Cross-Sectional Evidence of Generational Progress

Figure 2 shows that the occupational gaps between first-generation Mexicans and non-Mexican whites were large. The widest gap was for unskilled work, where immigrants in 1880 were 49 percentage points, or 169 percent, more likely to hold an unskilled job (78 vs. 29 percent). Note that unskilled work includes farm laborers, an occupation that Mexican immigrants frequently held. This positive gap for unskilled work contrasts with a negative gap for white-collar (–14), farmer (–27), and semi-skilled work (–8). These wide gaps for the 1880 cohort were similar in direction and magnitude for the 1910 cohort of first-generation immigrants.

Figure 2

OCCUPATIONAL CONVERGENCE ACROSS GENERATIONS BETWEEN 1880 AND 1940

Notes: Population is 30- to 44-year-old males who claim an occupation. The gap is measured relative to U.S.-born non-Mexican whites. The 1880 cohort follows the first generation in 1880, the second generation in 1910, and the third generation in 1940. The 1910 cohort follows the first generation in 1910 and the second generation in 1940.

Sources: 1880 and 1910 full-count censuses and 1910–1940 linked samples (Ruggles et al. Reference Ruggles, Flood, Goeken, Grover and Meyer2019).

The initial occupational gaps were strongly persistent across three generations for white-collar and semi-skilled work, but they were less persistent for unskilled and farming work.Footnote ¹² For example, the white-collar gap increased from –14 percentage points in 1880 to –16 percentage points for the third generation in 1940. Therefore, it appears that the highest-skilled occupations were not accessible to many Mexican Americans who were U.S. born and had two U.S.-born parents. On the other hand, there was generational convergence in unskilled and farming work. For instance, the gap in unskilled work halved from 49 percentage points in the first generation to 25 percentage points in the third generation. A similar pattern held for the 1910 cohort.

Farming is the unique category where convergence was nearly complete across three generations. Convergence (from –27 to –3 percentage points) partially occurred because non-Mexican whites were trending toward zero, reflecting structural transformation between 1880 and 1940.Footnote ¹³ Mexicans also left farming across three generations but did so more slowly. For example, the percent of Mexican Americans who farmed dropped from 15 to 10 percent, while it dropped from 42 to 12 percent for non-Mexican whites. In fact, second-generation Mexicans entered farming (increasing from 15 to 21 percent), which led to substantial convergence between the first and second generations.Footnote ¹⁴ The difference in entering or exiting farming does not appear to be due to geography: If one controls for state, then the Mexican/non-Mexican white gap converged from 36 to 4 percentage points.Footnote ¹⁵

If one uses income scores instead of occupational categories, then there was zero convergence across three generations. Figure 3 shows that Mexican immigrants held jobs that paid 53 percent of non-Mexican white jobs in 1880, which dropped to 52 percent for the second generation in 1910 and returned to 53 percent in 1940. The subsequent cohort who entered by 1910 had a slightly smaller gap with non-Mexican whites (income ratio of 54 percent) but also did not converge toward non-Mexican whites by 1940 (ratio of 55 percent). The lack of convergence is surprising because not only was the second generation reared in the United States, but the third generation was raised by two parents who were raised in the American environment—an environment that benefited first- and second-generation Europeans (Alexander and Ward Reference Alexander and Ward2018; Abramitzky et al. Reference Abramitzky, Boustan, Jácome and Pérez2019; Collins and Zimran 2019b). Further, the second and third generations should have enjoyed the benefits of citizenship, though it is unclear whether these benefits were accessible to Mexican Americans (Catron Reference Catron2019).

Figure 3

SLOW GENERATIONAL CONVERGENCE IN INCOME SCORES BETWEEN 1880 AND 1940

Notes: Population is 30- to 44-year-old males who claim an occupation. Non-Mexican white Americans and African Americans are native born. The 1880 cohort follows the first generation in 1880, the second generation in 1910, and the third generation in 1940. The 1910 cohort follows the first generation in 1910 and the second generation in 1940. The income score is as discussed in Appendix C. See Table 2 for estimates after controlling for geography.

