Expected Earnings and Migration: The Role of Minimum Wages

Expected Earnings and Migration: The Role of Minimum Wages Ernest Bo y-ramirez University of California Santa Barbara March 2010 Abstract Does increasing a state s minimum wage induce migration into the state? Previous literature has shown that people are mobile in response to welfare bene t di erentials across states, yet none have examined the minimum wage as a cause of mobility. Focusing on relatively mobile recent immigrants, this paper empirically examines the e ect of minimum wages on immigrant earnings and location decisions using data from the American Community and Current Population Surveys. This paper expands upon the current minimum wage, welfare magnet, and immigration literatures in two ways; rst by demonstrating that state minimum wages are factors central to an immigrant s earnings, and secondly by showing that the choice of destination is sensitive to minimum wage changes in certain scenarios. 1

1 Introduction Since at least the late 1960s, labor economists have attempted to measure the extent to which generous welfare bene ts induce low-income individuals to migrate between states (see Cebula 1979, Mo tt 1992). Of particular interest have been the location decisions of recent immigrants to the United States and their responsiveness to state-level welfare and other bene ts (e.g. Borjas 1999). The extent to which immigrants move within the U.S. in response to such public bene t levels is clearly of importance to state and national policymakers. At the same time, is it well known that the vast majority of working-age immigrants to the U.S. are employed. Thus, a full understanding of immigrants location choices needs also to focus on the relative attractiveness of working in each state. Perhaps surprisingly, relatively little work has been conducted on the e ects of exogenous changes in o ered wages on the location decisions of immigrants to the U.S. The purpose of this paper is to examine this question, by asking whether state-level minimum wage increases attract recent immigrants to the states that enact them. Wages constitute a larger share of most immigrants income than welfare bene ts do, so the focus on expected earnings as a determinant of immigrant location seems natural. In addtion, since recent immigrants are likely more mobile within the 50 U.S. states than are natives, immigrants are again a natural choice for examining if minimum wage changes have any e ect on labor migration decisions. Using a state panel, I estimate the e ect of within-state changes in the minimum wage on state immigrant ows taking into account key economic indicators 2

in the calculation of expected earnings and controlling for state and national trends. Data on immigrant ows and wages comes from the 2000 Census 5% sample, the 2001-2007 American Community Surveys (ACS), and the Current Population Survey s Monthly Outgoing Rotation Groups (MORG) for the years 1994-2007. In the next section I will review key literature in more depth. Before I analyze immigrant ows, I will rst establish the importance of the minimum wage for the earnings of low-skilled immigrants using the CPS MORG data. Following this, I explain how I measure net immigration counts. For this paper s purposes, the di erence between state in ows and out ows (or the net ow) is referred to simply as the ow or in ow. I then present my estimation strategy and I nd that overall, minimum wages are a highly signi cant component of immigrant earnings. The estimates also show that immigrants who have been in the U.S. for two to four years are mobile in response to changes in minimum wages, though the magnitudes are small. I follow up with robustness checks, then discuss why we only expect to see an e ect within a narrow window of years since U.S. arrival. I conclude by brie y discussing the policy implications of my results. 2 Immigration and Magnets Previous studies of minimum wage e ects have taken advantage of the minimum wage s greater prevalence among teenagers or service workers. In this paper I 3

will change the focus to a population which has received relatively less attention in the minimum wage literature, that of immigrants. Recent immigrants are a useful choice of subsample for a few reasons. Recent immigrants have less labor market attachments and are on average less educated than natives or established immigrants, tending to work in occupations where minimum wages bind. 1 Immigrants are often not eligible for welfare, and thus more focused on earning opportunities. Contained within migration decisions are implicit evaluations of expected costs and bene ts. Orrenius and Zavodny (2005) show that undocumented immigrant ows are highly responsive to changes in the returns to migration. They model the migration decision as a function of "push" and "pull" factors (those that cause people to leave a location versus those that attract them), nding that U.S. based pull factors are important to the lower-skilled immigrants. Though their sample is not representative of the emigrating population, the ndings are suggestive of a pull e ect resulting from a minimum wage increase. The concept of state labor market or income support policy attracting individuals is not new (Mo tt 1992 provides a brief overview the early literature). More recently, Gelbach (2004) has asked whether di erences in welfare generosity across states induce migration. Unlike previous work which nds very small e ects (Levine and Zimmerman 1999, Meyer 2000), Gelbach takes a sample of single mothers likely to be on welfare and nds signi cant movements into high 1 Orrenius, Zavodny, and Lukens (2008) nd that over the last few decades the percentage of immigrants in agricultural occupations has been steadily declining from 60 to 10%, while sectors legally bound by minimum wage law like service and sales have been growing. 4

