Labor Market Concerns and Attitudes Towards Immigrants

Transcription

1 Labor Market Concerns and Attitudes Towards Immigrants Ingar Haaland Christopher Roth July 19, 2017 Abstract We provide causal evidence on how people s beliefs about the labor market consequences of immigration affect their attitudes toward immigration. Using a representative sample of the US population, we elicit beliefs about how the Mariel boatlift which caused a large influx of low-skilled Cuban immigrants to Miami, Florida affected the Miami labor market. Then, drawing on Card (1990), half of the subjects receive information about the actual labor market impact of the Mariel boatlift. We find that respondents who receive the information display more optimistic views on the labor market impact of immigrants and develop more positive attitudes towards immigration. Respondents who receive the information are also more likely to sign a real online petition in favor of increasing the annual cap on non-agricultural guest workers to the US. Finally, treatment effects persist one week later in an obfuscated follow-up study. We thank Ned Augenblick, Roland Bénabou, Alexander Cappelen, David Card, Rachel Cassidy, Jack Citrin, Jon de Quidt, Dirk Engelmann, Armin Falk, Claudio Ferraz, Alexis Grigorieff, Jens Hainmueller, Lukas Hensel, Simon Jaeger, Peter Neary, Muriel Niederle, Simon Quinn, Matthew Rabin, Gautam Rao, Kenneth Scheve, Sebastian Schweighofer-Kodritsch, Daniela Scur, Stefanie Stantcheva, Adam Szeidl, Bertil Tungodden, Jonas Tungodden, and David Yang for valuable comments. Moreover, we would like to thank seminar audiences in Bergen, Berlin, and Oxford for valuable comments. The experiment is registered in the AEA RCT Registry as trial IRB approval was obtained from the University of Oxford. The usual disclaimer applies. NHH Norwegian School of Economics and The Choice Lab ( Ingar.Haaland@nhh.no) University of Oxford and CSAE ( Christopher.Roth@economics.ox.ac.uk) 1

2 JEL Codes: C91, D83, F22, J15 Keywords: immigration, belief updating, labor market concerns 1 Introduction Anti-immigrant sentiment in the Western world is high (Hainmueller and Hiscox, 2010; Card et al., 2012), and Americans consistently rank immigration as one of the country s most important problems (Gallup, 2016). The negative sentiment towards immigrants has given rise to the success of right-wing candidates with a nationalistic stance on immigration policy in both the US and Europe (Halla et al., 2016). The relevance of economic concerns in forming attitudes about immigration is controversial, and the empirical evidence is mixed (Scheve and Slaughter, 2001; Mayda, 2006; Card et al., 2012; Hainmueller and Hopkins, 2014; Gerber et al., 2017). Current evidence on the relevance of economic concerns faces two shortcomings: (i) it is reliant on correlational evidence and (ii) it mostly uses data on objective measures of labor market competition with immigrants rather than subjective beliefs about the economic impact of immigrants. In this paper, we use a pre-registered randomized experiment to shed light on the importance of economic concerns in forming attitudes toward immigration. In particular, we provide causal evidence on how people s beliefs about the labor market effects of immigration affect their attitudes toward immigration. To do this, we elicit people s beliefs about the labor market effects of the Mariel boatlift, which was an unexpected mass immigration of Cubans to the United States that increased the size of the workforce in Miami by 8 percent almost at once. In a between-subjects design, we then randomly allocate subjects either to a treatment group or a control group. In the treatment group, we provide subjects 2

3 with information about the results from a seminal study about the labor market consequences of the Mariel boatlift. The study finds that the mass immigration of Cubans to Miami had virtually no adverse effects on local labor market outcomes (Card, 1990). 4 Following Hainmueller et al. (2015), we then measure people s self-reported attitudes toward increasing the number of both low-skilled and high-skilled immigrants to the United States. In addition, we give our subjects the opportunity to sign two real online petitions suggesting changes to the annual cap on non-agricultural guest workers (the H-2B visa program). The first petition suggests to increase the annual cap on H2-B visas and the other petition suggests to decrease the annual cap. We then ask our respondents a series of questions about the economic and non-economic impact of immigration to shed light on different mechanisms through which the treatments may affect attitudes. We also conduct an obfuscated follow-up study where subjects are not told about the relationship between the main study and the follow-up to investigate whether treatment effects persist over time and to measure treatment effects in a setting where concerns about differential experimenter demand effects across treatment arms are no longer an issue. Turning to results, we first present a few stylized facts on people s beliefs about the labor market impact of immigrants in the context of the Mariel boatlift. Most of our respondents believe that the inflow of Cuban immigrants significantly decreased wages and increased unemployment among low-skilled workers in Miami. Moreover, most of our subjects think that wages of and unemployment among high-skilled 4 The findings by Card (1990) are consistent with a large body of economic research on the labor market consequences of immigration (Card, 2005; Dustmann et al., 2005; Kugler and Yuksel, 2008; Card, 2012; Manacorda et al., 2012; Ottaviano and Peri, 2012), and the Mariel boatlift is known as the one historical event that has most shaped how economists view immigration (Clemens, 2017). However, the ultimate effects of immigration on labor market conditions are still being debated (Borjas, 2014; Card and Peri, 2016). 3

