Self-Selection of Mexican Migrants in the Presence of Random Shocks:Evidence from the Panic of 1907

Self-selection of Mexican migrants in the presence of random shocks:Evidence from the Panic of 1907

David Escamilla-Guerrero*

and Moramay Lopez-Alonso†

[email protected] November 29, 2018

Abstract Using a unique data set consisting of individual border crossings, military recruitment records and passport applications, this paper estimates the selection pattern of Mexican migrants in the beginnings of the US-Mexico migration flow (1884-1910); and exploits the presence of two idiosyncratic shocks to evaluate their impact on short-run shifts in selection patterns. To estimate the initial selection pattern of Mexican migrants, we use height as a proxy for human capital. The results suggest that the first Mexican migrants belonged to the upper tail of the height distribution of the Mexican working class: migrants were positively self-selected. Additionally, the financial crisis of 1907 in the US and the presence of regional droughts in Mexico during 1908 modified significantly the selection patterns. Before the crisis, immigrants were 1.5 centimeters taller than the average soldier. During the panic, immigrants became less positively self-selected (0.9 centimeters taller), but return to a strong positive self-selection (2.1 centimeters taller) after the crisis. We argue that the stronger positive selection in the post-crisis period was influenced by droughts in Mexico.

Keywords: labor migration, migrant self-selection, Panic of 1907, Mexico JEL Classiﬁcation Numbers: F22, J61, N36, O15

Acknowledgments of DEG: I am especially grateful to my PhD supervisors Eric Schneider and Joan Rosés for their guidance and invaluable comments. I also want to thank Daniela Gutiérrez and Marco Villeda for their assistance on the data transcription. This research was developed with the ﬁnancial support of: the Mexican National Council for Science and Technology (2015-2018) - Scholarship No. 409165; the Mexican Ministry of Education Scholarship (2015-2016); the Radwan Travel and Discovery Fund (2016) - London School of Economics; the Pre-Dissertation Exploratory Grant (2017) - Economic History Association; the Research Fund for Graduate Students (2017) - Economic History Society; and my fellowship at the World Institute of Development Economics Research - United Nations. This research was beneﬁted from the academic visit at El Colegio de México in 2016. I am specially grateful to Professor Graciela Márquez and Professor Gerardo Esquivel for their support during this period. I am grateful to Dr. Fernando Pérez and Dr. León Fernández for his comments during my academic visit at The Bank of Mexico in 2018. I am grateful for the mentorship of Dr. Miguel Niño-Zarazúa during my fellowship at the World Institute of Development Economics Research. All errors are mine.

*London School of Economics - Department of Economic History †Rice University - Department of History.

1 1. Introduction

Since Borjas(1987) formalized the basic Roy model (Roy, 1951), migrant selection has become a proliﬁc topic in labor economics. Precisely, it provides a framework to understand the economic gains or loses from migration, which has been a core topic in the public sphere since the Age of Mass Migration (1850-1913) (Spitzer & Zimran, 2018). While the literature on migrant selection acknowledges that the decision to migrate depends on systemic and idiosyncratic reasons, the factors that may explain shifts in selection patterns over time remain empirically unexplored (Abramitzky & Boustan, 2017, p.1325). Therefore, for more than one century, we have implemented immigration policies without understanding completely the mechanics behind migrant selection.

This paper estimates the selection pattern of Mexican migrants in the beginnings of the US-Mexico migration ﬂow (1906-1908) and exploits the presence of two idiosyncratic shocks to evaluate their impact on short-run shifts in selection patterns.

To estimate the initial selection pattern of Mexican migrants, we use heights (as a proxy for human capital) of migrants, soldiers and passport holders. The migrant sample comes from individual border crossings from 1906 to 1908. The soldiers sample comes from military records of common soldiers and elite forces that proxy for two extreme points of the height distribution of the Mexican working class. The passport holders sample consists of passport applications representing the Mexican upper social class. Together, the military and passport records are used as comparison samples to determine whether the ﬁrst Mexican migrants were drawn from the upper or lower tail of the height distribution in Mexico.

The results suggest that the ﬁrst Mexican migrants were positively self-selected: they belonged to the upper tail of the height distribution of the Mexican working class. Indeed, migrants were 1.7 centimeters taller than the average soldier; almost as tall (0.5 centimeters shorter) as the military elite forces; and 2.3 centimeters shorter than the passport holders. We also document that Mexican migrants might have ended up in the upper tail of the height distribution of a pool of migrant workers from different source countries that competed for low-skill jobs. Therefore, following the

2 basic Roy model, the beginnings of the Mexico-US migration ﬂow were characterized by a positive selection of Mexican migrants on the basis of height. To evaluate if a random shock could inﬂuence short-run shifts in selection patterns, we focus on two events.

First, we use the Panic of 1907 as a exogenous labor demand shock in the US side. The Panic of 1907 was one of the most severe financial crises in the US before the Great Depression (Frydman et al., 2015, p.928; Moen & Tallman, 1992, p.611; Odell & Weidenmier, 2004, p.1003), and it severely affected the credit system in the US. During this crisis, banks and financial institutions of many cities limited or suspended their cash payments (Andrew, 1908, p.497), and around 2 thousand firms and over one hundred state banks failed (Markham, 2002, p.32).

The Panic of 1907 was a phenomena conﬁned to the US (Johnson, 1908, p.455), which severely affected industries such as railways and mining where Mexican immigrants were mostly employed. In addition, the crisis occurred during a period when Mexicans could migrate to the US without restrictions. These facts provide a unique opportunity to understand how idiosyncratic demand shocks might modify migrant selection patterns in a free mobility setting.

The results reveal that, before the panic, migrants were 1.5 centimeters taller than the average soldier. During the panic, migrants became less positively self-selected (0.9 centimeters taller), but returned to a strong positive self-selection (2.1 centimeters taller) after the crisis. This “U” pattern holds with all comparison samples (common soldiers, elite forces and passport holders) and when controlling for unobserved factors within migration regions and over time that might have inﬂuenced the individuals’ heights.

Using the basic Roy model as theoretical framework, we show that the change in the selection pattern during the panic period might have occurred because the Mexican government implemented a repatriation program. This policy aimed to assist immigrants that suddenly experienced economic hardship in the US due to the Panic of 1907. The program covered the transportation of immigrants back to Mexico eliminating the costs of returning home. As a consequence, the “best” individuals that initially emigrated to the US might have returned to Mexico, displacing relatively shorter

3 laborers. This excess of labor might have pushed displaced-individuals (relatively shorter) to emigrate. The fact that we observe migration flows (less positively selected) during the panic period implies that the shock devalued the observed and unobserved characteristics of Mexican migrants in the US, but not enough to discourage emigration completely. This suggests that Mexican immigrants might have not returned, unless mobility costs were sufficiently low. By eliminating the mobility costs, the repatriation program implemented during the Panic of 1907 influenced the immigrant’s decision to return, and therefore the migrant selection patterns during this period. In other words, government interventions can shift selection patterns over time (at least in the short-run), just as Abramitzky & Boustan(2017, p.1325) intuited.

On the other hand, to explain the stronger positive selection observed in the post-panic period (relative to the pre-panic period), we exploit the presence of climate shocks in Mexico. Contreras(2005, p.123), Clark(1908, p.473) and Mayet et al.(1980, p.757) document the presence of regional droughts in Mexico from 1907 to 1908. As a consequence, there were important crop losses in some areas (Cardoso, 1980, p.12). We conﬁrm the presence of such droughts using the Mexican Drought Atlas (Stahle et al., 2016) that provides reconstructions of the self-calibrating Palmer Drought Severity Index (PDSI). According to the index scale, these climate shocks can be considered as mild or moderate droughts, but their persistence for two consecutive years might have pushed taller individuals to migrate during the post-panic period.

Precisely, the results suggest that, in the post-panic period, the average migrant was as tall as the military elite, however those whose last permanent residence was in a state experiencing droughts were 0.5 centimeters taller than the military elite. This also holds for other groups. We believe that this stronger positive selection pattern might be a result of two overlapping forces: the reactivation of the American ﬁnancial system and the persistence of regional droughts in Mexico since 1907.

The contributions of this paper are two-fold. Firstly, we present the initial selection pattern of Mexican migrants. In contrast to migrants from the European periphery, who were negatively selected (Abramitzky et al., 2012; Cohn, 1995; Spitzer & Zimran, 2018),1 in the Age of Mass migration, the ﬁrst Mexican migrants were positively

1See Abramitzky & Boustan(2017) for a review.

