Concerns About Automation and Negative Sentiment Toward Immigration

Abstract

Across 12 studies (N = 31,581), we examined how concerns about the rise of automation may be associated with attitudes toward immigrants. Studies 1a to 1g used archival data ranging from 1986 to 2017 across both the United States and Europe to demonstrate a robust association between concerns about automation and more negative attitudes toward immigrants. Studies 2a, 2b, 2c, and 3 employed both correlational and experimental methods to demonstrate that people’s concerns about automation are linked to increased support for restrictive immigration policies. These studies show this association to be mediated by perceptions of both realistic and symbolic intergroup threat. Finally, Study 4 experimentally demonstrated that automation may lead to more discriminatory behavior toward immigrants in the context of layoffs. Together, these results suggest that concerns about automation correspond to perceptions of threat and competition with immigrants as well as consequent anti-immigration sentiment.

Keywords

automation intergroup threat immigration open data preregistered

The rise of sophisticated technology has altered our daily lives considerably (Bonnefon, Shariff, & Rahwan, 2016) and has begun affecting employment. Automation has already spurred job losses across industries and threatens to replace the jobs of millions in the coming years (Arntz, Gregory, & Zierahn, 2017; Frey & Osborne, 2017). Given its economic effects, automation will be a key political issue in forthcoming years.

Beyond its economic effects, automation has several downstream social and psychological consequences, including, as we suggest here, shaping attitudes toward immigration. Recent studies have linked exposure to automation to support for political candidates and parties associated with restrictive immigration positions in the United States (Frey, Berger, & Chen, 2018) as well as in Europe (Anelli, Colantone, & Stanig, 2019; Dal Bó, Finan, Folke, Persson, & Rickne, 2018; Zhen & Mayer, 2017). In this article, we directly examine the relationship between concerns about automation and attitudes toward immigrants.

Despite automation’s outsized role in shaping society, little empirical work has examined its influence on social and intergroup relations. Sociological research has documented how during America’s industrialization period in the late 1800s and early 1900s, automation increased competition in labor markets and hampered relations between native workers and minority groups (Brown, 1998; Olzak, 1989). These historical trends suggest that automation may influence intergroup attitudes by intensifying perceptions of economic competition with immigrants, contributing to resentment.

However, advances in artificial intelligence and automation not only affect perceptions of economic resources (i.e., jobs) but also drive perceptions of cultural change. Concerns about technological unemployment have already led to the proposal of new policies (such as universal basic income) that could yield profound social transformation. In addition, automation changes the nature and value of work, putting into question basic values on which American culture and identity are built. Much of Western society subscribes to a value system, the Protestant work ethic, that prizes hard work and self-reliance (Weber, 1905/2013; for the “American work ethic,” see Applebaum, 1998). Therefore, automation may influence intergroup attitudes by intensifying the perception that existing cultural and societal values are under threat.

Past work has shown that threats not directly related to a specific out-group (e.g., economic downturn, unemployment, natural disasters) can trigger threat perceptions toward out-groups (e.g., Andrighetto, Vezzali, Bergamini, Nadi, & Giovannini, 2016; Brambilla & Butz, 2013; Butz & Yogeeswaran, 2011; Diaz, Saenz, & Kwan, 2011). Automation may represent a previously unidentified macrolevel driver of anti-immigrant sentiment. On the basis of previous work on intergroup threat (Esses, Dovidio, Jackson, & Armstrong, 2001; Esses, Jackson, & Armstrong, 1998; Rios, Sosa, & Osborn, 2018; Stephan, Ybarra, & Morrison, 2009), we propose the following process: People may perceive automation as a tax on their group’s resources—both on their material resources and on their group’s more intangible resources (Yogeeswaran et al., 2016; Złotowski, Yogeeswaran, & Bartneck, 2017). The prospect of automation’s impact on employment may destabilize people’s general sense that their group has economic and cultural security, leading people to perceive that other entities that compete for these same resources are also more threatening. Specifically, automation may increase people’s perceptions of group threat from immigrants—a salient social group that is stereotyped as producing economic competition (i.e., taking jobs) and cultural competition (i.e., promoting competing values).

Therefore, we hypothesized that automation may be associated with anti-immigrant sentiment through two potential routes. First, automation may increase perceptions of realistic threat toward immigrants arising from competition for economic resources. Second, automation may increase perceptions of symbolic threat toward immigrants arising from changes to group values, identity, and status. Both types of intergroup threats have previously been shown to be exacerbated by macrolevel factors in the environment or social context. For instance, work shows that macroeconomic conditions such as economic scarcity can elicit realistic threat and heighten out-group hostility and prejudice (Butz & Yogeeswaran, 2011; Krosch & Amodio, 2014). Other work has demonstrated that exposure to demographic change can increase Whites’ concerns over losing group status, engendering greater endorsement of policies and ideologies associated with negative sentiment toward racial out-groups (Craig & Richeson, 2014; Willer, Feinberg, & Wetts, 2016) and support for a more anti-immigrant presidential candidate (Mutz, 2018). In addition, other work shows that simple exposure to the idea of multiculturalism activates more symbolic threats, increasing prejudice toward ethnic minorities (Morrison, Plaut, & Ybarra, 2010).

As a potential source of threat to economic resources, automation may activate experiences of realistic threat from immigrants. Perceiving that an important resource (e.g., wealth, jobs) is limited, combined with awareness of a potentially competitive out-group, can produce perceptions of group competition for resources (e.g., Butz & Yogeeswaran, 2011; Esses et al., 1998). This perceived intergroup competition can increase negative feelings toward members of the other group and efforts to thwart that group’s access to those resources (Sherif, 1966).