4 Americans in Miami were not affected by the inflow of Cuban immigrants. Next, we turn to the causal effect of labor market concerns on attitudes towards immigrants. Respondents who receive research information about the effects of the Mariel boatlift significantly change their views on the economic impact of the migration on American workers. They also significantly increase their support for increasing the number of low-skilled immigrants measured both with self-reports and real signatures of a petition advocating for increasing the number of H2B visas. These results also persist in an obfuscated follow-up one week after the main experiment where differential experimenter demand across treatment arms is of no concern. Our estimates allow us to define an upper bound of demand effects in this context as below 0.05 of a standard deviation or one third of the effect size in the main survey. Our key object of interest is how elastic policy preferences regarding immigration are to changes in economic concerns related to immigration. We find that a one standard change in perceptions about the economic effects of immigrants changes people s willingness to admit unskilled immigrants by about between 0.5 and 0.6 of a standard deviation. These effects stand in stark contrast to the previous literature which has argued that labor market concerns are not quantitatively important (Hainmueller and Hopkins, 2014; Card et al., 2012). We also use our rich data on beliefs about the effects of immigrants to shed light on mechanisms. We find evidence that people significantly update their beliefs about the effects both low-skilled and high-skilled immigrants have on wages and unemployment of most Americans. Surprisingly, we find that our respondents do not on average change their beliefs about the impact immigration will have on the wages and employment prospects of their own household. Our evidence thus highlights that economic concerns do not necessarily operate through beliefs how people s own household is affected by immigrants, but rather can operate through 4

5 people s beliefs about the effects immigration has on most Americans. Furthermore, we find no evidence that our information treatment shifted perceptions of whether immigrants constitute a burden for tax payers and how they affect American culture. Our evidence emphasizes the importance of a psychological mechanism related to in-group out-group bias (Brewer, 1979): that beliefs about how in-group members ( most Americans ) are economically affected by out-group members ( newly arriving immigrants ) are a quantitatively important determinant of attitudes towards immigrants. Overall, we extend the existing literature in several important ways. First, we collect novel data to capture people s mental models of how a labor market shock, an immigrant inflow, affects labor market conditions of incumbent workers. Thereby, we introduce novel data which seeks to measure people s mental models of the functioning of labor markets in the face of supply shocks. We thus contribute to a growing literature trying to measure economic agents mental models of the economy (Cappelen et al., 2017; Roth and Wohlfart, 2017; Haaland and Roth, 2017). Second, by randomly treating some subjects with information about the labor market consequences of the Mariel boatlift, we provide the first causal evidence on the role of people s beliefs about the labor market effects of immigrants on their attitudes toward immigrants. We also employ a behavioral measure that captures an important type of real-world political behavior by measuring people s willingness to sign an actual petition in favor of introducing a guest worker program to increase the legal flow of immigrants to the United States. Our evidence which highlights the importance of economic concerns for attitudes towards immigrants stands in contrast to the previous literature which mostly finds little support for an important role of labor market concerns (Hainmueller et al., 2015; Hainmueller 5

6 and Hopkins, 2014). 5 There are a few experiments which have examined the role of beliefs about the share of immigrants or the characteristics of immigrants in shaping attitudes toward immigration (Sides and Citrin, 2007; Grigorieff et al., 2016; Hopkins et al., 2016). These papers find overall muted responses of policy preferences to information about immigrants. We differ from these experiments by giving people research information about the causal effects of immigrants on labor markets rather than providing them with summary statistics about how many immigrants there are (Hopkins et al., 2016) or what their characteristics are (Grigorieff et al., 2016). Closely related to our work is also Facchini et al. (2016) who use a large sample in Japan and expose them to stories, e.g. that immigration can help alleviate a potential labor shortage in some professions. 6 Our results also add to the literature which studies determinants of attitudes towards immigrants (Citrin et al., 1997; Dustmann and Preston, 2001, 2006; Facchini et al., 2009; Hainmueller and Hiscox, 2010; Card et al., 2012; Bursztyn et al., 2017). This literature has mostly emphasized that non-economic concerns are particularly important determinants of attitudes towards immigrants (Hainmueller et al., 2015; Hainmueller and Hopkins, 2014). From a methodological point of view, we contribute to the literature on experimenter demand effects (List, 2007; Zizzo, 2010; de Quidt et al., 2017). We provide a generally feasible way of dealing with differential experimenter demand effects across treatment arms in online experiments by conducting obfuscated follow-up surveys in which the participants are not aware of the connection between the 5 There are some notable exceptions which show correlations in support of a relationship between labor market concerns and anti-immigrant attitudes (Scheve and Slaughter, 2001; Mayda, 2006; Gerber et al., 2017). 6 More generally, we add to the broader literature on how information affects people s policy preferences (Kuklinski et al., 2000; Gilens, 2001; Nyhan and Reifler, 2010; Lawrence and Sides, 2014; Kuziemko et al., 2015). 6