4 selected. Moreover, Mexican migrants might have displaced European immigrants in diverse sectors of the American Southwest (Clark, 1908).2 This ﬁnding lines up with literature arguing that Mexican migrants were not drawn from the lower tail of the educational, ability or height distribution (Chiquiar & Hanson 2005; Orrenius & Zavodny 2005; Kosack & Ward 2014). In other words, in the beginnings of the twentieth century, Mexicans sent its best people to the US, who in fact might have been key for building up two of the most dynamic economic regions in the US nowadays: the American Southwest and California.

Secondly, due to immigration policies in place, contemporary contexts complicate the assessment of factors driving the migrant selection patterns. However, the historical context of this research, a free labor mobility period, helps us to conﬁrm that government interventions (migration policies) and climate shocks could shape migrant selection in the short-run. Furthermore, shifts in selection patterns might be a result of overlapping effects from the demand and supply sides.

The paper is organized as follows, the next section describes the Panic of 1907 in a historical perspective. Subsequently, we present the characteristics of the data we use. The fourth and fifth sections present the identification strategy to estimate the selection patterns and the interpretation of the results, respectively. The sixth section develops the migrant selection mechanisms of adjustment. The last section offers some final remarks.

2The only quantitative benchmark we know is the research of Spitzer & Zimran(2018) which studies migrant selection of Italians to the US between 1907 and 1925. According to their statistics, the average height of Italian immigrants was 163.8 centimeters, but immigrants from the North of Italy were taller (165.6 centimeters) (Spitzer & Zimran, 2018, p. 231). In our sample the average Mexican immigrant was 167.8 centimeters tall, but immigrants from the North of Mexico were 168.8 centimeters tall. Therefore, on average, Mexican migrants might have been considerably taller than Italian immigrants.

5 2. Historical background

By the end of the nineteenth century, the United States emerged as the world´s leading manufacturing nation and its economy experienced average growth rates of 4.5 to 5%. Two factors were key to understand this robust economic growth.

Firstly, the constant labor supply that the US received from international migration ﬂows. From 1820 to 1920, about 36 million Europeans migrated to the United States looking for better life conditions. Mexicans joined this mass migration from 1880s, and in 1910, the stock of Mexican migrants in the US was about 222 thousand, and it doubled by the end of the Mexican Revolution (1910-1920) (García y Griego, 1983).3 The Mexican migrants were mostly rural unskilled laborers moving to the American Southwest and Midwest. Moreover, Mexican migration was unrestricted before 1910, representing an inexhaustible source of labor for the US economy.

Secondly, the American financial system was expanding rapidly by the end of the nineteenth century, facilitating investment for the creation of new firms in all economic sectors. Indeed, in 1907 there were 16 thousand financial institutions, which facilitated capital for the economy´s buoyant growth (Bruner & Carr, 2007, p. 116).4 These institutions were small unit banks, fiduciary trust companies, clearing houses and exchange houses that provided financial services at the local-level. Most of these financial institutions were supported by small companies and some others by the Bank of England and/or the United States Treasury. Furthermore, this fractionalized financial system operated without a Central Bank (Bruner & Carr, 2007). This condition along with the increasing optimism, engendered by the robust performance of the economy, fueled the tendency of the public to take on more risk and invest in companies from dynamic industries, e.g. the railways and mining. Therefore, the access to capital was relatively unconstrained for the US economy.

3Mexican migrants were employed mainly in farms, mines and the construction of railroads (Clark, 1908). By 1909, Mexicans migrants represented 17% of the labor force of the American railway industry (Verduzco, 1995). 4To dimension the size of the US ﬁnancial system at the time, in 2007 existed 7,500 ﬁnancial institutions.

6 2.1 The Panic of 1907

Besides its economic growth, the US economy was characterized by ﬁerce speculative processes in the ﬁrst years of the twentieth century. As a sign of this phenomenon, the Dow Jones index doubled between 1904 and 1906. At the end of 1905, the call money rates were 25% and it was foreseen to increase to 60% the following year (Markham, 2002, p. 29). In fact, stock prices were rising rapidly on Wall Street, peaking in September 1906 (Bruner & Carr, 2007, p. 115). This speculative process occurred within a period of increasing long-term investments. National and state banks increased their bond and stock assets from 50 million in 1892 to 487 million in 1907. Similarly, trust institutions increased their holdings of debt securities by more than 500%, reaching a value of 785 million dollars in 1907 (Johnson, 1908, p. 457). This environment blocked the liquidity of the economy and without a Central Bank, a market crash was a matter of time, as argued by Johnson(1908, p. 454).

In March 1907, a wave of panic left losses of 2 billion in stocks. Major players like the railway company Union Paciﬁc saw their shares devalued by 25 dollars in a single day. This ﬁnancial panic caused some major mining producers such as United States Steel Corporation to temporarily suspend dividend payments, however these were resumed in the following months (Markham, 2002, p. 29). To neutralized the panic, companies and city governments increased the interest rates of their bonds, however the wave of selling continued, pushing down the stocks´prices.5

As the speculative process developed, more and more fiduciary institutions reduced the required reserve of 10% of their deposits, because they used them to invest in financial and speculative activities. In October, the Knickerbrocker Trust Company, the third largest trustee in New York, went into bankruptcy due to speculative actions to control the price of copper. This event increased the panic among the public, sank the financial market and triggered a series of suicides among depositors. Throughout August to December 1907, two thousand companies went bankrupt as did 100 state banks and 30 national banks (Markham, 2002, p. 32).

5This phenomenon was record by the American press throughout 1907. For instance: ”New York. Aug. 12 - The wildest break in the stock market since the present wave of selling occurred today. I carried stocks down from 1 to 17.5 points. In some cases to new low records. About one-half of the entire number of issues dealt on the exchange rate were sold at new low prices for the year.” (The Washington Post, 1907).

7 This financial crisis, known as The Panic of 1907, was one of the most severe financial crises in the US before the Great Depression (Frydman et al., 2015, p. 928; Moen & Tallman, 1992, p. 611; Odell & Weidenmier, 2004, p. 1003). In fact, contemporaries argued that it was "probably the most extensive and prolonged breakdown of the country´s credit mechanism which has occurred since the establishment of the national banking system" (Andrew, 1908, p. 497). This financial crisis affected consumers and producers in all sectors, first through the speculation process and then through the suspension of payments that constrained basic transactions.

Therefore, in 1907 the US economy ceased to expand since credit for manufacturing or industrial companies was considerably constrained. As a consequence some industries curtailed operations, jobs were destroyed, and trade was considerably depressed (Frydman et al., 2015, p. 912; Johnson, 1908, p. 454).6 To restore the operation of the ﬁnancial system, substitutes for cash were emitted and rationalized to the population, but full convertibility of deposits by the nation´s banks was not restored until January 1908 (Frydman et al., 2015, p. 912).

The fact that the Panic of 1907 was a phenomena conﬁned to the US (Johnson, 1908, p. 455), which severely affected industries where Mexican immigrants were mostly employed; and that the crisis occurred during a period when Mexicans could migrate to the US without restrictions, provides a unique opportunity to understand how idiosyncratic demand shocks might modify migrant self-selection patterns in a free mobility setting.

3. Data

3.1 Immigrant Sample

The registration of aliens arriving at Mexican-US land border ports began in 1906. American authorities used different types of documents to collect information about immigrants. These documents are known as Mexican Border Crossing Records, and at the time were conducted by the Bureau of Immigration and Naturalization. The immigrant

6The American industrial production peaked in July 1907 and then fell 30% in the second half of the year (Hansen, 2014, p. 555).

8 sample used in this research comes from the publication A3365,7 which contains two-sheet manifests that provide rich and diverse information about immigrants. Demographics as age, sex, marital status, occupation, ability to read and write, citizenship and race were reported. The manifests include anthropometric data of the immigrant (height, complexion and color of eyes and hair), and geographical information regarding his/her birthplace, ﬁnal destination and last residence. In addition, these documents recorded information regarding the immigrant’s current and previous migration spells.

We reviewed the five rolls of the publication to identify the total data (population size) in the microfilms.8 Data from 1909 was not considered to guarantee capturing only labor migrants and not refugees from the Mexican Revolution (1910-1920). Therefore, we delimited the transcription process to the period from July 1906 to December 1908. Using as reference the port of entrance, the data was transcribed following a stratified random sampling strategy. The final sample size is 9,083 individuals. Figure 5 shows the eight ports of entrance along the Mexico-US border that were identified in the microfilms.

