Earnings Growth of Mexican Immigrants: New Versus Traditional Destinations

We study the earnings of Mexican immigrants in their traditional and newer destinations in the US. Analysis based on longitudinal data suggests that during 2001-2009, the real wage of Mexican immigrants increased 1-2% a year at the traditional destinations, but remained mostly statistically insignificant at the newer destinations. Mexicans at the traditional destinations exhibited greater residential stability: internal migration, non-follow up in the longitudinal data, and predicted return migration were higher among immigrants at the newer destinations than among immigrants at the traditional destinations. Predicted return migration was found to be selective on past earnings among men, but not among women. For men, a 10 percentage point increase in predicted probability of return migration was associated with a 0.3-0.5% lower wage in the year prior to return.


Introduction
The United States has experienced an unprecedented geographic dispersion of Mexican immigrants in last two decades (Massey 2008). Since 1990, Mexicans have migrated to states such as North Carolina, Georgia, Tennessee, Nevada, Utah, Oregon, and Wisconsin, which not only had a negligible presence of Mexican immigrants at that time but also had never received immigrants from any country in significant numbers. In 1990, 85% of the immigrants from Mexico lived in just three states: California, Texas, and Illinois. By 2010, this proportion fell to 57%. News media, almost on a daily basis, report the travails of Mexican immigrants in the new destinations and how residents, local communities, and state governments are responding to the immigrant influx. However, there are no national-level studies of the selection (entry-level characteristics) and earnings growth of Mexican immigrants in the newer versus traditional destinations.
The objective of this paper is to use nationally representative cross-sectional and longitudinal data to investigate the selection pattern and earnings growth of Mexican immigrants at the newer and traditional destinations. A unique contribution of this paper is to predict the probability of return migration of Mexican immigrants, and investigate if predicted return migration is influenced by past US earnings. Our study of these three inter-related processesselection, earnings assimilation, and return migration -is likely to provide a more thorough understanding of Mexican immigration than studies that have focused on only one or two of these processes.
Mexican immigrants have a growing and critical presence in the US economy. As of 2008, they constituted 6% of the country's working-age population and 23% of the working-age 3 population without a high-school degree. 1 They are the most disadvantaged in terms of education, earnings, and legal residence status in the US (Duncan et al. 2006;Passel and Cohn 2009;Ramirez 2004;and Rumbaut 2006). Previous research has found that Mexican immigrants experience much slower convergence in earnings than other immigrant groups causing fears that Mexican immigrants may be becoming the new underclass (Blau and Kahn 2007;Borjas and Katz 2007;Lazear 2007). These studies used Census data from 2000 or earlier years and did not distinguish between Mexican immigrants living in newer versus traditional destinations. 2 We use more recent data and study entry level earnings and earnings growth at traditional and newer destinations. In addition, our analysis also addresses some of the key weaknesses in previous research. For instance, previous research on Mexican earnings assimilation is based on repeated cross-sectional data, and does not adjust for potential bias on account of selection in immigration and emigration (see discussion in Borjas 1994;Kaushal 2011;Lubotsky 2007). 3 We address this issue in a number of ways. First, in the cross-sectional analysis, we compare the earnings of Mexican immigrants who arrived in the US during the same period but settled in newer versus traditional destinations after controlling for a rich set of variables including the period of arrival, age at arrival, and year of observation. The cross-sectional analysis thus provides estimates of the relative earnings of Mexicans at different destinations at any single point in time since immigration.
Second, we use longitudinal data to study earnings growth. This analysis includes personfixed effects to eliminate bias resulting from return migration. Finally, we predict the probability 1 Mexican immigrants in the United States: http://pewhispanic.org/files/factsheets/47.pdf. 2 Bohn (2009) and Kochhar et al. (2005) used more recent data, but both have a regional focus and neither has examined Mexican immigrants per se. 3 Lubotsky (2007) and Kaushal (2011) used panel data, but their analysis was not specific to Mexican immigrants. 4 of return migration of Mexicans and investigate if the predicted propensity to return differs by destination and if it is associated with the lagged earnings (earnings prior to return).