7 follow-up survey and the main survey in which the treatment is administered. Our estimates imply that under reasonable assumptions experimenter demand effects in our study are at most 0.05 of a standard deviation or less than one third of the original effect size in the main study. 2 Experimental Design 2.1 Main Experiment First, we ask subjects to complete a questionnaire on demographics, which includes questions on gender, age, income, race, geography, and political affiliation. Second, we elicit people s beliefs about the effects of the Mariel boatlift an unexpected mass immigration of Cubans to Miami, Florida, which increased the size of the workforce in Miami by 8 percent almost at once. We first describe the context of the Mariel boatlift to our respondents. We tell them that in 1980, Cuba s then President, Fidel Castro, suddenly announced that Cubans wishing to emigrate to the United States were free to do so and that this led to an unexpected mass immigration to Miami, Florida, where most of the Cuban immigrants arrived by boat. We also make it very clear to our respondents that the immigrant inflow of low-skilled workers was substantial. We explain to them that with the arrival of the new Cuban immigrants, Miami s workforce grew by 55,000, or 8 percent, almost at once. The new immigrants were mostly low-skilled, which meant that the low-skilled workforce increased by 20 percent. We then emphasize to our respondents that the Mariel boatlift constitutes an ideal setting to study the labor market impact of immigration. Specifically, we tell them that the large, unexpected addition of 55,000 new immigrants to the Miami workforce has allowed researchers to study the impact of immigration on the 7

8 labor market. To do so, the researchers studied wage and unemployment changes in Miami after the mass immigration relative to other US cities that, because of geographic distance, were not affected by the mass immigration of Cubans. We ask our subjects what they think happened to wages and unemployment in Miami relative to wages and unemployment in other comparable US cities that did not experience large inflows of low-skilled immigrants. Specifically, we ask our subjects what they think happened to the wages and the unemployment of (i) high-skilled workers and (ii) low-skilled workers. Since a large fraction of Americans tend to think that immigrants mostly hurt the economy by lowering wages for Americans (Gallup, 2005), we expect this treatment to alleviate people s labor market concerns about immigration. We inform subjects in the treatment group about the results from a seminal study about the labor market consequences of the Mariel boatlift (Card, 1990). 7 Specifically, we truthfully inform the subjects that this study found that the mass immigration of Cubans to Miami had virtually no adverse effects on the labor market. Subjects in the control group do not receive any information and go straight from the belief elicitation questions to the outcome questions. Following Hainmueller et al. (2015), we measure people s self-reported attitudes toward increasing or decreasing the number of low-skilled and high-skilled immigrants to the United States. We also vary whether the immigrants are highly familiar or not familiar with American values and traditions. Following the self-reported measures, all subjects get the opportunity to sign one 7 Card (1990) provides compelling evidence on the labor market consequences of the Mariel boatlift by conducting a difference-in-differences analysis. The findings by Card (1990) are consistent with a large body of economic research on the labor market consequences of immigration (Card, 2005; Dustmann et al., 2005; Kugler and Yuksel, 2008; Card, 2012; Manacorda et al., 2012; Ottaviano and Peri, 2012), and the Mariel boatlift is known as the one historical event that has most shaped how economists view immigration (Clemens, 2017). However, the ultimate effects of immigration on labor market conditions are still being debated (Borjas, 2014; Card and Peri, 2016). 8

9 of two real online petitions on the White House web-page, petitions.whitehouse.gov. We tell subjects that Congress is debating whether to change the annual cap on non-agricultural guest workers to the US the H-2B visa program and that they will be given the opportunity to sign one of two petitions related to this debate. The first petition suggests to increase the annual cap on H2-B visas and the other petition suggests to decrease the annual cap. For both petitions, we ask whether subjects intend to sign the petition. We then provide them with the link to the petition in case they intend to sign the petition. Since we provide the treatment and control groups with different links to the same petition, we can also observe the proportion of signatures in the treatment and the control group. Only participants with a link can see the petition until at least 150 people sign it, after which it becomes public. In practice, this means that petitions employed in our study never became public. Moreover, if the petition reaches 100,000 signatures in 30 days, it is entitled to get an official reply from the White House. We chose to create a very concrete petition with potentially concrete consequences, which attenuates concerns about its external validity. Then, we examine people s perceptions about the economic impact of increased immigration of low-skilled and high-skilled immigrants on their own household as well as for most Americans. We also examine people s beliefs about the effects of increased immigration of high-skilled and low-skilled immigrants on American culture and society. Half of our respondents are asked a series of questions about the impacts of high-skilled immigrants, while the other half of our respondents are asked about the impacts of low-skilled immigrants. Subsequently, we ask subjects whether they thought that the survey was politically biased. In addition, we check whether our subjects remember that most of the immigrants coming from Cuba were low-skilled. We also elicit our respondents beliefs on whether most of the Cuban immigrants coming to Florida 9

