I need 4 pages with 1.5 space on historical papers.

Write a reference reports of the attached paper ( article 5 )

one guideline file attached, You can see how it has to be: How to write a referee report-4.pdf

There are 2 examples here:

refereereport(example1).pdf

Explorations in Economic History 64 (2017) 37–52

..........................................................................................................................................................................

Attachment 1:

How to write a referee report

Referee report:

Summary: The first section of the report (often one paragraph) should contain a concise summary of the paper’s claimed results, contributions, and general line of reasoning. The editor is typically not an expert in the paper’s subfield, so it is important for this summary to be clear.

 Only after this, turn to substantive issues about the importance and validity of the claimed results

Content of the report: In writing a report, you can divide your comments into the following sections.

1) The importance of the paper. This is the most subjective part of the report. The editor needs to assess the importance of the contribution aided by your report. The report should contain an argument that supports your assessment of the importance of the work and detail the considerations that bear upon your judgment. Ask yourself, is the paper answering an interesting economic question that merits publishing in a top general interest journal/top field journal & does it improve the existing literature substantially?

2) Problems with the paper that render it unpublishable. In this case, provide a scientifically convincing argument for why these problems render the paper unpublishable (i.e. you believe the problems are serious enough that they are unlikely to be fixable in the next round). What you should NOT do is simply rejecting the paper because you don’t believe the results. If you recommend rejection, make sure it is based upon strong arguments rather than strong priors

 In science, it is important that the debate is based on evidence and objective criteria rather than personal preconceptions. If you decide to make a revise and resubmit recommendation, this section must be empty.

When you make a revise-and-resubmit recommendation, you are actually making three statements: 1) the paper is of sufficient importance in terms of scope and findings that you believe it is suitable for the journal at hand; 2) there are problems with the paper that currently make it unpublishable in its current form; and 3) these problems are correctable.

3) Problems with the paper that currently make it unpublishable. BUT, which you believe could be corrected. In this case be very clear why the paper is unpublishable (see above) and what a correction to the problem would look like. If the author addresses your points you should recommend publication.

4) Problems with the paper that do not render the paper unpublishable. Here you do not need to provide reasons for your opinion, but you cannot hold up publication if the authors do not address these problems. In many cases people disagree about what should and should not go into the paper.

 Ultimately, the author’s name goes on the paper, not yours. It is the author’s decision on how best to write the paper, not yours.

Further points:

Take a scientific stance in your report. Do not insult the authors, or use overly emotional or accusatory language. Avoid ascribing bad intent to authors (“The authors were trying for a cheap publication,” “The authors were trying to brush past literature/conflicting findings under the rug…”) and focus on the substance of the paper

Do not dismiss papers that attack larger issues merely because flaws can be found—weigh the pros as well as the cons. The important question that you need to assess is whether the flaws invalidate the contribution Try to impartially ask yourself the following question: Flaws and all, would I have been proud to write such a paper? If the answer is yes, that gives a strong hint that it should be published.

Please reread your report carefully and think about what a critic might make of your arguments. Too often, reports make inconsistent arguments in different parts of the report. This is just one symptom of the common problem of not thinking things through adequately

........................................................................................................................................................................Attachment 2:

Birds of passage: Return migration, self-selection and immigration MARK quotas

Zachary Ward

Research School of Economics, HW Arndt Building 25A, College of Business and Economics, The Australian National University, Canberra, ACT 2600, Australia

A R T I C L E I N F O A B S T R A C T

JEL: A key feature of migration in the late 19th and early 20th century is that many migrants returned F22 to Europe after a few years in the United States. A common view is that most temporary migrants J61 planned, upon entry, to eventually return home, yet there is little direct evidence to support this N30 claim. I collect the first dataset on migrants' intentions to stay or return home from Ellis Island Keywords: arrival records between 1917 and 1924. I find that fewer migrants planned to return home than Return migration actually did; many migrants, especially from Eastern and Southern Europe, left the United States Self-selection unexpectedly. The high rate of unplanned returns implies that the first few years after arrival Immigration quotas were more di fficult than expected. However, this high rate of unexpected returns lowered after

Planned returns

the 1920s migration quotas, suggesting improved outcomes for those lucky enough to enter.

“The steerage passengers may be roughly divided into two classes: those who go home because they have succeeded, and those who go home because they have failed.”

– Edward A. Steiner, On the Trail of the Immigrant (1906)

1. Introduction

Fundamentally, migration is about the flow of people across borders, but a fact often overlooked is that flows occur both into and out of a country. Return flows can be substantial: in the early 20th century United States they were up to three-forths of the size of inflows, suggesting that the average incoming migrant was temporary rather than permanent. Numerous scholars have claimed that many of these returns were planned from the beginning, as migrants purposely stayed a few years to accumulate savings for investment or consumption at home (Baines, 1995; Piore, 1979; Wyman, 1993).1 However, these claims are based on indirect evidence; an equally likely hypothesis is that migrants' return trips were decided upon after arrival, perhaps because conditions in the United States were worse than expected (Borjas and Bratsberg, 1996; Cerase, 1974).

The answer to this question is key for understanding migrant assimilation: if temporary migrants planned to return home, then they would have exerted little effort to assimilate. For example, planned temporary migrants are less likely to invest in United States-

E-mail address: Zach.A.Ward@gmail.com.

1 There are a range of estimates for the ratio of out-migrants to in-migrants in the early 20th century, from 39 percent (Willcox, 1931, Table 17), 45 percent (Kuznets and Rubin, 1954, Table 6), and 60 to 75 percent (Bandiera et al.,, 2013, Table 4). These are demographic estimates based on the change in foreign-born population from census to census; the change in population should reflect additions from new arrivals and subtractions from departures and deaths. Most recently Bandiera et al. (2013) aim to improve on prior estimates by using newly digitized data on the universe of Ellis Island arrivals, which contain more accurate counts of arrivals than the official US data.

http://dx.doi.org/10.1016/j.eeh.2016.09.002

Received 19 April 2016; Received in revised form 9 August 2016; Accepted 24 September 2016

Available online 28 September 2016

0014 -4983 / © 2016 Elsevier Inc. All rights reserved.

specific human capital, such as the ability to speak English, which has little return at home. On the other hand, if most returned home unexpectedly, then it would suggest that the United States was not a “land of opportunity” because outcomes for migrants were poor despite efforts to assimilate. Alternatively, unplanned returners may have had unrealistic expectations about the United States because information about the migrant experience was biased (McKenzie et al., 2013).

I aim to determine whether most return migration in the early 20th century was planned or unplanned by collecting a unique dataset; in 1917, the Bureau of Immigration started to ask entering migrants whether they planned to stay permanently or to return home. Migrants who planned to return home were also required to report the length of time, whether days, weeks or years, that they intended to stay.

Using a sample of 27,000 Ellis Island arrivals between 1917 and 1924, I can answer numerous questions about temporary migration including at what rate migrants planned to return, who selected into planned temporary migration, and how migration policy, such as the 1920s migration quotas, affected planned returns. I can also compare the rates and characteristics of planned returners to those of actual returners by using a dataset on departures back to Europe. This out-going data is unique because the early 20th century was the only time period in United States history when the government collected information on departures, a practice that no longer exists today.

I find that most migrants did not plan to return home at arrival: in the years prior to the migration quota of 1921, only 15 percent of entrants intended to eventually return home. This can be contrasted with estimates that at least 40 percent of migrants actually returned home, suggesting that most eventual returns prior to the immigration quotas were unexpected at the time of arrival. It appears that the discrepancy between planned and actual return rates is mostly due to single males from Southern and Eastern Europe unexpectedly returning home at high rates; this could reflect a large amount of failures in the labor market, perhaps due to unemployment after arrival. In contrast with the Southern and Eastern European experience, many Northern and Western European countries unexpectedly stayed in the United States, perhaps because outcomes were better than expected.

After the implementation of the quotas in 1921 and 1924, which caused a 60% fall of inflows over a four-year period, the pattern of a high rate of unplanned returns for Southern and Eastern Europeans completely changed. After the quotas, the rate of actual return migration decreased (Greenwood and Ward, 2015); but in this paper, I show that the quotas had no effect on the rate of planned return migration. A combination of these two results suggests that the migration quotas lowered unplanned returns, especially of those holding low-skilled jobs in the United States. The fact that more low-skilled migrants remained in the United States helps to explain why the self-selection of return migrants became less negative for the countries most restricted, as seen in Fig. 1.

A simple reason for the results could be that the labor market was difficult prior to the quotas and then became easier for workers as the quotas reduced incoming competitors. In other words, the quotas improved outcomes for those lucky enough to enter. Evidence of a tightening labor market can be clearly seen when examining the rapid increase of inflows of those with highly substitutable characteristics following the quotas: black migration from southern states, Mexican migration from across the border, and Canadian migration from the north (Collins, 1997; Kosack and Ward, 2014). Thus, this paper supports recent arguments that a voluminous inflow affects other migrants more strongly than natives (Ottaviano and Peri, 2012).

Another reason for the decreased rate of unplanned returns is that the quotas screened out those most likely to unexpectedly return: single males from Southern and Eastern Europe. Prior to the quotas, these migrants had high return rates but relatively low planned return rates, suggesting that they often changed duration plans after arrival – likely because it was cheaper for them to switch decisions compared with the costs for other migrants with families. Following the quotas, there were fewer single male entrants, reducing the number of unplanned returns.

This paper complements and extends prior work on return migration during this time period by Greenwood and Ward ( GW, 2015) in many ways. Primarily, GW are silent on whether migrants planned to return home or stay at arrival; this paper's main contribution is the data on migrants' duration plans, which is the first data, to my knowledge, on migrants' return intentions at any point prior to the 1980s. Further, GW limit their main analysis to the rate of return migration and never compare returners to the migrant stock on observables (i.e., the selection of returners). This paper estimates that return migrants were negatively selected using data that directly observes out-migrants; this additionally contributes to the literature by verifying the use of indirect methods to estimate return migrant selection (Abramitzky et al., 2014). Finally, this paper complements GW's results on the effects of the migration quotas by examining how the policies influenced return intentions; the results in this paper stress that the decrease in the overall return rate during the 1920s was primarily driven by a decrease in unplanned returns.

Fig. 1. Self-selection of return migrants, 1910–1930. Notes: Data is from the Annual Reports of the Commissioner General of Immigration (1911, 1921, and 1931) and IPUMS (1910, 1920, and 1930). Self-selection is the difference between return migrants' average logged occupational score and the migrant stock's average logged occupational score; a negative number implies return migrants were lower skilled than the stock. The average skills for the foreign-born in the Census reflect weighting as described in text, which in effect places less weight on those staying more than ten years and more weight for those staying less than ten years. New source countries are those who were more restricted by the 1921 and 1924 quotas, typically from Eastern and Southern Europe. Old source countries are those who were less restricted by the quotas which traditionally includes those from Western and Northern Europe, but also includes countries from the Western Hemisphere. Ethnicities from Asia and Africa are dropped. Confidence intervals are shown at a 95% level.

2. Understanding temporary migration

The decision to return to a poorer country may be baffling to the economist who believes in income maximization, but it has long been recognized that migrants have a variety of motives for moving temporarily (Dustmann and Görlach, 2016). For example, a migrant may move to temporarily increase earnings power for consumption or investment at home, or to acquire human capital that receives a large premium after returning. On the other hand, temporary migrants may return for cultural or family reasons, such as to provide care for older family members or to raise children in the source country (Gibson and McKenzie, 2011).

Early models motivated temporary migration by including a taste preference for consumption at home, the main way to explain returning to a poorer country (e.g., Hill, 1987; Galor and Stark, 1990); however, these models also assumed that the return trip was anticipated at the time of arrival, which is not always the case (Dustmann, 1996). Only recently have models started to consider the dynamic nature of migration as new information about life in the host country is revealed after arrival (e.g., Adda et al., 2015 ; Görlach, 2016; Thom, 2010). In response to an unexpected shock, the duration decision can update in different directions: one who planned to stay permanently could now decide to return home or a temporary migrant might be pushed to the corner solution and expect a permanent stay.

Yet how temporary migrants respond to unexpected income shocks is not straightforward; the response depends on whether a migrant is a “target saver” who aims to hit a pre-determined amount prior to returning home, or if a migrant is a standard “life-cycle” earner who responds to shocks in more conventional ways. For example, if wages unexpectedly increase, a life-cycle earner would remain longer to take advantage of better opportunities, while a target saver would hit his goal more quickly and return home sooner. On the other hand, wages may unexpectedly decrease, which could lead a life-cycle earner to update his optimal duration on the extensive margin (permanent to temporary stay) or on the intensive margin (longer to shorter stay); this is the scenario where return migrants “fail” in the labor market and return home. Alternatively, an unexpected wage decrease may lead a target saver to stay longer than originally intended.

Most models assume temporary migrants follow a life-cycle model instead of a reference-dependent one as in a target-saving model, which is consistent with evidence from Filipino migrants in the 1990s (Yang, 2006). Further, in the first half of the 20 th century, return flows increased rapidly during United States recessions (Biavaschi, 2013b; Jerome, 1926) – a relationship consistent with life-cycle “failures” returning due to negative income shocks. However, there is also evidence that recent Mexican migrants in the United States are target savers (Nekoei, 2013). Ultimately, the return flow is likely a mixture of both successful and unsuccessful life-cycle earners and target savers.

Perhaps the most well-known model that considers both the duration decision changing after arrival and the selection of temporary migrants is Borjas and Bratsberg (1996) extension of the original Borjas (1987) selection model to account for return migration. An important insight from this extension is that the migrants most likely to return home are those on the “margin” – migrants who are just barely attracted to living abroad by the slightly higher income. Of course, after arrival migrants may discover that actual wages differed from expected wages, which could change selection after arrival depending on how unexpected shocks are correlated with skill level – something that is typically unobserved.

The patterns of migrant selection based on relative skill premia as in a Roy–Borjas model may have held during the Age of Mass Migration (Stolz and Baten, 2012), but there has been much criticism of these models. In one strand, some argue that the standard Roy–Borjas selection model may not hold if the costs of migration vary across the skill distribution (Borjas, 1991; Chiquiar and Hanson, 2005). For example, if costs are highest for the least skilled, then they would not move despite a potentially large increase in income. A second criticism of the Roy–Borjas selection model is that migrants may move based on absolute wage differences rather than the relative return to skill (Grogger and Hanson, 2011). In this case, if temporary migrants are those on the margin, then they should be the ones with the smallest difference in absolute wages, while permanent migrants should have the largest differences in absolute wages. Ultimately, due to the differing predictions from these models, the true pattern of selection into temporary migration is an empirical question.

3. Temporary migration in the early 20th century

During the early decades of the Age of Mass Migration (1850–1913), rates of return migration were relatively low due to the high costs of the trip: traveling via sailing ship took months, fares were expensive and mortality rates were high (Bailey, 1912; Cohn, 1984). Return migration started to become more common following the Civil War, when shipping technology shifted from sail to steam and travel time shortened (Cohn, 2009). Following the diffusion of steam to travel, the source country composition changed rapidly in the late 1800s from “Old” sources in Northern and Western Europe to “New” sources in Southern and Eastern Europe; soon, migrant flows shifted from a “family” movement in which entire households moved to an influx of young, single males who were more mobile (Gould, 1980; Baines, 1995).10

The American experience for these new migrants was vastly different from that at home: many settled in densely populated areas and worked in construction, manufacturing or mining, different from their more agrarian source countries (Wyman, 1993). Migrants worked long and hard hours – “like animals” as one Italian returnee put it (Cerase, 1974, p. 250) – but with hard work and thrift, migrants could return home with sizable savings given the higher real wages in America (Williamson, 1995). Despite relatively high wages, many returned home for reasons of social isolation and alienation (Cerase, 1974): immigrants encountered substantial discrimination and racism from natives (Abramitzky et al., 2016; Higham, 2002; Moser, 2012). After World War I, the ensuing xenophobic sentiment and “Americanization” policies in the 1920s perhaps led some migrants, like Germans, to return home at higher rates (Fouka, 2015; Greenwood and Ward, 2015; Lleras-Muney and Shertzer, 2015).

Despite difficulties in the United States, hundreds of thousands kept migrating year after year, which is taken as indirect evidence that most temporary migrants did not fail in the United States (Baines, 1995). Besides letters, return migrants were the main source of information about life abroad for potential migrants: a successful return migrant could have been the envy of others, perhaps leading to more first-time migration. Indeed, inflows fluctuated strongly with United States business cycles, likely because successful migrants spread positive information back home, but also because failed return migrants spread negative information (Hatton and Williamson, 1998; Jerome, 1926).

