Title Education, region, and american attitudes toward immigrants: a multilevel and machine-learning analysis of the gss 1996–2024
Translation of Title Išsilavinimas, regionas ir amerikiečių požiūris į imigrantus: daugialygė ir mašininio mokymosi GSS 1996–2024 m. analizė.
Authors Balulis, Julia
Full Text Download
Pages 40
Keywords [eng] immigration attitudes, education, multilevel modelling, contextual effects, citizen ship attachment, machine learning, General Social Survey
Abstract [eng] This thesis examines how education and regional context relate to American attitudes toward immigrants and national identity from 1996 to 2024. Using pooled data from five waves of the Gen eral Social Survey, four outcomes are analyzed: whether immigrants are seen as taking jobs, being bad for the economy, whether immigration should be reduced, and whether respondents report very strong attachment to American citizenship. The main method is multilevel logistic regression with individuals nested within nine census regions. Education is split into a within–region (individual) component and a between–region (con textual) component. I add robustness checks with standardized income and compare the results to simple machine–learning models implemented in Python. Across all outcomes, higher education within a region is strongly linked to more positive views of immigrants: better educated respondents are less likely to see immigrants as a jobs or economic threat, and less likely to support restricting immigration. Living in more educated regions points in the same direction, though contextual effects are smaller. In contrast, education is associated with weaker reported attachment to American citizenship. Income explains only part of this pattern: it improves the citizenship models but does not remove the education effects. Over time, perceived threats decline between 1996 and 2014, then rise again by 2024, while strong citizenship attachment falls. The findings suggest that education and context shape both immigration attitudes and national identity, and that multilevel models combined with machine–learning tools provide a useful frame work for studying these relationships in large social surveys.
Dissertation Institution Vilniaus universitetas.
Type Master thesis
Language English
Publication date 2026