A concern about these manifests is that they record only immigrant crossings in ofﬁcial entrance ports. Estimations of undocumented Mexican migration are scarce for the period, mainly because Mexicans were not consider immigrants at the time (Durand, 2016). Yet, these data can be considered unique since it is the only migration data at individual level for the Porﬁrian period (1876-1911). Hence, it allows to know who migrated before the Mexican Revolution, and to identify the initial dynamics of the Mexican mass migration.9

To estimate the selection of Mexican migrants, this study uses as core data the immigrant’s age, height, birthplace, occupation, and literacy. The anthropometric data was recorded by a sworn physician and surgeon, who examined each immigrant at the port of entrance. The ability to read and write was evaluated by the immigration ofﬁcial on duty. However, the immigrant’s age, birthplace and occupation were

7The title of the publication is: Lists of Aliens Arriving at Brownsville, Del Rio, Eagle Pass, El Paso, Laredo, Presidio, Rio Grande City, and Roma, Texas, May 1903-June 1909, and at Aros Ranch, Douglas, Lochiel, Naco, and Nogales, Arizona, July 1906-December 1910. 8We did not ﬁnd data for years prior 1906 and for entrance ports in California. 9Gratton & Merchant(2015) argue that mass migration of Mexicans started from 1900.

9 self-reported, and consequently subject to bias.

3.2 Comparison samples: Military and Passport Records

In order to compare migrants with the population that chose to remain in Mexico, we use military recruitment ﬁles and passport records. This data is the result of an extensive and unique archival work of López-Alonso(2015), who uses height to study secular trends of living standards in Mexico from 1850 to 1950.10

For the purpose of this research the military sample has two components:11

1. Federales: Soldiers of the Mexican army (cavalry, infantry and artillery), who served and retired or lost their lives on duty; and soldiers of the Mexican army, who left their service without authorization before the ending of their contracts, i.e. deserters. At the time, there were minimum requirements to enlist in the army such as age (be between 18 and 45 years of age), height (be at least 160 centimeters tall), be able to understand Spanish language, be a Mexican citizen, among other health requirements. While these requirements might have introduced systematic biases to the sample, López-Alonso(2015, p.112) shows that neither of these requirements were enforced during the period.12 The sample size is 7,088 male individuals born between 1840 and 1950. The source of this data are the archives of the Ministry of National Defense (Secretaría de la Defensa Nacional - SEDENA).

2. Rurales: A militia known as the rural police, which was created in 1860 as an armed group loyal to the president. The members of this militia received a higher salary than the regular soldiers (federales), and in its beginnings they should bring their own horse and weapons. This changed during the Porﬁrian period when they had to pay for their equipment. The rurales often received additional monetary rewards and political favors to maintain the stability in the country. The sample size is 6,820 individuals born between 1840 and 1900.13 This sample covers all the enlistment

10López-Alonso(2015, p.107) provides a detailed description of the archival worked developed. 11López-Alonso(2015, p.111) breaks down the federales sample in three subsamples: deserters, non- deserters and women of the military. 12Moreover, it is not clear if the enrollment into the army was completely voluntary. Although the military did not required service until 1939 (Kosack & Ward, 2014), there is evidence that forced recruitment mechanisms might have been implemented at the time (Durand, 2016) 13The desertion rates in this militia were high since its members could sell their equipment at any time

10 records of this militia, and the source of this data are the archives of the General National Archive - Public Administration Section (Archivo General de la Nación - AGN).

We considered the previous samples separately because clearly the rurales were not representative of the average Mexican soldier (federales). The fact that they received a higher salary and had to bring their own equipment suggests that they might have been relatively richer than the average soldier. Moreover, they received extra monetary and non-monetary rewards for their service, hence along this research we consider the rurales as an elite militia of that time. In other words, each of the military samples might represent different parts of the height distribution of the Mexican working class.

Finally, the passport records consist of all the passport applications made from 1910 to 1942 that contain the height of the applicant. We believe that this sample might represent the Mexican upper social class since the passport holders might be individuals with the economic means to travel abroad for business, leisure or education purposes (López-Alonso & Condey, 2003). Nevertheless, two important characteristics of this data should be noticed. On the one hand, the height was self-reported by the applicant. On the other hand, the records capture all the emitted passports but not all the travel permits emitted by other regional ofﬁces for applicants that could not travel to Mexico City. The sample size is 6,746 male individuals born between 1860 and 1922. The source of this data are the archives of the Ministry of Foreign Affairs (Secretaría de Relaciones Exteriores - SRE) (López-Alonso, 2015, p. 121-122).

Besides the potential sources of bias described above, we acknowledge that the three comparison samples could be selected for different reasons.14 For example, since the military samples record volunteers rather than conscripts, they contain only individuals who choose to enlist in the military. Following Bodenhorn et al.(2017, p. 201), the decision to join the army reﬂects the individual’s evaluation of his future based on his accumulated human capital (height). Thus, the decision to enlist in the military becomes less attractive for taller individuals, specially in a growing economy.

and locating deserters was costly López-Alonso(2015, p.117-121). 14Despite its limitations, these data arise as the most suitable information to measure living standards during the Porﬁriato. For the Mexican case, long term data series on income, wages, prices and mortality is unreliable and scattered (López-Alonso, 2007).

11 To test for the presence of selection, we performed Bodenhorn et al.(2017) test for the three comparison samples.15 We found that our samples might present height-based selection considering the individual’s year of birth and year of registration. This might imply that our samples are not representative for the Mexican population. However, the objective of this research is not to estimate secular trends in height of the Mexican population, but to use the comparison samples to identify whether the individuals that emigrated were taller or shorter relative to the average soldier (federales), the average elite militia (rurales), and the average passport holder. In other words, we use the comparison samples to identify to which part of the height distribution of the Mexican population the emigrants might have belonged.

3.3 Data Reﬁnements

To obtain the best results possible, we impose a series of reﬁnements to the data. Regarding the migrant sample, only males reporting their town and state of birth in Mexico were considered.16 This allows to estimate accurately the migrant selection at a regional level. In addition, we keep individuals that might had reached their terminal height, i.e. individuals between 18 and 65 years old. This controls for growing and shrinkage effects.

We applied two different refinement rules for the comparison samples. For the rurales sample, we keep individuals that enlisted before the Mexican Revolution and had reached their terminal height. This avoids confounding any effect of the conflict without affecting the sample size significantly. Regarding the federales and passports samples, we keep individuals that had reached their terminal height before the Mexican Revolution regardless the year of registration. We decided to apply this partial refinement because keeping only those individuals registered before the revolution reduces significantly the sample sizes. Therefore, some effects of the revolution might be capture.

Table 1 presents the main characteristics of the reﬁned samples samples. On average,

15The test consists on including all possible interactions of birth-year and recruitment-year effects in a regression in which height is the dependent variable. 16The sample was constrained to males because the female military sample (comparison group) does not count with any geographic information.

12 emigrants were 167.8 centimeters tall, slightly taller than the elite militia, three centimeters taller than the average soldier, and two centimeters shorter than the passport holders.This initial ﬁnding suggests that emigrants did not belonged to the lower tail of the height distribution of the Mexican working class. However, almost all emigrants were unskilled laborers and had the lowest literacy rates among the samples, implying that they might have emigrated to perform activities with high returns to physical strength or health human capital. In contrast, 62% of the passport holders reported to be professionals, conﬁrming that they might have belonged to the upper social class.

The regional distribution of the samples shows that emigration occurred mostly from the North and Bajio regions, while military recruitment might have taken place mainly in the Bajio and Center regions.17 The passports sample concentrates in the Central region, reafﬁrming that most passport holders might have lived in Mexico City or nearby states, where the social elite resided. Observations from the South region are relatively scarce, complicating any estimation for this region.

Table 2 shows that among emigrants, the ones from the Bajio were the shortest: 2.1 centimeters shorter than emigrants from the North. Furthermore, the difference in height between the migrant and the military samples widens in the Central and South regions (Figure 1). This conﬁrms the intuition of Abramitzky & Boustan(2017, p. 20) about the existence of substantial variation in migration patterns across regions of sending countries.