Background and Theoretical Framework
Historically, new immigrants have followed earlier arrivals from the same country.
Living in co-ethnic communities provides access to and information about the local labor, housing, and credit markets. Social networks and cultural and linguistic affinity with the community also help the migration process (Amuedo-Dorantes and Mundra 2007; Aguilera and Massey 2003;Munshi 2003;Zhou and Logan 1989).
Until 1990, the migration pattern from Mexico to the US was typical of the historical trend -with 89 % of all Mexican immigrants settling in four states-California, Texas, Illinois and Arizona 4 . Over the past two decades, however, Mexican immigrants have displayed unprecedented geographic dispersion. 5 Researchers have expounded several theories to describe this phenomenon. Massey (2008) argues that the initial change began with California becoming a less attractive place for Mexicans due to a series of state and federal policy changes including Proposition 187 that barred undocumented persons from utilizing public services, and the tightening of the US-Mexico border that diverted Mexican immigrants from California to other border states. Card and Lewis (2007) found that county-level demand pull factors and city-level supply push factors were significant predictors of Mexican immigrant inflows. Kaushal and Kaestner (2010), on the other hand, found that economic factors to be only weakly associated with the geographic dispersion of Mexicans. 4 Traditionally a large proportion emigrated from Mexico's central west plateau, but during the past two decades Mexicans are emigrating from all across the country. 5 There is a large literature on immigrant dispersion in the recent decades. Our focus here is Mexican immigration. Thus, for brevity, we do not discuss those studies.
The choice of destination is not random (Borjas 1994). Immigrants move to new destinations because they expect the economic and noneconomic benefits of migration, net of costs, to be higher at the newer destinations than at the traditional ones. Because newer destinations provide fewer ethnic amenities and limited co-ethnic support, immigrants would move to these destinations only if net economic benefits compensate for the loss of network externalities. Massey (1987) argues that immigrants become less positively selected with each successive wave of immigration as expanding networks help reduce the risk of migration.
In short, due to these various selection factors, initial earnings of immigrants should be higher at the newer destinations than at traditional ones. However, it is not clear how earnings will grow over time. At the newer locations, immigrants are more likely to develop US-specific skills (e.g. English language proficiency) since the demand for ethnic skills (to produce goods and services for Mexican immigrants) would be lower at these newer destinations and the demand for US-specific skills higher. Acquisition of US-specific skills will improve eligibility for better paid jobs, facilitating assimilation. Community support and network externalities at traditional destinations also increase assimilation. The relative earnings growth at traditional versus newer destinations will therefore depend on network externalities, post-migration investments in skill development as well as relative opportunities at these destinations.
Immigration to the newer destinations is more likely to be for economic factors and less likely to unite with the family since by definition these destinations have fewer Mexicans (family members) who arrived in earlier cohorts. If so, compared to Mexicans at the traditional destinations, those at the newer destinations face lower costs (economic and non-economic) of return and internal migration, and thus they will have a higher propensity to return to Mexico or move within the US.

6
In the empirical analysis, we test these hypotheses with regard to immigrant selection, earnings growth, and return migration. A comprehensive analysis of these inter-related processes is critical to understand the earnings assimilation of Mexicans in the US.