As a potential source of cultural change, automation may also increase experiences of cultural threat from immigrants. Automation will transform work, which could diminish the importance of traditional Western values that emphasize devotion to work, discipline, and self-reliance (Katz & Hass, 1988; Weber, 1905/2013). Awareness of automation’s impact on the future of work may lead people to perceive that essential aspects of identity and culture are vulnerable, thereby increasing the perception that immigrants with differing values also pose a threat to the dominant culture. Considering automation’s potential to trigger both realistic and symbolic threat, we predicted that awareness of automation would be associated with anti-immigrant sentiment.

The Present Research

We tested our hypothesis across 12 studies. Studies 1a to 1g used archival data to examine the relationship between these constructs over the past 30 years across the United States and Europe. Studies 2a to 3 then used both correlational and experimental methods to investigate how automation influences perceptions of group threat toward immigrants and support for restrictive immigration policy. Finally, Study 4 examined whether people are more likely to discriminate against immigrants when automation results in workplace downsizing. Data and analysis scripts for all studies in this article are available on the Open Science Framework at https://osf.io/z28rq/.

Studies 1a to 1g

In Studies 1a to 1g, we tested whether people who perceive automation as a greater threat to employment are likely to also hold more negative perceptions about immigrants.

Method

We collected archival data from seven surveys conducted in the United States and Europe during the period ranging from 1986 to 2017. Additional details about survey selection and details of each survey can be found in the Supplemental Material available online. Summary demographic statistics for each survey are presented in the Supplemental Material.

Results

Descriptive statistics and zero-order correlations between all key variables and demographic statistics for each of the seven time periods are presented in the Supplemental Material.

For each of the seven surveys, we constructed a four-step linear regression model predicting attitudes toward immigrants; automation threat was entered in the first step, political ideology was added in the second step, demographic variables were added in the third step, and other employment-related threat variables were added in the fourth step (these variables were included to test the distinctive effect of automation concerns compared with general concerns about employment). Across Studies 1a to 1g, we used probability weights provided by the polling agency, and we estimated robust standard errors to account for heteroscedasticity in the errors.

Results supported our prediction: As shown in Table 1, there was a significant relationship between concerns about automation and attitudes toward immigrants across all surveys and time periods (all βs = 0.05–0.39, ps < .05). People who perceived automation as having a more harmful impact on workers also tended to have more negative attitudes toward immigrants. This relationship persisted across all time periods even after we adjusted for political ideology (all βs = 0.05–0.39, ps < .05). Furthermore, with the exception of the 1999 survey data, in which the association was no longer significant (β = 0.03, p = .189), adjusting for other demographic variables (i.e., nationality, European survey only, household income, race, education level, employment status, gender, and age) did not change the relationship between perceptions about automation and attitudes toward immigrants (all βs = 0.05–0.36, ps < .05). Finally, adjusting for people’s perceptions of other employment-related threats (e.g., unions, inflation, companies sending jobs overseas), which were assessed in six of the seven surveys, concerns about automation were still associated with more negative attitudes toward immigrants across surveys (βs = 0.04–0.29, ps < .001), with the exception of the 1995 survey data (β = 0.02, p = .528). These findings suggest that automation concerns are associated with anti-immigrant sentiment, irrespective of concerns about employment more broadly. These results provide preliminary evidence for the association between concerns about automation and negative attitudes toward immigrants in both American and European contexts. Additional exploratory analyses to account for other factors (presented in the Supplemental Material) show that these findings are robust.

Table 1.

Results From Hierarchical Regression Analyses for Automation Threat Predicting Negative Perceptions of Immigrants (Studies 1a–1g)

Step and variable	Time period
Step and variable	1986	1995	1996	1999	2006	2016	2017
Step 1
Automation threat	0.38*** (0.03)	0.10** (0.03)	0.32*** (0.03)	0.06*(0.03)	0.23*** (0.03)	0.26*** (0.03)	0.10*** (0.01)
Step 2
Automation threat	0.38*** (0.03)	0.10** (0.03)	0.32*** (0.03)	0.06* (0.03)	0.23*** (0.03)	0.25*** (0.03)	0.10*** (0.01)
Political ideology	0.05^† (0.03)	0.03(0.03)	0.11*** (0.03)	0.09*** (0.03)	0.16*** (0.03)	0.28*** (0.02)	0.10*** (0.01)
Step 3
Automation threat	0.35*** (0.03)	0.08* (0.03)	0.26*** (0.03)	0.03(0.03)	0.19*** (0.03)	0.21*** (0.02)	0.08*** (0.01)
Political ideology	0.03(0.03)	0.00(0.03)	0.11*** (0.03)	0.08** (0.03)	0.12*** (0.03)	0.21*** (0.03)	0.10*** (0.01)
Step 4
Automation threat	0.28*** (0.03)	0.04(0.03)	0.18*** (0.03)		0.14*** (0.03)	0.17*** (0.03)	0.05*** (0.01)
Political ideology	0.05(0.03)	0.00(0.03)	0.13*** (0.03)		0.15*** (0.03)	0.23*** (0.03)	0.11*** (0.01)
Employment threats	0.19*** (0.03)	0.16*** (0.03)	0.21*** (0.03)		0.18** (0.03)	0.16*** (0.03)	0.37*** (0.01)
N	1,250	1,014	1,347	1,483	1,225	1,704	21,228

Note: The table shows standardized regression coefficients, with robust standard errors in parentheses. Coefficients for the demographic variables added in Steps 3 and 4 are available in the Supplemental Material available online.

†

p < .10. *p < .05. **p < .01. ***p < .001.

Studies 2a and 2b

Studies 2a and 2b were targeted tests of the association between automation concerns and support for more restrictive immigration policy that also addressed factors driving this relationship. In both studies, we assessed participants’ perceptions of immigrants as a realistic and symbolic threat and hypothesized that concerns about automation would correspond to increased support for more restrictive immigration policy, driven by both threats. Study 2b was a replication of Study 2a with a population outside of the United States who are encountering similar immigration issues.