Similarly, based on the amount of cash held at the crossing (Table 1), migrants from the Center region were considerably richer than the rest. They reported to have 20 dollars, two times the amount reported by the migrants from the North. Bajio migrants had less than one dollar at the crossing, thus they might have been the poorest as argued by Durand(2016).

17Considering the population levels in 1907, the Bajio states were among the most populated. The states of Guanajuato, Jalisco and Michoacan were more populated than Mexico City. Therefore, the recruitment of soldiers could have been common in this regions.

13 Table 1: Summary Statistics for Migrant, Military and Passports Samples (Males)

Migrant Rurales Federales Passport

Average Height (cm) 167.8 167.3 164.5 170.1 Average Age (years) 28.5 28.4 35.7 49.0 Labor Class (%) Unskilled 90.4 42.1 72.8 3.7 Skilled 7.1 55.0 24.6 33.6 Professional 2.6 3.0 2.6 61.9 Literacy Rate (%) Literate 38.3 56.2 45.4 100.0 Marital Status (%) Married 48.9 NA NA NA Single 49.7 NA NA NA Widowed 1.4 NA NA NA Region of Birth (%) North 46.3 3.0 19.8 13.3 Bajio 51.8 57.2 26.8 30.2 Center 1.6 35.8 42.1 47.8 South 0.3 4.0 11.2 8.7 Cash on hand - US dollars (median) North 10 NA NA NA Bajio 0.5 NA NA NA Center 20 NA NA NA South 15 NA NA NA

Observations 4,651 4,408 1,130 1,231

Source: Immigrant sample from the Mexican Border Crossing Records - Microﬁlm publication number A3365. Military and Passport samples from López-Alonso(2015). Note: The immigrant’s regions of birth and occupations were classiﬁed following López-Alonso(2015, p. 127 & 128). The sample was constrained to males because the female military sample (comparison group) does not count with any geographic information. We consider individuals that had reached their terminal height, i.e. individuals between 18 and 65 years old.

14 Figure 1: Distribution of Heights by Region

(a) Complete Sample (b) North

Table 2: Average Heights (centimeters) across Regions and Samples

North Bajio Central South

Migrant 168.9 166.8 168.0 167.2 (5.9) (5.9) (6.9) (5.89 Rurales 167.8 167.6 166.9 166.6 (6.1) (5.8) (5.9) (5.6) Federales 167.0 165.1 163.7 161.7 (6.8) (6.8) (6.0) (5.2) Passports 171.3 171.1 169.4 168.6 (7.6) (7.6) (7.3) (6.9) Observations 2,674 5,604 2,718 424

Source: Immigrant sample from the Mexican Border Crossing Records - Microﬁlm publication number A3365. Military and Passport samples from López-Alonso(2015). Note: Standard deviations in parenthesis. The sample was constrained to males because the female military sample (comparison group) does not count with any geographic information. We consider individuals that had reached their terminal height, i.e. individuals between 18 and 65 years old.

15 Based on the individual’s year of birth, Figure 2 shows trends in height of the migrant and military samples during the second half of the nineteenth century. The rurales and migrant samples show similar levels in height, suggesting that both groups faced similar conditions of health care, nutrition, disease environment and work assignments some 10 to 50 years before enlistment.18 To some extent, this suggests that those individuals that decided to emigrate could have enlisted in the elite militia and vice versa, i.e. they might have belonged to the same part of the height distribution. Conversely, the average height of the federales remains below the migrants and rurales average height for most of the period. Therefore, the average Mexican soldier was exposed to worse child’s early nutrition/health conditions than the rest of the individuals, meaning that they might have belong to the lowest social class in Mexico.

Figure 2: Differences in Height by Year of Birth

Source: Migrant sample from the Mexican Border Crossing Records - Microﬁlm publication number A3365. Military and Passport samples from López-Alonso(2015). Mexican Border Crossing Records (1906-1908).

18Schneider & Ogasawara(2018, p.64) argue that disease environment, proxied by infant mortality rates, have economically meaningful effects on child height at ages 6-11.

16 4. Empirical Strategy

To estimate the self-selection of Mexican migrants during the last years of the Porﬁriato, we regress the height of individual i (heighti) on a dummy variable that takes the value of 1 if the individual belongs to the immigrant sample and zero otherwise

(migranti), a vector of individual characteristics (Xi) that includes age categories,

skill level categories and region of birth, year of birth ﬁxed effects (αy) and state of

destination ﬁxed effects (γd):

0 heighti = β + Φmigranti + Xiθ + αy + γd + ei. (1)

We estimate Equation 1 with each of the comparison samples, hence the estimated coefficient Φ reflects the average difference in height between migrants and rurales, federales or passport holders, respectively. Since final adult height might not be reached until 24 years of age, the three age categories (18-20 years, 21-23 years and over 24 years) control for individuals that might still be growing at the moment of registration (Kosack & Ward, 2014, p. 1030). The region of birth categories (North, Bajio, Center and South) control for environmental factors such as food availability, dietary patterns or presence of diseases that might influence height at the regional level. The regional classification is based on López-Alonso(2015) and are consistent with (Durand, 2016) migration regions. We include skill level categories (unskilled, skilled and professional) to control for the potential relationship between cognitive abilities acquired in early childhood and height (Bleakley et al., 2014, p. 124).19

The year of birth fixed effects control for any factor influencing height across years, e.g. structural and idiosyncratic shocks effecting the living standards of the population along time. Finally, the destination fixed effects control for differences in the US demand composition at the state-level that might have influenced the self-selection of migrants.

The estimated coefﬁcients of Equation 1 consist on average estimates of the period October 1906 - December 1908. However, from August 1907 to January 1908 the US

19We assume that skilled and professional occupations at the time demanded more training or education relative to unskilled occupations.

17 economy suffered one of the most severe ﬁnancial crises before the Great Depression (Moen & Tallman, 1992, p. 611).20

As argued previously, the Panic of 1907 affected severely the credit system in the US. During this crisis, banks and ﬁnancial institutions of many cities limited or suspended their cash payments (Andrew, 1908, p. 497), and around 2 thousand ﬁrms and over one hundred state banks failed (Markham, 2002, p. 32). This shock affected every industrial sector in the US destroying employment sources (Bruner & Carr, 2007, p. 115).

We extend equation Equation 1 to capture shifts in the self-selection patterns as a consequence of this crisis:

post heighti =β + Φ1migranti + Φ2migranti × panic + Φ3migranti × panic (2) 0 + Xiθ + αy + γd + ei.

The estimated coefﬁcients Φ2 and Φ3 capture the difference in height of individuals that emigrated during the panic period (August 1907 - January 1908) or after the panic (February 1908 - December 1908), respectively. These estimates are relative to those individuals that emigrated before the panic (October 1906 - July 1907). The difference in height between pre-panic emigrants and the different comparison samples (non-

immigrants) is reﬂected in Φ1. Everything else equal, the estimated self-selection pattern during the Panic of 1907 is Φ1 + Φ2.

5. Results

5.1 Self-selection of Mexican Emigrants before 1910

Column 2 of Table 3 shows that migrants were positively self-selected relative to the average soldier: migrants were 1.7 centimeters taller than the federales. In contrast, columns 1 and 3 show that immigrants were 0.5 centimeters shorter than the rurales and 2.3 centimeters shorter than the passport holders. In other words, immigrants

20There is no consensus in the literature about the ending month of the crisis. Yet, the scholarship on the matter agree that normalcy in the ﬁnancial market was restored in January 1908 (Frydman et al., 2015, p. 937).

18 were almost as tall as the military elite forces, but clearly negatively selected relative to the Mexican upper classes.

Table 3: Self-selection of Mexican Migrants. Dependent variable: height (centimeters)

Comparison Samples Rurales Federales Passports (1) (2) (3) Migrant -0.529*** 1.725*** -2.346*** (0.200) (0.376) (0.508) Age categories 18-20 years -1.055*** -1.024** -1.221** (0.317) (0.432) (0.481) 21-23 years -0.456** -0.177 -0.578 (0.209) (0.404) (0.458) Skill level categories Skilled -0.041 0.864*** 0.752** (0.159) (0.257) (0.308) Professional 0.891** 0.850 1.668*** (0.425) (0.530) (0.420) Region of birth North 2.166*** 4.557*** 3.039*** (0.455) (0.520) (0.675) Bajio 0.721* 2.906*** 1.595** (0.424) (0.516) (0.668) Center 0.178 1.859*** 0.609 (0.436) (0.514) (0.675) Year of birth FE YES YES YES Destination FE YES YES YES Constant 175.043*** 159.615*** 165.128*** (1.479) (4.377) (4.392)

Observations 9,005 5,727 5,818 R-squared 0.041 0.107 0.074 Source: Mexican Border Crossing Records - Microfilm publication number A3365 and López-Alonso (2015). Notes: * = Significant at 10% level; ** = Significant at 5% level; *** = Significant at 1% level. Robust standard errors in parenthesis. The omitted categories are individuals over age 24, born in the South region and unskilled workers.