10 in 1980 were familiar with American values. For subjects in the treatment group, we also ask subjects whether they thought the information provided is (i) trustworthy, (ii) accurately reflects the labor market effects of the mass immigration of Cubans in 1980, and (iii) is relevant for assessing the costs of benefit of allowing more or less immigrants into the US today. At the end of the survey, we ask subjects some additional background questions on household size, education, employment status, zip code, place of birth, and parents place of birth. 2.2 Obfuscated follow-up About one week after the main study, we perform a follow-up study with the same subjects to see whether any possible treatment effects persist. Furthermore, we obfuscate the purpose of the follow-up study to measure treatment effects in a setting where concerns about experimenter demand effects are no longer an issue. To make the follow-up seem like an independent study, we first ask our subjects a series of demographic questions. To further obfuscate the purpose of the followup, we then ask a series of questions about government spending, taxation, and redistribution. Finally, we ask the main outcome questions on immigration. To minimize the risk that subjects realize the relationship between the two studies, we use different wordings for the questions on immigration in the follow-up compared to the main study. In addition, using different questions circumvents the potential confound that treatment effects persist because of people s taste for consistency in survey responses (Falk and Zimmermann, 2012). 10

11 2.3 Setting and sample size We recruit subjects using Research Now, a digital data collection agency in the US. 8 We recruited 3130 subjects that are representative of the US population in terms of gender, age, income, race, and regions. We collected data from 2075 respondents in the obfuscated follow-up survey one week after the main survey. This latter sample is also representative of the US population in terms of several observables. Moreover, there is not differential selection into the follow-up survey depending on treatment status. 2.4 Summary statistics and balance Table A.4 describes the most important characteristics of respondents from our sample. The median age in our sample is 45, and the median income in the sample is $62, percent of the sample are male, 46 percent of the sample are full-time employed. 75 percent of the sample consider themselves high-skilled. 78 percent of our respondents completed at least a college degree. Table A.2 highlights that the treatment and control group are balanced in terms of observables. In the follow-up we recruit 2075 respondents which are still representative in terms of the US population in terms of observables, such as income, region, age, and gender. As is evidenced by Table A.10, we also find no difference in response rate to the follow-up survey across the different treatment groups. Table A.3 shows that the treatment and control group are still balanced in the follow-up survey. 8 Research Now has been used in previous academic research by, e.g., Almås et al. (2016) and de Quidt et al. (2017). 11

12 3 Results 3.1 Stylized facts: Beliefs We start by providing a series of stylized facts about people s mental models of how immigration inflows affect the economy. In Figure 1 we show that ca. 70 percent of our respondents think that wages of low-skilled workers decreased in response to the inflow of unskilled Cuban migrants. Most respondents also think that unemployment among low-skilled workers increased in response to the immigrants inflow. Moreover, most of our subjects think that high-skilled workers were unaffected by the inflow of unskilled immigrants workers. In Table A.1 we provide evidence on correlates of prior beliefs about the wage and employment impacts of immigration. Republicans have more negative views about the economic effects of immigrants. This is particularly true for Republicans beliefs about the impact of the Cuban immigrants on wages and unemployment of low-skilled workers. High-skilled workers have more pessimistic views about the effect of Cuban immigrants on high-skilled workers, but do not hold more pessimistic beliefs about the impact of the Cuban immigrants on low-skilled workers. Respondents with higher incomes have more optimistic views on the effects the Cuban immigrants had on high-skilled workers wages and employment. In Figures A.1 to A.4 we illustrate the average pre-treatment beliefs about the economic effects of Cuban immigrants on wages and unemployment of high-skilled and low-skilled workers across US states. 3.2 Stylized facts: Attitudes towards immigrants In Figure 2, we show the distribution of attitudes towards different types of immigrants. As in Hainmueller et al. (2015), we find that people have more positive attitudes towards high-skilled compared to low-skilled immigrants and that they are 12

13 more supportive of immigrants who are familiar with American values compared to immigrants not familiar with American values. 3.3 The causal effect of labor market concerns Next, we investigate whether beliefs about the labor market impact of immigrants affect average attitudes towards the four different types of immigrants. Specifically, we estimate the following equation: y i = α 0 + α 1 T i + A T X i + ε i y i is the outcome of interest, T i is an indicator for whether subject i received information about the labor market impacts of the Mariel boatlift. X i is a vector of controls and ε i is an individual-specific error term. For all of the specifications, we use robust standard errors. Table 1 and Figure show the main results of our information treatment on selfreported attitudes towards immigrants. Respondents who receive the information increase their support for increasing the number of low-skilled immigrants not (highly) familiar with American values by 0.17 (0.11) of a standard deviation. Attitudes towards increasing the number of high-skilled immigrants not (highly) familiar with American values by 0.11 (0.05) of a standard deviation. This shows that people s attitudes are most responsive for the type of immigrants on which they received information. Figure 2 displays the distribution of responses for the treatment group and control group separately and allows us to study which parts of the attitude distribution are shifted. For low-skilled immigrants not familiar with American values, the information treatment shifts people away from having a very negative attitude towards immigrants, while for low-skilled immigrants highly familiar with American 13