Method

Participants

On the basis of the estimates of the relationship between automation concerns and attitudes toward immigrants observed in Studies 1a to 1g, we expected to find a small to medium effect size. A G*Power analysis (Faul, Erdfelder, Lang, Buchner, 2007) revealed that we would have 99% power to detect a medium effect (f² = 0.15), with alpha set to .05, if we recruited a total of at least 125 participants. However, we decided to recruit upward of 250 participants in Studies 2a and 2b to achieve more stable estimates (Schönbrodt & Perugini, 2013). Data collection was stopped when we attained the prespecified target number.

Exclusions were decided a priori (for details, see the Supplemental Material). The final sample in Study 2a consisted of 265 participants (age: M = 44.93 years, SD = 15.80; 73% female; 79% White, 8% Black, 9% Asian, 4% other), and the final sample in Study 2b consisted of 399 participants (age: M = 38.56 years, SD = 11.27; 66% female; 95% White, 3% Asian, 2% other).

Procedure

Participants read that they would be completing a survey about their opinions and attitudes about the workplace. Participants first answered questions about their concerns regarding the impact of automation and were then given scales assessing perceived realistic threat and symbolic threat as a measure of group threat. Finally, participants were asked about six political issues, two of which were directly related to immigration. Demographic characteristics were assessed at the end of the studies.

Measures

The primary measures of interest are described below. More specific details about the primary measures and the exploratory measures included are described in the Supplemental Material.

Concerns about automation

Concerns about automation were measured using six items (Study 2a: α = .78, Study 2b: α = .82). Sample items included “How threatened do you feel by the prospect of automation in the workplace?” and “Do you think the automation of jobs will do more to hurt or help American workers?”

Group threat

Perceived group threat was measured using items from realistic-threat (e.g., “Immigrants should be eligible for the same health care benefits received by Americans who cannot pay for their health care”) and symbolic-threat (e.g., “The values and beliefs of immigrants regarding moral and religious issues are not compatible with the beliefs and values of most Americans”) subscales adapted from previous research (Stephan, Ybarra, & Bachman, 1999). Responses were collected on a 7-point scale ranging from 1 (strongly disagree) to 7 (strongly agree). Items were averaged together to create a composite score for each of the realistic-threat and symbolic-threat subscales (Study 2a: αs > .88, Study 2b: αs > .89).

Policy attitudes

Participants’ attitudes toward immigration policy were measured using two items (Study 2a: r = .38, p < .001; Study 2b: r = .47, p < .001) adapted from previous research (Craig & Richeson, 2014). These items asked participants to indicate the extent to which the required time to be eligible for citizenship in the participant’s country should be decreased or increased and the extent to which foreign immigration to the country should be decreased or increased (reverse scored). Participants’ attitudes toward nonimmigration-related conservative social policies were measured using six items (Study 2a: α = .75, Study 2b: α = .58) that asked about political issues not related to immigration, such as support for increasing funding of the military and defense departments and preferential hiring of racial minorities. Participants responded to all items on a scale from 1 (decreased a lot) to 5 (increased a lot). Given that some of these scales demonstrated relatively lower correlations or reliabilities, we analyzed all items separately; these analyses (shown in the Supplemental Material) yielded conclusions identical to those obtained when using the composites.

Results

Main analyses

Correlations among automation concerns, intergroup threat, and policy attitudes in Studies 2a and 2b are presented in Tables 2 and 3, respectively (descriptive statistics and zero-order correlations between key variables and demographic statistics are presented in the Supplemental Material). We constructed a two-step linear regression model predicting support for restrictive immigration policy; concerns about automation were entered in the first step, and political ideology and other demographic variables (i.e., gender, age, and education level) were entered in the second step. Results revealed that the more concerned participants were about automation, the more they endorsed restrictive immigration policy in both Study 2a (β = 0.28, SE = 0.09, p = .001) and Study 2b (β = 0.16, SE = 0.07, p = .017). Concerns about automation remained a significant predictor of immigration policy even after we controlled for political ideology and other demographic characteristics (i.e., household income, race, education level, employment status, gender, and age) in both Study 2a (β = 0.30, SE = 0.08, p < .001) and Study 2b (β = 0.16, SE = 0.06, p < .009). In addition, concerns about automation did not predict support for the other nonimmigration-related social policies across both Study 2a (β = −0.01, SE = 0.06, p = .866) and Study 2b (β = −0.01, SE = 0.04, p = .732). This result suggests that the effect does not simply represent a broad shift toward conservative policy attitudes.

Table 2.

Correlations Among Automation Concerns, Intergroup Threat, and Policy Attitudes in Study 2a

Variable	1	2	3	4
1. Automation concerns	—
2. Realistic threat	.25***	—
3. Symbolic threat	.24***	.78***	—
4. Immigration policy	.19**	.70***	.69***	—
5. Other social policies	.01	.68***	.62***	.55***

Note: Higher values for immigration policy indicate support for more restrictive immigration policy. Higher values for other social policies indicate more conservative policy positions.

p < .01. ***p < .001.

Table 3.

Correlations Among Automation Concerns, Intergroup Threat, and Policy Attitudes in Study 2b

Variable	1	2	3	4
1. Automation concerns	—
2. Realistic threat	.14**	—
3. Symbolic threat	.16**	.82***	—
4. Immigration policy	.12*	.70***	.68***	—
5. Other social policies	−.02	.62***	.63***	.57***

Note: Higher values for immigration policy indicate support for more restrictive immigration policy. Higher values for other social policies indicate more conservative policy positions.

p < .05. **p < .01. ***p < .001.

Test for mediation

Next, we examined whether realistic and symbolic threats mediated the effect of automation concerns on support for restrictive immigration policy. Although realistic and symbolic threats were strongly correlated in our studies, previous work has posited that these should be distinct constructs, and we therefore included them in our model separately (e.g., Rios et al., 2018; Stephan et al., 1999).¹ Multiple mediation was conducted to test each proposed mediator while accounting for the shared variance between them (Preacher & Hayes, 2008; for a schematic of the multiple mediation models in Studies 2a and 2b, see Fig. 1). Detailed results for tests of all components of the indirect effects can be found in the Supplemental Material.