Sources: 1880 and 1910 full-count censuses and 1910-1940 linked samples (Ruggles et al. Reference Ruggles, Flood, Goeken, Grover and Meyer2019).

It is important to note that income scores could underestimate actual income convergence but also underestimate the actual income gap. For example, the income score methodology estimates zero convergence for African Americans and a 56 percent Black–white income score gap in 1940, which contrasts with the (small) per capita income convergence and larger 62 percent income gap estimated using a different methodology in Margo (Reference Margo2016). Looking instead at changes in the occupational distribution, there were some gains in occupational status at the lower end of the distribution, which is likely the result of structural transformation in the economy (Collins and Zimran 2019a).Footnote ¹⁶

Differences in geographical location across Mexicans and non-Mexican whites do not explain the lack of income score convergence (see Table 1). In fact, there is often evidence for divergence after controlling for geographic location. For instance, the income score ratio dropped from 59 to 48 percent for those living in border states (Panel B) and from 71 to 65 percent for those living in non-border states (Panel C).Footnote ¹⁷ We also limit the sample to Texas and California (Panels D and E), given that California was at the forefront of educational expansion in the early twentieth century and speculation in the literature that barriers to progress were greater in Texas (del Castillo Reference del Castillo1985; Duncan et al. Reference Duncan, Grogger, Leon and Trejo2020). However, note that educational expansion in California might have left Mexican Americans behind if they were unable to attend higher quality schools. Indeed, the gap was smaller and generational progress was better in California since the ratio stagnated at 62 percent from the first to third generations. In contrast, the Texan income score ratio fell from 59 to 49 percent. In general, controlling for state or county of residence (Panels F and G) narrows the first-generation gap but lessens the assimilation rate.

Table 1

MEXICAN AMERICAN GENERATIONAL PROGRESS IN INCOME SCORE AND RANKS

Notes: Target population is 30- to 44-year-old males who claim an occupation. Non-Mexican white and African Americans are native born; 1880 only includes first-generation Mexicans, 1910 only includes second-generation Mexican Americans, and 1940 only includes third-generation Mexican Americans. Standard errors are not reported since 1880 and 1910 data include population of 30- to 44-year-old males. See Table A1 for the number of observations for the entire country.

Sources: 1880 and 1910 full-count censuses and 1910–1940 linked samples (Ruggles et al. Reference Ruggles, Flood, Goeken, Grover and Meyer2019).

While there was little generational convergence, the children and grandchildren of Mexican immigrants were upwardly mobile in absolute terms. When measured with income scores, the third generation held jobs that earned 15 percent more than the first generation.Footnote ¹⁸ Alternatively, the proportion of workers in white-collar jobs more than quadrupled from 4 percent of the first generation to 15.5 percent of the third generation (see Online Table A1). However, since non-Mexican whites experienced similar gains, the gaps between the two groups persisted.

Was there convergence after the third generation? A unique feature of the data is that since we can observe children in the household, we can observe completed years of education for the great-grandchildren of first-generation immigrants (the “fourth generation”).Footnote ¹⁹ In Figure 4 we show that the non-Mexican white and Mexican education gaps were wide for the fourth generation. However, fourth-generation Mexican Americans had similar education levels as African Americans. As early as age 7, fourth-generation Mexican Americans received about 0.4 fewer years of education relative to non-Mexican white Americans. This gap grew constantly throughout childhood, and by age 17, the gap was 1.6 years. To the extent that differences in human capital caused wage disparities in adulthood, more research is needed on why Mexican American schooling attainment lagged non-Mexican white American attainment after four generations. Further, note that the historical 1.6-year gap is larger than the modern-day gap of 0.9 years for third-generation Mexicans (Duncan and Trejo Reference Duncan and Trejo2011), which provides the first hint that modern-day Mexican assimilation is faster than historical assimilation.

Figure 4

EDUCATIONAL GAPS PERSISTED FOR “FOURTH-GENERATION” MEXICAN AMERICANS IN 1940

Notes: The sample includes the children observed in the household from the 1910–1940 linked sample. The figure shows the simple differences in means of years of education across fourth-generation Mexican Americans, African Americans and non-Mexican white Americans. We refer to children of third-generation of Mexican American fathers as fourth-generation Mexican Americans, although we do not observe the grandparental birthplace on the maternal side. 95 percent confidence interval for Mexican Americans.

