Open Collections

Gyetvay, Samuel

University of British Columbia

The first two chapters of this thesis study how worker sorting across firms shapes the wage effects of immigration using administrative data from Germany. The third chapter studies algorithmic and human discrimination in hiring. Chapter 1 documents how immigrant workers are sorted across firms in Germany over the period 1975-2018. The main findings are: (1) changes in firm sorting can explain nearly all of the increase in the native-migrant wage gap since the 1990s, (2) migrant workers are highly concentrated in low-paying firms, particular in the first 5-10 years after arrival, and the negative sorting is stronger for more recent cohorts, (3) migrant workers are highly segregated from native-born workers in the labor market, even within narrowly defined labour markets, and (4) immigrant workers--in particular Central and East Europeans--are more likely to find jobs at firms that previously employed workers from their country of origin. Chapter 2 investigates, both theoretically and empirically, how the sorting of migrant workers across firms documented in the first chapter shape the wage effects of immigration. The main theoretical result characterizes how the effect of an immigration shock on average native-born workers' wages depends on migrants' sorting across in a model of worker sorting across heterogeneous firms. Segregation reduces within-firm competition between immigrants and native-born workers, and when immigrants enter low-wage firms, workers exposed to competition reallocate to higher-wage firms. The empirical analysis tests and corroborates these theoretical predictions by estimating the effects of firm- and market-level immigration shocks on firms and workers. Chapter 3 introduces a new methodology to quantify degree of bias in the initial stages of the hiring process in both humans and Large Language Models (LLMs) such as ChatGPT. To do so, we reanalyze data from a previous resume audit study in which researchers submitted fictitious resumes with racially distinctive names to real job ads. We reconstruct each resume from the original study and conduct a synthetic experiment by asking LLMs who they would hire. Our results suggest that LLMs are less racially biased than their human counterparts. However, LLMs are not perfectly unbiased: they do use information contained in names.

Text

2024-08-15

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/88953

Economics

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/