Fig. 1.

Parameter estimates for the mediation model testing the effect of automation concerns on support for restrictive immigration policy, as mediated by both symbolic and realistic threat in (a) Study 2a and (b) Study 2b. Asterisks indicate significant paths (*p < .05, **p < .01, ***p < .001). On the paths from automation concerns to immigration-policy support, the values below the arrows (c) show the total effect, and the values above the arrows (c′) show the direct effect after controlling for the mediators.

Indirect effects in Study 2a

Results indicated that the total combined indirect effect was significant (β = 0.27, SE = 0.08, 95% confidence interval, or CI = [0.122, 0.423]) and that the individual indirect effects were significant for both realistic (β = 0.15, SE = 0.05, 95% CI = [0.058, 0.262]) and symbolic (β = 0.12, SE = 0.04, 95% CI = [0.046, 0.220]) threats.

Indirect effects in Study 2b

Results indicated that the total combined indirect effect was significant (β = 0.16, SE = 0.06, 95% CI = [0.052, 0.276]) and that the specific indirect effects were also significant for realistic (β = 0.09, SE = 0.04, 95% CI = [0.021, 0.164]) and symbolic (β = 0.07, SE = 0.03, 95% CI = [0.022, 0.134]) threats.

These results suggest that realistic and symbolic threats collectively mediate the relationship between automation concerns and support for immigration policy. Further, they support our hypothesized mechanism: Concerns about automation may increase perceptions of both realistic and symbolic threats from immigrants, increasing support for restrictive immigration policy. However, given the limitations of mediational analyses (Bullock, Green, & Ha, 2010; Fiedler, Schott, & Meiser, 2011), these results cannot definitively prove a causal relationship. Results from Study 2b provide converging support for our hypothesis in another country encountering similar immigration issues. In addition, exploratory analyses (described in the Supplemental Material) tested whether items from the automation-concerns measure assessing societal automation concerns (about automation’s societal effects) or personal concerns about automation (about automation’s personal effects) differentially drive this effect. This analysis found that items assessing general concerns more than personal concerns drove the relationship between automation concerns and anti-immigrant sentiment, a finding that we examined further in Study 2c.

Study 2c

Study 2c was a preregistered conceptual replication of Studies 2a and 2b that included a more precise measure of concerns about automation’s impact that captured both general societal concerns and personal concerns (https://aspredicted.org/mr6d4.pdf). We initially hypothesized that concerns about automation would intensify perceptions of threat toward immigrants. Exploratory analyses in Studies 2a and 2b found that general societal concerns about automation drove this relationship more than personal concerns about automation. However, we did not a priori seek to distinguish between these two concern types with the automation-concern measures in Studies 2a and 2b. Therefore, in Study 2c, we included items intended to assess these two distinct automation concerns in a more targeted fashion.

We had an ancillary aim in Study 2c of examining the role of uncertainty in explaining the relationship between automation concerns and anti-immigrant sentiment. Analyses (described in the Supplemental Material) found that automation’s effect on support for restrictive immigration policy was not being driven by participants’ discomfort with uncertainty.

Method

Participants

Our final sample consisted of 397 participants (age: M = 36.06 years, SD = 10.95; 55% female; 75% White, 13% Black, 7% Asian, 5% other). As in Studies 2a and 2b, we recruited upward of 250 participants to achieve more stable estimates. Data collection was stopped when we attained the prespecified target number. Exclusions were decided a priori (for details, see the Supplemental Material).

Procedure

Participants read that they would be completing a survey on their opinions about the workplace and were asked to answer questions about their concerns regarding the impact of automation on society in general and on their own job. Following this, participants answered the questions assessing perceived realistic and symbolic threats from Study 2. Finally, participants were asked about four immigration policies. Demographic characteristics were assessed at the end of the study.

Measures

Concerns about automation

Concerns about automation’s impact on participants’ own job were measured using three items (α = .97). Sample items included “How concerned are you personally about losing your occupation to automation?” and “How worried are you that you may have to look for a new career because of automation?” Concerns about automation’s impact on society were measured using three items (α = .94). Sample items included “How concerned are you about the general impact of automation in the workplace on society?” and “How worried are you about the impact automation will have on employment in society in general?”

Participants responded to all three items on a scale from 1 (not at all) to 7 (very much). An exploratory factor analysis and a parallel analysis (described in the Supplemental Material) confirmed that personal concerns and societal concerns about automation mapped onto distinct constructs.

Group threat

Perceived group threat was measured using the same realistic-threat (α = .93) and symbolic-threat (α = .87) subscales as in Studies 2a and 2b.

Policy attitudes

Attitudes toward immigration policy were measured using four items (α = .79) that assessed participants’ views on issues such as whether “Dreamers” (i.e., unauthorized immigrants who entered the country at a young age) should be granted citizenship and whether security of the southern U.S. border should be increased through construction of a physical wall.

Results

Main analyses

Correlations among societal and personal automation concerns, intergroup threat, and immigration-policy attitudes are presented in Table 4 (descriptive statistics and zero-order correlations between key variables and demographic statistics are presented in the Supplemental Material). Societal and personal concerns about automation were moderately correlated (r = .52, p < .001); therefore, to determine the relative power of societal and personal concerns about automation to predict support for restrictive immigration policy, we performed a regression with both societal and personal concerns as simultaneous predictors (for separate analyses of personal and societal automation concerns, see the Supplemental Material). The result of this analysis revealed that societal concerns were a significant predictor of support for more restrictive immigration policy (β = 0.19, SE = 0.09, p = .036) but that personal concerns were not a significant predictor (β = 0.04, SE = 0.09, p = .617). This suggests that any effect of personal concerns about automation on immigration attitudes was captured by concerns about the societal impact of automation.

Table 4.