Yet the narrative that increasing return migrant flows spurred on more first-time migration could be spun around: perhaps the rising number of first-time migrants drove others back home. Indeed, there is a consensus in the literature that new arrivals are substitutes for migrants already in the labor market (Friedberg, 2001; Ottaviano and Peri, 2012). The early 20th century had one of the highest rates of in-migration in American history (Abramitzky and Boustan, 2016), leaving it unclear whether the United States could have fully absorbed all arrivals. Ultimately, the return flows were likely a combination of failures and successes; as noted by Steiner (1906), “the steerage passengers may be roughly divided into two classes: those who go home because they have succeeded, and those who go home because they have failed.”

4. Data

4.1. Planned return migrants at arrival: ship manifests

The primary contribution of this paper comes from a new dataset on return migrant intentions. This dataset is a 1% random sample of ships that originated in Europe and arrived at Ellis Island from 1917 to 1924.11 The ship records used in this study are the same source used in several historical studies of immigration across the 19th and 20th century (e.g., Cohn, 2009; Ferrie, 1999; Massey, 2016). I start my sample in 1917 as this is the first year an important question was added: “Whether alien intends to return to country whence he came after engaging temporarily in laboring pursuits in the United States.”12

One concern is whether migrants truthfully revealed their intentions to stay. There was no benefit or penalty to stating a

(footnote continued) skilled receive a larger premium abroad, then the relatively less skilled of this high-skilled group return home. This is because the less skilled of the highly skilled group earn a smaller wage premium and thus are only marginally attracted abroad. Therefore, return migrants are negatively selected from the positively selected migrant pool.

10 There are several historical studies of return migration for specific countries during this time period. See Sarna (1981) for Jewish return migration, Kraljic (1978) for Croatian, and Saloutos (1956) for Greek return migration. Also, see Balch (1910) and Steiner (1906) for contemporary accounts of temporary migration. Finally, Gmelch has an anthropological survey of return migration (1980).

11 I randomly sample the ships from the Statue of Liberty Ellis-Island Foundation.

12 Another entry on the manifests was “Length of time alien intends to remain in the United States,” which I use to determine whether a migrant plans to leave by 1930 for the linked dataset.

permanent or a return plan; further, these ship manifests were filled out by ship captains rather than United States border officials, which potentially reduced any misrepresentation if migrants would be more intimidated by border officials. However, ship captains may have been careless when recording information; for example, this has been found for manifests from the 19th century when some ship captains listed all migrants as farmers (Erickson, 1972). For the sample collected for this paper, careless compiling does appear to be a small problem: two of the 70 ships in the original sample had the entire ship recorded as planning to stay permanently despite holding hundreds of migrants. I drop these ships from the analysis.

As a further check on whether migrants were truthful, I attempt to find arrivals in the 1930 United States Census; if migrants did return home, then they should not be located. The linking process follows standard practice where links are based on first name, last name, year of birth (plus or minus 2 years), and country of birth. While the overall linking rates are low, which is common in historical studies (see, for example, Abramitzky et al., 2014), the linking rate for planned returners (17.9%) is lower than for planned permanent migrants (29.4%), as expected if migrants were truthful at arrival. Of course, many migrants could have remained in the United States despite planning to return home at arrival because life was better than expected; this behavior is common for currentday migrants (Dustmann, 1996).

Yet failing to find a migrant is not a perfect measure of returning because there could be other reasons for failing to link. For example, failing to link could be related to death, changing one's name after arrival, or having a careless ship captain who haphazardly recorded names or return intentions. Further, a migrant may be linked to a wrong individual because he has a common name. To control for these and other reasons for failing to link, such as country of birth or year of arrival, I use a regression to test whether planning to return home is associated with failing to be found in the later census. Even with controlling for various other reasons for failing to find a migrant in the 1930 Census, including ship fixed effects to account for idiosyncratic recording by ship captains, I consistently find that planning to return home is negatively associated with a successful link. This provides some confidence that migrants truthfully revealed their duration plans and many returned home; see Appendix C for more detail on this analysis.

Some migrants (641 to be exact) listed their intention to stay as “indefinite” or “uncertain.” I allocate these uncertain migrants to the planned return migrant group, leading to a total of 3002 planned return migrants. Alternatively allocating the uncertain migrants to the planned permanent group would make the conclusion of this paper stronger, since that would lower the estimated planned return rate and thus increase the unexpected return rate. For the main analysis, I drop those who planned to stay for less than a year in order to remove tourists and also make the comparison between planned return and actual return data (discussed in the next section) consistent; this is because the actual return data only records the occupations of return migrants who had been in the United States for at least one year. However, results are robust to keeping those planning to stay less than one year in the sample. Since the final sample contains 26,058 individuals (from 68 ships), the planned return rate is 11.5%, which is significantly lower than the actual return rate, with some estimating it to be above 60% (Bandiera et al., 2013). This provides the first indication that unexpected departures were a significant fraction of outflows in the early 20th century.

4.2. Return migrants at departure: administrative data and IPUMS

The Ellis Island records allow me to estimate the selection of planned return migrants at arrival; I compare these results to estimates of the selection of actual return migrants at departure to understand who was more likely to switch plans after arrival. Data on out-migrants are found in the Annual Report of the Commissioner General of Immigration (RCI) between 1908 and 1932 (United States Bureau of Immigration, 1908–1932). Out-going ships had to deliver passenger lists, much like incoming ships, to port officials. These were forwarded to the Bureau of Immigration, which aggregated them into tables and reported them annually to Congress. Importantly, these are the only data from the United States that systematically observe departures; however, the data provide only aggregations of those who leave, making analysis of micro-determinants of out-migration impossible. Fortunately, the reports list the occupations of out-migrants by ethnicity, where occupation is the last one they had in the United States.

In order to determine whether return migrants were higher or lower quality than permanent migrants, I use occupational scores, which is standard amongst other historical self-selection papers (Abramitzky et al., 2012, 2014; Collins and Wanamaker, 2014). Ideally, one would compare wages or education instead of occupations, but administrative data only record the occupations of returnees. Lacking individual-specific wages, I assign an occupational score to each occupation to reflect its earnings, where all individuals claiming an occupation receive the same score. Accordingly, self-selection estimates are based on how temporary and permanent migrants differed on the occupational ladder. Similar to Collins and Wanamaker (2014), I use income data in the fullcount 1940 IPUMS data to assign each occupation the mean wage based purely on migrants' earnings in 1940. For more information on the creation of the occupational score, please see Appendix A.2.

To determine patterns of selection, I compare return migrants' characteristics with the population they were drawn from: the migrant stock as observed in the 1% IPUMS samples from 1910 to 1930 (Ruggles et al., 2010). One important adjustment that I make to the data reflect the fact that temporary migrants were different from the migrant stock in terms of years of stay: approximately 90% of out-migrants lived in the United States for less than ten years, while the corresponding number for the migrant stock is 30%. To make a better comparison between out-migrants and permanent migrants, I reweight the migrant stock to match the years of stay in the out-migrant data, which in effect places less weight on those staying more than ten years and more weight on those staying less than ten years.

These data on out-migrants are far from perfect. In particular, it is well known that the data likely under-counted the total number of departures (Bandiera et al., 2013). Thus, the RCI data can be thought as a sample of total out-migrants, just as the IPUMS is a sample of the entire population – however, while IPUMS is a random sample, the representativeness of the RCI data is unknown. The main reason suggested by Bandiera et al. for the under counting is careless compiling of ship manifests by the Bureau of Immigration. This type of measurement error would not necessarily bias the representativeness; however, there is no way to verify whether this is true as the original out-going manifests were not archived. However, most results follow from New source countries having higher actual return rates than Old Source countries prior to the quotas, a result well known from the RCI data; fortunately, Bandiera et al. (2013) estimated out-migration rates by country has a similar ranking as in the RCI data.

5. Return migration prior to the migration quotas

I show the raw differences between actual return migrants and stayers prior to the migration quotas in Table 1; these results will later be contrasted with data on planned returners. First, the average return migrant was less skilled than the rest of the migrant population, holding jobs that paid about 9.0 percent less. This finding supports the conclusion that return migrants were negatively self-selected on occupation by Abramitzky et al. (ABE, 2014), and validates the use of residual methods to estimate return migrant selection. However, note that I can only present raw measures of selection without adjusting for observables such as age because I do not observe individual return migrants.

The main reason for out-migrants' lower earnings is that return migrants were more likely to be from the New source countries of Italy, Greece and Russia; most migrants from these countries did not hold high-paying occupations in the United States. However, not only were return migrants less skilled when using variation across countries, but also they were mostly less-skilled when comparing return migrants to the migrant stock within-ethnicity; this also coincides with ABE's results using indirect methods. According to the demographic characteristics, return migrants were also more likely to be male, single, over the age of 45, and from the Northeast.

The differences in skill levels and demographics between return and permanent migrants could be for a simple reason: these were the migrants who had always planned to return home. One can check this using the descriptive statistics of planned return migrants, as shown for those who arrived prior to the migration quotas in Table 2. In addition to the raw differences between planned return and permanent migrants, I show the adjusted difference in the last column by controlling for country of birth, age, and sex.

There are a couple of key points to draw from this table. First, as mentioned prior, the planned return rate was 11.5% for the entire sample between 1917 and 1924; for the years prior to the implementation of the first quota it was 15.4%, much smaller than the estimated actual return rates during the same time period. Second, while the actual out-migrant data showed that they held jobs that paid 9.0 percent less than the migrant stock, planned return migrants held jobs that were statistically similar to planned permanent migrants – this result holds when controlling for country of birth, age and sex, as shown in the fourth column. Thus, prior to the migration quotas two results appear to hold: more migrants returned than initially planned to, and the selection of actual returners at departure was more negative than the selection of planned returners at arrival, suggesting that it was either the lowskilled or those who downgraded their occupation upon arrival who switched plans from staying to returning home.

It is important to note that the occupation listed by incoming migrants is likely their job in the source country rather than in the

1

Self-selection of return migrants prior to quotas, 1920.

Characteristics Return migrants Foreign-born in census Self-selection

Log (occupational score), if have job 6.859 6.949 −0.090*

 (0.155) (0.289) (0.00080)

Claim an occupation 0.765 0.572 0.193*

 (0.424) (0.495) (0.00134)

Male 0.796 0.543 0.253*

 (0.403) (0.498) (0.00131)

Less than 16 0.040 0.114 −0.074*

 (0.195) (0.318) (0.00069)

Age 16–45 0.710 0.755 −0.045*

 (0.454) (0.430) (0.00135)

Age over 45 0.250 0.131 0.119*

 (0.433) (0.337) (0.00125)

Married 0.481 0.605 −0.124*

 (0.500) (0.489) (0.00150)

New source country 0.813 0.439 0.374*

Region of last residence (0.390) (0.496) (0.00132)

Northeast 0.607 0.499 0.108*

 (0.489) (0.500) (0.00148)

Midwest 0.263 0.300 −0.037*

 (0.440) (0.458) (0.00133)

South 0.045 0.0548 −0.010*

 (0.206) (0.228) (0.00063)

West 0.086 0.146 −0.060*

 (0.281) (0.353) (0.00089)

Notes: Out-migrant data is from the Annual Reports of the Commissioner General of Immigration (1920–1921). The Census is from the 1920 IPUMS samples. The left column represents averages from 532,770 individuals; however, these individuals are not individually observed. The middle column represents 140,259 individuals observed in IPUMS. The third column is the difference between the out-migrants and the Census where a positive number indicates that return migrants have more of that characteristic.

** p-value of less than 0.05. *** p-value less than 0.10. * p-value of less than 0.01.

United States. Thus, self-selection estimates at arrival may not be the same comparison as self-selection estimates at departure, which are based on occupations held in the United States. A preferred metric for self-selection would be a permanent level of human capital, such as education, that does not change across borders. While there is no such metric in the actual out-migrant data, the incoming data does list an individual's height, which correlates with productivity, health, and standard of living; further, it does not change to match a country's economic structure, unlike occupation (Kosack and Ward, 2014). Importantly, there was no difference in height between planned return and planned permanent migrants, verifying that there was no selection on average skill into planned return migration at arrival.

The demographic characteristics give a further indication of who was more likely to unexpectedly return home prior to the quotas being put in place. This can be done by comparing the selection of actual return migrants (the last column in Table 1) to the selection of planned return migrants (the second to last column in Table 2). Planned returners were a little more likely to be male than planned stayers, in contrast with actual returners who were much more heavily male than actual stayers. Actual return rates were also higher for singles and those from the Northeast relative to planned return rates. Thus, unexpected returners were more likely to be male, single, and from the Northeast. Interestingly, these migrants were the ones who could more easily switch their decision from staying to returning home given their location and marital status.

Note that the planned return migrant data highlights the importance of family and network variables: being married, having children, joining family in the United States and listing no contact in the source country all have large magnitudes for selection into a permanent stay. Further, listing any nuclear family in the source country correlates with intending tol return home, although this become statistically insignificant after controlling for age, country of birth and sex. Unfortunately, many of these variables cannot be compared to actual return migrants since the administrative data did not collect this information.

The most important difference between actual and planned returners is that migrants from New source countries had a much lower planned return rate than migrants from Old source countries, which is surprising given that New source countries had a much higher actual return rate. Table 3 shows estimates of both the planned and actual return rates prior to the migration quotas by ethnicity (note again that tourists have been dropped from the data); the table is sorted by the difference between planned return and

2

Descriptives of planned return migrants prior to quotas (1917–1921).

Characteristics Planned return Planned permanent Raw difference Adjusted difference

Log(Occ score), if have job 6.814 6.848 −0.0342 0.00102

 (0.316) (0.286) (0.0328) (0.0239)

Height (cm), if recorded 164.5 163.1 1.410 0.289

 (10.51) (11.95) (1.062) (0.793)

Claim an occupation 0.683 0.548 0.136*** 0.0677***

 (0.465) (0.498) (0.0280) (0.0181)

Male 0.550 0.510 0.0398

 (0.498) (0.500) (0.0422)

Age 25.91 25.26 0.649

 (11.86) (13.50) (0.605)

Married 0.293 0.340 −0.0474** −0.0443***

 (0.455) (0.474) (0.0226) (0.0119)

New source ethnicity 0.543 0.680 −0.137* −0.00245

 (0.498) (0.467) (0.0675) (0.00632)

Northeast 0.616 0.628 −0.0119 −0.0158

 (0.486) (0.483) (0.0476) (0.0425)

West 0.151 0.0665 0.0844** 0.0499**

 (0.358) (0.249) (0.0352) (0.0237)

South 0.0406 0.0411 −0.000499 0.00438

 (0.197) (0.198) (0.00728) (0.00444)

Midwest 0.193 0.265 −0.0720** −0.0385

 (0.394) (0.441) (0.0296) (0.0256)

Repeater 0.190 0.167 0.0226 −0.00126

 (0.392) (0.373) (0.0219) (0.0150)

Number of accompanying children 0.242 0.474 −0.232*** −0.150**

 (0.668) (0.984) (0.0772) (0.0637)

Join family in United States 0.758 0.851 −0.0935** −0.0563**

 (0.429) (0.356) (0.0384) (0.0254)

Traveling alone 0.793 0.727 0.0659 0.0439

 (0.406) (0.446) (0.0552) (0.0370)

Nuclear family in source country 0.757 0.665 0.0919*** 0.0290

 (0.429) (0.472) (0.0185) (0.0243)

No contact in source country 0.0929 0.163 −0.0703*** −0.0419*

 (0.290) (0.370) (0.0188) (0.0241)

Traveling to big city (> 100,000) 0.597 0.608 −0.0105 0.0391

 (0.491) (0.488) (0.0345) (0.0236)

Observations 1948 10,669

Notes: Data is a random sample of ships from Ellis Island Records (1917–1924). The rightmost column controls for country of birth, age, and sex. Heights were generally not recorded for children.

*** p-value of less than 0.01.

** p-value of less than 0.05. * p-value less than 0.10.

actual return rates, which reflects the degree of unplanned returns. Further, note that the estimated actual return rate is a lower bound of the true return rate given that out-migrants in the return migrant data are under counted (Bandiera et al., 2013).

This table shows that many New source ethnicities had a high rate of unplanned returns; interestingly, unplanned returns were much more common among Eastern European ethnicities rather than places with an established flow like Italy. While the planned return rates for New source countries may be surprising because they were so low, to a certain extent it is much more surprising that the actual return rates were so high: given the premium for real wages in the United States, the economic return to migration was likely higher for these poorer source countries. Thus, it appears that the higher actual return rate could be due to unexpected factors prior to arrival, such as job losses or distaste for living in the United States, or unexpectedly better conditions in the source country.