14 values the treatment moves people from neutral attitudes to more positive ones. Table 2 and Figure 4 examine how the information affects people s willingness to sign petitions regarding the number of H2B visas. Column (1) of Table 2 shows that people s willingness to sign a petition in favor of increasing the number of H2B visas rises by 4.5 percentage points or about 15 percent. Column (2) of Table 2 highlights that people s willingness to sign a petition suggesting a decrease in H2B visas falls by 5.7 percentage points or about 18 percent. In columns (4) and (5) of Table 2 we compare actual signatures of the online petitions across the treatment groups. A proportion test indicates that the proportion of signatures of petitions in favor of increasing the number of H2B visas increases significantly by almost 3 percentage points or 80 percent compared to a control group mean of 4 percentage points. However, we find no statistically significant differences across the treatment and control group in signatures of petitions in favor of decreasing the number of H2B visas. 3.4 Obfuscated follow-up to deal with demand effects In Figure 5 and Table 3 we display results from the obfuscated follow-up survey. We find strong evidence that one week after the information was administered, respondents in the treatment group still think that immigrants have more beneficial economic impacts, and still display more positive attitudes towards immigrants with little or no education. We find smaller and statistically insignificant effects for attitudes towards high-skilled immigrants. In Figure 6, we display the distribution of attitudes towards low-skilled and high-skilled immigrants across treatment arms. As previously, we find that the information provision shifts people away from wanting to decrease the number of immigrants to increasing the number of immigrants. In the obfuscated follow-up survey, respondents were not aware that the follow- 14

15 up survey and the main survey in which the information was administered are linked. This means that differential experimenter demand effects are no longer a concern in the follow-up study. Our obfuscated follow-up survey allows us to find an upper bound for the size of demand effects. Specifically, under the assumption that there is no forgetting of information and that the information is not less salient in the follow-up study, differential experimenter demand effects across treatment arms is given by the difference in treatment effects for the index of attitudes towards low-skilled immigrants and the measure of attitudes towards low-skilled immigrants in the obfuscated follow-up survey. 9 Our data suggests that demand effects are likely very small and no larger than 0.05 of a standard deviation. These estimates are in line with recent evidence showing that subtle deliberately induced experimenter demand has rather moderate effects on people s behavior (de Quidt et al., 2017). 4 Mechanisms and heterogeneity 4.1 Heterogeneous treatment effects Next, we investigate heterogeneous responses to the information treatment. To do so, we estimate the following equation: y i = β 0 + β 1 T i + β 2 interact i + β 3 T i interact i + B T X i + ε i where the interaction term, interact i, is (i) pre-treatment standardized beliefs about the labor market impact of immigration 10, (ii) self-perceived skill level (a 9 Of course, we need to assume that the slight differences in wording of the outcome questions make little difference for our measurement of attitudes towards immigrants. 10 We construct this belief measure by creating a weighted index of people s standardized beliefs about the effect of immigration on (i) wages and (ii) unemployment

16 dummy taking value one for people who think they are high-skilled), (iii) political affiliation (a dummy taking value one for Republicans) as well as (iv) attention to the instructions on the Mariel boatlift Prior belief Panel A of Table A.8 and Table A.9 show heterogeneous effects depending on our respondents prior beliefs about the labor market impacts of immigrants. As expected we find that people with more positive pre-treatment beliefs about the labor market impacts of immigrants, respond less strongly to the information treatment. However, this heterogeneity is not statistically significantly different from zero. This could be a result of measurement error of prior beliefs or because pre-treatment beliefs are correlated with other covariates (such as political affiliation, attention or education) for which we may expect patterns of heterogeneity going in the opposite direction Self-interested labor market competition If people s own self-interest mattered for their attitudes towards immigrants, we would expect to see larger effects on attitudes towards low-skilled (high-skilled) immigrants for low-skilled (high-skilled) workers compared to high-skilled workers. As can be seen in Panel C of Table A.8 and Table A.9 we find no significant treatment heterogeneity in line with the idea that people adjust their policy preferences as a result of self-interest. This corroborates evidence from the previous literature that self-interested labor market concerns are not an important determinant of attitudes towards immigrants. 16

17 4.1.3 Political ideology Panel C of Table A.8 and Table A.9 show that Republicans do not respond differently to the treatment compared to non-republicans. This could reflect two opposing factors: one the one hand Republicans are more likely to dismiss the evidence from the research study; one the other hand, for Republicans there is more scope for increases in positive attitudes towards immigrants, as they have more negative attitudes to begin with Attention We asked all of our respondents whether Cuban immigrants who came to Miami were mostly high-skilled or low-skilled in order to test whether our respondents paid attention to the instructions in the belief elicitation. We find that 77.6 percent of respondents correctly remembered that most Cuban immigrants who arrived in Miami were low-skilled. This underlines that most of our respondents actually paid attention to the belief elicitation. We use this measure of attention to examine heterogeneous treatment effects. 12 Panel D of Table A.8 and Table A.9 show that the estimated treatment effects for respondents who correctly remember this information are significantly higher than the treatment effects for respondents who did not correctly remember the information. 4.2 Mechanisms General labor market concerns Figure 7 and Table 5 depict people s beliefs about the overall economic impact of immigrants on the US as a whole. We find evidence that people strongly 12 This analysis was not pre-specified. 17