Correlations Among Automation Concerns, Intergroup Threat, and Immigration-Policy Attitudes in Study 2c

Variable	1	2	3	4
1. Societal automation concerns	—
2. Personal automation concerns	.52***	—
3. Realistic threat	.14**	.04	—
4. Symbolic threat	.15**	.11*	.81***	—
5. Immigration policy	.14**	.09^†	.78***	.76***

Note: Higher values for immigration policy indicate support for more restrictive immigration policy.

†

p < .10. *p < .05. **p < .01. ***p < .001.

Test for mediation

Next, as specified in our preregistration plan, we examined whether realistic and symbolic threats mediated the effect of automation concerns on support for restrictive immigration policy. Multiple mediation analysis was conducted to test whether realistic and symbolic threats mediated the effect of both personal and societal automation concerns on support for restrictive immigration policy. Direct and indirect effects for both personal and societal threats were estimated simultaneously using structural equation modeling. (Detailed results for tests of all components of the indirect effects can be found in the Supplemental Material.) As shown in Table 5, the specific indirect effects of realistic and symbolic threats were both significant for societal automation concerns, but the indirect effects of realistic and symbolic threats were not significant for personal automation concerns. These results further suggest that concerns about automation’s societal impact (more than personal impact) are linked to greater perceptions of intergroup threat and support for restrictive immigration policy. In other words, societal concerns (about Americans as a group), more than personal concerns, were primarily associated with anti-immigrant attitudes.

Table 5.

Total, Direct, and Indirect Effects From a Multiple Mediation Model With Realistic and Symbolic Threat Mediating the Association Between Societal and Personal Automation Concerns and Support for Restrictive Immigration Policy (Study 2c)

Effect and variable	β
Total effect of automation concerns
Personal concerns	0.04 (0.09)
Societal concerns	0.19* (0.09)
Direct effect of automation concerns
Personal concerns	0.06 (0.05)
Societal concerns	−0.00 (0.05)
Indirect effect for personal automation concerns
Realistic threat	−0.03 (0.04)
Symbolic threat	0.02 (0.03)
Indirect effect for societal automation concerns
Realistic threat	0.12** (0.05)
Symbolic threat	0.07* (0.03)

Note: Standard errors are given in parentheses. Higher values for immigration policy indicate support for more restrictive immigration policy.

p < .05. **p < .01.

These findings are consistent with those reported in prior literature on intergroup relations indicating that group threat, but not individual threat, is linked to negative intergroup attitudes (Maoz & McCauley, 2005; Stephan et al., 2009). As a specific example, Huddy, Feldman, Taber, and Lahav (2005) found that perceived group threat regarding terrorist attacks predicts support for aggressive military actions toward terrorists, whereas personal feelings of anxiety toward terrorist attacks does not. Similarly, our findings suggest that the perception of automation as a general societal threat, rather than an individual threat, is primarily associated with negative attitudes toward immigrants.

Study 3

In Study 3, we employed an experimental manipulation of automation concerns to examine causality and investigated whether variation in high versus low susceptibility to automation affected the association between automation concerns and anti-immigrant sentiment.

Method

In an initial study, we attempted to manipulate participants’ vulnerability to automation, but the data were uninterpretable because of a methodological complication (for details, see the Supplemental Material).

Participants

A G*Power analysis revealed that we would have 99% power to detect a medium-size effect (f = 0.25), with alpha set to .05, if we recruited a total of at least 348 participants. We aimed to exceed this number to allow for exclusions based on attention checks and employment criteria (for exclusion criteria, see the Supplemental Material). Given this goal, our final sample was 445 participants (age: M = 35.55 years, SD = 10.47; 45% male, 54% female; 76% White, 13% Black, 7% Asian, 4% other).

Procedure

Participants first read that they would be completing a study investigating people’s attitudes about real issues related to employment today. Each participant was then randomly assigned to one of three conditions. In the control condition, participants answered questions about their current employment and read about research conducted by the Bureau of Labor Statistics demonstrating the industries in the United States expected to experience the biggest losses in employment over the next few years.

In the two automation-salience conditions, participants read a brief description about research done by University of Oxford researchers on the susceptibility of today’s jobs to automation. Participants were then told that the researchers had created an algorithm that could calculate the probability that any occupation would be automated over the next few years. They were then redirected to a different website to obtain their probability score. Once redirected, participants entered information about their employment and received feedback that their job had a high probability of being automated (85%) or a low probability of being automated (4%); this feedback constituted the manipulations for the high-probability-automation and low-probability-automation conditions, respectively. The full text for both conditions is available in the Supplemental Material. In both conditions, participants answered three attention checks about the text that they had just read.

Measures

Group threat

Perceived group threat was measured using the same realistic-threat (α = .91) and symbolic-threat (α = .82) subscales as in Study 2. The two scales were significantly correlated (r = .78, p = .004).

Immigration-policy attitudes

Attitudes toward restrictive immigration policy were assessed using the same items as in Study 2c (α = .79).

Results

A manipulation check included at the end of the study measured how threatened participants felt by automation, and analyses (described in detail in the Supplemental Material) suggest that our manipulation was effective in eliciting feelings of vulnerability to automation. (Descriptive statistics and zero-order correlations between key variables and demographic statistics are presented in the Supplemental Material.)

Main analyses

Support for restrictive immigration policies

There were significant differences in restrictive immigration policy across conditions, F(2, 442) = 4.42, MSE = 1.54, p = .013, η² = .02. Two planned contrasts tested the effect of level of personal automation threat (high automation probability vs. low automation probability) and the overall effect of exposure to automation on support for restrictive immigration policy (high and low automation probability vs. control). As shown in Figure 2, participants in the high-probability condition (M = 3.72, SD = 1.52) did not significantly differ in their support for restrictive immigration policy from those in the low-probability condition (M = 3.68, SD = 1.59), F(1, 442) = 0.04, p = .848, η² = .00. In addition, participants exposed to information about the impact of automation supported restrictive immigration policy more than participants in the control condition did (M = 3.24, SD = 1.50), F(1, 442) = 8.76, p = .003, η² = .02.