Traditional and New Destinations
We and 33% of all Mexican immigrants in our sample lived in these low-growth destinations. The CPS-ORG provides information on individual characteristics such as age, sex, educational attainment, country of birth, and labor-market outcomes, which include employment status, usual hours worked per week, usual weekly earnings, hourly wage for hourly paid workers, and industry of employment. These data are used to create the outcome and control variables. Consumer price index from the Bureau of Labor Statistics is applied to convert the wage data to constant dollars (base year 1982-1984=100). Observations with real wages of less than $2 or more than $250 are dropped from the wage analysis. The CPS provides data on period of arrival at two to three years intervals for those who arrived in the US in 1980 or later, which is used to assign immigrants to years-since-immigration categories. PMSA unemployment rates computed from CPS-ORG are used as a control in some model specifications. Real wage of second generation Mexicans (with at least one parent born in Mexico), by age (18-39 and 40-64), education (less than high-school, high-school, some college, and BA or more), destination (traditional, new high-growth, and new low-growth), gender, and year of the survey, constructed from the CPS-ORG, are used as control in some models.
The CPS interviews persons living within the same housing unit for four consecutive months, drops them from the survey for the next eight months, and re-enters them into the survey for the following four months. The CPS public-use data provide identifiers that can be used to match individuals in two consecutive years. Because the CPS sampling frame is residences and not people, we use a number of additional variables such as respondent's age, sex, race/ethnicity, nativity, state of residence, and period of arrival in the US to match individuals in years t-1 and t.

9
The CPS has a few limitations that may affect the analyses. The data on year of arrival are based on the question, "In which year did the respondent move to the US permanently?" The question is likely to be subject to different interpretations by repeat migrants; some may provide the year of first entry to the US and others may provide the year of the most recent entry (Redstone and Massey 2004). We assume that their responses refer to the year of permanent entry as specified in the question. 8 In the longitudinal sample, used in our preferred analysis, response to the above question is consistent for all respondents in years t-1 and t, suggesting low measurement error on this account. The second data issue relates to the length of the longitudinal panel. Theoretically, we cannot observe a difference in earnings over longer periods (e.g.10 years), without observing changes in earnings between short periods (e.g. two years). Thus, the issue is not whether observing a person one additional year is a sufficiently long time, which it is, but rather whether there is sufficient statistical power to detect potentially small changes.

Empirical Strategy: Earnings Analysis
We first study the selection patterns of Mexican immigrants in the traditional, new highgrowth and low-growth destinations. For this, we study the descriptive data on the demographic and labor-market characteristics of Mexico-born persons who have been in the US for five or fewer years. Next, we study earnings trajectories of Mexican immigrants at the three destinations using the following model on a sample of Mexicans, who arrived in the US in 1980 or later:  -1989,1990 -1980 , the log real wage of individual (i) of age (j) in year (t) is a function of the individual's characteristics (X), namely age (a dummy variable for each year of age), education (< high school, high school, some college, and a bachelor's degree or higher), whether married, whether US citizen, industry of work, and location specific variables (Z), namely, PMSA unemployment rate, the real wage of second generation Mexicans 9 (by age, education, Because there are fewer co-ethnic groups for social support, Mexicans at the newer destinations are more likely to be temporary migrants and more likely to return if they do poorly in the labor market compared with Mexicans at traditional destinations. A cross-sectional comparison of the earnings trajectories across destinations, as specified in equation (1), is therefore likely to be affected by selective return migration. We address this issue by using longitudinal data that follow the same individuals over time. Equation (2) describes the longitudinal analysis carried out on a sample of Mexican immigrants: There are three things to note about equation (2). First, the equation includes person-specific fixed effects ( i π ). Second, each person is in the sample for two periods: t-1 and t, and the value of years since immigration in the US (YSI) is fixed at year t-1. Third, we allow the effect of YSI to differ by whether the observation is from year t-1 or t. In equation (2)  The use of individual fixed effects, however, does not control for the time-varying effect of characteristics that may be correlated with residential choice and earnings.

Selection: New versus Traditional Destinations
We first study if there are any distinct selection patterns among recently arrived Mexican immigrants at the three destinations. Descriptive data in It is also likely that to some extent the boom was facilitated by the influx of low-skilled labor. with an additional year of stay in the US.  Next, we investigate the effect of an additional year of residence in the US on the real wage of Mexico-born men and women, using person-fixed-effects models based on equation (2) ( Table 4). In this analysis we explicitly adjust for local economic conditions by including controls for the real wage of second generation Mexicans and the PMSA unemployment rate.