The evidence so far shows the differences in the means between returners and stayers, but selection may differ along the distribution of human capital. For example, other studies show a U-shaped pattern of self-selection on income; the logic here is that those with the highest income hit a savings target quickly, while those on the lower end “fail” and return home (Bijwaard and Wahba, 2014). To provide an idea about the pattern of return migrant selection, especially over the 1920s before and after the migration quotas were put into place, I plot the skill densities of actual return migrants and the migrant stock, after controlling for ethnicity, in Fig. 2. Actual return migrants, particularly in 1920, were from a concentrated set of occupations which included laborers, farm laborers, and miners. This negative pattern of self-selection became less strong by 1930, as seen in the right hand panels. Rather than a mass of return migrants leaving as laborers, there was a wider variety of occupations; yet the selection of return migrants was still strongly negative.

While the actual out-migrant data show that return migrants were often negatively selected and came from a concentrated set of occupations, the planned return migrant data tell a different story. The upper panels of Fig. 3 plot the residual logged occupational

3

Return rates for those who arrived between 1917 and 1921.

Ethnicity I II (II − I) N in ship records

Planned return rate LB actual return rate LB unplanned return rate

Romanian 0.03 0.30 0.26 254

Russian 0.04 0.28 0.24 56

Greek 0.15 0.36 0.21 116

Polish 0.01 0.16 0.15 731

Syrian 0.00 0.11 0.10 282

English 0.13 0.16 0.04 1310

Italian 0.16 0.18 0.01 4310

Armenian 0.02 0.02 0.00 153

German 0.10 0.09 −0.01 248

Hebrew 0.02 0.01 −0.02 1011

Finnish 0.13 0.11 −0.02 229

Slovak 0.05 0.04 −0.02 585

Dutch and Flemish 0.19 0.14 −0.05 213

Scottish 0.15 0.07 −0.08 216

Scandinavian 0.22 0.14 −0.08 1142

Welsh 0.23 0.08 −0.15 57

French 0.31 0.15 −0.16 363

Spanish 0.57 0.32 −0.24 391

Irish 0.34 0.06 −0.28 718

Notes: Data is from Ellis Island passenger manifests and the Annual Reports of the Commissioner General of Immigration (1917–1928). The expected return rate is the percent of incoming migrants who planned to return home. See online Appendix B for the calculation of lower-bound (LB) actual return rate, which approximates the return rate after ten years.

score, after controlling for ethnicity, of planned return and planned permanent migrants before and after the 1921 migration quota. Prior to the migration quotas, planned return and planned permanent migrants had relatively similar occupational score densities, with the planned return migrant density having a slightly wider set of occupations. Due to this higher standard deviation, planned

Fig. 2. Actual return migrant densities, 1920 and 1930. Notes: Data is from Report of the Commissioner General of Immigration (1921, 1931) and IPUMS (1920 , 1930). The panels plot the residual occupational score after controlling for ethnicity and year of departure. The bottom two panels plot the difference in the return migrant and migrant stock densities.

Fig. 3. Planned return migrant densities, pre and postmigration quota. Notes: Data is from Ellis Island Records (1917–1924). The panels plot the residual occupational score after controlling for ethnicity and year of arrival. The bottom two panels plot the difference in the planned return migrant and planned permanent migrant densities.

return migrants were self-selected in the way found by other papers for current-day actual return migrants: a U-shaped pattern where the upper and lower ends of the skill distribution planned to return at higher rates. The U-shaped pattern of self-selection for planned return migrants holds after the 1921 migration quotas, perhaps suggesting little effect on those who planned to return home.

6. The effect of the 1920s migration quotas

6.1. Empirical strategy

The visual evidence suggests that the quotas had no impact on selection into planned return migration on skill; here, I more rigorously test how the 1920s migration quotas affected planned return rates. Descriptions of these laws can be found elsewhere (Greenwood and Ward, 2015; Massey, 2016), but of primary importance for this paper is that the migration quotas were unexpectedly implemented, varied across countries in absolute quota limits, and varied across time with the 1921 and 1924 quota laws.25 These quotas were restrictive enough that they led to a 60% drop in immigrant flows in a four-year time span.

To estimate how the quotas affected temporary migration plans, I create a measure that uses the variation in quota limits across the different iterations of the law to gauge the law's restrictiveness on a country. I proxy the potential migrant flow for a country using its flow from July 1920 to June 1921, or the fiscal year prior to the quota law fully going into effect:

QuotaLimitjt QuotaRestrictionjt =1−

Immigrantsj,FiscalYear1921 (1)

This measure is specific for the country j and time period t due to the changes in quota laws in 1921 and 1924. The variable takes values between zero and one, where 0.98 is its highest value for Italian migrants after 1924, implying that 98% of potential migrants were restricted from entry.

I test how the migration quotas affected return intentions by running variations of the following linear probability model:

PlannedReturnijt = β0 + β1QuotaRestrictionjt + β2LowSkillijt + β3MediumSkillijt

+ β4QuotaRestrictionjt × LowSkillijt + β5QuotaRestrictionjt × MediumSkillijt

+ ηt + εijt + ΓXijt + γj (2)

where PlannedReturnijt is a zero-one variable where one indicates that migrant i of birth country j arriving in fiscal year t planned to return home. The regression splits migrants into three different skill groups, based on whether they are in the top, middle or bottom third of occupational scores within the dataset to test the U-shaped pattern of self-selection. The most common occupations in the lowest third are servants and farm laborers, the middle third are laborers and farmers and the top third are merchants and dealers. Those without jobs are dropped from the regression.

The main regressor of interest is QuotaRestrictionjt and its interactions with each of the different skill categories. In the format of the regression, HighSkillijt is the excluded skill group, so the main effect of QuotaRestrictionjt, β1, gives the estimate of restricting migration on the high skilled group. The coefficients of the other interactions are β4 and β5, which estimate whether the effect of quotas is statistically different from the high-skilled category.

Also in the regression I include dummy variables for country j to control for country-specific heterogeneity that is time-invariant, such as culture and language, and fiscal year of arrival dummy variables t which control for general time trends. I also control for various individual observables in Xijt, which include a migrant's sex, age, network, and other observables listed in Table 2. Importantly, I also control for the source country's GDP changes to account for factors drawing migrants back to Europe.

For the estimated effects of QuotaRestrictionjt to have a causal interpretation, there must not be any time-varying unobservables that are also correlated with the intensity of quota limits. To justify the parallel trends assumption, I plot the planned return rates by New and Old source countries in Fig. A3, which show similar trends between the two groups. However, given the differences in migrant characteristics between Old source countries and New source countries, the effect of the migration quotas can only be generalized to restricting migration of relatively poorer source countries rather than restricting all countries; yet one may argue that this is a more relevant effect to estimate as interest groups often aim to restrict migration from poorer countries.

6.2. Results for planned return migration

Table 4 shows results from the regression. The first column tests the main effect of a more restrictive quota on planning to return, before allowing heterogeneity to exist across skill groups. The coefficient on quota restriction is statistically insignificant and close to zero, suggesting that migration quotas did not cause the planned return rate to decrease. This result contrasts with the fall in actual return rates caused by quotas found by Greenwood and Ward (2015). Thus, the policy seems to have had little influence on return migration due to expectations of migrants at arrival, but rather changed return migration by altering a migrant's experience against expectations.

The second column adds the lower and medium skilled categories and shows that those who were middle skilled were less likely to plan to return home relative to the high skilled by about 4.3 percentage points. Further, the low-skilled group has statistically the same return rate as the high-skilled category. These two results empirically confirm that planned return migrants had a U-shaped pattern of self-selection within country, where the highest and least skilled were the most likely to plan to return home.

The third column tests the interactions between a more restrictive quota and the different skill groups. The coefficients on quota restriction and the interactions with skill groups are all statistically insignificant; further, the total effect of quota restrictions are not statistically significant for the medium and low-skilled group. These results suggest that the quotas did not drastically alter the rate of planned return migration across skill groups, at least for the first couple years of the quotas from 1921 to 1924. Once again, this contrasts with other evidence that the migration quotas caused more low-skilled migrants to stay. Finally, the fourth column includes control variables for various observables, and the results still hold.

When one contrasts these results with the estimated effects on actual return rates from Greenwood and Ward (2015), the main difference is that actual return rates dropped while the planned return rates did not. There are two reasons why this may have occurred: first, there were fewer unexpected returns, perhaps due to fewer failures in the labor market, or second that there were

Table 4

Self-selection into planned return migration, 1917–1924.

 I II III IV

Quota restriction −0.00552 −0.0131 −0.0229 −0.0123

 (0.0648) (0.0645) (0.0626) (0.0604)

Low skilled 0.00477 0.0103 0.00116

  (0.0153) (0.0188) (0.0191)

Medium skilled −0.0435* −0.0515* −0.0461*

  (0.0129) (0.0149) (0.0151)

Quota restriction × Low-skilled −0.0321 −0.0260

   (0.0365) (0.0375)

Quota restriction × medium-skilled 0.0431 0.0425

   (0.0362) (0.0350)

Age 0.000746

(0.00284)

Age squared −1.64e-06

(3.18e-05)

Male −0.0179 (0.0149)

Repeater −0.0116 (0.0116)

Join family −0.0296**

(0.0137)

Ever married −0.0177 (0.0134)

Traveling alone 0.0100

(0.0140)

Number of children −0.0102***

(0.00577)

Midwest −0.0106 (0.0103)

West 0.0403

(0.0273)

South 0.0118

(0.0152)

Nuclear family at home 0.0306**

(0.0121)

Extended family at home 0.0246

(0.0192)

Big City (> 100,000) 0.0148*

(0.00831)

GDP in source country −5.86e−05

(6.23e−05)

Country of birth indicators X X X X

Fiscal year indicators X X X X

Observations 15,246 15,246 15,246 15,246

R-squared 0.159 0.162 0.163 0.168

Notes: Data is from incoming passenger manifests from Ellis Island (1917–1924). The dependent variable is whether or not an individual planned to return home.

* p-value of less than 0.01.

** p-value of less than 0.05. *** p-value less than 0.10.

more unexpected stays, perhaps due to greater success in the labor market. In Appendix C, I provide suggestive evidence that there were both fewer unexpected returns and more unexpected stays. Using the sample linked to 1930, I estimate that the quotas are correlated with successfully finding more planned stayers and planned returners in the country by 1930. This suggests that planned returners were more likely to stay following the quotas (i.e., more unexpected successes), and planned stayers were also more likely to stay (i.e., fewer unexpected failures). Yet these results are still only suggestive due to other reasons for failing to link besides return migration.

6.3. Results for expected years of stay

The previous results show that the migration quotas did not change the rate at which incoming migrants planned to return or stay, but this is only one way to measure how quotas affected return migration. In this section, I explore a different margin: the years of intended stay. Using this data, besides being able to test the effects of the quotas, I can also test whether planned return migrants exhibited behavior consistent with holding a savings target. If all migrants held a common savings target, then those who were more skilled would plan to stay fewer years as it would take fewer years to achieve the savings target. Another factor consistent with a target savings hypothesis is that those who had higher costs, such as those traveling further inland from New York City, would also stay longer in order to recuperate the traveling costs.

I run the same estimating framework as Eq. (2) but with the number of years of stay as the dependent variable. Note that the mean years of expected stay is 4.3, and that those staying under one year have already been dropped from the data set to exclude tourists. Further, those who planned to stay permanently have been dropped from the following analysis.

The results are presented in Table 5. The first column shows the main effect of a more restrictive quota on the planned years of stay, and finds that the quotas also had no overall effect on this variable. Just like the results for planned return rates, this result can also be compared to Greenwood and Ward (2015) as they argue that the quotas led individuals to stay longer in the United States.

The second through fourth columns add variables controlling for the skill level and other observables of entering migrants. The second column shows evidence consistent with a target savings hypothesis: low- and medium- skilled migrants planned to stay approximately one year more than high-skilled migrants, perhaps because high-skilled migrants were more easily able to hit a savings target. When adding the extra observables in column four, we see that those traveling to the Midwest and West planned to stay longer than those traveling to the Northeast, by approximately one year, also consistent with migrants taking longer trips to cover higher traveling costs.

The third column examines the heterogeneous effect of the immigration quotas and, unlike the effect of quotas on planned return rates, finds different effects for length of stay across skill groups. In particular, the results show that the more restrictive a quota, the fewer years the least skilled migrants planned to stay; the estimate suggests that a 60% restriction lowered the planned length of stay by one year relative to the high-skilled group. It is unclear what is driving this result: on one hand, if the costs of entry were higher due to the policy change, then migrants would stay longer to cover the higher entry costs; however, for the migrants who did plan to return home and had a savings target, it could be that they expected to hit the savings target more quickly.

6.4. Discussion of the effects of the 1920s quotas

This paper shows that in the transition from a free to a restricted migration system, migrants were less likely to unexpectedly return home and more likely to unexpectedly stay longer. Particularly, it appears that single males from poorer countries returned at unexpectedly high rates prior to the quotas, and following the quotas there were fewer of these types of returns. Here I discuss possible reasons for these patterns.

One possibility is that the return migrants who planned to take multiple trips in and out of the United States were trapped inside due to increased re-entry costs. This suggestion, while it likely did have an impact, is probably not the main driver of the selection patterns for a few reasons. First, most repeated entries were those who left and re-entered within a year, and such short trips back home were actually not restricted by the migration quotas as long as one applied for a permit. Accordingly, repeat entrants were more prevalent in the migrant flow after the implementation of the migration quotas (Ward, 2016). Second, the probability of a return migrant re-entering was similar across New and Old source countries (about 15–20 percent); therefore, quotas were not strongly correlated with repeat migration patterns.

Rather, one reason for fewer unexpected returns is that the quotas screened out those who were both less committed to staying and had low switching costs. The new process of entering the country was extensive: after the 1924 quota, migrants now had to apply for entrance at a consular office in the source country, supply documents such as a birth certificate, military, and criminal record, and sign the application and swear an oath administered by the consular officer. One had to do all of this and also be one of the first applications; indeed, the quotas filled rapidly and forced individuals to apply quickly. The process was far different from an earlier time when one could simply buy a ticket and arrive ten days later. This increase in requirements likely screened out those less certain about life abroad, and further led to a large drop in the number of single males migrating from Southern and Eastern Europe – those who could easily switch their decision from staying permanently to returning home.

While the quotas may have screened out the less certain, they also improved the experience in the United States for those lucky enough to enter. The quotas eliminated numerous competitors in the labor market: the inflow dropped by 60% over four years and further shifted the composition towards fewer labor force participants. Given the large responses of outflows to unemployment (Bijwaard et al., 2014), it is possible that the migration quotas led to less unemployment specifically for other migrants in the labor market. Many migration studies have found a high degree of substitutability between migrants and other migrants, as they have similar locations and skills, as opposed to migrants and natives. For example, Beaman (2012) finds that having more recentlyarrived migrants in a network actually worsens outcomes for newly-arriving migrants, as both groups compete for the same jobs. Indeed, those with highly substitutable characteristics tended to leave more rapidly in the early 20th century: prime-aged males living in the Northeast faced substantial competition from inflows of migrants, and they were also driven out.

Table 5

Expected years of stay, 1917–1924.

I II III IV

Quota restriction −0.873 −0.701 0.212 0.0351

(0.660) (0.664) (0.919) (1.005)

Low skilled 1.009* 1.139* 1.236*

(0.243) (0.244) (0.161)

Medium skilled 1.227* 1.244* 1.054*

(0.259) (0.279) (0.272)

Quota restriction × low-skilled −1.898* −1.913*

(0.594) (0.567)

Quota restriction × medium-skilled −0.612 −0.216

(0.791) (0.802)

Age −0.0468

(0.0342)

Age squared 0.000127

(0.000452)

Male 0.476**

(0.194)

Repeater 0.203

(0.185)

Join family 0.448*

(0.141)

Ever married 0.348***

(0.200)

Traveling alone 0.143

(0.188)

Number of children 0.304

(0.335)

Midwest 0.573*

(0.202)

West 1.252*

(0.324)

South −0.0773

(0.413)

Nuclear family at home −0.255

(0.562)

Extended family at home −0.860

(0.689)

Big city (> 100,000) 0.138

(0.124)

GDP in source country 0.000577

(0.000883)

Country of birth indicators X X X X

Fiscal year indicators X X X X

Observations 1398 1398 1398 1398

R-squared 0.265 0.291 0.295 0.336

Notes: Data is from incoming passenger manifests from Ellis Island (1917–1924). The dependent variable is the years a return migrant planned to stay prior to returning. *

p-value of less than 0.01. **

p-value of less than 0.05. ***

p-value less than 0.10.