Fig. 2.

Mean support for restrictive immigration policy as a function of condition. Error bars represent 95% confidence intervals (Study 3).

Group threat

Although realistic threat was not significantly different across condition, F(2, 442) = 2.22, MSE = 1.38, p = .110, η² = .01, the same two planned contrasts described above were constructed for both components of group threat. Participants in the high-probability (M = 3.43, SD = 1.24) and low-probability (M = 3.55, SD = 1.48) conditions did not significantly differ in perceptions of realistic threat, F(1, 442) = 0.61, p = .434, η² = .00. Participants in both the high-probability and low-probability conditions perceived immigrants as more of a realistic threat compared with participants in the control condition (M = 3.21, SD = 1.42), F(1, 442) = 3.89, p = .049, η² = .01 (see Fig. 3).

Fig. 3.

Mean rating of realistic threat as a function of condition. Error bars represent 95% confidence intervals (Study 3).

Symbolic threat was significantly different across conditions, F(2, 442) = 4.26, MSE = 1.16, p = .015, η² = .01. As shown in Figure 4, planned contrasts revealed that participants in the high-probability condition (M = 3.72, SD = 1.11) did not significantly differ in their perceptions of symbolic threat from participants in the low-probability condition (M = 3.77, SD = 1.19), F(1, 442) = 0.14, p = .708, η² = .00. Participants in both the high-probability and low-probability conditions perceived immigrants as more of a symbolic threat than did participants in the control condition (M = 3.40, SD = 1.18), F(1, 442) = 8.43, p = .004, η² = .02.

Fig. 4.

Mean rating of symbolic threat as a function of condition. Error bars represent 95% confidence intervals (Study 3).

Test for mediation

Finally, multiple mediation analyses were conducted to examine whether realistic and symbolic threats mediated the effect of the automation-salience conditions on support for restrictive immigration policy (see Fig. 5; detailed results for tests of all components of the indirect effects can be found in the Supplemental Material). In these analyses, we collapsed across the high-probability and low-probability conditions to compare exposure to automation with the control condition. Results indicated that the individual indirect effect was not significant for realistic threat (β = 0.07, SE = 0.04, 95% CI = [−0.001, 0.159]) but was significant for symbolic threat (β = 0.13, SE = 0.05, 95% CI = [0.040, 0.229]). These results demonstrate that simple exposure to information about automation’s effect on jobs can increase perceptions of intergroup threat, which is linked to increasing support for restrictive immigration policies. This study is consistent with findings from Study 2c and suggests that general exposure to automation rather than personal vulnerability to automation may have been driving these effects.

Fig. 5.

Parameter estimates for the mediation model testing the effect of automation salience on support for restrictive immigration policy, as mediated by both symbolic and realistic threat (Study 3). Asterisks indicate significant paths (^†p < .10, **p < .01, ***p < .001). On the path from automation-salience condition to immigration-policy support, the value below the arrow (c) shows the total effect, and the value above the arrow (c′) shows the direct effect after controlling for the mediators.

Study 4

In Study 4, we moved away from a personally relevant context to test the effects of exposing people to automation information in a personally unrelated context.

Method

Participants

Given the effect size of overall automation exposure in Study 3, a power analysis using G*Power indicated that we would have 99% power to detect an effect corresponding to a Cohen’s d of 0.30, with alpha set to .05, if we recruited a total of at least 820 participants. We aimed to exceed this number to allow for potential exclusions based on attention checks (for exclusion criteria, see the Supplemental Material), and the final sample consisted of 824 participants (age: M = 38.13 years, SD = 12.44; 61% female; 81% White, 10% Black, 4% Asian, 5% other).

Procedure

Participants first completed a demographics questionnaire, and each participant was then randomly assigned to one of two conditions in which he or she read about a company that has to lay off about 500 employees. In the control condition, participants read that the company needs to lay off workers because it wants to cut costs by restructuring and by downsizing certain departments. In the automation-threat condition, participants read that the company needs to lay off workers because it wants to cut costs by adopting new technology that will automate many tasks currently done by human workers. The full text for both conditions is available in the Supplemental Material. In both conditions, participants answered three attention-check questions about the text they had just read.

After this threat manipulation, participants were then asked to provide their opinions about (a) what kind of workers the company should lay off and (b) personal automation threat. Then they completed exploratory measures assessing personal policy preferences (described in the Supplemental Material).

Company layoff decisions

Participants were asked to indicate what percentage of different kinds of workers should be laid off by the company. Participants made layoff decisions about four categories of workers (presented in random order): our category of interest—workers who immigrated to the United States from other countries versus workers from the United States—as well as three additional categories of workers (male vs. female, older vs. younger, less educated vs. more educated). We included these additional three categories to disguise our category of interest (immigrants vs. nonimmigrants). Participants were told that the total percentage of people laid off in each group should add up to 100%. Participants were asked what percentage of workers should be laid off and to indicate their response on a sliding scale ranging from 0% to 100% for each type of worker.

Personal automation threat

After the layoff-decisions measure and exploratory measure, participants completed a measure of feelings of threat about automation consisting of two items (r = .75, p < .001): “How concerned are you that automation (the use of technology and robots to perform labor) could make your job obsolete?” and “How threatened do you personally feel by the prospect of automation?” (1 = not at all, 7 = very much). This score did not differ significantly between the automation condition (M = 2.84, SD = 1.66) and the restructuring condition (M = 3.00, SD = 1.78), t(821) = 1.38, p = .167, d = 0.10, suggesting that, consistent with the results of Study 3, exposure to automation acts independently of personal vulnerability to automation in affecting attitudes toward immigrants.

