Comparison with the Population and Dwelling Count

that the log wage by itself is a very noisy measure of underlying skill. In Column 4, we add the normalized U.S. wage residual residual divided by the standard deviation of residuals to the Mexican earnings specifi cation. The coeffi cient on the U.S. wage residual suggests that a one standard deviation increase in the U.S. wage residual is associated with a 3.8 percent increase in earnings back in Mexico, and this estimate is signifi cant at the 0.10 level. This relationship suggests that the U.S. wage does contain information about an individual’s level of skill that is not captured by our observables. Crucially, when the U.S. wage residual is added, the coeffi cient on U.S. migration experience is virtually identical to the estimate in the previous column. If endogeneity were driving our results, we would expect the coeffi cient on years of migration experi- ence to drop substantially with the inclusion of a the U.S. wage residual as a control. This exercise provides some evidence against the claim that selection and endogeneity bias are driving our main results.

C. Comparison with the Population and Dwelling Count

Two further concerns naturally present themselves related to the empirical specifi ca- tions in Tables 3–4. First, it might be the case that the communities selected by the Table 4 Regressions with U.S. Wage Residual Dependent variable 1 Wage U.S. 2 Earnings Mexico 3 Earnings Mexico 4 Earnings Mexico USExp 0.015 0.023 0.023 0.023 0.007 0.009 0.009 0.009 Age –0.024 0.048 0.048 0.047 0.013 0.014 0.014 0.014 Age2 0.033 –0.054 –0.054 –0.054 0.018 0.016 0.016 0.016 Educ 0.022 0.046 0.045 0.046 0.006 0.006 0.006 0.006 Married 0.099 0.153 0.153 0.156 0.056 0.073 0.074 0.074 U.S. wage log 0.012 0.051 U.S. wage residual 0.038 0.022 Observations 1,240 1,240 1,240 1,240 R 2 0.538 0.350 0.350 0.352 Note: Stars signify the following: signifi cant at the 0.01 level, signifi cant at the 0.05 level, signifi cant at the 0.1 level. Standard Errors are reported in parentheses. All specifi cations include MMP community ef- fects which subsume survey time effects. In Column 1, we also include a full set of U.S. state and year effects for the last migration, along with their interactions. MMP are not generally representative of Mexico, or even of the states in which they are located. Secondly, the MMP is only able to provide data on monthly labor market earnings and not hourly wage rates. Thus, we might be concerned that the estimates presented in Table 3 do not reveal much about the effect of migration experience on productivity as measured through wages, but rather refl ect a combination of the effect of migration on wages and hours worked. To assess these concerns, and more generally test the robustness of the fi ndings, we compare the patterns found here with those present in the 1995 Population and Dwelling Count. The 1995 Population and Dwelling Count hereafter abbreviated PDC is a nation- ally representative survey that contains some questions on the amount of time that return migrants spent abroad before returning to Mexico. 10 We defi ne our main PDC sample by applying the same sample selection criteria used for the MMP sample. Specifi cally, we restrict our attention to male household heads, aged 18–65, who are employed and do not own a business. We would like to restrict the PDC sample so that it covers similar communities as those in the MMP. To do this, we initially restrict the PDC sample to only include those individuals that live in the same State- Community Size cells as those represented in the MMP. 11 Table 5 presents summary statistics related to the PDC sample. While the detailed life- histories in the MMP allow us to construct a measure of total lifetime migration experience, the PDC contains a more limited set of migration questions. The PDC survey asks respondents if they have previously resided in a different location, and then asks individuals how long they resided in that previous location. We count as return migrants those individuals that indicate previously residing in the U.S., and we measure USExp in the PDC as the duration of stay in the previous location. It is note- worthy that there are substantially fewer return migrants in the PDC than in the MMP. While about 28 percent of the MMP sample has some past migration experience, only about 5 percent of individuals in the PDC are classifi ed as return migrants. Some of this gap is due to the fact that the PDC asks about the place of last residence, which could be a domestic location. Thus, if people move around in Mexico after returning from a trip to the United States, they will not be counted as return migrants. In the MMP sample, we can observe whether such internal migration takes place. Classifying post- return internal movers as nonmigrants reduces the fraction of return migrants in the MMP sample to about 22 percent. Using the measure of US and USExp available in the PDC, we replicate our ba- sic regression in Column 1 of Table 6. The estimated coeffi cient on USExp is 0.015, which is imprecisely estimated and smaller than our baseline estimate of 0.022 from the MMP. Furthermore, the coeffi cient on the US dummy is now very close to zero, whereas it was negative in the MMP specifi cations. However, it may be inappropriate to directly compare the PDC estimates with those of Table 3, because the PDC and the MMP use distinct measures of migration experience. While we observe total migration experience in the MMP, we only observe migration experience accumulated on the last U.S. trip in the PDC. Furthermore, the PDC classifi es internal movers as nonmigrants. 