Earnings Analysis: Longitudinal data
These models also allow the year effects to differ across destinations.
Estimates suggest that the average real wage of Mexican men, after adjusting for a rich set of variables, changes by -1.8 to 1.2 % with one additional year of US residency, and the estimates are always statistically insignificant. The increase in the real wages of Mexican women with an additional year in the US is1 to 4%, and is statistically significant for women who have been in the country for 11 to 20 years. There is no noticeable trend in wage growth with time in the US (rising or falling). In models that compute the adjusted annual earnings by place of residence, statistical tests fail to reject the hypothesis that the wage growth is the same in the traditional and newer destinations.

Return Migration-Empirical Strategy
Next, we study whether the propensity to return to Mexico differs for Mexican immigrants across the three destinations using a somewhat modified version of the methodology We use the National Health Interview Surveys-National Death Index (NHIS-NDI) to compute the probability of death of first-and second-generation Mexicans by each year of age and sex. The CPS-ORG does not provide data on internal migration. We use the March CPS, which provides data on whether the respondent changed residences between t-1 and t, and impute this outcome for second-generation Mexicans in the CPS-ORG for year t-1 using the following set of regression variables: age (a dummy variable for each year of age), education (< high school, high school, some college, and a bachelor's degree or higher), sex, whether married, whether employed, industry of work, year of observation, and state of residence in year t-1. 13 Four additional variables are added in imputing whether moved residence for first-generation Mexicans: whether US citizen, period of arrival, age at arrival, and years-since-immigration categories. Assuming that the probability of outmigration for the second generation is zero 14 , we arrive at the residual nonmatch rate (for other reasons) for second-generation Mexicans who live in the traditional destinations: Using the March CPS data, we apply a logit regression with whether the respondent changed residences between years t-1 and t as the dependent variable and the explanatory variables mentioned in the text. The coefficients on the regression variables are used to predict the internal migration for first-and second-generation Mexicans in the CPS-ORG. The March-CPS does not provide the PMSA of residence in year t-1 for those who moved but does provide the state of residence at t-1, which is controlled in this analysis. 14 We make this assumption following Van Hook et al. (2006). To examine its validity, we investigated the country of birth of individuals who had returned to Mexico in the past five years in the 2000 Mexican Census and found that 14% of all return migrants were born outside of Mexico. Some of them are likely to be US-born. Thus although a nontrivial number of US-born individuals return to Mexico every year, relative to Mexican-born returnees, their number is small. Further assuming that conditional on demographic characteristics, the probability of a residual nonmatch in the traditional destinations is the same for the first and second generation immigrants, we predict the outmigration rate of the first generation Mexicans in the traditional destinations (equation 5). 15 In the same manner, we predict the outmigration rate of first generation Mexicans living in the new high growth and low growth destinations. Finally, we test whether the predicted probability of out-migration of Mexicans is selective on their wages in year t-1.

Table 5 provides a summary statement of predicted return migration of first-generation
Mexicans and the variables used in its computation. Mexicans in the traditional destinations exhibited greater residential stability: internal migration, non-follow up in the longitudinal data, and predicted return migration were higher among immigrants at the newer destinations than among immigrants at the traditional destinations.
The estimated return migration rate is 5.3% for Mexican men in the traditional destinations, 11% for Mexican men at high-growth destinations and 10% for Mexican men at low-growth destinations. The estimated outmigration rate is 1.5% for Mexican women living in the traditional destinations, 5.4% for Mexican women living in the new high growth destinations, and 5.1% for Mexican women at the low-growth destinations. 16 Predicted outmigration, for both Mexican men and women, declines with time in the US (Figure 3).