Indeed, soon after the 1921 quota was put into place, many contemporaries suggested that there should be looser migration restrictions to reduce rising labor costs (Jerome, 1934). If one compares the post-restrictions migration totals to a simple counterfactual, the 1900s decade, the 1920s about 2.8 million fewer net immigrants (53% less); therefore, the reduction in labor supply caused by the migration quotas left open a significant gap in the labor market (Collins, 1997). The reduction of inflows from Europe encouraged the migration of those elsewhere: for example, the Great Migration of about 900,000 blacks from the South to the North in the 1920s has been linked to decreased migrant inflows due to the quotas (Collins, 1997). Further, indirect evidence for a better labor market for migrants can also be seen in the rise of inflows from countries not limited by quotas: Mexico and

Canada's average migrant flows to the United States increased by 67% and 62% for the years following the 1921 quota. 7. Concluding remarks

This paper establishes that most return migrants unexpectedly returned home in the early 20th century. This contrasts with a common understanding of temporary migration during this time period and paints a more pessimistic picture of migration to the United States. Assimilation into the United States was particularly difficult in the years prior to the migration quotas for migrants from poorer source countries in Eastern and Southern Europe. Combined with evidence that migrants rarely upgraded their occupations (Abramitzky et al., 2014) and that the return to migration was relatively low compared with today (Abramitzky et al., 2012), it appears that outcomes for migrants were often worse than expected and many returned home. Importantly, the rate of occupational upgrading was low, but not for lack of an incentive to invest in United States-specific human capital as many planned to stay permanently.

However, this pessimistic view applies only to the average migrant; there were certainly plenty of successful temporary and permanent migrants. Moreover, this pessimistic view of the Southern and Eastern European migrant experience may not apply to the period following the implementation of the migration quotas: in particular, this paper shows that there were fewer unplanned returns, perhaps due to improved outcomes in the United States. This suggests that if migration policy creates a large and dramatic shock to incoming flows, the group most affected by this policy are prior migrants who have already entered, or those lucky enough to enter under the policy. These migrants are in less competition with inflows, and thus will be more likely to stay permanently within the United States. On the other hand, if the United States liberalizes its migration policy, it is possible that more migrants would be driven to return home due to intense competition among entering cohorts.

Acknowledgments

I would like to thank Ann Carlos for numerous conversations and insightful guidance on this project. Other helpful comments were given by Ran Abramitzky, Brian Cadena, Dustin Frye, Mike Greenwood, Myron Gutmann, Tim Hatton, Murat Iyigun, Ian Keay, Priti Kalsi, Edward Kosack, Amber McKinney, Terra McKinnish, Steven Smith, John Tang, Matthew Van Wyhe, Marianne Wanamaker, and several anonymous referees. Special thanks go to Lee Alston for helping gain access to the Census data. Part of this paper was previously circulated under the title of “The U-Shaped Self-Selection of Return Migrants.” I would also like to thank those individuals at the Australian National University, World Cliometrics Conference, University of Colorado Economic History Workshop, the ANU Height and Development Workshop, 2014 WEAI meetings, and 8th Annual Conference for Migration and Development who offered helpful comments. All errors are my own.

Appendix A. Supplementary data

Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.eeh.2016.09. 002.

References

Abramitzky, R., Boustan, L.P., 2016. Immigration in American economic history. J. Econ. Lit., doi: 10.3386/w21882.

Abramitzky, R., Boustan, L.P., Eriksson, K., 2012. Europe's tired, poor, huddled masses, self-selection and economic outcomes in the age of mass migration. Am. Econ. Rev. 102 (5), 1832–1856.

Abramitzky, R., Boustan, L.P., Eriksson, K., 2014. A nation of immigrants, assimilation and economic outcomes in the age of mass migration. J. Polit. Econ. 122 (3), 467–506.

Abramitzky, R., Boustan, L.P., Eriksson, K., 2016. Cultural Assimilation During the Age of Mass Migration. Working Paper 22381, National Bureau of Economic Research. July. 〈http://www.nber.org/papers/w22381〉

Adda, J., Dustmann, C., Görlach, J.-S., 2015. The dynamics of return migration, human capital accumulation, and wage assimilation.

Angrist, J., 2002. How do sex ratios affect marriage and labor markets? evidence from America's second generation. Q. J. Econ., 997–1038.

Bailey, W., 1912. The bird of passage. Am. J. Sociol. 18 (3), 391–397.

Baines, D., 1995. Emigration from Europe vol. 11. Cambridge University Press, Cambridge, 1815–1930.

Balch, E.G., 1910. Our Slavic Fellow Citizens. Charities Publication Committee, New York.

Bandiera, O., Rasul, I., Viarengo, M., 2013. The making of modern America: migratory flows in the age of mass migration. J. Dev. Econ. 102, 23–47.

Beaman, L.A., 2012. Social networks and the dynamics of labour market outcomes: evidence from refugees resettled in the us. Rev. Econ. Stud. 79 (1), 128–161. Biavaschi, C., 2013a. Fifty Years of Compositional Changes in US Out-migration, 1908–1957. Technical Report, Discussion Paper Series, Forschungsinstitut zur Zukunft der Arbeit.

Biavaschi, C., 2013b. The labor demand was downward sloping: disentangling migrants' inflows and outflows, 1929–1957. Econ. Lett. 118 (3), 531–534.

Biavaschi, C., Giulietti, C., Siddique, Z., 2013. The Economic Payoff of Name Americanization. Technical Report, IZA Discussion Paper.

Bijwaard, G.E., Schluter, C., Wahba, J., 2014. The impact of labor market dynamics on the return migration of immigrants. Rev. Econ. Stat. 96 (3), 483–494.

Bijwaard, G.E., Wahba, J., 2014. Do high-income or low-income immigrants leave faster? J. Dev. Econ. 108, 54–68.

Borjas, G.J., 1987. Self-selection and the earnings of immigrants. Am. Econ. Rev. 77 (4), 531–553.

Borjas, G.J., 1991. Immigration and self-selection. In: Immigration, Trade, and the Labor Market. University of Chicago Press, Chicago, pp. 29–76.

Borjas, G.J., Bratsberg, B., 1996. Who leaves? The outmigration of the foreign-born. Rev. Econ. Stat. 78 (1), 165–176, URL 〈http://www.jstor.org/stable/2109856〉.

Carter, Susan B., Gartner, Scott S., Haines, Michael R., Olmstead, Alan L., Sutch, Richard, Wright, Gavin, 2006. Historical statistics of the United States: Earliest times to the present. Cambridge University Press, Cambridge.

Cerase, F.P., 1974. Expectations and reality: a case study of return migration from the United States to Southern Italy. Int. Migr. Rev. 8 (2), 245–262.

Chiquiar, D., Hanson, G.H., 2005. International migration, self-selection, and the distribution of wages: evidence from Mexico and the United States. J. Polit. Econ. 113 (2), 239–281.

Cohn, R.L., 1984. Mortality on immigrant voyages to New york, 1836–1853. J. Econ. Hist. 44 (2), 289–300, URL 〈http://www.jstor.org/stable/2120706〉.

Cohn, R.L., 1995. Occupational evidence on the causes of immigration to the United States, 1836–1853. Explor. Econ. Hist. 32 (3) , 383–408, URL 〈http://www.sciencedirect.com/science/article/pii/S0014498385710170〉.

Cohn, R.L., 2009. Mass Migration Under Sail: European Immigration to the Antebellum United States. Cambridge University Press, New York.

Collins, W.J., 1997. When the tide turned: immigration and the delay of the great black migration. J. Econ. Hist. 57 (03), 607–632.

Collins, W.J., Wanamaker, M.H., 2014. Selection and economic gains in the great migration of African Americans: new evidence from linked census data. Am. Econ. J.: Appl. Econ. 6 (1), 220–252.

Dustmann, C., 1996. Return migration, the European experience. Econ. Policy, 213–250.

Dustmann, C., Görlach, J.-S., 2016. The economics of temporary migrations. J. Econ. Lit. 54 (March (1)) , 98–136, URL〈http://www.aeaweb.org/articles/?doi=10.1257/jel.54.1.98〉.

Erickson, C.J., 1972. Who were the English and Scots Emigrants to the United States in the Late Nineteenth Century?, pp. 347–381.

Ferrie, Joseph P., 1999. Yankeys Now: Immigrants in the Antebellum US 1840–1860. Oxford University Press, New York.

Fouka, Vasiliki, 2015. Backlash: The Unintended Effects of Language Prohibition in US Schools after World War I. Working Paper.

Friedberg, Rachel M., 2001. The impact of mass migration on the Israeli labor market. Q. J. Econ. 116 (4), 1373–1408.

Fryer, R.G., Levitt, S.D., 2004. The causes and consequences of distinctively black names. Q. J. Econ., 767–805.

Galor, Oded, Stark, Oded, 1990. Migrants' savings, the probability of return migration and migrants performance. Int. Econ. Rev. 31 (2), 463–467.

Gibson, J., McKenzie, D., 2011. The microeconomic determinants of emigration and return migration of the best and brightest, evidence from the Pacific. J. Dev. Econ. 95 (1), 18–29.

Goldin, C., 1994. The political economy of immigration restriction in the United States, 1890 to 1921. In: The Regulated Economy: A Historical Approach to Political Economy. University of Chicago Press, Chicago, pp. 223–258.

Görlach, J.-S., 2016. Borrowing Constraints, Migrant Selection, and the Dynamics of Return and Repeat migration. Mimeo.

Gould, J.D., 1980. European inter-continental emigration. The road home: return migration from the USA. J. Eur. Econ. Hist. 9 (1), 41–112.

Greenwood, M.J., Ward, Z., 2015. Immigration quotas, world war I, and emigrant flows from the United States in the early 20th century. Explor. Econ. Hist. 55 (0) , 76–96, URL 〈http://www.sciencedirect.com/science/article/pii/S0014498314000199〉.

Grogger, J., Hanson, G.H., 2011. Income maximization and the selection and sorting of international migrants. J. Dev. Econ. 95 (1), 42–57.

Hatton, Timothy J., Williamson, Jeffrey G., 1998. The Age of Mass Migration: Causes and Economic Impact. Oxford University Press, New York.

Higham, J., 2002. Strangers in the Land: Patterns of American Nativism, 1860–1925. Rutgers University Press, New Brunswick.

Hill, John K., 1987. Immigrant decisions concerning duration of stay and migratory frequency. J. Dev. Econ. 25 (1), 221–234.

Jerome, Harry, 1926. Migration and Business Cycles. NBER Books, Cambridge.

Jerome, H., 1934. Mechanization in Industry. NBER Books, New York.

Kosack, E., Ward, Z., 2014. Who crossed the border? Self-selection of mexican migrants in the early twentieth century. J. Econ. Hist. 74 (04), 1015–1044. Kraljic, F., 1978. Croatian migration to and from the United States 1900–1914.

Kuznets, S., Rubin, E., 1954. Immigration and the Foreign Born. NBER Books.

Lleras-Muney, A., Shertzer, A., 2015. Did the americanization movement succeed? an evaluation of the effect of english-only and compulsory schooling laws on immigrants. Am. Econ. J.: Econ. Policy 7 (3), 258–290, URL 〈http://www.aeaweb.org/articles.php?doi=10.1257/pol.20120219〉.

Massey, C.G., 2016. Immigration quotas and immigrant selection. Explor. Econ. Hist. 60 , 21–40, URL 〈http://www.sciencedirect.com/science/article/pii/S0014498315000455〉.

McKenzie, D., Gibson, J., Stillman, S., 2013. A land of milk and honey with streets paved with gold, do emigrants have over-optimistic expectations about incomes abroad? J. Dev. Econ. 102, 116–127.

Moser, P., 2012. Taste-based discrimination evidence from a shift in ethnic preferences after wwi. Explor. Econ. Hist. 49 (2), 167–188.

Nekoei, A., 2013. Immigrants' labor supply and exchange rate volatility. Am. Econ. J.: Appl. Econ. 5 (4), 144–164.

Ottaviano, G.I., Peri, G., 2012. Rethinking the effect of immigration on wages. J. Eur. Econ. Assoc. 10 (1), 152–197.

Piore, M.J., 1979. Birds of Passage: Migrant Labor and Industrial Societies. Cambridge University Press, New York.

Ruggles, S., Alexander, T., Genadek, K., Goeken, R., Schroeder, M., Sobek, M., 2010. Integrated Public Use Microdata Series (ipums): Version 5.0 [Machine-Readable Database]. University of Minnesota, Minneapolis.

Saloutos, T., 1956. They Remember America: The Story of the Repatriated Greek-Americans. University of California Press, Berkeley.

Sarna, J.D., 1981. The myth of no return – Jewish return migration to eastern Europe, 1881–1914. Am. Jew. Hist. 71 (2), 256–268. Steiner, E.A., 1906. On the Trail of the Immigrant. FH Revell Company, New York.

Stolz, Yvonne, Baten, Joerg, 2012. Brain drain in the age of mass migration: Does relative inequality explain migrant selectivity? Explor. Econ. Hist. 49 (2), 205–220.

Thom, K., 2010. Repeated Circular Migration: Theory and Evidence from Undocumented Migrants. New York University, mimeo.

United States Bureau of Immigration, 1908–1932. Annual Report of the Commissioner General of Immigration. U.S. Government Printing Office.

Ward, Z., 2016. There and back (and back) again: Repeat migration to the United States 1897–1936. Unpublished Manuscript

Williamson, J.G., 1995. The evolution of global labor markets since 1830: background evidence and hypotheses. Explor. Econ. Hist. 32 (2), 141–196.

Willcox, Walter F., 1931. Immigration into the United States. In: International Migrations, vol. II: Interpretations. NBER, Cambridge, pp. 83–122.

Wyman, M., 1993. Round-Trip to America: The Immigrants Return to Europe. Cornell University Press, Ithaca, 1880–1930.

Yang, D., 2006. Why do migrants return to poor countries? evidence from Philippine migrants' responses to exchange rate shocks. Rev. Econ. Stat. 88 (4), 715–735.

Explorations in Economic History 64 (2017) 37–52

Birds of passage: Return migration, self-selection and immigration MARK quotas

Zachary Ward

Research School of Economics, HW Arndt Building 25A, College of Business and Economics, The Australian National University, Canberra, ACT 2600, Australia

A R T I C L E I N F O A B S T R A C T

JEL: A key feature of migration in the late 19th and early 20th century is that many migrants returned F22 to Europe after a few years in the United States. A common view is that most temporary migrants J61 planned, upon entry, to eventually return home, yet there is little direct evidence to support this N30 claim. I collect the first dataset on migrants' intentions to stay or return home from Ellis Island Keywords: arrival records between 1917 and 1924. I find that fewer migrants planned to return home than Return migration actually did; many migrants, especially from Eastern and Southern Europe, left the United States Self-selection unexpectedly. The high rate of unplanned returns implies that the first few years after arrival Immigration quotas were more di fficult than expected. However, this high rate of unexpected returns lowered after

Planned returns

the 1920s migration quotas, suggesting improved outcomes for those lucky enough to enter.

“The steerage passengers may be roughly divided into two classes: those who go home because they have succeeded, and those who go home because they have failed.”

– Edward A. Steiner, On the Trail of the Immigrant (1906)

1. Introduction

Fundamentally, migration is about the flow of people across borders, but a fact often overlooked is that flows occur both into and out of a country. Return flows can be substantial: in the early 20th century United States they were up to three-forths of the size of inflows, suggesting that the average incoming migrant was temporary rather than permanent. Numerous scholars have claimed that many of these returns were planned from the beginning, as migrants purposely stayed a few years to accumulate savings for investment or consumption at home (Baines, 1995; Piore, 1979; Wyman, 1993).1 However, these claims are based on indirect evidence; an equally likely hypothesis is that migrants' return trips were decided upon after arrival, perhaps because conditions in the United States were worse than expected (Borjas and Bratsberg, 1996; Cerase, 1974).

The answer to this question is key for understanding migrant assimilation: if temporary migrants planned to return home, then they would have exerted little effort to assimilate. For example, planned temporary migrants are less likely to invest in United States-

E-mail address: Zach.A.Ward@gmail.com.