bene t states. Kennan and Walker (2009) also use a sample of unskilled single mothers but instead nd that despite large di erences in Aid to Families with Dependent Children (AFDC) bene ts across states, the magnet e ect is weak. Income di erentials appear to be a more important incentive. Borjas (1999) asks whether immigrants and female-headed households with children cluster into states which provide greater welfare bene ts. Once the xed costs of migration have been internalized, migrants will naturally gravitate to states where the expected value of earnings and bene ts are higher. Using 1980 and 1990 household level Census data, Borjas takes advantage of variation in bene ts across states and nds that not only does California act as magnet for immigrants but it also attracts a disproportionate amount of low-skilled, loweducated workers. According to his model, low-skilled workers cluster where bene ts are higher but the returns to skill are lower. A puzzle remains though; in many leading immigration states bene t levels were static as a percentage of the median state s bene ts. One advantage of focusing on immigrants is that they are by de nition mobile, thus more likely to respond to income di erentials. 2 But despite clear evidence of clustering in Borjas study, the importance of welfare bene ts as the mechanism inducing clustering is tenuous. As de ned, welfare bene ts do not account for non-monetary bene ts like food stamps, and it is probable that di erent types of people rely on di erent types of welfare bene ts. In an earlier 2 Lutbosky (2007) outlines the importance of non-random outmigration on the composition of immigrant cohorts. My sample mitigates the e ect of return migration by restricting the sample of immigrants to those who entered the US in narrow windows. 5

study, Borjas and Hilton (1996) show that participation rates for cash-bene t welfare programs are the same for immigrants and natives, but di er greatly if food stamps, Medicaid, and housing assistance are included. Familiarizing oneself with state welfare policies and comparing their bene ts involves incurring information costs, which may be especially high for non-english speaking immigrants. Because programs can take di erent forms and have di erent requirements, calculating welfare bene ts as a part of an implicit utility maximization problem is nuanced. This study makes an important distinction from previous literature by focusing on minimum wages. Since the majority of immigrants work, wages (and minimum wages) seem more likely to a ect location decisions than welfare bene- ts. Compared to the costs associated with understanding state welfare systems, knowing a minimum wage in one state versus another is less costly and directly translates into earnings. 3 Immigrant earnings and minimum wages Before analyzing how immigrant ows respond to di erences in the minimum wages across states, I examine the connection between minimum wage laws and immigrant earnings. Previous literature by Lee (1999) and DiNardo, Fortin, and Lemieux (1996) nds a strong e ect of minimum wages on wage distributions for the whole US population. Both studies look at the role of minimum wages in explaining the polarization of the wage distribution during the 1980s when the 6

real value of the minimum wage sustained its longest decline in 30 years. Lee nds that most of the gap between the 10-50 percentiles can be explained when he takes the erosion in the real value of the minimum wage into account (for women 70%, and 25 to 70% for men depending on the speci cation). DiNardo et al. construct counterfactual wage densities by simulating wage distributions holding the value of the minimum wage at 1979 levels. To make the connection between immigrant earnings and the minimum wage in my context, I use hourly wage data from the Current Population Survey s Monthly Outgoing Rotation Groups for the years 1994 through 2007. Lemieux (2006) argues that the MORG data is better suited for measuring hourly wages because it reports wages at a point in time, while the March CPS is computed by dividing yearly earnings by annual hours worked. Signi cant measurement error in the reporting of yearly earnings make using the March CPS unreliable. In addition, topcoding is substantially more prevalent in the March supplement. Observations included are limited to persons age 16-65, who have at most a high school degree or GED. Individuals are all foreign born non-citizens. Students, the self employed, and agricultural workers are also excluded. 3 For workers not paid by the hour their weekly earnings are divided by their usual weekly hours to construct averages, following Lemieux (2006). My results are robust to the exclusion of these observations nonetheless. Though trivial in number, topcoded weekly earnings are adjusted by a factor of 1.4. After dropping all 3 Certain agricultural work is paid using a seasonal schemes exempt from minimum wages. In my study I abstract from all agricultural work. 7