We acknowledge that the federales sample is selected because these individuals were not conscripts but volunteers; and it is expected that in a growing economy, as was Proﬁrian Mexico, the outside option of military service becomes less attractive for productive and tall individuals (Bodenhorn et al., 2017, p. 173). Therefore, the federales soldiers might have belonged to the lower tail of the population height distribution. The rurales were volunteers as well; however they were taller than the federales in the population height distribution. The estimated intercepts show that, on average, the rurales were 15 centimeters taller than the federales, considering the omitted categories

19 in Table 3. This large difference in the estimated height between military samples along with the estimated Φs are strong evidence that immigrants might have belong to the upper tail of the height distribution of the Mexican working class.

As argued by Kosack & Ward(2014, p. 1023), height is strongly correlated with wages in countries where large sectors of the economy rely on the physical productivity of labor.21 Even though the rurales and federales samples are not representative of the Mexican population by themselves, jointly, they allow us to infer that the migrants’ potential wage was very close to that of the military elite. More importantly, those individuals that decided to emigrate had unobserved individual-speciﬁc factors that reveal even a higher human capital accumulation (Bodenhorn et al., 2017, p. 201). Therefore, it is unlikely that, the ﬁrst Mexican immigrants were negatively self-selected relative to the average laborer.

The age categories confirm that individuals between 18 to 20 years of age were relatively shorter and might have been still growing. In addition, individuals from the North and the Bajio were considerably taller than the rest, confirming that regional environmental factors influenced height in Mexico.

As a robustness check, we include state of birth ﬁx effects instead of region dummies in the models for which we count with desegregated geographic data (rurales and passports). Table 6 in the Annex show that the previous results hold in magnitude and signiﬁcance.

5.2 The Effect of the Panic of 1907

Table 4 shows the effect of the Panic of 1907: individuals that migrated during the crisis were approximately 0.5 centimeters shorter than their pre-panic counterparts, i.e. migrants became less positively self-selected during this period. However, those who migrated after the crisis were approximately 0.6 centimeters taller than pre-panic migrants. These effects modiﬁed consistently the self-selection patterns observed before the panic.

21Schultz(2002) estimate that a one centimeter increase in height leads to an 8 to 10 percent increase in wages in Brazil and Ghana. Mexico was basically an agrarian country in the beginning of the twentieth century (Rosenzweig, 1965), thus physical strength was the principal source of human capital for the average laborer.

20 Table 4: Impact of the Panic of 1907 on Self-Selection Patterns. Dependent variable: height (centimeters)

Comparison Samples Rurales Federales Passports (1) (2) (3) Migrant -0.612*** 1.499*** -2.609*** (0.215) (0.402) (0.531) Migrant × Panic -0.549** -0.552** -0.524** (0.253) (0.253) (0.253) Migrant × Post Panic 0.676*** 0.628*** 0.642*** (0.236) (0.238) (0.239) Age categories 18-20 years -0.922*** -0.873** -1.097** (0.322) (0.437) (0.487) 21-23 years -0.431** -0.106 -0.520 (0.209) (0.404) (0.459) Skill level categories Skilled -0.069 0.805*** 0.668** (0.159) (0.257) (0.307) Professional 0.876** 0.837 1.600*** (0.424) (0.529) (0.420) Region of birth North 2.017*** 4.459*** 2.923*** (0.458) (0.521) (0.677) Bajio 0.720* 2.949*** 1.622** (0.425) (0.517) (0.670) Center 0.172 1.870*** 0.604 (0.436) (0.514) (0.676) Year of birth FE YES YES YES Destination FE YES YES YES Constant 175.061*** 160.161*** 165.707*** (1.475) (4.430) (4.440)

Observations 9,005 5,727 5,818 R-squared 0.043 0.110 0.076 Source: Mexican Border Crossing Records - Microfilm publication number A3365 and López-Alonso (2015). Notes: * = Significant at 10% level; ** = Significant at 5% level; *** = Significant at 1% level. Robust standard errors in parenthesis. The omitted categories are individuals over age 24, born in the South region and unskilled workers.

Column 1 of Table 4 reveals that before the crisis, immigrants were negatively self- selected relative to the military elite (rurales). This pattern deepens during the crisis, when immigrants were 1.1 centimeters shorter, but shifts to a neutral self-selection after the crisis.

Column 2 shows the same pattern relative to the federales sample. Before the panic, immigrants were 1.5 centimeters taller than the average soldier (federales). During the panic, immigrants became less positively self-selected (0.9 centimeters taller), but

21 return to a strong positive self-selection (2.1 centimeters taller) after the panic.

This "U" pattern holds considering the passports sample as well (column 3). We include state of birth fixed effects to test if these findings were driven by unobserved factors within migration regions. Table 7 in the Annex show that our results are unaffected. Therefore, the findings suggest that in the beginnings of the twentieth century, when migrants were able to cross the border without restrictions, the composition of the Mexican migration flow adjusted to changes in the labor demand of the US market.

This can be appreciated more clearly in Figure 3, which depicts the predicted height of immigrants during the period under analysis (October 1906 - December 1908). Two things can be noticed.

First, in March 1907 the ﬁrst strong drop in stock prices occurred. In the following months, the speculation and uncertainty continued and by May 1907 the US had fallen into a short but severe recession (Odell & Weidenmier, 2004, p. 1003). This might explain the sharply fall in the predicted height from May to August 1907.

However, in August 1907 the Secretary of the Treasury announced the deposit of 28 million dollars to banks across the US for relieving the expected stringency in money supply and bring back confidence to the financial system (Markham, 2002, p. 31). This measure only delayed the financial crash of October, but along with substitutes for legal currency and the creation of "legal holidays", prevented even more bankruptcies during the panic period (Andrew, 1908, p. 516). These events might explain why the predicted height slightly increased after August and fells later on.

Second, the predicted height increased signiﬁcantly after January 1908, when the payments to depositors of commercial banks were fully restored. After May 1908, the predicted height returns to pre-crisis levels. This suggests the presence of additional factors driving the self-selection pattern after the crisis.

6. Adjustment Mechanisms of Self-selection Patterns

We have presented evidence showing that migrants were positively selected from the height distribution of the Mexican working class. Yet, to understand through

22 Figure 3: Effect of the Panic of 1907

Source: Mexican Border Crossing Records (1906-1908). Note: The estimates correspond to individuals over age 24, born in the South region and unskilled workers (year of birth and state of destination fixed effects were included). The predicted values were estimated for each individual based on year-month fixed effects. May-07: By May 1907, the US had fallen into a short but severe recession. Aug-07: In August 1907, the Secretary of the Treasury announced the deposit of 28 million dollars to banks across the US for relieving the expected stringency in money supply and bring back confidence to the financial system. Jan-08: In January 1908, the payments to depositors of commercial banks were fully restored. which mechanisms the selection patterns adjusted during and after the panic, it is necessary to know whether Mexican emigrants ended up in the upper or lower tail of the US height distribution.22 However, using the US height distribution would not be appropriate because most immigrants faced different levels of labor discrimination relative to native whites (Clark, 1908, p. 478), i.e. the US labor market was segmented by ethnic or racial characteristics which were penalized.

In fact, Mexican migrants belonged to a pool of migrant workers from different source countries that competed for underpaid low-skill jobs. We found evidence suggesting that Mexicans immigrants might have belonged to the upper tail of the height distribution of such pool of immigrants.

Clark(1908, 479-501) extensively describes that Mexican immigrants were preferred to Greeks or Italians, and were just next to Americans in the railways industry. Similarly,

22The latter case would be inconsistent with income maximization and theoretically impossible in the selection model of Borjas(1987).

23 from 1900 Mexicans started to displace Japanese workforce in California, and in the beet and melon ﬁelds (New Mexico), Mexican immigrants were as preferred as Russian and Japanese laborers. Furthermore, in domestic service jobs, Mexicans gradually displaced African-American labor force in Arizona, New Mexico and Texas according to Fuller(1928, p. 66).