18 update their beliefs about the economic impacts of both high-skilled and low-skilled immigrants. People develop more positive views on the likely economic effects of immigrants, with the strongest effects for immigrants that are low-skilled and not familiar with American values; i.e., those immigrants most closely resembling the immigrants from the Mariel boatlift. We also observe a shift in beliefs about the overall economic impact of high-skilled immigrants. Next, we examine how the research evidence affected people s beliefs about the effect of low-skilled and high-skilled immigrants on (i) wages, (ii) unemployment, and (iii) the tax burden of most Americans. In Table 4 we show that people in the treatment group strongly change their beliefs about the effects immigrants have for wages and unemployment of most Americans, but do not update their beliefs about the tax burden created by immigrants. These patterns are very similar for people s beliefs about the impact of both high-skilled and low-skilled immigrants. Moreover, in Table 4, we show that the information treatment did not affect people s views on how low-skilled or high-skilled immigrants affect American culture and society. This indicates that the information about the Mariel boatlift did not affect people s views on how immigrants integrate into American society or how they affect American culture Self-interested labor market concerns In Table 4, we also examine whether people shift their beliefs about the economic effects of low-skilled and high-skilled immigrants on their own household. Treated respondents do not change their beliefs about how low-skilled or high-skilled immigrants affect the wages, unemployment and tax burden of their own household. This highlights that the changes in policy preferences regarding immigration are not the result of self-interested labor market concerns, but rather that people care about outcomes at the group level. This evidence is also consistent with the absence 18

19 of treatment heterogeneity by people s self-perceived skill as described in subsection Ingroup-out-group bias and economic concerns Our results showcase the importance of a psychological mechanism related to ingroup out-group bias: that beliefs about how in-group members ( most Americans ) are economically affected by out-group members ( newly arriving immigrants ) are a quantitatively important determinant of attitudes towards immigrants. Indeed, our evidence suggests people s social identity and their identification with an in-group ( American citizens ) are an important ingredient to understand the relevance of economic concerns. Our evidence reconciles seemingly contradictory evidence which had concluded the economic concerns (and in particular self-interested labor market concerns) are not an important determinant of attitudes towards immigrants Beliefs about the research evidence We also collected rich data among respondents in the treatment group to measure people s perception of the research evidence. 13 First, we examine people s levels of trust in the research information. 57 percent of respondents deemed the research information fully trustworthy or somewhat trustworthy. 35 percent of respondents considered the information neither trustworthy nor unstrustworthy, while the remaining 7 percent of respondents deemed the research information untrustworthy. 49 percent of respondents agree or strongly agree that the research information about the Mariel boatlift is relevant for evaluating the labor market impacts of immigrants today, while 15 percent strongly disagree or disagree. 36 percent of respondents neither agree nor disagree. 49 percent of respondents trust in the 13 The analysis in this section is exloratory and was not pre-specified. 19

20 accuracy of the study, while 5 percent do not trust in the accuracy of the study. As one would expect, Table A.6 shows that beliefs about the internal and external validity of the study as well as people s beliefs about the trustworthiness of the study are strongly correlated with attitudes towards immigrants in the expected direction: people more skeptical of the research information display lower support for immigration conditional on a large set of control variables. In Table A.5, we examine the correlates of trust in the information provided, as well as beliefs about the internal and external validity of the research information provided among respondents in the treatment group. Republicans, Independents and older respondents display significantly lower levels of trust in the research information, lower agreement that the study is externally valid, and that the study is accurate. Surprisingly, education is not correlated with any of these beliefs. 4.3 Interpreting effect sizes: IV estimates In order to interpret effect sizes of our main effects, we examine the responsiveness of policy preferences with respect to randomly assigned changes in perceptions about the economic impacts of immigrants. Specifically, we estimate the effect of beliefs about the economic impact of immigrants by instrumenting beliefs with the randomly assigned exposure to research information. Our instrumental variable estimates reveal substantial elasticities of policy preferences with respect to perceptions about the economic impact of immigrants. In Table A.7 we display these instrumental variables estimates. A one standard deviation change in perceptions of the labor impacts of immigrants of low-skilled immigrants not familiar with American values translates into a 0.65 standard deviation change in attitudes towards immigrants. 20

21 5 Conclusion We provide evidence that beliefs about the labor market effects of immigrants strongly causally affect people s attitudes towards immigrants as measured by self-reports and signatures of real online petitions. Our estimated effect sizes are large and imply that economic concerns are a quantitatively important determinant of attitudes towards immigrants: a one-standard deviation change in perception of the economic impact of immigrants results in a 0.5 to 0.6 change in attitudes towards immigrants. Our results also hold up in an obfuscated follow-up study in which experimenter demand is no longer a concern. Our evidence, which highlights the importance of economic concerns in forming attitudes towards immigrants, stands in stark contrast to the current consensus in the literature. For instance, in a recent review article, Hainmueller and Hopkins (2014) describe labor market concerns as a zombie theory that has repeatedly failed to find empirical support. Interestingly, we show that economic concerns do not matter because of self-interested labor markets concerns, but through group level concerns. This explains why previous literature concluded economic concerns are unimportant. The results shows that high-credibility research information can be effective in changing hearts and minds over an important policy question. Our research therefore has important policy implications. Disseminating research results can persistently change people s beliefs about economic issues and persistently change people s policy preferences. 21