1 There are a range of estimates for the ratio of out-migrants to in-migrants in the early 20th century, from 39 percent (Willcox, 1931, Table 17), 45 percent (Kuznets and Rubin, 1954, Table 6), and 60 to 75 percent (Bandiera et al.,, 2013, Table 4). These are demographic estimates based on the change in foreign-born population from census to census; the change in population should reflect additions from new arrivals and subtractions from departures and deaths. Most recently Bandiera et al. (2013) aim to improve on prior estimates by using newly digitized data on the universe of Ellis Island arrivals, which contain more accurate counts of arrivals than the official US data.

http://dx.doi.org/10.1016/j.eeh.2016.09.002

Received 19 April 2016; Received in revised form 9 August 2016; Accepted 24 September 2016

Available online 28 September 2016

0014 -4983 / © 2016 Elsevier Inc. All rights reserved.

specific human capital, such as the ability to speak English, which has little return at home. On the other hand, if most returned home unexpectedly, then it would suggest that the United States was not a “land of opportunity” because outcomes for migrants were poor despite efforts to assimilate. Alternatively, unplanned returners may have had unrealistic expectations about the United States because information about the migrant experience was biased (McKenzie et al., 2013).

I aim to determine whether most return migration in the early 20th century was planned or unplanned by collecting a unique dataset; in 1917, the Bureau of Immigration started to ask entering migrants whether they planned to stay permanently or to return home. Migrants who planned to return home were also required to report the length of time, whether days, weeks or years, that they intended to stay.

Using a sample of 27,000 Ellis Island arrivals between 1917 and 1924, I can answer numerous questions about temporary migration including at what rate migrants planned to return, who selected into planned temporary migration, and how migration policy, such as the 1920s migration quotas, affected planned returns. I can also compare the rates and characteristics of planned returners to those of actual returners by using a dataset on departures back to Europe. This out-going data is unique because the early 20th century was the only time period in United States history when the government collected information on departures, a practice that no longer exists today.

I find that most migrants did not plan to return home at arrival: in the years prior to the migration quota of 1921, only 15 percent of entrants intended to eventually return home. This can be contrasted with estimates that at least 40 percent of migrants actually returned home, suggesting that most eventual returns prior to the immigration quotas were unexpected at the time of arrival. It appears that the discrepancy between planned and actual return rates is mostly due to single males from Southern and Eastern Europe unexpectedly returning home at high rates; this could reflect a large amount of failures in the labor market, perhaps due to unemployment after arrival. In contrast with the Southern and Eastern European experience, many Northern and Western European countries unexpectedly stayed in the United States, perhaps because outcomes were better than expected.

After the implementation of the quotas in 1921 and 1924, which caused a 60% fall of inflows over a four-year period, the pattern of a high rate of unplanned returns for Southern and Eastern Europeans completely changed. After the quotas, the rate of actual return migration decreased (Greenwood and Ward, 2015); but in this paper, I show that the quotas had no effect on the rate of planned return migration. A combination of these two results suggests that the migration quotas lowered unplanned returns, especially of those holding low-skilled jobs in the United States. The fact that more low-skilled migrants remained in the United States helps to explain why the self-selection of return migrants became less negative for the countries most restricted, as seen in Fig. 1.

A simple reason for the results could be that the labor market was difficult prior to the quotas and then became easier for workers as the quotas reduced incoming competitors. In other words, the quotas improved outcomes for those lucky enough to enter. Evidence of a tightening labor market can be clearly seen when examining the rapid increase of inflows of those with highly substitutable characteristics following the quotas: black migration from southern states, Mexican migration from across the border, and Canadian migration from the north (Collins, 1997; Kosack and Ward, 2014). Thus, this paper supports recent arguments that a voluminous inflow affects other migrants more strongly than natives (Ottaviano and Peri, 2012).

Another reason for the decreased rate of unplanned returns is that the quotas screened out those most likely to unexpectedly return: single males from Southern and Eastern Europe. Prior to the quotas, these migrants had high return rates but relatively low planned return rates, suggesting that they often changed duration plans after arrival – likely because it was cheaper for them to switch decisions compared with the costs for other migrants with families. Following the quotas, there were fewer single male entrants, reducing the number of unplanned returns.

This paper complements and extends prior work on return migration during this time period by Greenwood and Ward ( GW, 2015) in many ways. Primarily, GW are silent on whether migrants planned to return home or stay at arrival; this paper's main contribution is the data on migrants' duration plans, which is the first data, to my knowledge, on migrants' return intentions at any point prior to the 1980s. Further, GW limit their main analysis to the rate of return migration and never compare returners to the migrant stock on observables (i.e., the selection of returners). This paper estimates that return migrants were negatively selected using data that directly observes out-migrants; this additionally contributes to the literature by verifying the use of indirect methods to estimate return migrant selection (Abramitzky et al., 2014). Finally, this paper complements GW's results on the effects of the migration quotas by examining how the policies influenced return intentions; the results in this paper stress that the decrease in the overall return rate during the 1920s was primarily driven by a decrease in unplanned returns.

Fig. 1. Self-selection of return migrants, 1910–1930. Notes: Data is from the Annual Reports of the Commissioner General of Immigration (1911, 1921, and 1931) and IPUMS (1910, 1920, and 1930). Self-selection is the difference between return migrants' average logged occupational score and the migrant stock's average logged occupational score; a negative number implies return migrants were lower skilled than the stock. The average skills for the foreign-born in the Census reflect weighting as described in text, which in effect places less weight on those staying more than ten years and more weight for those staying less than ten years. New source countries are those who were more restricted by the 1921 and 1924 quotas, typically from Eastern and Southern Europe. Old source countries are those who were less restricted by the quotas which traditionally includes those from Western and Northern Europe, but also includes countries from the Western Hemisphere. Ethnicities from Asia and Africa are dropped. Confidence intervals are shown at a 95% level.

2. Understanding temporary migration

The decision to return to a poorer country may be baffling to the economist who believes in income maximization, but it has long been recognized that migrants have a variety of motives for moving temporarily (Dustmann and Görlach, 2016). For example, a migrant may move to temporarily increase earnings power for consumption or investment at home, or to acquire human capital that receives a large premium after returning. On the other hand, temporary migrants may return for cultural or family reasons, such as to provide care for older family members or to raise children in the source country (Gibson and McKenzie, 2011).

Early models motivated temporary migration by including a taste preference for consumption at home, the main way to explain returning to a poorer country (e.g., Hill, 1987; Galor and Stark, 1990); however, these models also assumed that the return trip was anticipated at the time of arrival, which is not always the case (Dustmann, 1996). Only recently have models started to consider the dynamic nature of migration as new information about life in the host country is revealed after arrival (e.g., Adda et al., 2015 ; Görlach, 2016; Thom, 2010). In response to an unexpected shock, the duration decision can update in different directions: one who planned to stay permanently could now decide to return home or a temporary migrant might be pushed to the corner solution and expect a permanent stay.

Yet how temporary migrants respond to unexpected income shocks is not straightforward; the response depends on whether a migrant is a “target saver” who aims to hit a pre-determined amount prior to returning home, or if a migrant is a standard “life-cycle” earner who responds to shocks in more conventional ways. For example, if wages unexpectedly increase, a life-cycle earner would remain longer to take advantage of better opportunities, while a target saver would hit his goal more quickly and return home sooner. On the other hand, wages may unexpectedly decrease, which could lead a life-cycle earner to update his optimal duration on the extensive margin (permanent to temporary stay) or on the intensive margin (longer to shorter stay); this is the scenario where return migrants “fail” in the labor market and return home. Alternatively, an unexpected wage decrease may lead a target saver to stay longer than originally intended.

Most models assume temporary migrants follow a life-cycle model instead of a reference-dependent one as in a target-saving model, which is consistent with evidence from Filipino migrants in the 1990s (Yang, 2006). Further, in the first half of the 20 th century, return flows increased rapidly during United States recessions (Biavaschi, 2013b; Jerome, 1926) – a relationship consistent with life-cycle “failures” returning due to negative income shocks. However, there is also evidence that recent Mexican migrants in the United States are target savers (Nekoei, 2013). Ultimately, the return flow is likely a mixture of both successful and unsuccessful life-cycle earners and target savers.

Perhaps the most well-known model that considers both the duration decision changing after arrival and the selection of temporary migrants is Borjas and Bratsberg (1996) extension of the original Borjas (1987) selection model to account for return migration. An important insight from this extension is that the migrants most likely to return home are those on the “margin” – migrants who are just barely attracted to living abroad by the slightly higher income. Of course, after arrival migrants may discover that actual wages differed from expected wages, which could change selection after arrival depending on how unexpected shocks are correlated with skill level – something that is typically unobserved.

The patterns of migrant selection based on relative skill premia as in a Roy–Borjas model may have held during the Age of Mass Migration (Stolz and Baten, 2012), but there has been much criticism of these models. In one strand, some argue that the standard Roy–Borjas selection model may not hold if the costs of migration vary across the skill distribution (Borjas, 1991; Chiquiar and Hanson, 2005). For example, if costs are highest for the least skilled, then they would not move despite a potentially large increase in income. A second criticism of the Roy–Borjas selection model is that migrants may move based on absolute wage differences rather than the relative return to skill (Grogger and Hanson, 2011). In this case, if temporary migrants are those on the margin, then they should be the ones with the smallest difference in absolute wages, while permanent migrants should have the largest differences in absolute wages. Ultimately, due to the differing predictions from these models, the true pattern of selection into temporary migration is an empirical question.

3. Temporary migration in the early 20th century

During the early decades of the Age of Mass Migration (1850–1913), rates of return migration were relatively low due to the high costs of the trip: traveling via sailing ship took months, fares were expensive and mortality rates were high (Bailey, 1912; Cohn, 1984). Return migration started to become more common following the Civil War, when shipping technology shifted from sail to steam and travel time shortened (Cohn, 2009). Following the diffusion of steam to travel, the source country composition changed rapidly in the late 1800s from “Old” sources in Northern and Western Europe to “New” sources in Southern and Eastern Europe; soon, migrant flows shifted from a “family” movement in which entire households moved to an influx of young, single males who were more mobile (Gould, 1980; Baines, 1995).10

The American experience for these new migrants was vastly different from that at home: many settled in densely populated areas and worked in construction, manufacturing or mining, different from their more agrarian source countries (Wyman, 1993). Migrants worked long and hard hours – “like animals” as one Italian returnee put it (Cerase, 1974, p. 250) – but with hard work and thrift, migrants could return home with sizable savings given the higher real wages in America (Williamson, 1995). Despite relatively high wages, many returned home for reasons of social isolation and alienation (Cerase, 1974): immigrants encountered substantial discrimination and racism from natives (Abramitzky et al., 2016; Higham, 2002; Moser, 2012). After World War I, the ensuing xenophobic sentiment and “Americanization” policies in the 1920s perhaps led some migrants, like Germans, to return home at higher rates (Fouka, 2015; Greenwood and Ward, 2015; Lleras-Muney and Shertzer, 2015).

Despite difficulties in the United States, hundreds of thousands kept migrating year after year, which is taken as indirect evidence that most temporary migrants did not fail in the United States (Baines, 1995). Besides letters, return migrants were the main source of information about life abroad for potential migrants: a successful return migrant could have been the envy of others, perhaps leading to more first-time migration. Indeed, inflows fluctuated strongly with United States business cycles, likely because successful migrants spread positive information back home, but also because failed return migrants spread negative information (Hatton and Williamson, 1998; Jerome, 1926).

Yet the narrative that increasing return migrant flows spurred on more first-time migration could be spun around: perhaps the rising number of first-time migrants drove others back home. Indeed, there is a consensus in the literature that new arrivals are substitutes for migrants already in the labor market (Friedberg, 2001; Ottaviano and Peri, 2012). The early 20th century had one of the highest rates of in-migration in American history (Abramitzky and Boustan, 2016), leaving it unclear whether the United States could have fully absorbed all arrivals. Ultimately, the return flows were likely a combination of failures and successes; as noted by Steiner (1906), “the steerage passengers may be roughly divided into two classes: those who go home because they have succeeded, and those who go home because they have failed.”

4. Data

4.1. Planned return migrants at arrival: ship manifests

The primary contribution of this paper comes from a new dataset on return migrant intentions. This dataset is a 1% random sample of ships that originated in Europe and arrived at Ellis Island from 1917 to 1924.11 The ship records used in this study are the same source used in several historical studies of immigration across the 19th and 20th century (e.g., Cohn, 2009; Ferrie, 1999; Massey, 2016). I start my sample in 1917 as this is the first year an important question was added: “Whether alien intends to return to country whence he came after engaging temporarily in laboring pursuits in the United States.”12

One concern is whether migrants truthfully revealed their intentions to stay. There was no benefit or penalty to stating a

(footnote continued) skilled receive a larger premium abroad, then the relatively less skilled of this high-skilled group return home. This is because the less skilled of the highly skilled group earn a smaller wage premium and thus are only marginally attracted abroad. Therefore, return migrants are negatively selected from the positively selected migrant pool.

10 There are several historical studies of return migration for specific countries during this time period. See Sarna (1981) for Jewish return migration, Kraljic (1978) for Croatian, and Saloutos (1956) for Greek return migration. Also, see Balch (1910) and Steiner (1906) for contemporary accounts of temporary migration. Finally, Gmelch has an anthropological survey of return migration (1980).

11 I randomly sample the ships from the Statue of Liberty Ellis-Island Foundation.

12 Another entry on the manifests was “Length of time alien intends to remain in the United States,” which I use to determine whether a migrant plans to leave by 1930 for the linked dataset.

permanent or a return plan; further, these ship manifests were filled out by ship captains rather than United States border officials, which potentially reduced any misrepresentation if migrants would be more intimidated by border officials. However, ship captains may have been careless when recording information; for example, this has been found for manifests from the 19th century when some ship captains listed all migrants as farmers (Erickson, 1972). For the sample collected for this paper, careless compiling does appear to be a small problem: two of the 70 ships in the original sample had the entire ship recorded as planning to stay permanently despite holding hundreds of migrants. I drop these ships from the analysis.

As a further check on whether migrants were truthful, I attempt to find arrivals in the 1930 United States Census; if migrants did return home, then they should not be located. The linking process follows standard practice where links are based on first name, last name, year of birth (plus or minus 2 years), and country of birth. While the overall linking rates are low, which is common in historical studies (see, for example, Abramitzky et al., 2014), the linking rate for planned returners (17.9%) is lower than for planned permanent migrants (29.4%), as expected if migrants were truthful at arrival. Of course, many migrants could have remained in the United States despite planning to return home at arrival because life was better than expected; this behavior is common for currentday migrants (Dustmann, 1996).

Yet failing to find a migrant is not a perfect measure of returning because there could be other reasons for failing to link. For example, failing to link could be related to death, changing one's name after arrival, or having a careless ship captain who haphazardly recorded names or return intentions. Further, a migrant may be linked to a wrong individual because he has a common name. To control for these and other reasons for failing to link, such as country of birth or year of arrival, I use a regression to test whether planning to return home is associated with failing to be found in the later census. Even with controlling for various other reasons for failing to find a migrant in the 1930 Census, including ship fixed effects to account for idiosyncratic recording by ship captains, I consistently find that planning to return home is negatively associated with a successful link. This provides some confidence that migrants truthfully revealed their duration plans and many returned home; see Appendix C for more detail on this analysis.

Some migrants (641 to be exact) listed their intention to stay as “indefinite” or “uncertain.” I allocate these uncertain migrants to the planned return migrant group, leading to a total of 3002 planned return migrants. Alternatively allocating the uncertain migrants to the planned permanent group would make the conclusion of this paper stronger, since that would lower the estimated planned return rate and thus increase the unexpected return rate. For the main analysis, I drop those who planned to stay for less than a year in order to remove tourists and also make the comparison between planned return and actual return data (discussed in the next section) consistent; this is because the actual return data only records the occupations of return migrants who had been in the United States for at least one year. However, results are robust to keeping those planning to stay less than one year in the sample. Since the final sample contains 26,058 individuals (from 68 ships), the planned return rate is 11.5%, which is significantly lower than the actual return rate, with some estimating it to be above 60% (Bandiera et al., 2013). This provides the first indication that unexpected departures were a significant fraction of outflows in the early 20th century.

4.2. Return migrants at departure: administrative data and IPUMS

The Ellis Island records allow me to estimate the selection of planned return migrants at arrival; I compare these results to estimates of the selection of actual return migrants at departure to understand who was more likely to switch plans after arrival. Data on out-migrants are found in the Annual Report of the Commissioner General of Immigration (RCI) between 1908 and 1932 (United States Bureau of Immigration, 1908–1932). Out-going ships had to deliver passenger lists, much like incoming ships, to port officials. These were forwarded to the Bureau of Immigration, which aggregated them into tables and reported them annually to Congress. Importantly, these are the only data from the United States that systematically observe departures; however, the data provide only aggregations of those who leave, making analysis of micro-determinants of out-migration impossible. Fortunately, the reports list the occupations of out-migrants by ethnicity, where occupation is the last one they had in the United States.