Sources: 1940 United States Census (Ruggles et al. Reference Ruggles, Flood, Goeken, Grover and Meyer2019)

Robustness

One limitation of the analysis so far is that the first generation includes potential return migrants. If return migrants were negatively selected, then we would estimate a generational assimilation profile that is too steep (Abramitzky, Platt Boustan, and Eriksson 2014).Footnote ²⁰ Note that we already estimate a flat generational profile, so correcting for return migration is unlikely to overturn the result of “slow generational progress.” Nevertheless, in Online Figure A1, we provide alternative estimates to address return migration: limiting the sample to adults with children.Footnote ²¹ Single migrants returned at higher rates than families, so limiting the sample to fathers skews the sample to more permanent migrants (Greenwood and Ward 2015). After this sample limitation, we find the same flat assimilation profile, suggesting that return migration is not strongly biasing results. A more direct test can be done for the 1910 cohort since we can restrict the first generation to long-term stayers with the year of immigration variable, but this method also finds a similar assimilation profile.Footnote ²² Overall, we do not believe that selective return migration would overturn the result of slow generational convergence.

Another limitation arises from border changes following the Treaty of Guadalupe Hidalgo in 1848. Since the border changed, there is some ambiguity over whether those who claimed to be, for example, Californian born were actually of Mexican origin. One way to address this issue is to include those born in former Mexican territories in the Mexican group. The states that had some parts in Mexico were Arizona, California, Colorado, New Mexico, Nevada, Texas, Utah, and Wyoming.Footnote ²³ Of course, individuals in these areas could have been American families who arrived prior to 1848 and, therefore, should not be included in the Mexican-origin group, especially in Texas.Footnote ²⁴ Nevertheless, if one includes these states in the Mexican-origin group, then generational convergence was still slow (from 61 to 70 percent, see Online Figure A2). This method is a blunt and perhaps unsatisfying way to handle the border problem, so we make an alternative check: limiting the sample to those born after the 1848 Mexican Cession (30- to 31-year-olds in 1880). We continue to find no generational progress for this group, consistent with our main finding (see Online Figure A3). Finally, the border change problem does not exist for the second cohort of Mexico born in 1910, which also has slow generational progress between 1910 and 1940. Overall, we are confident that border changes do not alter our main result that Mexican-origin outcomes converged slowly toward non-Mexican white outcomes.

Bias from Pre-1980 Hispanic Identifier

We can identify third-generation Mexican Americans by grandparent’s country of birth, which may serve as an improvement over an indirect way of identifying higher-order Mexican Americans. Prior to 1980, when the Hispanic self-identifier question was added to the Census, Hispanics were identified in IPUMS data based on a variety of rules discussed by Gratton and Gutmann (Reference Gratton and Gutmann2000) and primarily by whether one had a Spanish surname.Footnote ²⁵ However, this strategy may miss descendants who lose Spanish surnames through the maternal line.Footnote ²⁶ Since we have the IPUMS Hispanic identifier in our 1940 data, we can compare outcomes of the entire set of third-generation Mexican Americans (identified by grandparents’ country of birth) with those identified by the Hispanic variable.

In Table 2, we split our 1940 third-generation Mexican Americans into those who are identified as Hispanic versus those not identified as Hispanic by IPUMS. First, about 49.5 percent of our (weighted) third-generation Mexican Americans are not identified as Hispanic, indicating that the variable misses almost half of third-generation Mexican Americans. Second, there are large economic differences between the group that can be identified as Hispanic and the one that cannot. For example, the Hispanic identified have 5 years of education, while those non-identified have 7.6 years of education. The difference in log wage income for wage workers is similarly large, where those identified as Hispanics have 58 percent less wage income than non-Mexican whites and those not identified have 38 percent less. Part of the differences in outcomes between those identified and those not identified as Hispanic are due to where people lived; those identified as Hispanic were much more likely to live in a border state, perhaps because their father was more likely to be a descendant of Mexico (and have a Spanish-sounding surname).

