Computational analysis of 140 years of US political speeches reveals more positive but increasingly polarized framing of immigration | Proceedings of the National Academy of Sciences

Immigration is one of the most important and divisive topics in American public life. From the rise of vocal antiimmigrant politicians in recent years, it is tempting to conclude that attitudes toward immigration are more negative—or at least more polarized—than ever before. However, resistance to newcomers has always been a central part of our public discourse about immigration. From anti-Chinese fearmongering in the 1880s to concerns about Southern and Eastern European immigrants in the 1920s to the antiimmigration rhetoric of the Trump administration (2017 to 2020), claims that certain types of immigrants can never truly join American society have been a perennial part of our discourse. For example, Senator Henry Cabot Lodge, an architect of antiimmigrant legislation, declared a century ago, “[Immigration] is bringing to the country people whom it is very difficult to assimilate” (1, p. 35) because immigrants are from “races most alien to the body of the American people” (1, p. 32).

We seek to move beyond individual anecdotes to ask, how have attitudes toward immigrants in the United States changed over the past century? How does recent political debate over immigration compare to the long sweep of US history? This question is a challenge because public opinion polls that asked about attitudes toward immigration only began in the 1960s and were then only asked about immigration sporadically until recent years. We instead turn to the Congressional Record and other sources of political speech, using quantitative text analysis methods to systematically investigate the language used in congressional and presidential speeches about immigration over the past 140 y.

Our paper considers the full corpus of more than 17 million congressional speeches from 1880 to the present, of which we identify ∼200,000 speeches relevant to the topic of immigration. We also incorporate presidential communications from the same time period, making this a comprehensive quantitative analysis of American political speech about immigration at the federal level, covering the entire time period from the Chinese Exclusion Act of 1882 to the present day.

Numerous studies have analyzed the political history of US immigration using qualitative approaches and historical archives (2–7); quantitative work on immigration has also used data such as migration and census records (8, 9). Rhetorical aspects of immigration debates have been studied qualitatively—especially the use of dehumanizing language and metaphors such as “vermin” and “cargo” (10–13)—but these authors have not rigorously quantified how common such language is over time. Last, other scholars have applied computational methods from natural language processing to study coverage of immigration in news media and Congress (14–18), but none have used these tools to investigate such a long time span or comprehensive corpus of speeches about US immigration with a consistent methodology.

Our analysis is based on a combination of methods. To identify relevant speeches, along with a corresponding tone (proimmigration, antiimmigration, or neutral), we make use of automated text classification based on extensive human annotations. Using a semiautomated process, we also curate and apply a set of lexicons for analyzing relevant frames (i.e., ways of characterizing immigrants and immigration). Finally, to quantify implicit dehumanizing metaphors in speeches, we develop an approach using neural contextual embedding models to measure if references to immigrants are suggestive of various metaphorical categories (Materials and Methods).

We find that political speeches about immigration today are far more likely to be positive than in the past, with the shift from negative to positive mostly taking place between World War II (WWII) and the passage of the 1965 Immigration and Nationality Act, and being net positive on average in nearly all sessions of Congress since the early 1950s. Extending this analysis to presidential communications, we find President Trump to be a stark exception, as the first president in modern American history to express sentiment toward immigration that is more negative than the average member of his own party. As with many political issues, the two parties have become increasingly polarized over time, and we find a linear increase in polarization on immigration, beginning in the late 1970s under President Carter. Today, Democrats are unprecedentedly positive about immigration, whereas Republicans are as negative as the average legislator was in the 1920s during the push for strict immigration quotas. This divergence is clearly part of a broader trend toward polarization on many issues (Discussion); for immigration specifically, our analysis reveals the beginnings of this, predating the rise in generic political polarization observed in Gentzkow et al. (19) by more than a decade.

Along with the polarization by party, nationality of immigrants continues to matter greatly, with speeches mentioning Mexican immigration being consistently more negative than the average (dramatically so in comparison to European groups). Moreover, there is a striking similarity between how Mexican immigrants are framed today and how Chinese immigrants were framed during the period of Chinese exclusion in the 19th century: more negative in tone; greater explicit emphasis on frames such as “crime,” “labor,” and “legality”; and significantly greater use of implicit dehumanizing metaphors, in comparison to European groups.