In order to determine whether return migrants were higher or lower quality than permanent migrants, I use occupational scores, which is standard amongst other historical self-selection papers (Abramitzky et al., 2012, 2014; Collins and Wanamaker, 2014). Ideally, one would compare wages or education instead of occupations, but administrative data only record the occupations of returnees. Lacking individual-specific wages, I assign an occupational score to each occupation to reflect its earnings, where all individuals claiming an occupation receive the same score. Accordingly, self-selection estimates are based on how temporary and permanent migrants differed on the occupational ladder. Similar to Collins and Wanamaker (2014), I use income data in the fullcount 1940 IPUMS data to assign each occupation the mean wage based purely on migrants' earnings in 1940. For more information on the creation of the occupational score, please see Appendix A.2.

To determine patterns of selection, I compare return migrants' characteristics with the population they were drawn from: the migrant stock as observed in the 1% IPUMS samples from 1910 to 1930 (Ruggles et al., 2010). One important adjustment that I make to the data reflect the fact that temporary migrants were different from the migrant stock in terms of years of stay: approximately 90% of out-migrants lived in the United States for less than ten years, while the corresponding number for the migrant stock is 30%. To make a better comparison between out-migrants and permanent migrants, I reweight the migrant stock to match the years of stay in the out-migrant data, which in effect places less weight on those staying more than ten years and more weight on those staying less than ten years.

These data on out-migrants are far from perfect. In particular, it is well known that the data likely under-counted the total number of departures (Bandiera et al., 2013). Thus, the RCI data can be thought as a sample of total out-migrants, just as the IPUMS is a sample of the entire population – however, while IPUMS is a random sample, the representativeness of the RCI data is unknown. The main reason suggested by Bandiera et al. for the under counting is careless compiling of ship manifests by the Bureau of Immigration. This type of measurement error would not necessarily bias the representativeness; however, there is no way to verify whether this is true as the original out-going manifests were not archived. However, most results follow from New source countries having higher actual return rates than Old Source countries prior to the quotas, a result well known from the RCI data; fortunately, Bandiera et al. (2013) estimated out-migration rates by country has a similar ranking as in the RCI data.

5. Return migration prior to the migration quotas

I show the raw differences between actual return migrants and stayers prior to the migration quotas in Table 1; these results will later be contrasted with data on planned returners. First, the average return migrant was less skilled than the rest of the migrant population, holding jobs that paid about 9.0 percent less. This finding supports the conclusion that return migrants were negatively self-selected on occupation by Abramitzky et al. (ABE, 2014), and validates the use of residual methods to estimate return migrant selection. However, note that I can only present raw measures of selection without adjusting for observables such as age because I do not observe individual return migrants.

The main reason for out-migrants' lower earnings is that return migrants were more likely to be from the New source countries of Italy, Greece and Russia; most migrants from these countries did not hold high-paying occupations in the United States. However, not only were return migrants less skilled when using variation across countries, but also they were mostly less-skilled when comparing return migrants to the migrant stock within-ethnicity; this also coincides with ABE's results using indirect methods. According to the demographic characteristics, return migrants were also more likely to be male, single, over the age of 45, and from the Northeast.

The differences in skill levels and demographics between return and permanent migrants could be for a simple reason: these were the migrants who had always planned to return home. One can check this using the descriptive statistics of planned return migrants, as shown for those who arrived prior to the migration quotas in Table 2. In addition to the raw differences between planned return and permanent migrants, I show the adjusted difference in the last column by controlling for country of birth, age, and sex.

There are a couple of key points to draw from this table. First, as mentioned prior, the planned return rate was 11.5% for the entire sample between 1917 and 1924; for the years prior to the implementation of the first quota it was 15.4%, much smaller than the estimated actual return rates during the same time period. Second, while the actual out-migrant data showed that they held jobs that paid 9.0 percent less than the migrant stock, planned return migrants held jobs that were statistically similar to planned permanent migrants – this result holds when controlling for country of birth, age and sex, as shown in the fourth column. Thus, prior to the migration quotas two results appear to hold: more migrants returned than initially planned to, and the selection of actual returners at departure was more negative than the selection of planned returners at arrival, suggesting that it was either the lowskilled or those who downgraded their occupation upon arrival who switched plans from staying to returning home.

It is important to note that the occupation listed by incoming migrants is likely their job in the source country rather than in the

1

Self-selection of return migrants prior to quotas, 1920.

Characteristics Return migrants Foreign-born in census Self-selection

Log (occupational score), if have job 6.859 6.949 −0.090*

 (0.155) (0.289) (0.00080)

Claim an occupation 0.765 0.572 0.193*

 (0.424) (0.495) (0.00134)

Male 0.796 0.543 0.253*

 (0.403) (0.498) (0.00131)

Less than 16 0.040 0.114 −0.074*

 (0.195) (0.318) (0.00069)

Age 16–45 0.710 0.755 −0.045*

 (0.454) (0.430) (0.00135)

Age over 45 0.250 0.131 0.119*

 (0.433) (0.337) (0.00125)

Married 0.481 0.605 −0.124*

 (0.500) (0.489) (0.00150)

New source country 0.813 0.439 0.374*

Region of last residence (0.390) (0.496) (0.00132)

Northeast 0.607 0.499 0.108*

 (0.489) (0.500) (0.00148)

Midwest 0.263 0.300 −0.037*

 (0.440) (0.458) (0.00133)

South 0.045 0.0548 −0.010*

 (0.206) (0.228) (0.00063)

West 0.086 0.146 −0.060*

 (0.281) (0.353) (0.00089)

Notes: Out-migrant data is from the Annual Reports of the Commissioner General of Immigration (1920–1921). The Census is from the 1920 IPUMS samples. The left column represents averages from 532,770 individuals; however, these individuals are not individually observed. The middle column represents 140,259 individuals observed in IPUMS. The third column is the difference between the out-migrants and the Census where a positive number indicates that return migrants have more of that characteristic.

** p-value of less than 0.05. *** p-value less than 0.10. * p-value of less than 0.01.

United States. Thus, self-selection estimates at arrival may not be the same comparison as self-selection estimates at departure, which are based on occupations held in the United States. A preferred metric for self-selection would be a permanent level of human capital, such as education, that does not change across borders. While there is no such metric in the actual out-migrant data, the incoming data does list an individual's height, which correlates with productivity, health, and standard of living; further, it does not change to match a country's economic structure, unlike occupation (Kosack and Ward, 2014). Importantly, there was no difference in height between planned return and planned permanent migrants, verifying that there was no selection on average skill into planned return migration at arrival.

The demographic characteristics give a further indication of who was more likely to unexpectedly return home prior to the quotas being put in place. This can be done by comparing the selection of actual return migrants (the last column in Table 1) to the selection of planned return migrants (the second to last column in Table 2). Planned returners were a little more likely to be male than planned stayers, in contrast with actual returners who were much more heavily male than actual stayers. Actual return rates were also higher for singles and those from the Northeast relative to planned return rates. Thus, unexpected returners were more likely to be male, single, and from the Northeast. Interestingly, these migrants were the ones who could more easily switch their decision from staying to returning home given their location and marital status.

Note that the planned return migrant data highlights the importance of family and network variables: being married, having children, joining family in the United States and listing no contact in the source country all have large magnitudes for selection into a permanent stay. Further, listing any nuclear family in the source country correlates with intending tol return home, although this become statistically insignificant after controlling for age, country of birth and sex. Unfortunately, many of these variables cannot be compared to actual return migrants since the administrative data did not collect this information.

The most important difference between actual and planned returners is that migrants from New source countries had a much lower planned return rate than migrants from Old source countries, which is surprising given that New source countries had a much higher actual return rate. Table 3 shows estimates of both the planned and actual return rates prior to the migration quotas by ethnicity (note again that tourists have been dropped from the data); the table is sorted by the difference between planned return and

2

Descriptives of planned return migrants prior to quotas (1917–1921).

Characteristics Planned return Planned permanent Raw difference Adjusted difference

Log(Occ score), if have job 6.814 6.848 −0.0342 0.00102

 (0.316) (0.286) (0.0328) (0.0239)

Height (cm), if recorded 164.5 163.1 1.410 0.289

 (10.51) (11.95) (1.062) (0.793)

Claim an occupation 0.683 0.548 0.136*** 0.0677***

 (0.465) (0.498) (0.0280) (0.0181)

Male 0.550 0.510 0.0398

 (0.498) (0.500) (0.0422)

Age 25.91 25.26 0.649

 (11.86) (13.50) (0.605)

Married 0.293 0.340 −0.0474** −0.0443***

 (0.455) (0.474) (0.0226) (0.0119)

New source ethnicity 0.543 0.680 −0.137* −0.00245

 (0.498) (0.467) (0.0675) (0.00632)

Northeast 0.616 0.628 −0.0119 −0.0158

 (0.486) (0.483) (0.0476) (0.0425)

West 0.151 0.0665 0.0844** 0.0499**

 (0.358) (0.249) (0.0352) (0.0237)

South 0.0406 0.0411 −0.000499 0.00438

 (0.197) (0.198) (0.00728) (0.00444)

Midwest 0.193 0.265 −0.0720** −0.0385

 (0.394) (0.441) (0.0296) (0.0256)

Repeater 0.190 0.167 0.0226 −0.00126

 (0.392) (0.373) (0.0219) (0.0150)

Number of accompanying children 0.242 0.474 −0.232*** −0.150**

 (0.668) (0.984) (0.0772) (0.0637)

Join family in United States 0.758 0.851 −0.0935** −0.0563**

 (0.429) (0.356) (0.0384) (0.0254)

Traveling alone 0.793 0.727 0.0659 0.0439

 (0.406) (0.446) (0.0552) (0.0370)

Nuclear family in source country 0.757 0.665 0.0919*** 0.0290

 (0.429) (0.472) (0.0185) (0.0243)

No contact in source country 0.0929 0.163 −0.0703*** −0.0419*

 (0.290) (0.370) (0.0188) (0.0241)

Traveling to big city (> 100,000) 0.597 0.608 −0.0105 0.0391

 (0.491) (0.488) (0.0345) (0.0236)

Observations 1948 10,669

Notes: Data is a random sample of ships from Ellis Island Records (1917–1924). The rightmost column controls for country of birth, age, and sex. Heights were generally not recorded for children.

*** p-value of less than 0.01.

** p-value of less than 0.05. * p-value less than 0.10.

actual return rates, which reflects the degree of unplanned returns. Further, note that the estimated actual return rate is a lower bound of the true return rate given that out-migrants in the return migrant data are under counted (Bandiera et al., 2013).

This table shows that many New source ethnicities had a high rate of unplanned returns; interestingly, unplanned returns were much more common among Eastern European ethnicities rather than places with an established flow like Italy. While the planned return rates for New source countries may be surprising because they were so low, to a certain extent it is much more surprising that the actual return rates were so high: given the premium for real wages in the United States, the economic return to migration was likely higher for these poorer source countries. Thus, it appears that the higher actual return rate could be due to unexpected factors prior to arrival, such as job losses or distaste for living in the United States, or unexpectedly better conditions in the source country.

The evidence so far shows the differences in the means between returners and stayers, but selection may differ along the distribution of human capital. For example, other studies show a U-shaped pattern of self-selection on income; the logic here is that those with the highest income hit a savings target quickly, while those on the lower end “fail” and return home (Bijwaard and Wahba, 2014). To provide an idea about the pattern of return migrant selection, especially over the 1920s before and after the migration quotas were put into place, I plot the skill densities of actual return migrants and the migrant stock, after controlling for ethnicity, in Fig. 2. Actual return migrants, particularly in 1920, were from a concentrated set of occupations which included laborers, farm laborers, and miners. This negative pattern of self-selection became less strong by 1930, as seen in the right hand panels. Rather than a mass of return migrants leaving as laborers, there was a wider variety of occupations; yet the selection of return migrants was still strongly negative.

While the actual out-migrant data show that return migrants were often negatively selected and came from a concentrated set of occupations, the planned return migrant data tell a different story. The upper panels of Fig. 3 plot the residual logged occupational

3

Return rates for those who arrived between 1917 and 1921.

Ethnicity I II (II − I) N in ship records

Planned return rate LB actual return rate LB unplanned return rate

Romanian 0.03 0.30 0.26 254

Russian 0.04 0.28 0.24 56

Greek 0.15 0.36 0.21 116

Polish 0.01 0.16 0.15 731

Syrian 0.00 0.11 0.10 282

English 0.13 0.16 0.04 1310

Italian 0.16 0.18 0.01 4310

Armenian 0.02 0.02 0.00 153

German 0.10 0.09 −0.01 248

Hebrew 0.02 0.01 −0.02 1011

Finnish 0.13 0.11 −0.02 229

Slovak 0.05 0.04 −0.02 585

Dutch and Flemish 0.19 0.14 −0.05 213

Scottish 0.15 0.07 −0.08 216

Scandinavian 0.22 0.14 −0.08 1142

Welsh 0.23 0.08 −0.15 57

French 0.31 0.15 −0.16 363

Spanish 0.57 0.32 −0.24 391

Irish 0.34 0.06 −0.28 718

Notes: Data is from Ellis Island passenger manifests and the Annual Reports of the Commissioner General of Immigration (1917–1928). The expected return rate is the percent of incoming migrants who planned to return home. See online Appendix B for the calculation of lower-bound (LB) actual return rate, which approximates the return rate after ten years.

score, after controlling for ethnicity, of planned return and planned permanent migrants before and after the 1921 migration quota. Prior to the migration quotas, planned return and planned permanent migrants had relatively similar occupational score densities, with the planned return migrant density having a slightly wider set of occupations. Due to this higher standard deviation, planned

Fig. 2. Actual return migrant densities, 1920 and 1930. Notes: Data is from Report of the Commissioner General of Immigration (1921, 1931) and IPUMS (1920 , 1930). The panels plot the residual occupational score after controlling for ethnicity and year of departure. The bottom two panels plot the difference in the return migrant and migrant stock densities.

Fig. 3. Planned return migrant densities, pre and postmigration quota. Notes: Data is from Ellis Island Records (1917–1924). The panels plot the residual occupational score after controlling for ethnicity and year of arrival. The bottom two panels plot the difference in the planned return migrant and planned permanent migrant densities.

return migrants were self-selected in the way found by other papers for current-day actual return migrants: a U-shaped pattern where the upper and lower ends of the skill distribution planned to return at higher rates. The U-shaped pattern of self-selection for planned return migrants holds after the 1921 migration quotas, perhaps suggesting little effect on those who planned to return home.

6. The effect of the 1920s migration quotas

6.1. Empirical strategy

The visual evidence suggests that the quotas had no impact on selection into planned return migration on skill; here, I more rigorously test how the 1920s migration quotas affected planned return rates. Descriptions of these laws can be found elsewhere (Greenwood and Ward, 2015; Massey, 2016), but of primary importance for this paper is that the migration quotas were unexpectedly implemented, varied across countries in absolute quota limits, and varied across time with the 1921 and 1924 quota laws.25 These quotas were restrictive enough that they led to a 60% drop in immigrant flows in a four-year time span.

To estimate how the quotas affected temporary migration plans, I create a measure that uses the variation in quota limits across the different iterations of the law to gauge the law's restrictiveness on a country. I proxy the potential migrant flow for a country using its flow from July 1920 to June 1921, or the fiscal year prior to the quota law fully going into effect:

QuotaLimitjt QuotaRestrictionjt =1−

Immigrantsj,FiscalYear1921 (1)

This measure is specific for the country j and time period t due to the changes in quota laws in 1921 and 1924. The variable takes values between zero and one, where 0.98 is its highest value for Italian migrants after 1924, implying that 98% of potential migrants were restricted from entry.

I test how the migration quotas affected return intentions by running variations of the following linear probability model:

PlannedReturnijt = β0 + β1QuotaRestrictionjt + β2LowSkillijt + β3MediumSkillijt

+ β4QuotaRestrictionjt × LowSkillijt + β5QuotaRestrictionjt × MediumSkillijt

+ ηt + εijt + ΓXijt + γj (2)

where PlannedReturnijt is a zero-one variable where one indicates that migrant i of birth country j arriving in fiscal year t planned to return home. The regression splits migrants into three different skill groups, based on whether they are in the top, middle or bottom third of occupational scores within the dataset to test the U-shaped pattern of self-selection. The most common occupations in the lowest third are servants and farm laborers, the middle third are laborers and farmers and the top third are merchants and dealers. Those without jobs are dropped from the regression.

The main regressor of interest is QuotaRestrictionjt and its interactions with each of the different skill categories. In the format of the regression, HighSkillijt is the excluded skill group, so the main effect of QuotaRestrictionjt, β1, gives the estimate of restricting migration on the high skilled group. The coefficients of the other interactions are β4 and β5, which estimate whether the effect of quotas is statistically different from the high-skilled category.