Table 2

BIAS IN THIRD-GENERATION OUTCOMES BASED ON THE PRE-1980 IPUMS HISPANIC IDENTIFIER

Notes: Data are of third-generation Mexican Americans. The column “ID as Hispanic” indicates the observation was identified as Hispanic according to the HISPAN variable at IPUMS. This variable lists one as Hispanic based on whether they or someone they were related to was born in Mexico, had a parent born in Mexico, or had a Spanish surname (see variable HISPRULE at IPUMS for more details). Log wage income is only measured for wage workers; the income score is based on the one discussed in Appendix C. Finally, years of education is missing for 53 individuals. *p < 0.1, **p < 0.05, ***p < 0.01.

Source: 1910–1940 linked sample (Ruggles et al. Reference Ruggles, Flood, Goeken, Grover and Meyer2019).

Introduction of bias from the pre-1980 Hispanic identifier is seen most clearly in the comparison of the two Table 2 columns: “ID as Hispanic” and “All.” Using the identifier will underestimate the average level of education for third-generation Mexican Americans by about 1.3 years. Additionally, estimates from our sample show that using the identifier will significantly overestimate the size of the wage income gap with non-Mexican white Americans. Since Reference SmithSmith (2003; Reference Smith2006) uses a similar way to identify Hispanics as the IPUMS identifier, this suggests that his estimates of third- and higher-order Mexican American outcomes using pre-1980 data are downwardly biased. Absent the more accurate identification of third-generation Mexican Americans in this paper, we would have incorrectly estimated that there was generational divergence for the Mexican-origin population. Thus, the analysis highlights how important it is to accurately identify the third generation of Mexican Americans in the pre-1980 data, just as it has been shown to be important for the post-1980 identifier (Duncan and Trejo 2007; Reference Duncan and Trejo2011).

Evidence of Generational Progress from Father–Son Linked Data

With the evidence so far, we document that Mexican generational progress in income scores was slow to non-existent. Now we turn to understanding why. It could be that slow convergence was due to an inheritance of poverty or to factors outside of the household, such as labor market discrimination or the quality of the neighborhood. Note that for this section, we pool second- and third-generation Mexican Americans together since they both had limited convergence and similar rates of intergenerational mobility (see Online Figure A4). Further, note that we will use the hand-linked version of the third-generation data since we cannot reject that the hand-linked slope is different from a data set that is created using the same machine-learning methods for the non-Mexican white and Black groups (see Online Figure A5).

In Figure 5, we show that Mexican Americans were less likely to strictly improve on their father’s percentile rank than non-Mexican whites (Bhattacharya and Mazumder Reference Bhattacharya and Mazumder2011). About 91 percent of non-Mexican white Americans who grew up in the poorest decile improved on their father’s percentile rank, in contrast with 53 percent of Mexican Americans. Mexican Americans were less likely to improve on their father’s rank across all deciles, but note that most Mexican Americans were between the zero and 20^th income score deciles (see percentile rank distributions in Figure A6). Interestingly, the poorest Mexican Americans appear to have an even lower rate of upward rank mobility than African Americans in the bottom decile (53 vs. 59 percent). Our results for African Americans are consistent with Collins and Wanamaker (2017), who provide more detail on the limited upward mobility of African Americans.

Figure 5

MEXICAN AMERICAN UPWARD RANK MOBILITY BETWEEN 1910 AND 1940

Notes: Income scores are estimated incomes based on occupation, region, and whether the son is non-Mexican white, Black, or Mexican American (either second or third generation). See Appendix B for more details. This figure plots the share of sons to have a strictly greater percentile rank than their father by the father’s decile in the income score distribution. 95 percent confidence intervals are shown. Only deciles with more than 30 people by group are plotted.

Source: 1910–1940 linked census sample (Ruggles et al. Reference Ruggles, Flood, Goeken, Grover and Meyer2019).