10. The PDC is administered by the Mexican statistical agency INEGI Instituto Nacional de Estadística y Geografía. We use the version made available by IPUMS International Minnesota Population Center 2011 11. In both the MMP and the PDC, we observe the Mexican state that an individual lives in as well as the size of their community. In the PDC, we observe a categorical variable that indicates the following com- munity sizes: 2,500; 2,500 −14,999; 15,000−19,999; 20,000−99,999; ≥ 100,000. In Column 2 of Table 6, we estimate our basic specifi cation with MMP data, but now we impose the two kinds of measurement error that are present in the PDC’s USExp variable on our preferred measure in the MMP. That is, we only consider experience form the last trip, and we code internal movers as nonmigrants. When we do this, we get results that are quite consistent with those from the PDC data. The estimated coeffi cient on USExp is smaller and insignifi cant 0.011, and the coeffi cient on US is very close to zero. The measurement error present in the PDC’s measure of migration experience thus seems to decrease the precision of estimated coeffi cient on experience, and perhaps induce a downward bias. The lack of precision could also be due to the smaller number of return migrants in the PDC sample relative to the MMP sample. If the MMP data is any guide, then the 0.015 estimate is probably a lower bound of the return to migration experience in the PDC data. In Column 3 of Table 6, we expand our PDC sample to include all of Mexico. In this sample, we estimate a return to migration experience of 2.3 percent which is sta- tistically signifi cant at the 0.05 level. Again, this may be a lower bound on the return to migration experience given the presence of measurement error described above. Nevertheless, it is interesting that we get a return to migration experience here that is very similar in magnitude to our basic MMP estimate of 2.2 percent. In Column 4, we try to assess whether this relationship between migration experience and earnings is still present when we look at the log hourly wage as the dependent variable. Here we estimate a return of about 1.9 percent that is imprecisely estimated. Some of the im- precision in the wage specifi cation could be due to noisy hours and income measures for those individual who are more marginal participants in the labor force. In Columns 5–6, we repeat this exercise but now restrict the PDC sample to those working at least 20 hours. Here we estimate a return to migration experience of 2.5 percent when income is the dependent variable, and 2.1 percent when the wage is the dependent variable. Both coeffi cient estimates are signifi cant at the 0.10 level or higher. Thus, it appears that most of the relationship between income and migration experience in the PDC data is coming through wages. Table 5 Summary statistics—Population and Dwelling Count Survey Entire Sample Return Migrants Mean Standard Deviation Mean Standard Deviation Age 37.10 10.53 35.98 9.74 Education 7.30 4.81 6.72 4.32 Married 0.80 0.40 0.84 0.36 Monthly earnings 3232.95 2860.90 3252.16 3009.78 Return migrant 0.05 0.22 USExp last trip 1.82 2.36 Observations 10,570 615 T he J ourna l of H um an Re sourc es 788 Table 6 Comparison with the Population and Dwelling Count 1 2 3 4 5 6 Dependent variable Earnings Earnings Earnings Hourly Wage Earnings Hourly Wage Data source PDC MMP PDC PDC PDC PDC Regions MMP MMP All All All All Hours restriction None None None None ≥ 20 ≥ 20 USExp 0.015 0.011 0.023 0.019 0.025 0.021 0.011 0.009 0.011 0.012 0.011 0.013 US 0.000 0.007 –0.025 –0.011 –0.033 –0.005 0.036 0.027 0.037 0.040 0.038 0.039 Age 0.038 0.040 0.037 0.039 0.037 0.039 0.004 0.005 0.003 0.004 0.003 0.004 Age2 –0.039 –0.044 –0.037 –0.037 –0.037 –0.037 0.005 0.006 0.004 0.005 0.004 0.005 Educ 0.076 0.054 0.080 0.088 0.080 0.089 0.002 0.002 0.001 0.002 0.001 0.002 Married 0.051 0.046 0.061 0.049 0.055 0.055 0.017 0.026 0.013 0.015 0.013 0.014 Observations 10,570 6,200 25,280 25,190 24,709 24,619 R 2 0.526 0.369 0.500 0.464 0.506 0.488 Note: Stars signify the following: signifi cant at the 0.01 level, signifi cant at the 0.05 level, signifi cant at the 0.1 level. Standard errors are reported in parentheses. All specifi cations include MMP community effects, which subsume survey time effects. One caveat to this exercise is that the PDC and the MMP samples cover differ- ent time periods. This raises the possibility that the results from the MMP and the PDC may not be comparable because of changes over time in the relationship be- tween migration experience and earnings. Whereas the PDC data represent a single cross- section from 1995, the MMP data come from survey waves throughout the in- terval 1987–2009. One could imagine restricting the comparison to MMP data from 1995, but since only a small number of communities are surveyed in each year, and community characteristics change over time, the resulting sample would be quite small and would be representative of very few communities. Without richer data, it is diffi cult to learn more about such time trends. Although the different sampling frames and migration questions in the MMP and PDC make precise comparisons diffi cult, the exercises performed here suggest that the two data sets provide reasonably consistent evidence on the relationship between migration experience and income. Measuring experience only in terms of the last trip seems to decrease the precision of estimates and lower the estimated returns to migra- tion experience. In the larger nationally representative PDC sample, we get a return to migration experience that is comparable to our baseline MMP estimate, and in the presence of measurement error, this may be a lower bound on the average return.

D. Mechanisms