Thus, while far more members of Congress today express favorable attitudes toward immigration than in the past, there remains a strong and growing strain of antiimmigration speech, especially among Republicans, along with perennial references to threats, legality, and crime. Despite the elimination of country-specific immigration quotas in the 1960s, expressed opinions toward immigrants still vary greatly by country of origin, and enduring rhetorical strategies continue to be deployed against more marginalized groups.

Results

Tone of Immigration Speeches.

Starting with the complete record of 17 million congressional speeches from 1880 to 2020 (Data), we collected human annotations and trained machine learning classifiers to identify speeches relevant to immigration, along with an accompanying tone (proimmigration, antiimmigration, or neutral; Classification). Both panels of Fig. 1 show the average tone (percent proimmigration minus percent antiimmigration) expressed in congressional speeches over this time period (black line).* The trends for congressional speeches by Democrats and Republicans are also shown in Fig. 1, Top. A comparable time series for presidents is shown in Fig. 1, Bottom, by applying the same models to all presidential communications collected by the American Presidency Project (20). For alternative models, validity checks, and variation within parties, refer to SI Appendix.

Fig. 1.

OPEN IN VIEWEROPEN IN VIEWER

We begin by documenting a number of findings about political speech related to immigration. First, average sentiment toward immigration in Congress and the executive branch is negative throughout the late 19th and early 20th centuries, from the passage of the Chinese Exclusion Act (1882) through the advent of strict immigration quotas in the 1920s. The pervasiveness of negative sentiment can help make sense of the political context that gave rise to a suite of increasingly restrictive immigration regulations. It is particularly noteworthy that we do not find a rise in negative speeches leading up to the Emergency Quota Act of 1921. Rather, we find that political sentiment in Congress was staunchly antiimmigration for more than 4 decades, which is consistent with the political history that has recounted the many congressional attempts to pass antiimmigration legislation, all of which were struck down by the president, in the years before the successful passage of quotas (21). Second, attitudes toward immigration became more positive around the start of WWII, rising steadily from 1940 until the end of the Johnson administration (1969). The average tone in Congress has essentially been proimmigration since the beginning of the Eisenhower administration (1953), consistent with efforts by postwar presidents to reframe the public understanding of immigration as positive for the country.

Third, beginning about a decade after the reopening of the border with the 1965 Immigration and Nationality Act, there has been a growing partisan divide, larger year-to-year variations, and an overall decline in sentiment toward immigration among Republicans. Democrats, by contrast, have grown more positive about immigration over time, especially under Presidents Obama and Trump, with the exception of a temporary bipartisan drop in proimmigration speeches in the early 1990s, coinciding with the end of the Cold War and the passage of the North American Free Trade Agreement (NAFTA). By contrast, Republican legislators are now approximately as overtly antiimmigration in their speeches as the average legislator was during the Age of Mass Migration from Europe and the 1920s quota periods.

The trends for presidential attitudes toward immigration should be treated more cautiously as there is less text available from presidents overall and because these estimates involve a slight domain shift (from congressional speeches, on which our models were trained, to more varied types of presidential communications). Nevertheless, we document a similar pattern, whereby early presidents were more antiimmigration than modern presidents. In recent years, presidents have been uniformly more proimmigration than the average member of Congress, including both Republicans like Ronald Reagan and Democrats like Jimmy Carter. In historical comparison, President Trump was a stark exception: by his utterances, he was the most antiimmigration president to sit in office over the past 140 y, relative to the average attitude of the time expressed in Congress.

Although the difference in tone between the parties today is larger than at any point in the past, tone also varies dramatically depending on which groups of immigrants are being discussed. Fig. 2 shows the average tone when considering only those speeches that mention each of the three most commonly mentioned nationalities in immigration speeches—Mexican, Chinese, and Italian (Identifying Groups).

Fig. 2.

…