The only quantitative benchmark we know is the research of Spitzer & Zimran(2018) which studies migrant selection of Italians to the US between 1907 and 1925. According to their statistics, the average height of Italian immigrants was 163.8 centimeters, but immigrants from the North of Italy were taller (165.6 centimeters) (Spitzer & Zimran, 2018, p. 231). On the other hand, in our sample the average Mexican immigrant was 167.8 centimeters tall, but immigrants from the North of Mexico were 168.8 centimeters tall.

These average estimates suggest that Mexican migrants were considerably taller than Italian immigrants, and might explain the substitution of the Italian workforce described by Clark(1908). Consequently, Mexican migrants might have been positively selected based on the Borjas(1987, p. 533) model.

6.1 Government Intervention

As shown in Table 4, Mexican migrants became less positively self-selected in the crisis period (the estimated Φ2 in Equation 2). We use the Borjas(1987) selection model as theoretical framework to explain how a government intervention could have inﬂuence the self-selection patterns by eliminating the cost of migration.

Following Borjas(1987), the relevant earnings distributions that determine an immigrant’s decision to return home are:

lnw0 = µ0 + ε0, (3)

lnwm = µm + εm, (4) where the subscript 0 denotes the home country (Mexico) and the subscript m denotes

24 the segment of the US labor market in which the immigrant ended up. µ and ε are observed and unobserved socioeconomic characteristics, respectively, which jointly determine earnings (w).

Hence, the return migration function would be:

I = (µ0 − µm − π) + (ε0 − εm). (5)

The indicator variable I equals 1 if the selection condition is satisﬁed and 0 otherwise.

π captures the return mobility costs in a time-equivalent measure: π = C/wm, where C gives the level of mobility costs.23

Therefore, an individual will return to his/her home country if:

(µ0 − µm − π) + (ε0 − εm) > 0. (6)

The virtual positive selection of Mexican immigrants, argued previously, necessarily

implies that µm > µ0 and εm > ε0. However, in the event of an idiosyncratic shock like

the Panic of 1907, µm and εm loose value in the affected labor market, i.e. the marginal pre−panic panic return to labor of the immigrants decreases (wm > wm ). The magnitude of the devaluation would depend on several factors such as the duration and intensity of the shock, and the devaluation might vary between observed and unobserved characteristics.

panic Assume further that µm and εm are homogeneously devalued to the levels µm panic and εm , and that the earnings distribution in Mexico remained constant. Then immigrants face three cases based in Equation 6:

panic panic (a) µm + εm > µ0 + ε0: The shock devalues the immigrants human capital, but not enough to change the selection pattern in the US. Mexican immigrants do not return.

panic panic (b) µm + εm = µ0 + ε0: Immigrants do not return since π>0, implying that the differences in mean income levels do not net of return migration costs.

23As Borjas(1987) the mobility costs for returning are constant across all individuals.

25 panic panic (c) µm + εm < µ0 + ε0: The decision to return will depend on the mobility costs panic panic and the devaluation effect. For example, in the extreme case that µm + εm ' 0,

then immigrants will return only if π < µ0 + ε0.

The previous cases show that the decision to return is not straight forward in the event of a shock. To observe return migration, the shock must devalued signiﬁcantly the immigrant’s human capital and the mobility costs for returning must be reduced signiﬁcantly or neutralized. Our immigrant sample does not capture return migration, thus we search in Mexican newspapers whether return migration was observed as a consequence of the crisis.

The press of that time describes that federal and local authorities from Mexico ﬁnanced the repatriation of thousands of immigrants that became unemployed during the Panic of 1907. Similarly, the Mexican Consuls of Los Angeles, Tucson and El Paso arranged the transportation of immigrants back to Mexico (El Imparcial, 1908a; El Imparcial, 1908c; La Iberia, 1908).24

The repatriation took place through railways arriving to Juarez (Chihuahua), the twin city of El Paso (Texas), and it is estimated that 100 to 150 Mexican immigrants returned daily (El Imparcial, 1908d; La Patria, 1908). These repatriated migrants moved towards farms and mining centers across the north of Mexico, but others moved to states in the Bajio and the Center region (El Imparcial, 1908b).

As part of this improvised repatriation program, several Mexican and American railway companies reduced prices for repatriation tickets. However, the intervention of the Mexican Federal Government was significant and can be traced in the bulletins of the National Treasury that recorded extraordinary expenses under the concept of "transportation and maintenance of repatriated migrants" (Federal Government of Mexico, 1908).25 The flow of repatriated migrants ended in February 1908 (El Diario, 1908), when the US financial system was fully restored and the predicted height of Mexican emigrants became stable (Figure 3).

24El Imparcial, a newspaper from Mexico City, recorded:"... Mr. Lozano, Mexican Consul in Los Angeles, informs that one hundred families will be repatriated. Around three hundred unemployed Mexicans will be returning, for which transportation has been arranged." 25Other interventions at the local level were registered in treasury reports of some states. For example, the state of Chihuahua reported expenses for the payment of "transportation costs for the repatriated" (Government of Chihuahua, 1908).

26 The hemerographic evidence suggests that effectively during the panic, Mexican immigrants faced considerably low or no costs for returning (π ' 0). Consequently, the "best" individuals that initially emigrated to the US returned to their position in the earnings distribution of Mexico (Equation 3).26 In the context of our research, the tallest returned to the upper tail of the height distribution of the Mexican working class, affecting the relative position of those individuals marginally shorter than the return migrants.

The unanticipated return of migrants could generate excess supply of labor, which in turn pushes displaced individuals (relatively shorter) to emigrate. The results in Table 4 suggests that this might have happened, especially in those cities where initially the return migrants arrived.

Observing Mexican emigration (less positively selected) during the panic period implies that the shock devalued the observed and unobserved characteristics of Mexican immigrants in the US, but not enough to discourage emigration completely. This suggests that Mexican immigrants might have not returned, unless mobility costs were sufﬁciently low. By eliminating the mobility costs, the repatriation measures implemented during the Panic of 1907 inﬂuenced the immigrant’s decision to return, and therefore the self-selection patterns during this period.

6.2 Overlapping shocks

Results in Table 4 show that the self-selection pattern of Mexican emigrants changed after the Panic of 1907. However, it is not obvious why emigrants became more positively selected. On the one hand, as a result of the shock, the post-panic labor demand in the US could have been organically different. This new composition might have demanded taller immigrants relative to the pre-panic period. On the other hand, factors in Mexico might have pushed taller individuals to emigrate during the post-panic period.

We explore the latter scenario looking at climate shocks that might have inﬂuenced the observed self-selection patterns after the Panic of 1907. Contreras(2005, p. 123),

26If as a consequence of the migration spell, their human capital increased, they could have returned to a higher position.

27 Clark(1908, p. 473), and Mayet et al.(1980, p. 757) document that the states of Nuevo León, Quéretaro, San Luis Potosí and Zacatecas experienced droughts in 1907 and 1908. As a consequence, there were important crop losses in some areas (Cardoso, 1980, p. 12). We conﬁrm the presence of such droughts using the Mexican Drought Atlas (Stahle et al., 2016) that provides reconstructions of the self-calibrating Palmer Drought Severity Index (PDSI) on a 0.5° latitude/longitude grid centered over Mexico from AD 1400-2012. Figure 4 show that droughts affected part of the North region and the Yucatan peninsula. According to the index scale, these climate shocks can be considered as mild or moderate droughts,27 however their persistence for two consecutive years impacted the production of agricultural products.

Figure 4: Droughts in Mexico 1907-1908. Palmer Drought Severity Index (PDSI)

(a) Jan - Dec 1907 (b) Jan - Dec 1908 Source: Stahle et al.(2016). Note: The Palmer Drought Severity Index (PDSI) uses temperature and precipitation data to estimate relative dryness. It is a standardized index that spans from -6 (dry) to +6 (wet). However, values below -4.0 represent extreme droughts while values above +4.0 represent extreme wet spells. The panel shows the presence of regional droughts in 1907 and continues in 1908. The drought severity index in 1908 was -1.36 (Durango), -1.40 (Estado de Mexico), -1.31 (Nuevo León), 0.88 (Queretaro), -0.90 (San Luis Potosí), -1.79 (Sinaloa), -1.35 (Zacatecas) and -2.07 (Yucatan). According to the PDSI scale, these states experienced, on average, mild droughts but they were more severe in Sinaloa and Yucatan.