22 References Almås, Ingvild, Alexander Cappelen, and Bertil Tungodden, Cutthroat Capitalism versus Cuddly Socialism: Are Americans more Meritocratic and Efficiency-seeking than Scandinavians?, Anderson, Michael L., Multiple Inference and Gender Differences in the Effects of Early Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects, Journal of the American Statistical Association, 2008, 103 (484), Borjas, George J., Immigration Economics, Harvard University Press, Brewer, Marilynn B, In-group bias in the minimal intergroup situation: A cognitive-motivational analysis., Psychological bulletin, 1979, 86 (2), 307. Bursztyn, Leonardo, Georgy Egorov, and Stefano Fiorin, From Extreme to Mainstream: How Social Norms Unravel, Working Paper 23415, National Bureau of Economic Research May Cappelen, Alexander, Ingar Haaland, and Bertil Tungodden, Beliefs about Behavioral Responses to Taxation, Working Paper, Card, David, The Impact of the Mariel Boatlift on the Miami Labor Market, Industrial and Labor Relations Review, 1990, 43 (2), , Is the new immigration really so bad?, Economic Journal, 2005, 115 (507)., Comment: The Elusive Search for Negative Wage Impacts of Immigration, Journal of the European Economic Association, 2012, 10 (1), and Giovanni Peri, Immigration Economics by George J. Borjas: A Review Essay, Journal of Economic Literature, 2016, 54 (4),

23 , Christian Dustmann, and Ian Preston, Immigration, Wages, and Compositional Amenities, Journal of the European Economic Association, 2012, 10 (1), Citrin, Jack, Donald P Green, Christopher Muste, Cara Wong et al., Public Opinion Toward Immigration Reform: The Role of Economic Motivations, Journal of Politics, 1997, 59, Clemens, Michael A., What the Mariel Boatlift of Cuban Refugees Can Teach Us about the Economics of Immigration: An Explainer and a Revelation, de Quidt, Jonathan, Johannes Haushofer, and Christopher Roth, Measuring and Bounding Experimenter Demand, Working Paper 23470, National Bureau of Economic Research June Dustmann, Christian and Ian Preston, Attitudes to Ethnic Minorities, Ethnic Context and Location Decisions, Economic Journal, 2001, 111 (470), and, Is Immigration Good or Bad for the Economy? Analysis of Attitudinal Responses, Research in Labor Economics, 2006, 24 (1), 3 34., Francesca Fabbri, and Ian Preston, The impact of immigration on the British labour market, Economic Journal, 2005, 115 (507). Facchini, Giovanni, Anna Maria Mayda, and Riccardo Puglisi, Illegal immigration and media exposure: Evidence on individual attitudes, CEPR Discussion Papers 7593, 2009., Yotam Margalit, and Hiroyuki Nakata, Countering Public Opposition to Immigration: The Impact of Information Campaigns, Falk, Armin and Florian Zimmermann, A taste for consistency and survey response behavior, CESifo Economic Studies, 2012, 59 (1),

24 Gallup, Public Ambivalent About New Immigrants, 2005., Most Important Problem, Gerber, Alan S, Gregory A Huber, Daniel R Biggers, and David J Hendry, Self-Interest, Beliefs, and Policy Opinions: Understanding How Economic Beliefs Affect Immigration Policy Preferences, Political Research Quarterly, Gilens, Martin, Political Ignorance and Collective Policy Preferences, American Political Science Review, 2001, 95 (2), Grigorieff, Alexis, Christopher Roth, and Diego Ubfal, Does Information Change Attitudes Towards Immigrants? Evidence From Survey Experiments, IZA Discussion Paper No , Haaland, Ingar and Christopher Roth, Beliefs about Racial Discrimination: Representative Evidence, Working Paper, Hainmueller, Jens and Daniel J Hopkins, Public Attitudes Toward Immigration, Annual Review of Political Science, 2014, 17, and Michael J Hiscox, Attitudes Toward Highly Skilled and Low-skilled Immigration: Evidence from a Survey Experiment, American Political Science Review, 2010, 104 (01), ,, and Yotam Margalit, Do Concerns about Labor Market Competition Shape Attitudes Toward Immigration? New Evidence, Journal of International Economics, 2015, 97 (1), Halla, Martin, Alexander F Wagner, and Josef Zweimüller, Immigration and Voting for the Far Right, Journal of the European Economic Association,

25 Hopkins, Daniel J, John Sides, and Jack Citrin, The Muted Consequences of Correct Information About Immigration, Working Paper, Kugler, Adriana and Mutlu Yuksel, Effects of Low-Skilled Immigration on U.S. Natives: Evidence from Hurricane Mitch, NBER Working Paper Kuklinski, James H, Paul J Quirk, Jennifer Jerit, David Schwieder, and Robert F Rich, Misinformation and the currency of democratic citizenship, Journal of Politics, 2000, 62 (3), Kuziemko, Ilyana, Michael I Norton, Emmanuel Saez, and Stefanie Stantcheva, How Elastic are Preferences for Redistribution? Evidence from Randomized Survey Experiments, American Economic Review, 2015, 105 (4), Lawrence, Eric D and John Sides, The consequences of political innumeracy, Research & Politics, 2014, 1 (2), 1 8. List, John A, On the interpretation of giving in dictator games, Journal of Political Economy, 2007, 115 (3), Manacorda, Marco, Alan Manning, and Jonathan Wadsworth, The impact of immigration on the structure of wages: theory and evidence from Britain, Journal of the European Economic Association, 2012, 10 (1), Mayda, Anna Maria, Who is Against Immigration? A Cross-country Investigation of Individual Attitudes Toward Immigrants, Review of Economics and Statistics, 2006, 88 (3), Nyhan, Brendan and Jason Reifler, When Corrections Fail: The Persistence of Political Misperceptions, Political Behavior, 2010, 42,

