Abstract
Technology is expected to displace many workers in the future. The public generally supports government assistance for workers viewed as less responsible for their unemployment; thus, we ask whether individuals who lose their jobs to technology are perceived as less at fault and more deserving of government benefits, compared to those who lose their jobs to other workers. We conducted a survey experiment on a nationally representative sample in the United States, randomizing whether a hypothetical worker was replaced by technology, a foreign worker, or domestic worker, and asked questions about fault perception and support for unemployment benefits. We find that workers who lose jobs to technology (or foreign workers) are viewed as less at fault than those who lose jobs to domestic workers, and that fault attribution mediated support for unemployment benefits.
Millions of workers across the globe are at risk of being replaced by technology in the coming years. In the United States, nearly 50% of jobs could be automated in the relatively near future according to some estimates (Frey and Osborne, 2017). Thus far there has been a limited public or government response to this problem, but whether the public views those unemployed by technology as relatively blameless for their unemployment compared to being displaced by other factors of production is likely to be a major factor affecting public support for assistance programs for these workers (Petersen, 2012).
Technological Unemployment, Fault, and Support for Benefits
Research suggests that concerns about one’s own exposure to technology in the workplace can shape political behavior and policy attitudes (Gingrich, 2019; Heinrich and Witko, 2021a, 2021b; Im et al., 2019). How people respond to others’ joblessness due to technology is also an important question; if technological unemployment becomes more common and politically salient (e.g., Andrew Yang’s presidential nomination bid), public judgments about whether those unemployed by technology are blameless victims worthy of government support will be important to policy responses.
Petersen (2012) argues that individuals make quick judgments about whether the unemployed are jobless due to their own fault or are the victims of bad luck, which determine attitudes toward government support. Most subsequent research has provided vignettes or cues with information about individuals’ characteristics (laziness, race, ethnicity, etc.) (Aarøe and Petersen, 2014; Applebaum, 2001; DeSante, 2013) and examined how these shape perceptions of fault and deservingness. There has been little analysis of whether unemployment due to technology versus other factors of production lead to differing judgments of fault and deservingness.
People who were working should generally be viewed as more deserving of benefits than the non-working (Petersen, 2012). However, considering those who lose their job, which factor of production replaces them may also shape fault attribution. Because American workers compete on a somewhat level playing field, an American worker replaced by another American may be viewed as somewhat at fault for their predicament. In contrast, if an American worker is replaced by a foreign worker, they will likely be viewed as less at fault given that governments (and not workers) make immigration laws and enter into trade deals, and many Americans believe these policies lead to job losses. 1 It is pretty clear that workers cannot compete with many types of technology and there is a preference for workers to be replaced by other workers rather than technology (Granulo et al., 2019). Thus, we expect that workers replaced by technology would be viewed as less at fault than those replaced by other domestic workers.
To examine this question, we use a survey experiment presenting a vignette with a hypothetical unemployed worker and randomizing whether the source of job displacement is a new technology, a foreign worker, or a domestic worker. We can thus calibrate the effect of technological unemployment on attitudes to these other two sources of job loss reflecting a relatively lower and higher “fault” scenario.
Experimental Design
We fielded our survey on the YouGov platform in April 2017 on a nationally representative sample of adult citizens in the United States (n = 999). Respondents were presented with the following, in which the source of displacement was randomly assigned: “Suppose that an American worker lost his/her job because their employer determined that [a domestic worker (n = 342), foreign worker (n = 333), technology (n = 324)] could do the job more efficiently.” Respondents were then asked whether they “support or oppose this worker receiving unemployment benefits from the federal government?” on a 7-point Likert scale (1 = “Strongly oppose”; 7 = “Strongly support”). Next, respondents were asked to indicate whether they “agree or disagree it is the worker’s own fault that he/she got fired?”, also on a 7-point Likert scale (1 = “Strongly disagree”, 7 = “Strongly agree″). We also use demographics to increase the efficiency of the treatment effect estimates and in our causal mediation analysis. The question wordings and any transformations we applied are described in the Online Appendix.
Results
Figure 1 shows the proportion of people in each treatment condition selecting each response option for both outcomes. The distribution for each question was quite skewed, which may have subjected our treatment effects to ceiling and/or floor effects. In the upper panel, only a few respondents agreed that the worker was at fault, while in the lower panel, the two most frequent responses indicated the highest level of support for assistance. Proportion of responses by treatment condition and outcome options. Each panel gives the (weighted) proportion of respondents selecting each outcome option for the Fault (upper panel) and Support (lower) questions by treatment status (gray scale). 95% confidence intervals obtained from 1000 non-parametric bootstrap draws.
Coefficient estimates for main models (ordered probit).