To test whether these droughts inﬂuenced the self-selection patterns, we expand Equation 2 as follows:

27 The Palmer Drought Severity Index (PDSI) uses temperature and precipitation data to estimate relative dryness. It is a standardized index that spans from -6 (dry) to +6 (wet). However, values below -4.0 represent extreme droughts while values above +4.0 represent extreme wet spells (Wells et al., 2004).

28 post heighti =β + Φ1migranti + Φ2migranti × panic + Φ3migranti × panic (7) post 0 + Φ4migranti × panic × drought + Xiθ + αy + γd + ei, where drought is a dummy variable that takes the value of 1 if the emigrant’s location of last permanent residence (state) experienced droughts and zero otherwise.28

Everything else equal, the estimated self-selection pattern in the states experiencing

droughts during the post-panic period is Φ1 + Φ4.

Table 6 shows that migrants from states experiencing droughts were taller (at least one centimeter) than those from non-drought states. This implies that they were more positively selected, relative to the comparison samples. For example, on average, post- panic emigrants were as tall as the rurales (Φ1 + Φ3), however those emigrating from states experiencing droughts were 0.5 centimeters taller than the military elite (column 1). Similarly, while the average post-panic emigrant was 2 centimeters shorter than the average passport holder, post-panic emigrants from states experiencing droughts were only 1.4 centimeters shorter (column 3).

These differences in self-selection patterns across states during the post-panic period might explain why Mexican migrants became more positively selected after the crisis relative to their pre-panic counterparts. Indeed, since droughts impact all the population of a region, it is likely that laborers taller than pre-panic emigrants, were pushed to migrate during the post-panic period due to the poor harvests. Hence, the positive self-selection pattern might be a result of two overlapping forces: the reactivation of the American ﬁnancial system and the persistence of regional droughts in Mexico since 1907.

28The states that we considered that experienced droughts during the post-panic period were: Durango, Estado de México, Querétaro, San Luis Potosí, Sinaloa, Zacatecas and Yucatán. This states experienced mild or moderate droughts in 1907 and 1908 based on the PDSI scale; and/or had important losses of harvests according to the historical scholarship. The state of Nuevo Léon was not considered because the PDSI in this region was relatively low in 1907 and 1908. Moreover, it was an smelter and mining state which economic activities were unlikely to be affected by incipient droughts.

29 Table 5: Impact of Droughts on Self-Selection Patterns. Dependent variable: height (centimeters)

Comparison Samples Rurales Federales Passports (1) (2) (3) Migrant -0.621*** 1.518*** -2.586*** (0.215) (0.402) (0.531) Migrant × Panic -0.555** -0.559** -0.530** (0.253) (0.254) (0.253) Migrant × Post Panic 0.558** 0.494** 0.518** (0.247) (0.250) (0.250) Migrant × Post Panic × Droughts 1.087* 1.233** 1.139** (0.570) (0.569) (0.567) Age categories 18-20 years -0.923*** -0.871** -1.091** (0.322) (0.437) (0.486) 21-23 years -0.435** -0.118 -0.528 (0.209) (0.404) (0.459) Skill level categories Skilled -0.064 0.823*** 0.688** (0.159) (0.258) (0.308) Professional 0.884** 0.850 1.619*** (0.424) (0.528) (0.420) Region of birth North 2.065*** 4.491*** 2.958*** (0.459) (0.521) (0.677) Bajio 0.713* 2.918*** 1.599** (0.425) (0.517) (0.670) Center 0.170 1.866*** 0.600 (0.436) (0.514) (0.676) Year of birth FE YES YES YES Destination FE YES YES YES Constant 175.062*** 160.134*** 165.671*** (1.476) (4.436) (4.445)

Observations 9,005 5,727 5,818 R-squared 0.043 0.110 0.077 Source: Mexican Border Crossing Records - Microfilm publication number A3365 and López-Alonso (2015). Notes: * = Significant at 10% level; ** = Significant at 5% level; *** = Significant at 1% level. Robust standard errors in parenthesis. The omitted categories are individuals over age 24, born in the South region, unskilled workers and states without droughts.

7. Final Remarks

Using a unique data set consisting of individual border crossings (migrant sample), military recruitment records (soldier sample) and passport applications (passport holders sample), we estimate the selection pattern of Mexican migrants in the beginnings of the US-Mexico migration ﬂow (1884-1910). The results suggest that, on the basis of

30 height, the ﬁrst Mexican migrants might have belonged to the upper tail of the height distribution of the Mexican working class: migrants were positively self-selected. In other words, Mexicans sent its best people to the US, who in fact might have been key for building up two of the most dynamic economic regions in the US nowadays: the American Southwest and California.

The positive selection of Mexican migrants implies that Mexico experienced a constant loss of high quality human capital (tallest laborers) in the beginnings of the twentieth century. Therefore, the industrialization policy implemented during the Porﬁriato (1877–1911) might have not attained its full potential.29 Precisely, while machinery was introduced in key sectors of the Mexican economy, as the mining industry, manual labor predominated: the productivity depended largely on peons and unskilled workers (Von Mentz, 2001); and the best of them might have emigrated to the US. Hence, this initial selection pattern might be part of the story behind the development asymmetries between Mexico and the US during the twentieth century.

Additionally, we show that a random shock, as the US ﬁnancial crisis of 1907 (also known as the Panic of 1907), could modify the migrant selection patterns. In the context of this research, we observed that Mexican migrants became less positively self-selected during the crisis period because the Mexican government implemented a repatriation program to assist immigrants that suddenly experienced economic hardship in the US due to this ﬁnancial crisis. As a consequence, the “best” individuals that initially emigrated to the US might have returned to Mexico, displacing relatively shorter laborers, who in turn might have been pushed to emigrate.

This ﬁnding provides an intuition about what might have been happening nowadays. After four decades of strong growth, the Mexican immigrant population in the United States hit a turning point in 2010. Since then, more Mexican immigrants have returned to Mexico than have migrated to the United States, and apprehensions of Mexicans at the US-Mexico border are at a 40-year low (Zong & Betalova, 2018). If the positive selection of Mexican migrants in 1990, observed by Chiquiar & Hanson(2005), persists today, the current return migration ﬂows might be displacing laborers with relatively

29From 1877 to 1911, Mexico was ruled by General Porfirio Díaz under the slogan "Order and Progress" (Cosío Villegas & Bernal, 1973). Diaz’s objective was to “modernize” Mexico by bringing capitalism to the countryside and other key sectors of the economy. The dictatorship of Porfirio Diaz is known as the Porfiriato

31 lower human capital. Hence, this structural change in the migration ﬂow might be generating labor frictions in Mexico since displaced laborers cannot emigrate freely as in 1907.

Finally, we also argue that the stronger positive selection pattern observed after the Panic of 1907 was inﬂuenced by climate shocks. During 1907 and 1908 Mexico experienced regional droughts that caused major crop losses. This idiosyncratic shock might have pushed the “best” laborers to look for income sources in the US. This evidence suggests that climate shocks with long-lasting consequences can inﬂuence path dependence in economic development through shifts in migrant selection patterns.

32 References

Abramitzky, R. & Boustan, L. (2017). Immigration in American Economic History. Journal of Economic Literature, 55(4), 1311–45.

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. American Economic Review, 102(5), 1832–56.

Andrew, A. P. (1908). Substitutes for Cash in the Panic of 1907. The Quarterly Journal of Economics, 22(4), 497–516.

Bleakley, H., Costa, D., & Lleras-Muney, A. (2014). Health, Education, and Income in the United States, 1820–2000. In Human Capital in History: The American Record (pp. 121–159). University of Chicago Press.

Bodenhorn, H., Guinnane, T. W., & Mroz, T. A. (2017). Sample-selection biases and the industrialization puzzle. The Journal of Economic History, 77(1), 171–207.

Borjas, G. J. (1987). Immigration and Self-Selection. American Economic Review, 77(4), 531–553.

Bruner, R. F. & Carr, S. D. (2007). Lessons from the ﬁnancial crisis of 1907. Journal of Applied Corporate Finance, 19(4), 115–124.

Cardoso, L. A. (1980). Mexican emigration to the United States, 1897-1931. Tucson: University of Arizona Press.

