity lines, such as political ideology or religious views. Although online segregation should be specifically detrimental when appearing between ethnic groups in a post-conflict setting, to date we have no systematic evidence on the level of online ethnic segregation. To close this gap, the present study examines online ethnic segregation in a large ethnically mixed blogger network in a postconflict society, Bosnia and Herzegovina. Since politics has been found to enhance ethnic divides in the offline world, we additionally examine whether segregation is higher for bloggers engaging with political topics. Using large-scale web scraping, automated text analysis and Monte Carlo simulation, we find evidence for pronounced ethnic divisions. Furthermore, we find that political bloggers tend to have more ethnically segregated networks. The findings show that a broad public exchange transcending ethnic categories remains limited in the online context we study, and that those who dominate the online political debate tend to be those who in their social interactions put even more weight on ethnic categories than the average.