Our final objective is to investigate if return migration is selective on US earnings. For
this, we study the association between non-follow-up (and estimated probability of outmigration) and the real wages of Mexicans in period t-1 (Table 6). Using the sample of Mexico-born men and women in period t-1, we run regressions with the log of the real wage in year t-1 as the dependent variable. Estimates suggest that in the traditional destinations, the real wages of Mexico-born men in year t-1who are not in the sample in year t are 3.2 to 4% lower than the real wages of men who are in the sample in both years (columns 1-2). The coefficient for the interactions between non-follow-up and the new high-growth destination is negative and statistically significant, and of non-follow-up and the new low-growth destinations is close to zero and statistically insignificant.
Next we study the association between predicted outmigration between years t-1 and t and real wage in t-1 (columns 3-4). For this analysis, we predict the outmigration rate of Mexicans in two ways. First, we predict the outmigration rate of only those who were not matched in t-1 and t, and for the remaining individuals the outmigration rate is 0. This prediction is based on the assumption that non-follow-up is random, which is not true for our sample given the results in columns 1 and 2 (Cameron and Trivedi 2005). Therefore, next we predict outmigration for all observations in t-1(including those who were matched in t). These estimates suggest that a 10 percentage point increase in predicted outmigration is associated with a 0.3 to 0.5% lower average wage for Mexican men at t-1 in traditional destinations. Here too the coefficient of interaction between predicted outmigration and new-high-growth destination is negative and significant, but of predicted out-migration and new low-growth destinations is modest and statistically insignificant.

20
Given this evidence, our analysis thus suggests that the steeper rise in earnings of men at new high-growth destinations observed in Table 2 is at least partly due to the difference in negative selection of return migrants across destinations. In the women's analysis (columns 5-8), neither non-follow-up nor predicted outmigration has any statistically significant association with the lagged wage of Mexican women, suggesting that return migration among women is not selective on earnings.

We use the Current Population Survey, Outgoing Rotation Group data from 2001to 2009
to study the earnings growth and return migration of Mexican immigrants across destinations.
PMSAs are divided in three categories based on vintage Mexican presence in PMSA population and its growth during the 1990s.
Our analyses lead to three main findings. First, recently arrived Mexican men living in the newer destinations (high and low-growth) are two percentage points more likely to be employed and have a 4 to 5% higher average wage than recently arrived Mexican men in the traditional destinations. Mexican women at the new low growth destinations are 6 percentage points more likely to be employed than those at the traditional or new high growth destinations but there is no statistical difference in wages across destinations. Most of the difference in labor market outcomes across destinations disappear in regressions that adjust for demographics.
We also find that recently arrived Mexican men at new destinations were about 10 percentage points (about 38%) more likely to work in construction suggesting that the influx of Mexican men to new destinations could partly be driven by the construction boom of the past decade. It is also likely that the presence low cost Mexican labor to some extent contributed to the construction boom.