missing values for hourly wages I am left with 96,776 observations. While the MORG data allows me to look at hourly wages and provides a larger sample size, it does not allow the researcher to reliably distinguish recent immigrants from those who have been in the US longer. Willingness to move across states drops with years since arrival, so the arrival year of immigrants is important. The CPS takes the reported year of arrival and groups immigrants into bins of two or three years. Besides having bin of di erent sizes, the bins are also restrictive; by grouping observations into two or three intervals, there is no way to distinguish an immigrant who arrived in say 1999 versus 2001. Furthermore, the bin years often overlap, meaning observations are be in more than one bin. On the other hand, in the American Community Survey data we can consistently identify recent immigrants but we do not have consistent hourly wage data. Table 1 gives summary statistics for this data and brie y highlights my two samples side-by-side using equivalent sample selection criteria. The two samples show very similar demographic characteristics. The education pro les of the two samples are close, implying that immigrants do not not return to school after migration. Almost all persons in my sample work in the private sector in food services, construction, accommodation and hospitality industries. 4 The ethnic composition of the populations is somewhat di erent, with the majority of immigrants coming from Mexico and smaller amounts from Central America. The di erence can be explained by the timing of the data sets. 4 Due to di erent employment coding and levels of aggregation, classi cations in the CPS and ACS a not strict analogs. I group certain jobs into larger classi cations for rough comparability. 8

The ACS covers only the most recent years from 2000-07 when the immigrant share of Latinos has been larger than in previous years. Similarly, changes in the composition of immigrant cohorts are responsible for the di erences in number of children present. 5 To illustrate how the minimum wage impacts the distribution of immigrant worker s wages, I rst look at the wages of the bottom 10% of immigrants over time. Figures 1 and 2 follow the tenth percentiles in two important cases, California and Texas (states which together account for over one-third of the CPS sample). The vertical lines indicate when the minimum wage changes. In California, real wages rise in accordance with each increase in the minimum wage. Increases in real wages in Texas coincide with the Federal increases of 1997 and 1998, then remain relatively steady. 6 To get a sense of all states, Figure 3 plots two distributions of immigrants real hourly wages. The rst distribution groups all observations in which the minimum has not increased since the prior year, while the second distribution aggregates all observations where the e ective minimum did increase. Real hourly wages are normalized by the previous year s minimum wage to show the growth of wages relative to the previous minimum. Wages at parity with the minimum are at the 1 mark on the x-axis, while wages which are greater are to the right. 5 Results in Appendix A restrict the CPS sample to observations in years 2000-07, con rming that di erences between the data sets is almost entirely driven by the length of sampling years. 6 This deviation in 2004 is likely an artifact of the small size of the Texas sample, making it sensitive to slight changes in the 10th percentile. There are 8,832 observations over 14 years, so in taking the 10th percentile of each year s observations I narrow the sample signi cantly. 9

Compared to the previous year s minimum wage, states who increased their minima experienced a spike in density around the minimum wage and greater compression implying that minimum wages played an important role in the wages of immigrants. 7 The lack of di erences as we move further to the right of the minimum parity mark suggests that spillovers at the higher ends of the distribution are negligible, con rming that the minimum wage does not e ect the upper tail. This also assures us that the wage observations post-increase are not higher than the pre-increase by construction. Both distributions contain observations from all states, meaning that if a higher wage state is more likely to raise their minimum (or vice versa) it will not create bias. These ndings are intuitively reasonable, showing that minimum wages impact the wages of low-paid workers but not those of higher-paid workers. Though illustrative, Figure 3 begs further investigation into the relationship between immigrants and the minimum wage. 8 I now turn my attention to the association between the wages and changes in the minimum wage across and within states by constructing OLS estimates of the e ect of minimum wages on hourly wages. Speci cally, the hourly wage of individual i living in state s in 7 The hours weighted version of the distributions are equivalent. The version reported is preferred for ease of interpretation. 8 Like previous studies, this distributional change does not account for potential disemployment e ects since we only observe wages of those who are working. Preliminary probit estimates looking at the e ect of minimum wages on total employment give insigni cant results for most speci cations. The probability of being employed changes within a range of -0.09 to 0.3 percent. 10