Each number gives the mean estimate, the range below the 95% confidence interval. All 1000 observations were used, and estimates are pooled across all imputations. Worker’s Fault is measured by respondent choosing on a 7-point Likert scale (1 = Strongly disagree, 7 = Strongly agree) that the worker was at fault for losing their job. Support Benefits is measured by asking whether respondents support or oppose a hypothetical worker receiving unemployment assistant (1 = Strongly oppose; 7 = Strongly support).
To better demonstrate these results, we plot differences in the predicted probabilities for each outcome level across our three experimental conditions in Figure 2. Interpreting the results from our ordered probit models this way allows us to observe the margin at which any changes occur. We see that respondents were more likely to indicate that they “disagree” and “strongly disagree” that the worker was at fault in the automation and foreign worker conditions compared to the control condition, reflecting an intensely felt attribution, while respondents were much more likely to have a neutral view of fault for the worker replaced by another domestic worker. In contrast, there were no significant changes in support for government benefits and overall high levels of support for government benefits for unemployed workers. Does this mean fault attribution is irrelevant for support for benefits? We explore this more below. Treatment effects for support for government assistance (lower panel) and personal fault attribution (upper). The response options are given on the x-axis; the y-axis gives the change in probability for each response option (shown on the x-axis) when the replacement reason changes from domestic to domestic (black dot/line), foreign (light gray), or technology (dark gray). Lines give 95% confidence intervals.
To investigate whether fault attribution mediated the effect of the source of displacement on support for benefits, we follow the approach of Imai et al. (2011). The quantity of interest is the so-called (natural) indirect mediation effect. Technically, it is defined as the change in choosing one level of support for unemployment benefits when the value of the mediator (fault attribution) changes from its realization under the control (Domestic) to the treatment (Foreign, Technology) condition while holding the “direct” effect of the treatment constant.
To evaluate this causal chain, we needed to estimate a model of policy support that included levels of fault attribution. The third column in Table 1 shows that fault attribution is a significant determinant of support for unemployment benefits. Since less fault attribution leads to more support for benefits for workers and since foreign and technological sources of displacement lead to less fault attribution, we should see indirect effects. To determine the magnitude of these indirect effects and the margins at which they occur, we turn to the simulation approach recommended by Imai et al. (2011). Details of this approach can be found in the online appendix.
Figure 3 reports the estimates for indirect effects comparing a domestic worker to a foreign worker (left panel) and to new technology (right) as a source of displacement. The scale of the effects can be compared to the overall proportions of responses, as shown in Figure 1. We see that displacement by a foreign worker or new technology (compared to a domestic worker) increases support for benefits at the most enthusiastic level, among those who “strongly agree” the worker should receive unemployment assistance. Therefore, we conclude that the earlier absence of a (total) effect on policy support was an artifact of a significant positive indirect effect (at the highest level of support) and several individually insignificant direct effects (i.e., when the mediator is held constant) across all levels of support (see Figure A1 in the Online Appendix). Thus, the lower fault attribution associated with the foreign worker or technology replacement does matter for support for benefits, albeit in a somewhat complex manner and not in a linear way. Indirect effects for support for government assistance based on Technology (right panel) and Foreign (left) treatments compared to Domestic.
Finally, we evaluated several theoretically informed moderators of these causal effects, such as family income and political ideology, but we failed to observe a significant source of moderation in any of these variables (see Figures A2–A5 in the Online Appendix for details and results).
Discussion
Our results show that attitudes toward those displaced by technology are similar to those displaced by foreign workers in terms of fault and deservingness of government benefits. This potentially has important implications for policy in the future. Technological unemployment has not been as politically salient as jobs being “stolen” by foreign workers, but this will perhaps change if it becomes more common. Our results suggest that there is latent support for programs to assist those who lose jobs to technology.
On the other hand, even though mass publics recognize the threat that technology poses to jobs (Heinrich and Witko, 2021a), outside of the lab people exposed to technology often blame other factors for their weakened labor market position (Wu, 2021), and it is not clear that the public connects the abstract threat of technology to jobs to the actual unemployment or precarity of certain groups or individuals. Thus, future research should examine how people make judgments about the macroeconomic reasons why they and others are facing a weakened labor market position.
Supplemental Material
sj-pdf-1-rap-10.1177_20531680221093440 – Supplemental Material for Public support for assistance for workers displaced by technology
Supplemental Material, sj-pdf-1-rap-10.1177_20531680221093440 for Public support for assistance for workers displaced by technology by Seth Werfel, Christopher Witko and Tobias Heinrich in Research & Politics
Footnotes
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by a grant from the Laboratory for the Study of American Values at Stanford University.
Correction (June 2025):
Supplemental Material
Note
References
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.