21
Second, analysis based on multiple years of cross-sectional data, after controlling for a rich set of variables including period of arrival and age at arrival, shows a modest growth in Mexican immigrants' wages with time in the US: two to three decades of residency in the US is associated with a 10% increase in the hourly wage of Mexican men and an 8% increase in the hourly wage of Mexican women. This result is somewhat similar to previous research that used data for 2000 and earlier years (see, for example, Borjas and Katz 2007). Analysis based on cross-sectional data showed different earnings trajectories across destinations. However, subsequent analysis shows differences in selective return migration across-destinations leading us to conclude that the earnings trajectories based on cross-sectional analyses are misleading.
Third, the longitudinal analysis suggests whereas first-generation Mexican men and women experienced positive annual wage growth during 2001-2009, their wage growth was generally modest in comparison to the annual growth experienced by second-generation Mexican men and women. We also find that Mexicans in the traditional destinations exhibited greater residential stability: internal migration, non-follow up in the longitudinal data and predicted return migration were higher among immigrants at the newer destinations than among immigrants at the traditional destinations. Predicted return migration was found to be selective on past earnings among men, but not among women. For men, a 10 percentage point increase in predicted probability of return migration was associated with a 0.3 to 0.5% lower wage in the year prior to return. Statistical tests rejected the hypothesis that the selection pattern was the same for Mexican men in the traditional versus new high-growth destinations, underscoring the inherent weakness in estimates of earnings trajectories based on multiple cross-sections of data.
Further, this evidence thus suggests that studies on earnings assimilation without corresponding 22 knowledge of selection in return migration provide an incomplete picture of Mexican immigration.
The combined evidence on earnings growth and selective return migration thus suggests that concerns about Mexicans becoming the new underclass are somewhat exaggerated since those who do poorly in the labor market often choose to return to Mexico. There is also very high residential mobility among Mexican immigrants, in particular those living in non-traditional destinations, which also points towards high level of dynamism among Mexican immigrants. The analysis thus suggests that policies that create incentives for Mexicans to restrict their crossborder mobility (e.g. stricter border controls) are likely to limit the choices of Mexicans who are not successful in the US economy. These immigrants may decide to extend their stay in the US, due to restrictions on cross-border flows, even when they are better off returning to Mexico.    and Mexican women (columns 4-6) who arrived in the US in 1980 or later. See notes to table 1 for the definitions of destinations. All regressions control for age (a dummy variable for each year of age), period of arrival and age at arrival, PMSA unemployment rate, average real wage of second generation Mexicans (by age, education, destination, gender, and year of observation), PMSA and year of observation effects. The effects of year of observations in columns 2, 3, 5 and 6 are allowed to differ across destinations because statistical tests reject the restricted models. Models 3 and 6 also include controls for educational attainment, marital status, citizenship status and industry of work. Standard errors clustered around PMSA of residence are in parentheses. + indicates the coefficients for new high-growth and low-growth destinations are significantly different at 95% confidence interval. *0.05 < p ≤ 0.1, **0.01 < p ≤ 0.05, ***p ≤ 0.01.  1 and 5 are based on separate regressions with log real wage as the dependent variable. Years-since-immigration (YSI) is measured as of t-1 and is the same for an individual in both periods t-1 and t. All regressions control for individual fixed effects, age (a dummy variable for each year of age), education, whether married, whether citizen, industry of work, average real wage of second generation Mexicans (by age, education, destination, year of observation and gender), year of observation, and PMSA unemployment rate. Figures in columns 2-4 and 6-8 are also based on separate regressions, where the effect of years-since-immigration is allowed to differ across destinations with the inclusion of three way interactions of: years since immigration, whether the respondent lives in a traditional (or new high-growth or new low growth) destination and whether the observation is taken from year t. Similarly we also allow the effect of year of observation to differ across destinations in the regressions in columns 2-4 and 6-8. + indicates that the coefficients for new high-growth and new low growth destinations are significantly different at a 95% confidence interval. ~indicates that the coefficients for traditional and new destinations (high or low-growth) are significantly different at a 95 % confidence interval. Standard errors clustered on PMSA of residence are in parenthesis. *0.05 < p ≤ 0.1, **0.01 < p ≤ 0.05, ***p ≤ 0.01.  Note: See notes to table 1 for the definitions of destinations. Samples are restricted to Mexico-born men (or women) in t-1, who arrived in the US in 1980 or later. The dependent variable is log real wage in year t-1. In addition to the variables listed as row headings, all regressions control for age (a dummy variable for each year of age), education, whether married, whether US citizen, industry of employment, PMSA unemployment rate, PMSA and year of observation fixed effects, age at arrival, period of arrival and years since immigration. Standard errors clustered on PMSA of residence are in parenthesis. The regressions in columns 2 and 6 also control for imputed internal migration and residual non match and these effects are allowed to differ across destinations. See text for the differences in model specifications for columns 3 and 4 (or 7 and 8). *0.05 < p ≤ 0.1, **0.01 < p ≤ 0.05, ***p ≤ 0.01.