year t is, hourlywage i;s;t = + mnwage s;t + X i + s + t + s year + i;s;t (1) Where measures the marginal e ect of changes in state minima on wage earnings. X is a vector of demographic, family, employment covariates, and polynomials in age. s and t are state and year xed e ects respectively. I also include state speci c linear time trends. All speci cations contain state xed e ects to remove unobserved state characteristics constant over time speci c to each state that a ect wages, and year xed e ects to remove national trends common to all states over the sample period like aggregate economic shocks and in ation. Linear state speci c time trends are included because it is likely that wage trends varied across states with di ering minimum wage regimes. Demographic characteristics include age, sex, education, whether a person lives in a metropolitan area, race, ethnicity, and year of reported entry in to the US. Family characteristics are marital status, number of own children present in the household, and total family present. Employment controls are usual hours worked, if they are working in the private sector, if they are paid hourly, a dummy for work in agriculture, and a dummy for union membership. I also include a square, cubic, and quartic in age. Observations are weighted using CPS person weights and standard errors are clustered by state, allowing for the errors to be correlated within a state over time. Results are reported in Table 3. The rst row shows the e ect of using the full sample of immigrants. In all ve speci cations the minimum wage is signi cantly correlated with hourly 11

wages for immigrants. Adding the state-speci c time trend increases the coe - cient, perhaps due to the trends in unobserved state factors picking up variation from trends in the other covariates which were collinear with the minimum wage. Coe cients range from about 13 to 22 cents for a dollar increase in the minimum wage. Given the entire sample does not make the minimum wage, the coe cients make intuitive sense. If a worker earns slightly more than the minimum, they would not experience the entire increase by construction. If a worker earns much more than the minimum there should no direct e ect. Those who have hourly wages below the federal minimum are already earning an "unlawful" wage, so increases may have either no e ect or put upward pressure on wages through making outside o ers more appealing. My estimates allow for all scenarios. In the second row, I narrow my sample to only the largest immigrant group, Latin Americans. This yields larger estimates, signi cant well beyond the 1% level. To further isolate those workers most likely to be covered by the minimum wage, additional speci cations were run using samples of young workers (less than 25 years old) and separating workers by sex. Compared to the full sample, I nd that younger workers wages signi cantly increased by about 3 to 4 cents more for a dollar increase in the minimum. I also nd that the minimum wage has an insigni cant e ect on women s wages, while estimates for men are highly signi cant and 10 cents larger when compared to the full sample. The full sample contains 5,547 (5.7%) agricultural workers who may not be subject to the same minimum wage regime as other workers. Removing them from the sample increases the signi cance and magnitude of the coe cients by 12

about 2 cents for all speci cations, strengthening previous results. 9 4 Measuring Immigrant Flows So far, I have established that changes in minimum wage laws appear to a ect the wages of low-educated immigrants. The advantage of using the CPS was that we had a reliable measure of hourly wages; unfortunately the CPS is not well suited to study immigrant ows due to its inconsistent measurement of arrival years. To investigate recent immigrants more rigorously, I now look to the Census and American Community Surveys. Immigrants in my Census/ACS sample are de ned using the same stipulations as were used in the CPS MORG sample. The sample comprises persons of age 16-65 who have a high school degree or less, are foreign born, non-self employed, and are participating in the labor force (though many recent immigrants are likely not employed, we need to separate those who are working or looking from those who have no intention of working). The 2000 5% Census and 2001-2007 American Community Surveys contain information on how many years an immigrant reports living in the US, so I add criteria to identify the timing of immigrants. Similar to the CPS sample, I remove a small percentage of the sample who are citizens. Agricultural workers are removed because employment and pay 9 I run speci cations looking at the e ects of the minimum wages on agricultural worker s wages and employment. No e ect is found for wages, while probit estimates of employment on minimum changes indicate a $1 increase in the minimum may push workers into the agricultural sector. But the probability of 0.6% is trivially small. These results are reported in Appendix B1 and B2 for reference. 13

follow seasonal trends unrelated to the minimum wage. The ow statistics I calculate may include legal nonresidents, temporary workers, expected long-term residents, and unauthorized immigrants. Of the existing sample I group immigrants into intervals by the number of years they have been living in the United States. Collapsing immigrant counts by state and year, I get 408 observations (51 states over 8 years, District of Columbia included). By construction, immigrant counts by state and year are net of out ows. A single observation is the sum of all immigrants in the state that year, so it re ects both the in ows and out ows of people. As Lubotsky (2007) has shown, there is no precise way to determine how many times an immigrant has come into the US before. The ACS asks the year when the individual arrived, but speci es that if the person has entered the US more than once, to report the latest year he or she came to live in the US. This does not specify whether it was the rst time the immigrant had come to the US. If an immigrant has a set pattern, it is unlikely they will be responsive to changes in the minimum wage. But on the other hand, a person who migrates often is footloose, and if sending remittances abroad is a priority we predict stronger sorting. Immigrant ows into California and Texas account for over 50% of the total ows spanning the 8 year sample, while the next four leading destination states total around 21%. California consistently accounts for about 35% of the total in ows. During this period, California experienced two increases in its minimum wage (2001 and 2002). Minimum wages in Texas did not e ectively change from 14