Results

As predicted, participants in the automation condition decided to lay off a greater percentage of immigrants (M = 55.97, SD = 18.42) than did participants in the restructuring condition (M = 53.36, SD = 17.29), t(821) = 2.09, p = .037, d = 0.14. Looking across the other categories of workers that participants were asked to make layoff decisions about, we found no significant differences between conditions regarding how many female workers, t(818) = 1.91, p = .056, d = 0.13; less educated workers, t(821) = 1.18, p = .236, d = 0.08; or older workers, t(820) = 0.62, p = .534, d = 0.04, should be laid off. These results suggest that in a personally unrelated context, exposure to automation can produce discriminatory behavior toward immigrants by influencing decisions about layoffs.

Discussion

Twelve studies showed that exposure to automation is associated with and predictive of anti-immigrant sentiment. Archival analyses revealed that this pattern has persisted over the past 30 years in both the United States and Europe (Studies 1a–1g). Targeted correlational studies showed that concerns about automation’s impact on society are also associated with endorsing restrictive immigration policy, an effect mediated by stronger perceptions of immigrants as threats to both material resources and cultural values (Studies 2a and 2b) and distinct from general feelings of uncertainty (Study 2c). An experiment revealed that mere exposure to information about automation’s effect on unemployment increased support for restrictive immigration policy, an effect again triggered by perceptions of intergroup threat (Study 3). A final experiment showed that automation exposure increased discriminatory decision-making against immigrants (Study 4).

Our findings suggest that automation increases support for restrictive immigration policies through at least two pathways. Awareness of automation appears to increase the perception of immigrants as both threats to existing resources and threats to present cultural values. The findings on realistic threat suggest that exposure to automation can activate concerns about competition over scarce resources, spurring anti-immigrant sentiment. In addition, symbolic threat also consistently mediated the relationship between automation and support for restrictive immigration policy across our studies, even after we controlled for perceptions of realistic threat. This suggests that beyond the economic impact of automation, people may also construe it as a threat to dominant cultural values. The rise of automation, and the very idea of advanced technology, may make people feel that their current way of life is changing and that society’s shared values are eroding. These findings on symbolic threat may also speak to why the present work shows that automation has unique effects that are different from other employment threats—sophisticated technology capable of replacing humans represents not only a threat to resources but a fundamental change to life as we know it, affecting our culture and values. These feelings may then intensify the perception that immigrants are threatening because of perceived group differences in values and beliefs. Consistent with this, our findings also suggest that the effects of automation exposure do not depend on people’s personal perceived vulnerability to automation. Instead, our effects seem driven more by people’s perceptions of automation as a broad societal threat to group values, status, and resources.

One possible question regarding this work concerns the causal relationship between automation and attitudes toward immigrants. We have argued that concerns about automation can increase feelings of intergroup threat toward immigrants. However, exposure to (and negative feelings toward) immigrants might exacerbate people’s concerns about automation. Experimentally manipulating automation concerns in Studies 3 and 4 at least provided positive evidence for the hypothesis that exposure to automation causes anti-immigrant sentiment, although other studies may prove the other directional pattern to be true as well.

This work contributes to the burgeoning field of the psychology of technology. Much work in this field has focused on the human–technology relationship itself and examined questions related to people’s attitudes toward and adoption of new technologies such as smartphones, algorithms, or self-driving cars. The present body of work is one of the first to focus on the societal implications of new technology for intergroup relations.

Overall, our work identifies a previously unestablished link between technological development and social attitudes: that automation exposure can hamper intergroup relations. In addition, this work suggests that automation’s disruptive effects may further fuel divisive politics. Understanding how people respond to the economic and cultural changes precipitated by automation is important because public reaction to automation can shape future policy and messaging regarding the adoption and development of new technologies.

Supplemental Material

Gamez-Djokic_OpenPracticesDisclosure_rev – Supplemental material for Concerns About Automation and Negative Sentiment Toward Immigration

Supplemental material, Gamez-Djokic_OpenPracticesDisclosure_rev for Concerns About Automation and Negative Sentiment Toward Immigration by Monica Gamez-Djokic and Adam Waytz in Psychological Science

Supplemental Material

Gamez-Djokic_Supplemental_Material_rev – Supplemental material for Concerns About Automation and Negative Sentiment Toward Immigration

Supplemental material, Gamez-Djokic_Supplemental_Material_rev for Concerns About Automation and Negative Sentiment Toward Immigration by Monica Gamez-Djokic and Adam Waytz in Psychological Science

Footnotes

Acknowledgements

We thank Peter A. Hall, Frank Kachanoff, Nour Kteily, Filippo Mezzanotti, and Kimberly Rios for their helpful feedback on the manuscript for this article.

Transparency

Action Editor: Leaf Van Boven

Editor: D. Stephen Lindsay

Author Contributions

M. Gamez-Djokic and A. Waytz share first authorship; the authors are displayed alphabetically. Both authors developed the study concept, drafted the manuscript, and interpreted the data. M. Gamez-Djokic analyzed the data. Both authors approved the final manuscript for submission.

ORCID iDs

Monica Gamez-Djokic

Adam Waytz

Supplemental Material

Additional supporting information can be found at

Notes

References

Andrighetto

Vezzali

Bergamini

Nadi

Giovannini

(2016). Inside the earthquake: Perceived disaster exposure and helping intentions among Italian and immigrant victims of the 2012 Italian earthquakes. Group Processes & Intergroup Relations, 19, 753–768.

Anelli

Colantone

Stanig

(2019). We were the robots: Automation and voting behavior in Western Europe (IZA Discussion Paper No. 12485). Retrieved from http://ftp.iza.org/dp12485.pdf

Applebaum

H. A.

(1998). The American work ethic and the changing work force: An historical perspective. Westport, CT: Greenwood.

Arntz

Gregory

Zierahn

(2017). Revisiting the risk of automation. Economics Letters, 159, 157–160.

Bonnefon

J. F.

Shariff

Rahwan

(2016). The social dilemma of autonomous vehicles. Science, 352, 1573–1576.