Also in the regression I include dummy variables for country j to control for country-specific heterogeneity that is time-invariant, such as culture and language, and fiscal year of arrival dummy variables t which control for general time trends. I also control for various individual observables in Xijt, which include a migrant's sex, age, network, and other observables listed in Table 2. Importantly, I also control for the source country's GDP changes to account for factors drawing migrants back to Europe.

For the estimated effects of QuotaRestrictionjt to have a causal interpretation, there must not be any time-varying unobservables that are also correlated with the intensity of quota limits. To justify the parallel trends assumption, I plot the planned return rates by New and Old source countries in Fig. A3, which show similar trends between the two groups. However, given the differences in migrant characteristics between Old source countries and New source countries, the effect of the migration quotas can only be generalized to restricting migration of relatively poorer source countries rather than restricting all countries; yet one may argue that this is a more relevant effect to estimate as interest groups often aim to restrict migration from poorer countries.

6.2. Results for planned return migration

Table 4 shows results from the regression. The first column tests the main effect of a more restrictive quota on planning to return, before allowing heterogeneity to exist across skill groups. The coefficient on quota restriction is statistically insignificant and close to zero, suggesting that migration quotas did not cause the planned return rate to decrease. This result contrasts with the fall in actual return rates caused by quotas found by Greenwood and Ward (2015). Thus, the policy seems to have had little influence on return migration due to expectations of migrants at arrival, but rather changed return migration by altering a migrant's experience against expectations.

The second column adds the lower and medium skilled categories and shows that those who were middle skilled were less likely to plan to return home relative to the high skilled by about 4.3 percentage points. Further, the low-skilled group has statistically the same return rate as the high-skilled category. These two results empirically confirm that planned return migrants had a U-shaped pattern of self-selection within country, where the highest and least skilled were the most likely to plan to return home.

The third column tests the interactions between a more restrictive quota and the different skill groups. The coefficients on quota restriction and the interactions with skill groups are all statistically insignificant; further, the total effect of quota restrictions are not statistically significant for the medium and low-skilled group. These results suggest that the quotas did not drastically alter the rate of planned return migration across skill groups, at least for the first couple years of the quotas from 1921 to 1924. Once again, this contrasts with other evidence that the migration quotas caused more low-skilled migrants to stay. Finally, the fourth column includes control variables for various observables, and the results still hold.

When one contrasts these results with the estimated effects on actual return rates from Greenwood and Ward (2015), the main difference is that actual return rates dropped while the planned return rates did not. There are two reasons why this may have occurred: first, there were fewer unexpected returns, perhaps due to fewer failures in the labor market, or second that there were

Table 4

Self-selection into planned return migration, 1917–1924.

 I II III IV

Quota restriction −0.00552 −0.0131 −0.0229 −0.0123

 (0.0648) (0.0645) (0.0626) (0.0604)

Low skilled 0.00477 0.0103 0.00116

  (0.0153) (0.0188) (0.0191)

Medium skilled −0.0435* −0.0515* −0.0461*

  (0.0129) (0.0149) (0.0151)

Quota restriction × Low-skilled −0.0321 −0.0260

   (0.0365) (0.0375)

Quota restriction × medium-skilled 0.0431 0.0425

   (0.0362) (0.0350)

Age 0.000746

(0.00284)

Age squared −1.64e-06

(3.18e-05)

Male −0.0179 (0.0149)

Repeater −0.0116 (0.0116)

Join family −0.0296**

(0.0137)

Ever married −0.0177 (0.0134)

Traveling alone 0.0100

(0.0140)

Number of children −0.0102***

(0.00577)

Midwest −0.0106 (0.0103)

West 0.0403

(0.0273)

South 0.0118

(0.0152)

Nuclear family at home 0.0306**

(0.0121)

Extended family at home 0.0246

(0.0192)

Big City (> 100,000) 0.0148*

(0.00831)

GDP in source country −5.86e−05

(6.23e−05)

Country of birth indicators X X X X

Fiscal year indicators X X X X

Observations 15,246 15,246 15,246 15,246

R-squared 0.159 0.162 0.163 0.168

Notes: Data is from incoming passenger manifests from Ellis Island (1917–1924). The dependent variable is whether or not an individual planned to return home.

* p-value of less than 0.01.

** p-value of less than 0.05. *** p-value less than 0.10.

more unexpected stays, perhaps due to greater success in the labor market. In Appendix C, I provide suggestive evidence that there were both fewer unexpected returns and more unexpected stays. Using the sample linked to 1930, I estimate that the quotas are correlated with successfully finding more planned stayers and planned returners in the country by 1930. This suggests that planned returners were more likely to stay following the quotas (i.e., more unexpected successes), and planned stayers were also more likely to stay (i.e., fewer unexpected failures). Yet these results are still only suggestive due to other reasons for failing to link besides return migration.

6.3. Results for expected years of stay

The previous results show that the migration quotas did not change the rate at which incoming migrants planned to return or stay, but this is only one way to measure how quotas affected return migration. In this section, I explore a different margin: the years of intended stay. Using this data, besides being able to test the effects of the quotas, I can also test whether planned return migrants exhibited behavior consistent with holding a savings target. If all migrants held a common savings target, then those who were more skilled would plan to stay fewer years as it would take fewer years to achieve the savings target. Another factor consistent with a target savings hypothesis is that those who had higher costs, such as those traveling further inland from New York City, would also stay longer in order to recuperate the traveling costs.

I run the same estimating framework as Eq. (2) but with the number of years of stay as the dependent variable. Note that the mean years of expected stay is 4.3, and that those staying under one year have already been dropped from the data set to exclude tourists. Further, those who planned to stay permanently have been dropped from the following analysis.

The results are presented in Table 5. The first column shows the main effect of a more restrictive quota on the planned years of stay, and finds that the quotas also had no overall effect on this variable. Just like the results for planned return rates, this result can also be compared to Greenwood and Ward (2015) as they argue that the quotas led individuals to stay longer in the United States.

The second through fourth columns add variables controlling for the skill level and other observables of entering migrants. The second column shows evidence consistent with a target savings hypothesis: low- and medium- skilled migrants planned to stay approximately one year more than high-skilled migrants, perhaps because high-skilled migrants were more easily able to hit a savings target. When adding the extra observables in column four, we see that those traveling to the Midwest and West planned to stay longer than those traveling to the Northeast, by approximately one year, also consistent with migrants taking longer trips to cover higher traveling costs.

The third column examines the heterogeneous effect of the immigration quotas and, unlike the effect of quotas on planned return rates, finds different effects for length of stay across skill groups. In particular, the results show that the more restrictive a quota, the fewer years the least skilled migrants planned to stay; the estimate suggests that a 60% restriction lowered the planned length of stay by one year relative to the high-skilled group. It is unclear what is driving this result: on one hand, if the costs of entry were higher due to the policy change, then migrants would stay longer to cover the higher entry costs; however, for the migrants who did plan to return home and had a savings target, it could be that they expected to hit the savings target more quickly.

6.4. Discussion of the effects of the 1920s quotas

This paper shows that in the transition from a free to a restricted migration system, migrants were less likely to unexpectedly return home and more likely to unexpectedly stay longer. Particularly, it appears that single males from poorer countries returned at unexpectedly high rates prior to the quotas, and following the quotas there were fewer of these types of returns. Here I discuss possible reasons for these patterns.

One possibility is that the return migrants who planned to take multiple trips in and out of the United States were trapped inside due to increased re-entry costs. This suggestion, while it likely did have an impact, is probably not the main driver of the selection patterns for a few reasons. First, most repeated entries were those who left and re-entered within a year, and such short trips back home were actually not restricted by the migration quotas as long as one applied for a permit. Accordingly, repeat entrants were more prevalent in the migrant flow after the implementation of the migration quotas (Ward, 2016). Second, the probability of a return migrant re-entering was similar across New and Old source countries (about 15–20 percent); therefore, quotas were not strongly correlated with repeat migration patterns.

Rather, one reason for fewer unexpected returns is that the quotas screened out those who were both less committed to staying and had low switching costs. The new process of entering the country was extensive: after the 1924 quota, migrants now had to apply for entrance at a consular office in the source country, supply documents such as a birth certificate, military, and criminal record, and sign the application and swear an oath administered by the consular officer. One had to do all of this and also be one of the first applications; indeed, the quotas filled rapidly and forced individuals to apply quickly. The process was far different from an earlier time when one could simply buy a ticket and arrive ten days later. This increase in requirements likely screened out those less certain about life abroad, and further led to a large drop in the number of single males migrating from Southern and Eastern Europe – those who could easily switch their decision from staying permanently to returning home.

While the quotas may have screened out the less certain, they also improved the experience in the United States for those lucky enough to enter. The quotas eliminated numerous competitors in the labor market: the inflow dropped by 60% over four years and further shifted the composition towards fewer labor force participants. Given the large responses of outflows to unemployment (Bijwaard et al., 2014), it is possible that the migration quotas led to less unemployment specifically for other migrants in the labor market. Many migration studies have found a high degree of substitutability between migrants and other migrants, as they have similar locations and skills, as opposed to migrants and natives. For example, Beaman (2012) finds that having more recentlyarrived migrants in a network actually worsens outcomes for newly-arriving migrants, as both groups compete for the same jobs. Indeed, those with highly substitutable characteristics tended to leave more rapidly in the early 20th century: prime-aged males living in the Northeast faced substantial competition from inflows of migrants, and they were also driven out.

Table 5

Expected years of stay, 1917–1924.

I II III IV

Quota restriction −0.873 −0.701 0.212 0.0351

(0.660) (0.664) (0.919) (1.005)

Low skilled 1.009* 1.139* 1.236*

(0.243) (0.244) (0.161)

Medium skilled 1.227* 1.244* 1.054*

(0.259) (0.279) (0.272)

Quota restriction × low-skilled −1.898* −1.913*

(0.594) (0.567)

Quota restriction × medium-skilled −0.612 −0.216

(0.791) (0.802)

Age −0.0468

(0.0342)

Age squared 0.000127

(0.000452)

Male 0.476**

(0.194)

Repeater 0.203

(0.185)

Join family 0.448*

(0.141)

Ever married 0.348***

(0.200)

Traveling alone 0.143

(0.188)

Number of children 0.304

(0.335)

Midwest 0.573*

(0.202)

West 1.252*

(0.324)

South −0.0773

(0.413)

Nuclear family at home −0.255

(0.562)

Extended family at home −0.860

(0.689)

Big city (> 100,000) 0.138

(0.124)

GDP in source country 0.000577

(0.000883)

Country of birth indicators X X X X

Fiscal year indicators X X X X

Observations 1398 1398 1398 1398

R-squared 0.265 0.291 0.295 0.336

Notes: Data is from incoming passenger manifests from Ellis Island (1917–1924). The dependent variable is the years a return migrant planned to stay prior to returning. *

p-value of less than 0.01. **

p-value of less than 0.05. ***

p-value less than 0.10.

Indeed, soon after the 1921 quota was put into place, many contemporaries suggested that there should be looser migration restrictions to reduce rising labor costs (Jerome, 1934). If one compares the post-restrictions migration totals to a simple counterfactual, the 1900s decade, the 1920s about 2.8 million fewer net immigrants (53% less); therefore, the reduction in labor supply caused by the migration quotas left open a significant gap in the labor market (Collins, 1997). The reduction of inflows from Europe encouraged the migration of those elsewhere: for example, the Great Migration of about 900,000 blacks from the South to the North in the 1920s has been linked to decreased migrant inflows due to the quotas (Collins, 1997). Further, indirect evidence for a better labor market for migrants can also be seen in the rise of inflows from countries not limited by quotas: Mexico and

Canada's average migrant flows to the United States increased by 67% and 62% for the years following the 1921 quota. 7. Concluding remarks

This paper establishes that most return migrants unexpectedly returned home in the early 20th century. This contrasts with a common understanding of temporary migration during this time period and paints a more pessimistic picture of migration to the United States. Assimilation into the United States was particularly difficult in the years prior to the migration quotas for migrants from poorer source countries in Eastern and Southern Europe. Combined with evidence that migrants rarely upgraded their occupations (Abramitzky et al., 2014) and that the return to migration was relatively low compared with today (Abramitzky et al., 2012), it appears that outcomes for migrants were often worse than expected and many returned home. Importantly, the rate of occupational upgrading was low, but not for lack of an incentive to invest in United States-specific human capital as many planned to stay permanently.

However, this pessimistic view applies only to the average migrant; there were certainly plenty of successful temporary and permanent migrants. Moreover, this pessimistic view of the Southern and Eastern European migrant experience may not apply to the period following the implementation of the migration quotas: in particular, this paper shows that there were fewer unplanned returns, perhaps due to improved outcomes in the United States. This suggests that if migration policy creates a large and dramatic shock to incoming flows, the group most affected by this policy are prior migrants who have already entered, or those lucky enough to enter under the policy. These migrants are in less competition with inflows, and thus will be more likely to stay permanently within the United States. On the other hand, if the United States liberalizes its migration policy, it is possible that more migrants would be driven to return home due to intense competition among entering cohorts.

Acknowledgments

I would like to thank Ann Carlos for numerous conversations and insightful guidance on this project. Other helpful comments were given by Ran Abramitzky, Brian Cadena, Dustin Frye, Mike Greenwood, Myron Gutmann, Tim Hatton, Murat Iyigun, Ian Keay, Priti Kalsi, Edward Kosack, Amber McKinney, Terra McKinnish, Steven Smith, John Tang, Matthew Van Wyhe, Marianne Wanamaker, and several anonymous referees. Special thanks go to Lee Alston for helping gain access to the Census data. Part of this paper was previously circulated under the title of “The U-Shaped Self-Selection of Return Migrants.” I would also like to thank those individuals at the Australian National University, World Cliometrics Conference, University of Colorado Economic History Workshop, the ANU Height and Development Workshop, 2014 WEAI meetings, and 8th Annual Conference for Migration and Development who offered helpful comments. All errors are my own.

Appendix A. Supplementary data

Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.eeh.2016.09. 002.

References

Abramitzky, R., Boustan, L.P., 2016. Immigration in American economic history. J. Econ. Lit., doi: 10.3386/w21882.

Abramitzky, R., Boustan, L.P., Eriksson, K., 2012. Europe's tired, poor, huddled masses, self-selection and economic outcomes in the age of mass migration. Am. Econ. Rev. 102 (5), 1832–1856.

Abramitzky, R., Boustan, L.P., Eriksson, K., 2014. A nation of immigrants, assimilation and economic outcomes in the age of mass migration. J. Polit. Econ. 122 (3), 467–506.

Abramitzky, R., Boustan, L.P., Eriksson, K., 2016. Cultural Assimilation During the Age of Mass Migration. Working Paper 22381, National Bureau of Economic Research. July. 〈http://www.nber.org/papers/w22381〉

Adda, J., Dustmann, C., Görlach, J.-S., 2015. The dynamics of return migration, human capital accumulation, and wage assimilation.

Angrist, J., 2002. How do sex ratios affect marriage and labor markets? evidence from America's second generation. Q. J. Econ., 997–1038.

Bailey, W., 1912. The bird of passage. Am. J. Sociol. 18 (3), 391–397.

Baines, D., 1995. Emigration from Europe vol. 11. Cambridge University Press, Cambridge, 1815–1930.

Balch, E.G., 1910. Our Slavic Fellow Citizens. Charities Publication Committee, New York.

Bandiera, O., Rasul, I., Viarengo, M., 2013. The making of modern America: migratory flows in the age of mass migration. J. Dev. Econ. 102, 23–47.

Beaman, L.A., 2012. Social networks and the dynamics of labour market outcomes: evidence from refugees resettled in the us. Rev. Econ. Stud. 79 (1), 128–161. Biavaschi, C., 2013a. Fifty Years of Compositional Changes in US Out-migration, 1908–1957. Technical Report, Discussion Paper Series, Forschungsinstitut zur Zukunft der Arbeit.

Biavaschi, C., 2013b. The labor demand was downward sloping: disentangling migrants' inflows and outflows, 1929–1957. Econ. Lett. 118 (3), 531–534.

Biavaschi, C., Giulietti, C., Siddique, Z., 2013. The Economic Payoff of Name Americanization. Technical Report, IZA Discussion Paper.

Bijwaard, G.E., Schluter, C., Wahba, J., 2014. The impact of labor market dynamics on the return migration of immigrants. Rev. Econ. Stat. 96 (3), 483–494.

Bijwaard, G.E., Wahba, J., 2014. Do high-income or low-income immigrants leave faster? J. Dev. Econ. 108, 54–68.