2000 to 2007. This is shown in Figures 1 and 2. Table 2 lists the states that experienced changes in their minimum wage, the year they took e ect, and the resulting wage. Between 2000 and 2007, there were 18 states with consequential changes in the minimum wage. Not included are 9 states that experienced changes in 2007. They cannot be included since these wage changes become e ective January 2007 and I only have immigration data until April 2007. State minimum wages are collected by the Bureau of Labor Statistics Wage and Hour Division. I replace a state s minimum wage with the federal minimum if the federal minimum lies above it (many states have minimums which are vestiges of outdated law). In the cases where the minimum wage takes on a range of values based on a rm s annual receipts (i.e. Michigan), the lowest rate is used. 10 There are 13 observations in which the state minimum wage applies only to employers of at least either 2 (Vermont), 4 (Illinois), or 6 employees and have a rate above the federal. Because the state rates do not apply consistently across all employees, I use the federal rate as the default measure, thus dampening its potential e ect. The ACS is conducted before June 1st, so the midyear federal rate change of $5.15 to $5.85 in 2007 is not re ected in the 2007 immigrant counts. Data on real GDP per capita by state is from the Bureau of Economic Analysis and is chained to 2000 dollars. State unemployment rates are from the Bureau of Labor Statistics. To represent the value of possible earnings opportunities in non-minimum wage jobs, I use wages in the manufacturing sector. I calculate 10 There are only 3 observations which fall under this criteria. 15

average manufacturing wages by dividing total salary and wage disbursements by total wage and salary employment using the most inclusive NAICS manufacturing classi cation. 11 A comparison between states that changed their minimum wage as those that did not are reported in Table 4. 5 Estimation Intuitively, the decision to migrate must capture a state s macroeconomic conditions, migration networks, and social environment. My empirical model looks at the aggregation of individual migration outcomes and estimates the e ect of minimum wage increases using a reduced form approach taking into account employment prospects and other general economic indicators. Historical popularity, networks, and immigrant tolerance are also important determinants of the location choice. I will address these issues in my estimation by including state speci c xed e ects and linear time trends which I will talk about later in the section. The interaction of state changes in minimum wages with the federal minimum wage creates cross-state variation from which we can identify the e ect of minimum wages. Unlike studies that use a single change (Card and Krueger 1994, Card 1992b, Katz and Krueger 1992, Neumark and Wascher 1998), my study exploits multiple changes. Dube, Lester, and Reich (2008) note that case studies of a single change address immediate minimum wage e ects, but cannot address whether the e ects occur with a longer lag. Furthermore, the precision 11 NAICS codes SA07N and SA270. Measures contain both full and part-time workers. 16

of standard errors is overstated because case studies treat individual observations as independent when they are in fact spatially correlated. An important concern for estimation is the lag between a change in a minimum wage and its e ect on immigrant ows. Census and ACS data is collected every April while minimum wages change at the beginning of a new year. If we consider the state immigrant counts from the 2001 survey year, they re ect the number of immigrants entering the state starting in April 2000 and ending in April 2001. In my sample though, minimum wages only change in January. Because I do not expect the a minimum wage change in January 2001 to explain the previous 9 months of immigration, a change in 2001 must be matched with the immigrant data for the following year. This lagging is done for all sample years and for all covariates. 12 As seen in previous studies, anticipation may also be a concern. If an immigrant anticipates a rate increase, they may wait until after the new year to relocate and counted as responding to an incorrect minimum. Unrelated but suggestive evidence of anticipation comes from Neal (1999) who argues that young men make well thought out and calculated decisions with respect to education and occupational choices; this might apply to migration choices as well. Nonetheless, Dube et al. (2008) directly reject the possibility of signi cant minimum wage anticipation using a distributed lag model covering a six year window around minimum wage changes. They speci cally look at the employment ef- 12 It does not completely solve the problem though, as there is still the issue of the 3 month period after the a wage change and before the ACS data is conducted. 17