Chiquiar, D. & Hanson, G. H. (2005). International migration, self-selection, and the distribution of wages: Evidence from mexico and the united states. Journal of political Economy, 113(2), 239–281.

Clark, V. (1908). Mexican Labor in the US. Bulletin of the United States Bureau of Labor, XVII(78), 466–522.

Cohn, R. L. (1995). Occupational evidence on the causes of immigration to the united states, 1836-1853. Explorations in Economic History, 32(3), 383–408.

33 Contreras, C. (2005). The droughts in mexico during the XIX century. Investigaciones Geográﬁcas, (58), 118–133.

Cosío Villegas, D. & Bernal, I. (1973). Historia mínima de México. Mexico City: El Colégio de México.

Durand, J. (2016). Historia Mínima de la Migración México-Estados Unidos. Mexico City: El Colégio de México.

El Diario (1908). Year VI-No. 482. Feb 7, pp. 6.

El Imparcial (1908a). Tome XXIV-No. 4110. Jan 1, pp. 6.

El Imparcial (1908b). Tome XXIV-No. 4118. Jan 9, pp. 7.

El Imparcial (1908c). Tome XXIV-No. 4130. Jan 11, pp. 1.

El Imparcial (1908d). Tome XXIV-No. 4130. Jan 21, pp. 4.

Federal Government of Mexico (1908). Bulletin of the National Treasury. Jan 30, pp. 67.

Frydman, C., Hilt, E., & Zhou, L. Y. (2015). Economic effects of runs on early “shadow banks”: Trust companies and the impact of the panic of 1907. Journal of Political Economy, 123(4), 902–940.

Fuller, R. (1928). Occupations of the Mexican-Born Population of Texas, New Mexico and Arizona, 1900-1920. Journal of the American Statistical Association, 23(161), 64–67.

García y Griego, M. (1983). The Importation of Mexican Contract Laborers to the United States, 1942-1964: Antecedents, Operation, and Legacy. In The Border that Joins. Mexican Migrants and US Responsibility (pp. 32–49). Rowman and Littleﬁeld.

Government of Chihuahua (1908). Ofﬁcial Newspaper. Feb 2, pp. 13.

Gratton, B. & Merchant, E. (2015). An Immigrant’s Tale: The Mexican American Southwest 1850 to 1950. Social Sciences History, (39), 521–50.

Hansen, B. A. (2014). A failure of regulation? Reinterpreting the panic of 1907. Business History Review, 88(3), 545–569.

34 Johnson, J. F. (1908). The crisis and panic of 1907. Political Science Quarterly, 23(3), 454–467.

Kosack, E. & Ward, Z. (2014). Who Crossed the Border? Self-Selection of Mexican Migrants in the Early Twentieth Century. The Journal of Economic History, 74(4), 1015–1044.

La Iberia (1908). Year III-No. 504. Jan 11, pp. 1.

La Patria (1908). Year XXXII-No. 9313. Jan 15, pp. 2.

López-Alonso, M. (2007). Growth with inequality: Living standards in Mexico, 1850– 1950. Journal of Latin American Studies, 39(1), 81–105.

López-Alonso, M. (2015). Estar a la altura. Una historia de los niveles de vida en Mexico, 1850-1950. Mexico City: Fondo de Cultura Económica.

López-Alonso, M. & Condey, R. P. (2003). The ups and downs of Mexican economic growth: the biological standard of living and inequality, 1870–1950. Economics and Human Biology, 1(2), 169–186.

Markham, J. W. (2002). The Panic of 1907. In A ﬁnancial history of the United States, volume Volume II: From JP Morgan to the Institutional Investor (1900-1970) (pp. 29–41). Sharpe.

Mayet, E. F., y Cervera, J. S., & Arias, D. P. G. (1980). Las sequías en México: historia, características y efectos. Comercio Exterior, 30(7), 756.

Moen, J. & Tallman, E. W. (1992). The bank panic of 1907: the role of trust companies. The Journal of Economic History, 52(3), 611–630.

Odell, K. A. & Weidenmier, M. D. (2004). Real shock, monetary aftershock: The 1906 San Francisco earthquake and the panic of 1907. The Journal of Economic History, 64(4), 1002–1027.

Orrenius, P. M. & Zavodny, M. (2005). Self-selection among undocumented immigrants from mexico. Journal of Development Economics, 78(1), 215–240.

Rosenzweig, F. (1965). El Desarrollo Económico de México de 1877 a 1911. El Trimestre Económico, 32(127(3)), 405–454.

35 Roy, A. D. (1951). Some thoughts on the distribution of earnings. Oxford economic papers, 3(2), 135–146.

Schneider, E. B. & Ogasawara, K. (2018). Disease and child growth in industrialising Japan: critical windows and the growth pattern, 1917-39. Explorations in Economic History, 69, 64–80.

Schultz, T. P. (2002). Wage gains associated with height as a form of health human capital. American Economic Review, 92(2), 349–353.

Spitzer, Y. & Zimran, A. (2018). Migrant self-selection: Anthropometric evidence from the mass migration of Italians to the United States, 1907–1925. Journal of Development Economics, 134, 226–247.

Stahle, D. W., Cook, E. R., Burnette, D. J., Villanueva, J., Cerano, J., Burns, J. N., Grifﬁn, D., Cook, B. I., Acuna, R., Torbenson, M. C., et al. (2016). The Mexican Drought Atlas: Tree-ring reconstructions of the soil moisture balance during the late pre-Hispanic, colonial, and modern eras. Quaternary Science Reviews, 149, 34–60.

The Washington Post (1907). Stocks thrown over. Aug 13, 3.

Verduzco, G. (1995). La migración mexicana a Estados Unidos: recuento de un proceso histórico. Estudios Sociológicos, 13(39), 573–574.

Von Mentz, B. (2001). Trabajo minero y control social durante el porﬁriato. Los operarios de dos poblaciones contrastantes. Historia Mexicana, 555–607.

Wells, N., Goddard, S., & Hayes, M. J. (2004). A self-calibrating Palmer drought severity index. Journal of Climate, 17(12), 2335–2351.

Zong, J. & Betalova, J. (2018). Mexican Immigrants in the United States. Washington, DC: Migration Policy Institute.

36 Annex

Figure 5: Mexican Regions and Entrance Ports of Mexican Immigrants (1906-1908)

37 Table 6: Robustness check 1. Self-selection of Mexican Migrants including state of birth ﬁxed effects. Dependent variable: height (centimeters)

Comparison Samples Rurales Passports (1) (2) Migrant -0.460** -2.125*** (0.210) (0.522) Age categories 18-20 years -1.079*** -1.233** (0.317) (0.482) 21-23 years -0.455** -0.631 (0.209) (0.459) Skill level categories Skilled 0.005 0.819*** (0.160) (0.309) Professional 1.026** 1.618*** (0.423) (0.424) Year of birth FE YES YES State of birth FE YES YES Destination FE YES YES Constant 177.038*** 167.401*** (1.723) (4.327)

Observations 9,005 5,804 R-squared 0.048 0.085 Source: Mexican Border Crossing Records - Microfilm publication number A3365 and López-Alonso (2015). Notes: * = Significant at 10% level; ** = Significant at 5% level; *** = Significant at 1% level. Robust standard errors in parenthesis. The omitted categories are individuals over age 24 and unskilled workers.

38 Table 7: Robustness check 2. Impact of the Panic of 1907 including state of birth ﬁxed effects. Dependent variable: height (centimeters)

Comparison Samples Rurales Passports (1) (2) Migrant -0.564** -2.422*** (0.222) (0.540) Migrant×Panic -0.438* -0.440* (0.255) (0.256) Migrant×Post Panic 0.846*** 0.775*** (0.250) (0.257) Age cohorts 18-20 years -0.925*** -1.111** (0.321) (0.486) 21-23 years -0.425** -0.575 (0.209) (0.459) Labor class Skilled -0.022 0.755** (0.159) (0.308) Professional 1.012** 1.564*** (0.422) (0.423) Year of birth FE YES YES State of birth FE YES YES Destination FE YES YES Constant 176.954*** 167.865*** (1.679) (4.321)

Observations 9,005 5,804 R-squared 0.050 0.088 Source: Mexican Border Crossing Records - Microfilm publication number A3365 and López-Alonso (2015). Notes: * = Significant at 10% level; ** = Significant at 5% level; *** = Significant at 1% level. Robust standard errors in parenthesis. The omitted categories are individuals over age 24 and unskilled workers.