Borjas, G.J., 1987. Self-selection and the earnings of immigrants. Am. Econ. Rev. 77 (4), 531–553.

Borjas, G.J., 1991. Immigration and self-selection. In: Immigration, Trade, and the Labor Market. University of Chicago Press, Chicago, pp. 29–76.

Borjas, G.J., Bratsberg, B., 1996. Who leaves? The outmigration of the foreign-born. Rev. Econ. Stat. 78 (1), 165–176, URL 〈http://www.jstor.org/stable/2109856〉.

Carter, Susan B., Gartner, Scott S., Haines, Michael R., Olmstead, Alan L., Sutch, Richard, Wright, Gavin, 2006. Historical statistics of the United States: Earliest times to the present. Cambridge University Press, Cambridge.

Cerase, F.P., 1974. Expectations and reality: a case study of return migration from the United States to Southern Italy. Int. Migr. Rev. 8 (2), 245–262.

Chiquiar, D., Hanson, G.H., 2005. International migration, self-selection, and the distribution of wages: evidence from Mexico and the United States. J. Polit. Econ. 113 (2), 239–281.

Cohn, R.L., 1984. Mortality on immigrant voyages to New york, 1836–1853. J. Econ. Hist. 44 (2), 289–300, URL 〈http://www.jstor.org/stable/2120706〉.

Cohn, R.L., 1995. Occupational evidence on the causes of immigration to the United States, 1836–1853. Explor. Econ. Hist. 32 (3) , 383–408, URL 〈http://www.sciencedirect.com/science/article/pii/S0014498385710170〉.

Cohn, R.L., 2009. Mass Migration Under Sail: European Immigration to the Antebellum United States. Cambridge University Press, New York.

Collins, W.J., 1997. When the tide turned: immigration and the delay of the great black migration. J. Econ. Hist. 57 (03), 607–632.

Collins, W.J., Wanamaker, M.H., 2014. Selection and economic gains in the great migration of African Americans: new evidence from linked census data. Am. Econ. J.: Appl. Econ. 6 (1), 220–252.

Dustmann, C., 1996. Return migration, the European experience. Econ. Policy, 213–250.

Dustmann, C., Görlach, J.-S., 2016. The economics of temporary migrations. J. Econ. Lit. 54 (March (1)) , 98–136, URL〈http://www.aeaweb.org/articles/?doi=10.1257/jel.54.1.98〉.

Erickson, C.J., 1972. Who were the English and Scots Emigrants to the United States in the Late Nineteenth Century?, pp. 347–381.

Ferrie, Joseph P., 1999. Yankeys Now: Immigrants in the Antebellum US 1840–1860. Oxford University Press, New York.

Fouka, Vasiliki, 2015. Backlash: The Unintended Effects of Language Prohibition in US Schools after World War I. Working Paper.

Friedberg, Rachel M., 2001. The impact of mass migration on the Israeli labor market. Q. J. Econ. 116 (4), 1373–1408.

Fryer, R.G., Levitt, S.D., 2004. The causes and consequences of distinctively black names. Q. J. Econ., 767–805.

Galor, Oded, Stark, Oded, 1990. Migrants' savings, the probability of return migration and migrants performance. Int. Econ. Rev. 31 (2), 463–467.

Gibson, J., McKenzie, D., 2011. The microeconomic determinants of emigration and return migration of the best and brightest, evidence from the Pacific. J. Dev. Econ. 95 (1), 18–29.

Goldin, C., 1994. The political economy of immigration restriction in the United States, 1890 to 1921. In: The Regulated Economy: A Historical Approach to Political Economy. University of Chicago Press, Chicago, pp. 223–258.

Görlach, J.-S., 2016. Borrowing Constraints, Migrant Selection, and the Dynamics of Return and Repeat migration. Mimeo.

Gould, J.D., 1980. European inter-continental emigration. The road home: return migration from the USA. J. Eur. Econ. Hist. 9 (1), 41–112.

Greenwood, M.J., Ward, Z., 2015. Immigration quotas, world war I, and emigrant flows from the United States in the early 20th century. Explor. Econ. Hist. 55 (0) , 76–96, URL 〈http://www.sciencedirect.com/science/article/pii/S0014498314000199〉.

Grogger, J., Hanson, G.H., 2011. Income maximization and the selection and sorting of international migrants. J. Dev. Econ. 95 (1), 42–57.

Hatton, Timothy J., Williamson, Jeffrey G., 1998. The Age of Mass Migration: Causes and Economic Impact. Oxford University Press, New York.

Higham, J., 2002. Strangers in the Land: Patterns of American Nativism, 1860–1925. Rutgers University Press, New Brunswick.

Hill, John K., 1987. Immigrant decisions concerning duration of stay and migratory frequency. J. Dev. Econ. 25 (1), 221–234.

Jerome, Harry, 1926. Migration and Business Cycles. NBER Books, Cambridge.

Jerome, H., 1934. Mechanization in Industry. NBER Books, New York.

Kosack, E., Ward, Z., 2014. Who crossed the border? Self-selection of mexican migrants in the early twentieth century. J. Econ. Hist. 74 (04), 1015–1044. Kraljic, F., 1978. Croatian migration to and from the United States 1900–1914.

Kuznets, S., Rubin, E., 1954. Immigration and the Foreign Born. NBER Books.

Lleras-Muney, A., Shertzer, A., 2015. Did the americanization movement succeed? an evaluation of the effect of english-only and compulsory schooling laws on immigrants. Am. Econ. J.: Econ. Policy 7 (3), 258–290, URL 〈http://www.aeaweb.org/articles.php?doi=10.1257/pol.20120219〉.

Massey, C.G., 2016. Immigration quotas and immigrant selection. Explor. Econ. Hist. 60 , 21–40, URL 〈http://www.sciencedirect.com/science/article/pii/S0014498315000455〉.

McKenzie, D., Gibson, J., Stillman, S., 2013. A land of milk and honey with streets paved with gold, do emigrants have over-optimistic expectations about incomes abroad? J. Dev. Econ. 102, 116–127.

Moser, P., 2012. Taste-based discrimination evidence from a shift in ethnic preferences after wwi. Explor. Econ. Hist. 49 (2), 167–188.

Nekoei, A., 2013. Immigrants' labor supply and exchange rate volatility. Am. Econ. J.: Appl. Econ. 5 (4), 144–164.

Ottaviano, G.I., Peri, G., 2012. Rethinking the effect of immigration on wages. J. Eur. Econ. Assoc. 10 (1), 152–197.

Piore, M.J., 1979. Birds of Passage: Migrant Labor and Industrial Societies. Cambridge University Press, New York.

Ruggles, S., Alexander, T., Genadek, K., Goeken, R., Schroeder, M., Sobek, M., 2010. Integrated Public Use Microdata Series (ipums): Version 5.0 [Machine-Readable Database]. University of Minnesota, Minneapolis.

Saloutos, T., 1956. They Remember America: The Story of the Repatriated Greek-Americans. University of California Press, Berkeley.

Sarna, J.D., 1981. The myth of no return – Jewish return migration to eastern Europe, 1881–1914. Am. Jew. Hist. 71 (2), 256–268. Steiner, E.A., 1906. On the Trail of the Immigrant. FH Revell Company, New York.

Stolz, Yvonne, Baten, Joerg, 2012. Brain drain in the age of mass migration: Does relative inequality explain migrant selectivity? Explor. Econ. Hist. 49 (2), 205–220.

Thom, K., 2010. Repeated Circular Migration: Theory and Evidence from Undocumented Migrants. New York University, mimeo.

United States Bureau of Immigration, 1908–1932. Annual Report of the Commissioner General of Immigration. U.S. Government Printing Office.

Ward, Z., 2016. There and back (and back) again: Repeat migration to the United States 1897–1936. Unpublished Manuscript

Williamson, J.G., 1995. The evolution of global labor markets since 1830: background evidence and hypotheses. Explor. Econ. Hist. 32 (2), 141–196.

Willcox, Walter F., 1931. Immigration into the United States. In: International Migrations, vol. II: Interpretations. NBER, Cambridge, pp. 83–122.

Wyman, M., 1993. Round-Trip to America: The Immigrants Return to Europe. Cornell University Press, Ithaca, 1880–1930.

Yang, D., 2006. Why do migrants return to poor countries? evidence from Philippine migrants' responses to exchange rate shocks. Rev. Econ. Stat. 88 (4), 715–735.

.....................................................................................................................................................................

Example no 1:

Review and comments of “Closing Heaven’s Door: Evidence from the 1920s U.S. Immigration Quota Acts.”

This paper uses Census micro data and census of population and manufacturing statistics by detailed geography for the 1900‐1940 period to ask how a variety of labor market, demographic, and manufacturing outcome trends change after the imposition of national origin‐based immigration quotas in the 1924 differentially in places likely to be more affected by those quotas because of having pre‐existing large concentrations of immigrants from countries whose immigration levels were cut by the quotas. Specifically, the key right‐hand side variable is a kind of version of the usual “shift share” instrument used in local labor market studies of immigration, shown on page 10 (two equations, which I combine):

_ _ _

  max , 0

_ _

where _ is the stock of foreigners from origin n (living in U.S. location c – a county or a city depending on the regression) in 1910, and _ is the sum of gross arrivals from n over the years (y) specified. Mexico and Canada are assigned 0 in (and therefore excluded from) this sum, as are any other countries whose sum of annual immigrant arrivals were larger over 1922‐1930 than over 1910‐1914. It is not reported if there are any such countries.

This is the right‐hand side variable that is used in event‐study and difference‐in differences type regressions where, depending on the outcome, the “post” period is essentially after 1920 (see page 12, equations 3 and 4). Many outcomes are in changes: ln population, ln manufacturing value added and horsepower; the latter two are measured as 5 year changes and the former as 10 year changes. Other outcomes include the log occupational score (wage) and various demographic measures at the individual level in IPUMS data, which are measured in levels. Regressions often include area (county or city), time, state x year effects, and in some cases other controls interacted with year.

This quota exposure variable is associated with lower population growth after 1920 (Table 2) compared to during the 1910s. It is associated with lower value added growth and (manufacturing capital) horsepower growth after 1919 compared to 1914‐1919 (Table 3). These associations are sometimes larger in urban counties, and the horsepower results are only shown in a city‐ (rather than county‐) level analysis. This variable is also associated with declines in occupational scores of native workers, with relatively smaller declines or even increases for blacks (Table 5). It is also weakly associated with other demographic changes, including lower fertility among immigrants.

Comments

I think studying the impact of the early 20th century immigration restrictions is an important and timely topic. The authors have assembled and impressive dataset to do it.

Studying these restrictions is conceptually challenging. The “treatment” is the essentially unmeasurable “absence” of immigration flows that result from the quotas. I’m sure the authors struggled with how to capture this, and what they do is quite creative. But I don’t find it all that meaningful. Fundamentally, an approach like the authors’ is essentially still using comparisons of places that got more immigrants to fewer based on their initial country composition of immigrants, just like the existing literature on the labor market impact of immigration does, but then putting a negative sign in front of the instrument and calling it the effect of “quotas.” That description may sound uncharitable (but I do not mean it to be); another way to put what I am saying is there is fundamentally no way to identify the effect of restricting immigrant arrivals other than to study the impact of immigrant arrivals and saying restricting it would have the opposite effect. We don’t get to observe the universe in which the 1924 law failed to pass. I don’t know if this is just semantics; I don’t think it is.

That said, even the authors’ measure is not quite the negative of the impact of immigration. It has a number of features that deviate from standard instruments for studying the labor market impact of immigration:

• It excludes immigrant groups not subject to the quota, or whose immigration is actually increasing. What this means, though, is that at least some of the places with “low change in immigration” according to the authors’ measure may, in some cases actually have increases in immigration. For example, Mexican enclaves in the southwest may have been experiencing large increases in immigration due, to the surge in Mexican immigration in the 1920s. By imposing zero on this, I think the authors’ method potentially inflates the estimated impact of “non‐immigration.” It would be sort of like taking everyone with above 12 years of education in a wage regression and reassigning them 12 years of education; this imposition would make the “returns” to completing high school appear to be much larger. That said, I do not know how much of an issue this is in practice. o Besides the Canadians and Mexicans, are there other groups with positive immigration change between 1910‐1914 and 1922‐1930? As there are really only a handful of origin groups, it seems like the author should report the change in immigration for all of them individually, at least in an appendix table. o One additional strange feature of this measure is that it compares the flows over one four‐year period to a different eight‐year period. Perhaps the second number should be divided by two to make it more comparable?

• The authors’ measure is also not constructed to meaningfully capture how any theory says immigration should affect the labor market. It does not measure the population share of immigrants, nor the growth in immigrants, and it is not even necessarily proportional to those things. A more meaningful (and standard) measure would drop the denominator of one or the other term above, and also probably scale by some population size measure. For example a measure almost like the authors (and also incorporating some of my other suggestions) but would be more meaningfully measured in terms of potential labor market impacts would be:

_ /2

∑

The numerator of this measure is now a measure of predicted immigrant flows, which is scaled by population. (But even this is not done. Why would want to measure immigrant flows over a four‐year period on outcomes that are mostly decadal? So I might multiply this by 10/4 or something to make it in decade equivalent flows.)

o The fact that the authors’ measure is not scaled by population probably helps explain why the authors’ results are sensitive to controls for population size, and also whey they are “larger” in some cases in urban areas. It is not necessarily the case that the effects of immigration are larger in urban areas, it’s that a unit increase in this measure may translate into more immigration in urban than rural areas. o More generally, I find it hard to give a meaningful interpretation to the magnitude of authors’ measure. The authors do, too, apparently, which is why they interpret it in standard deviation units. But if these correspond to different actual amounts of immigration in different parts of the country, it remains not very meaningful even in standard deviation units.

So I am of the opinion that the authors should abandon this approach altogether and try studying the impact of immigrant flows in a more traditional way, and perhaps use that to simulate what impact the lost immigrants due to national origin restrictions had.

However, mine is quite a radical suggestion, and I realize the authors might not want to upend their whole paper just because some referee said so. So if they persist in taking their current approach, I have a few other comments about the authors’ current specification.

• First, they are essentially asking how outcomes change in places with lower inflows of immigrants to higher inflows. But in this era there was a lot of return migration, with which the inflows might be correlated. This is why many times we look at net changes in stocks rather than inflows.

• Levels vs. changes. Many of the outcomes the authors look at in changes. I guess this makes sense since it matches up with the immigrant flow measure.

But: o Wages are measured in levels. So the author is regressing the level of wages on the change in immigration, essentially. But wages levels are theoretically tied to the overall stock of immigration rather than the flows (e.g., Borjas, 2003). And immigrant inflows are correlated with stocks (that’s the basis of the authors’ measure, after all.) so this likely produces estimates that are upward biased in magnitude.

o The authors change specification also controls for fixed effects. But in this specification, this is equivalent to putting in county‐specific trends. So one way to make the wage and other results more comparable is to control for county‐specific trends in the wage regressions. (Another would be to aggregate the wage to a mean log wage by county‐year, and use that in the regression in changes.)

• The authors do a lot of subgroup analysis by splitting the sample. But notice that this implicitly allows all of the numerous controls that the authors’ have in the regression (e.g., the year or county effects) to have differential effects by the subgroups. These implicit controls are not in the specification that combines the groups, so it is hard to compare the two. o For example, this might account for the puzzling fact that there is some evidence of increases in black wages for both men and women separately (App Table 3), but not when you combine the genders (Table 5). The difference must be that results separately by gender implicitly add controls for WWI exposure x Ipost x gender, literacy act of 191 exposure x Ipost x gender individual controls x gender, county fixed effects x gender, gender‐state‐specific unrestricted time trends and more, none of which was in the regressions that combined genders. The same goes for split samples by place of birth.

• I recommend the authors replace their figures with plots of the event study coefficients. These could be shown for every outcome, instead of just selected outcomes as the paper is currently drafted. (Then the tables could then focus on the diff‐in‐diff results.) Showing that there are no pretrends is important. I am not surprised that there are, though, since immigration is a pretty trendy variable. And it seems to be less of an issue with the population controls, so it might also be alleviated by my suggestion of scaling the treatment measure by city size.

By the way, Appendix A is great. I recommend the authors continue to think carefully and systematically about the structure of the problem they are investigating like they do in this section before plowing into their empirical methods. Appendix Table 2 is also pretty good. So maybe the appendix should be the main paper.

Anyway, good luck to the authors. They are writing on an important topic, and are using rich and complex data in creative ways. I think this could be a very good paper if they were using more standard, or at least more carefully and clearly rationalized, analytical approaches.