fects of minimum wage changes using a pooled sample of all cross-border county pairs which have di erent minimums. Using a state panel over an 8 year period, I estimate the e ect of minimum wages on net immigrant ows by regressing the log of the number of recent immigrants in the state on the previous year s minimum wages and controls. 13 The controls are annual statewide macroeconomic indicators that are likely to in uence the host state of choice; they are real statewide GDP per capita chained on 2000 dollars, annual state unemployment rates, and average state manufacturing wages. GDP per capita is used as generic gauge of economic prosperity and how much money an average person makes per year. Because of their size, yearly immigrant ows are not large enough to signi cantly change aggregate GDP at the statewide level. Average manufacturing wages are used to capture a salient alternative or substitute to the minimum wage, and will also indirectly incorporate uctuations in the greater state economy at the expense of the other covariates precision. The decision to migrate is not only based on how much one expects to earn, but also the likelihood of nding work so I use state employment rates to control for this. Reverse causality might be a concern, but today s immigration is unlikely to in uence the previous year s economic conditions. I address this issue in next two sections. 13 Because the e ect of changes in the minimum wages on the number of immigrants are likely proportional to the state s size, I account for the di erential treatment of minimum wages on ows by logging the state ows allowing me to interpret the in ows relative to the state s population. 18

Speci cally, I estimate ln(imm s;t ) = + mnwage s;t 1 + X s;t 1 + s + t + s year + s;t (2) Where imm represents the number of new immigrants in state s in year t. measures the marginal e ect of changes in state minimums on the immigrant ratio and X is the vector economic covariates. and are state and year xed e ects respectively. Standard errors are clustered by state. All speci cations contain state xed e ects to remove unobserved state characteristics constant over time speci c to each state that e ect immigration, and year xed e ects to remove national immigration trends common to all states over the sample period. A state s overall popularity with recent immigrants will be accounted for in the state xed e ect, but the changing popularity of traditional destination states over time suggest the inclusion of linear state speci c time trends. It is likely that immigration trends varied across states with di ering minimum wage regimes during this period. It is important to note that this model is static in the sense that after immigrants and natives migrate according to wages, we do not observe changes in returns to skill or immediate feedbacks as a consequence of the new labor supply composition. This is a safer assumption in the short-run. 6 Results The results are reported in Table 5. Each speci cation varies the interval of the number of years an immigrant has been in the US. The intervals are used to identify which immigrant categories are more likely to be mobile in response to 19

minimum wages. The rst speci cation captures the most recent immigrants, those who have reported being the US for 2 years or less. For the most recent immigrants, minimum wages e ect is indistinguishable from zero as are the other factors. In the second speci cation with 2 to 4 years in the US, the minimum wage is signi cant. Beyond this window, the e ect dissipates. The last two speci cations take immigrants who are relatively established, with either 4 to 6 or over 10 years in the US. We expect these immigrants to be less mobile, thus less responsive to changes in minimums. The last two speci cations suggest that as the length of residence in the US increases, immigrants become insensitive to a state s minimum wage and other predictors of mobility. For immigrants with 2 to 4 years in the US, the minimum wage is positively associated with immigration ows and signi cant at the 5% level. A dollar increase in the minimum wage corresponds to a 26% increase in the ow of immigrants. Using California as an example, in 2007 there were 224,885 immigrant arrivals with 2-4 years in the US, and a state population of 36,553,215. An increase of a dollar in California s minimum wage would increase the ow by 26%, so at 2007 population levels the increase represents 58,470 additional individuals (or an increase of 0.16 percentage points). Holding all else equal, if minimum wages remained unchanged across states, within 5 years an additional 26% would represent an additional ow of 185,689 people. Cumulatively over the 5 years, total increases in ows caused by the higher minimum wage would add up to 674,992 people. In a state like New Mexico, where the immigrant share in 2007 totaled 0.37% of the total state population, that ratio increases 20