Brambilla

Butz

D. A.

(2013). Intergroup threat and outgroup attitudes: Macro-level symbolic threat increases prejudice against gay men. Social Psychology, 44, 311–319.

Brown

(1998). Introduction: Labor, race, and class conflict in 1919. In Bruchey

(Ed.), Racial conflict and violence in the labor market (pp. 3–17). New York, NY: Routledge.

Bullock

J. G.

Green

D. P.

S. E.

(2010). Yes, but what’s the mechanism? (don’t expect an easy answer). Journal of Personality and Social Psychology, 98, 550–558.

Butz

D. A.

Yogeeswaran

(2011). A new threat in the air: Macroeconomic threat increases prejudice against Asian Americans. Journal of Experimental Social Psychology, 47, 22–27.

10.

Craig

M. A.

Richeson

J. A.

(2014). On the precipice of a “majority-minority” America: Perceived status threat from the racial demographic shift affects White Americans’ political ideology. Psychological Science, 25, 1189–1197.

11.

Dal Bó

Finan

Folke

Persson

Rickne

(2018). Economic losers and political winners: Sweden’s radical right (NBER Working Paper No. 115534). Cambridge, MA: National Bureau of Economic Research.

12.

Diaz

Saenz

D. S.

Kwan

V. S.

(2011). Economic dynamics and changes in attitudes toward undocumented Mexican immigrants in Arizona. Analyses of Social Issues and Public Policy, 11, 300–313.

13.

Esses

V. M.

Dovidio

J. F.

Jackson

L. M.

Armstrong

T. L.

(2001). The immigration dilemma: The role of perceived group competition, ethnic prejudice, and national identity. Journal of Social Issues, 57, 389–412.

14.

Esses

V. M.

Jackson

L. M.

Armstrong

T. L.

(1998). Intergroup competition and attitudes toward immigrants and immigration: An instrumental model of group conflict. Journal of Social Issues, 54, 699–724.

15.

Faul

Erdfelder

Lang

A.-G.

Buchner

(2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39, 175–191.

16.

Fiedler

Schott

Meiser

(2011). What mediation analysis can (not) do. Journal of Experimental Social Psychology, 47, 1231–1236.

17.

Frey

C. B.

Berger

Chen

(2018). Political machinery: Did robots swing the 2016 US presidential election? Oxford Review of Economic Policy, 34, 418–442.

18.

Frey

C. B.

Osborne

M. A.

(2017). The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change, 114, 254–280.

19.

Huddy

Feldman

Taber

Lahav

(2005). Threat, anxiety, and support of antiterrorism policies. American Journal of Political Science, 49, 593–608.

20.

Katz

Hass

R. G.

(1988). Racial ambivalence and American value conflict: Correlational and priming studies of dual cognitive structures. Journal of Personality and Social Psychology, 55, 893–905.

21.

Krosch

A. R.

Amodio

D. M.

(2014). Economic scarcity alters the perception of race. Proceedings of the National Academy of Sciences, USA, 111, 9079–9084.

22.

Maoz

McCauley

(2005). Psychological correlates of support for compromise: A polling study of Jewish–Israeli attitudes toward solutions to the Israeli–Palestinian conflict. Political Psychology, 26, 791–808.

23.

Morrison

K. R.

Plaut

V. C.

Ybarra

(2010). Predicting whether multiculturalism positively or negatively influences White Americans’ intergroup attitudes: The role of ethnic identification. Personality and Social Psychology Bulletin, 36, 1648–1661.

24.

Mutz

D. C.

(2018). Status threat, not economic hardship, explains the 2016 presidential vote. Proceedings of the National Academy of Sciences, USA, 115, E4330–E4339.

25.

Olzak

(1989). Labor unrest, immigration, and ethnic conflict in urban America, 1880–1914. American Journal of Sociology, 94, 1303–1333.

26.

Preacher

K. J.

Hayes

A. F.

(2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods, 40, 879–891.

27.

Rios

Sosa

Osborn

(2018). An experimental approach to Intergroup Threat Theory: Manipulations, moderators, and consequences of realistic vs. symbolic threat. European Review of Social Psychology, 29, 212–255.

28.

Schönbrodt

F. D.

Perugini

(2013). At what sample size do correlations stabilize? Journal of Research in Personality, 47, 609–612.

29.

Sherif

(1966). In common predicament: Social psychology of intergroup conflict and cooperation. Boston, MA: Houghton Mifflin.

30.

Stephan

W. G.

Ybarra

Bachman

(1999). Prejudice toward immigrants. Journal of Applied Social Psychology, 29, 2221–2237.

31.

Stephan

W. G.

Ybarra

Morrison

K. R.

(2009). Intergroup threat theory. In Nelson

T. D.

(Ed.), Handbook of prejudice, stereotyping, and discrimination (pp. 43–59). Mahwah, NJ: Erlbaum.

32.

Weber

(2013). The Protestant ethic and the spirit of capitalism. Abingdon, England: Routledge. (Original work published 1905)

33.

Willer

Feinberg

Wetts

(2016). Threats to racial status promote Tea Party support among white Americans. SSRN. Retrieved from https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2770186

34.

Yogeeswaran

Złotowski

Livingstone

Bartneck

Sumioka

Ishiguro

(2016). The interactive effects of robot anthropomorphism and robot ability on perceived threat and support for robotics research. Journal of Human-Robot Interaction, 5(2), 29–47.

35.

Zhen

J. I.

Mayer

(2017). The losers of digitalization: A reservoir of votes for the far-right? In Rhee

(Ed.), The political economy of inequality and social integration (pp. 89–109). Seoul: Korea Economic Daily.

36.

Złotowski

Yogeeswaran

Bartneck

(2017). Can we control it? Autonomous robots threaten human identity, uniqueness, safety, and resources. International Journal of Human-Computer Studies, 100, 48–54.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.23 MB

1.25 MB