In Figure 6, we estimate the race/ethnic differences in rank–rank associations, which provide greater detail on where Mexican American sons ended up in the distribution. On average, Mexican Americans are predicted to end up 22.7 percentiles lower in the distribution, conditional on the father’s economic rank. This is a smaller gap than for African Americans, which is estimated at 25.7 percentiles. An alternative way of showing the same pattern is to use occupational categories; for example, Online Figure A7 shows that Mexican Americans were less likely to enter a professional job, conditional on the father’s category. Not only were Mexican Americans less upwardly mobile relative to non-Mexican whites, but Mexicans also were more downwardly mobile since they were more likely to enter an unskilled occupational category, conditional on the father’s category (Online Figure A8).

Figure 6

RANK-RANK MOBILITY BETWEEN 1910 AND 1940

Notes: Ranks are based on estimated income scores, as discussed in Appendix C. Mexican Americans are second- or third-generation sons.

Source: 1910–1940 linked census sample (Ruggles et al. Reference Ruggles, Flood, Goeken, Grover and Meyer2019).

Measurement error in the father’s economic status may bias intergenerational associations, so it may be that we are measuring mobility gaps with error (Solon Reference Solon1992; Ward Reference Ward2019).Footnote ²⁷ One way to account for measurement error is to instrument one father observation with a second one under the assumption that the error components are uncorrelated (Modalsli and Vosters Reference Modalsli and Vosters2019).Footnote ²⁸ We take this approach after linking the 1910 fathers to a second observation in 1920 and find that measurement error does not appear to drive the direction of the gaps. However, correcting for measurement error shrinks the magnitude of the gap from 23 to 13 percentiles (see Online Table D1). This pattern occurs because we are more accurately comparing poor non-Mexican white households to poor Mexican households. While the magnitude of the mobility gap may be subject to error, the more important result is the direction: given the father’s income score, Mexican American children ended up lower in the income score distribution.

A negative mobility gap for Mexican Americans contrasts a positive mobility gap found for nearly all European sources in Abramitzky et al. (Reference Abramitzky, Boustan, Jácome and Pérez2019).Footnote ²⁹ The one exception for European sources is Norway, but the Norway gap (4 percentiles) was much smaller than the Mexico gap (23 percentiles when estimated in a similar way). Relatedly, Ward (Reference Ward2020b) estimates that about half of the occupational income gaps across source countries persisted from the first generation in 1880 to the third generation in 1940; however, we find zero convergence for Mexican Americans. Therefore, Mexico appears to be a unique source in the late nineteenth and early twentieth centuries, which experienced limited gains relative to Europeans and U.S.-born whites.

Understanding Why Mexican Americans Had Worse Outcomes Relative to Non-Mexican White Americans

What factors outside of the father’s income score limited the upward mobility of Mexican Americans? In Table 3, we use the Gelbach (Reference Gelbach2016) decomposition to understand what mechanisms contributed to the Mexican/non-Mexican white gap. We are particularly interested in the importance of childhood environment since recent evidence suggests that neighborhood effects are important (Chetty and Hendren 2018a; Reference Chetty and Hendren2018b). Therefore, rather than focus on the gap for the entire population, we limit the sample to 1910 enumeration districts that have at least one non-Mexican white and one Mexican American son. This geographical restriction skews the sample toward the American Southwest, but the income score gap for children raised in these enumeration districts is the same as for the overall country (see the first and second columns in Table 3).

Table 3

DECOMPOSING THE GAP BETWEEN NON-MEXICAN WHITE AND MEXICAN AMERICANS IN 1940

Notes: The dependent variable is the log income score. Mexican Americans are either second or third generation. The columns under “Same Childhood Enumeration District” limit the same to enumeration districts with at least one Mexican and one non-Mexican white American. Childhood neighborhood are indicators for childhood enumeration district. Parental job and literacy include father and mother’s log income score and literacy levels. *p < 0.1, **p < 0.05, ***p < 0.01.

Source: 1910–1940 linked samples (Ruggles et al. Reference Ruggles, Flood, Goeken, Grover and Meyer2019).