26 Ottaviano, Gianmarco IP and Giovanni Peri, Rethinking the effect of immigration on wages, Journal of the European Economic association, 2012, 10 (1), Roth, Christopher and Johannes Wohlfart, How Do Expectations About the Aggregate Economy Affect Personal Expectations and Behavior? Experimental Evidence, Working Paper, Scheve, Kenneth F and Matthew J Slaughter, Labor Market Competition and Individual Preferences over Immigration Policy, Review of Economics and Statistics, 2001, 83 (1), Sides, John and Jack Citrin, How Large the Huddled Masses? The Causes and Consequences of Public Misperceptions About Immigrant Populations, in Annual Meeting of the Midwest Political Science Association, Chicago Zizzo, Daniel John, Experimenter Demand Effects in Economic Experiments, Experimental Economics, 2010, 13 (1),

27 Main Figures Figure 1: Prior beliefs about the labor Market impacts of the Mariel boatlift.7 Low skilled, wages.7 Low skilled, unemployment Fraction.4.3 Fraction Strongly decrease No effect Strongly increase Somewhat decrease Somewhat increase 0 Strongly increase No effect Strongly decrease Somewhat increase Somewhat decrease.7 High skilled, wages.7 High skilled, unemployment Fraction.4.3 Fraction Strongly decrease No effect Strongly increase Somewhat decrease Somewhat increase 0 Strongly increase No effect Strongly decrease Somewhat increase Somewhat decrease N=3130 Notes: The figure shows standardized treatment effects. 27

28 Figure 2: Distribution of attitudes towards immigrants: Treatment vs. control Mean ± s.e.m Control Treatment Mean ± s.e.m Control Treatment A lot less Less Same More A lot more Low-skilled, not familiar A lot less Less Same More A lot more Low-skilled, highly familiar 28 Mean ± s.e.m Control Treatment Mean ± s.e.m Control Treatment A lot less Less Same More A lot more High-skilled, not familiar A lot less Less Same More A lot more High-skilled, highly familiar Notes: The figure shows standardized treatment effects.

29 Figure 3: Main Treatment Effects.2 Treatment Effect (z-scored) Not Familiar Highly Familiar Not Familiar Highly Familiar Low-skilled Immigrants High-skilled Immigrants Notes: The figure shows standardized treatment effects. Figure 4: Online Petitions Increase annual cap Decrease annual cap Mean ± s.e.m Control Treatment 0.24 Control Treatment Notes: The figure shows standardized treatment effects. 29

30 Figure 5: Main treatment effects: follow-up.2 Treatment Effect (z-scored) Immigrants: Immigrants: Immigrants: Economic impact Low-skilled High-skilled Notes: The figure shows standardized treatment effects. Figure 6: Distribution of responses: follow-up.5 Control Treatment.5 Control Treatment.4.4 Mean ± s.e.m..3.2 Mean ± s.e.m..3.2 Strongly decreased Somewhat decreased Keep the same Somewhat increased Strongly increased The number of immigrants with little to no education should be... Strongly decreased Somewhat decreased Keep the same Somewhat increased Strongly increased The number of highly educated immigrants should be... Notes: The figure shows standardized treatment effects. 30

31 Figure 7: Beliefs about the overall economic impact of immigrants.3 Treatment Effect (z-scored) Not Familiar Highly Familiar Not Familiar Highly Familiar Low-skilled Immigrants High-skilled Immigrants Notes: The figure shows standardized treatment effects. 31

32 Main Tables Table 1: Self-reports: Attitudes towards immigrants Low-skilled High-skilled Not familiar Familiar Index Not familiar Familiar Index Panel A: With controls Treatment (0.033) (0.034) (0.030) (0.033) (0.034) (0.030) Adjusted p-value [0.001] [0.001] [0.002] [0.086] Panel B: No controls Treatment (0.036) (0.036) (0.033) (0.036) (0.035) (0.032) Adjusted p-value [0.001] [0.001] [0.004] [0.078] Observations Notes: The outcome variables are z-scored using the mean and standard deviation in the control group. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors in parentheses. 32

33 Table 2: Online Petition on increasing or decreasing H2B visas Intention: H2B Visas Actual signatures H2B Visas Increase Decrease Net support Increase Decrease Panel A: With controls Treatment (0.016) (0.016) (0.034) Panel B: No controls Treatment (0.016) (0.016) (0.036) (0.009) (0.009) Control Mean Observations Notes: The outcome variables are z-scored using the mean and standard deviation in the control group. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors in parentheses. 33

34 Table 3: Results from obfuscated follow-up Immigrants hurt Low-skilled High-skilled American Workers Immigrants Immigrants Panel A: With controls Treatment (0.040) (0.040) (0.042) Panel B: No controls Treatment (0.044) (0.044) (0.044) Observations Notes: The outcome variables are z-scored using the mean and standard deviation in the control group. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors in parentheses. 34