In the third column of Table 3, we estimate the importance of parental characteristics (father’s occupation, father’s literacy, and mother’s literacy), childhood enumeration district (with enumeration district fixed effects), having a Spanish-sounding surname, the number of children in the household, and homeownership (which is a proxy for wealth). After including these variables in the regression, the income score gap drops from 66 to 41 log points (or from 48 to 34 percent). The largest contributor to the gap is parental characteristics, which accounts for about half of the drop. Having a Spanish-sounding surname explains five log points, which may reflect labor market discrimination but also likely proxies for other unobservable characteristics. The full range of fixed effects for childhood enumeration district explains only five log points, less than parental characteristics, which is surprising given the evidence that childhood location matters for ethnic gaps (Abramitzky et al. Reference Abramitzky, Boustan, Jácome and Pérez2019; Alexander and Ward Reference Alexander and Ward2018; Ward Reference Ward2020b). It may be that other barriers to progress were strong enough that improvements to the childhood environment had a limited impact on upward mobility. The number of children in the household and homeownership also explain little of the gap.

We also control for education for interested readers; however, education is an outcome of the parental controls and neighborhood factors, so the interpretation of the estimated coefficient is tenuous. When including total years of education, Mexican Americans earned 36 log points less. The decomposition further suggests that most of the Mexican and non-Mexican white gap is due to educational differences. We prefer the results from the third column since education is endogenous to parental characteristics and childhood environment.

The residual 41 log point gap in the third column may be due to unobserved discrimination in the labor market. To provide a quantitative measure of discrimination, we follow the method discussed by Hilger (Reference Hilger2016) and construct measures of racist attitudes among whites using surveys from mostly college students between 1926 and 1956 (see Table 4). It is important to remember, however, that college students in the early twentieth century were not representative of the population in general. Nonetheless, information from these surveys can provide suggestive evidence. The resulting measures suggest that Mexicans were among the least favored race/ethnic group and faced a similar amount of prejudice as did African Americans. This result suggests that labor market discrimination might have contributed to poorer economic outcomes for Mexican Americans, but we cannot directly measure the effect in a regression.

Table 4.

MEASURES OF RACISM

Notes: This table follows the table in Hilger (Reference Hilger2016), presenting the rankings of racism toward Black and Mexican individuals. The scale goes from 0 (minimum prejudice) to 1 (maximum prejudice). Averages omit the rankings from New York U. because the sample was disproportionately Jewish-American, Howard University because the sample was disproportionately Black, and any studies where rankings were not available for Mexicans.

Sources: Shown in the first column of the table.

There are many other unobserved factors that may have led to a residual gap besides labor market discrimination, such as differences in unobserved human capital. One way to proxy for unobserved human capital is to use the Armed Forces Qualifying Test or its precursor, the Army General Classification Test (AGCT) (Neal and Johnson 1996; Collins and Wanamaker 2017; Hilger Reference Hilger2016). In Figure 7, we show that the Mexican American AGCT score distribution lies between the African American and non-Mexican white distribution. On average, Mexican Americans scored 0.6 standard deviations below non-Mexican whites but 0.8 standard deviations above African Americans.Footnote ³⁰ This group ordering is consistent with the mobility gaps we found earlier, which suggests that unobserved human capital differences partially explain the difference in mobility rates. However, we cannot include AGCT directly as an explanatory variable in the 1940 data due to limited sample sizes. Differences in human capital may come from differences in schooling quality, which has been shown to be important for the Black–white gap (Carruthers and Wanamaker 2017; Collins and Wanamaker 2017); future research could attempt to quantify the importance of schooling quality for Mexican Americans, to the extent that such schooling records exist.

Figure 7

DIFFERENCES IN ARMED GENERAL CLASSIFICATION TEST (AGCT) SCORES

Notes: The AGCT is a precursor to the more well-known Armed Forces Qualification Test. The test aims to measure the intelligence of recruits and has been used as a proxy of ability in the economics literature (Neal and Johnson 1996; Carruthers and Wanamaker 2017). There are 187,112 non-Mexican white Americans, 18,288 African Americans, and 2,302 second- and third-generation Mexican Americans.

Source: Linked sample of WWII records to the 1930 Census.