to 0.47% as a consequence of increasing the minimum wage by a dollar. An increase of 26% in the number of immigrants who have been in the US 2-4 years translates into approximately 1,909 people. Holding all else constant, if I total each year s additional in ows caused by the raised minimum wage, over those 5 years the number comes out to 22,037 additional immigrants. Overall, the projected ow 5 years later taking into account the 26% ow increases and the "baseline" immigration would translate into a total in ow of 23,317 people. Real GDP per capita and unemployment are signi cant at the 10% level in the second speci cation, but due to the collinearity between the controls, the inclusion of state xed e ects and time trends, the estimates are not robust. Accurate measurement of these coe cients is not the aim of this paper. More importantly, since identi cation depends on the exogeneity of minimum wages, examining collinearity between minimum wages and the controls is necessary. According to Dube et al. (2008) collinearity between minimum wage changes and trends in pre-existing trends in restaurant employment is not strong. Including leads in their estimation, they argue that unemployment e ects are either small and insigni cant or slightly declining. They also show no pre-trends for overall earnings. To examine collinearity in my context, I run the speci cations in Table 5 without any controls. I nd that the estimated coe cients and signi cance of the minimum wage are almost identical, meaning that collinearity between the controls and the minimum wage is inconsequential. These results are reported in Table 6. Concerned that the results are an artifact of the construction of my years- 21

since-us-arrival intervals, I perform a variety of alternate speci cations changing the size of the groupings. My results are robust when I group immigrants by either 1-5, 1-6, 2-5, 2-6, 2-7, or 3-6 years in the US. These results are summarized in Table 7. All speci cations are signi cant at either the 5% or 1% level and the coe cient magnitudes range between 11 and 21%. Further isolating immigrants by single year of arrival shows that most of the e ect observed is coming from immigrants with 2, 3, and 4 years in the US. To summarize, minimum wages have a signi cant e ect after the second year and dissipate after year 4. 7 Discussion The e ect of minimum wages on an immigrant s choice of state is not a monotonic function of how long the immigrant has resided in the US. Within the rst year and after the sixth year there is no e ect. The only signi cant e ect is for immigrants who have had time to adjust but are still relatively new and mobile. The results suggest a migration scenario in which an immigrant does not sort immediately upon arrival into the US, but rather moves after a brief period. This makes intuitive sense in light of the fact that certain states have traditionally been hubs for new immigrants. It seems plausible that an individual arrives at a migration hub in the US with information provided by a family member or a contact already in the US. The immigrant then revaluates their present value of moving against that of staying using a new information set. When immigrants sort after initial arrival instead of before, they are better suited to analyze the 22

di erent costs and bene ts across states. Immigrants are less likely to move when they nd an adequate state match or set roots within a community. Over time, if an immigrant accumulates human capital which increases wages, this would also reduce the importance of the minimum wage. Some studies of the e ects of state policies on individuals location decisions, including the work by Borjas(1999), do not address native cross-state mobility. Similarly, I only consider how a foreign immigrant responds to minimum wages. It is not clear if natives face the same decision as natives since we cannot assume mobility ex-ante as we did with immigrants. If natives are su ciently mobile and respond to minimum wages as they do to welfare bene t di erentials, this would further increase the magnet e ect. My identi cation depends on the exogeneity of minimum wage changes. I have shown that the wages immigrants earn are correlated with minimum wages, so it is possible that minimum wage laws could be a ected by the number of people working for the minimum wage (thus responding to immigrant ows indirectly). If this is a criterion by which legislation is made, there is the potential for my estimates to be biased. For example, if immigration causes real wages to decrease, this may motivate increases in the minimum wage. On the other hand, large numbers of existing minimum wage workers might make raising the minimum wage politically contentious. Previous work on this potential endogeneity issue has used instruments for minimum wages, some of those being the number congressional seats for a certain party or government political a liation (Lemos 2005). Yet, as mentioned in the previous section, Dube et al. (2008) 23

nd no evidence of any pre-existing trend in wages before a minimum increase. I partially address reverse causality with the lagged speci cation, and given that the size of yearly immigrant cohorts are small compared to total state populations (generally around half of a percent in this study), it is not obvious that the in ux would shock the labor supply or generate ripples in market wages. The legislative process faces long lags from the point when the wage hike is proposed, debated, passed, and implemented. If immigrant ows are in uencing this legislation at all, it is indirect and not plausible in the short run. Immigration is rarely explicitly mentioned as a reason for or against minimum wage legislation in policy discussion (though it may be an underlying factor). Common arguments against increases in the minimum wage have included job loss, labor substitution by outsourcing or mechanization, increased business costs, and higher prices to consumers. Factors cited as motivating minimum wage legislation have included o setting cost of living increases, raising standards of living, stimulating consumption, improving work ethic, and reducing wage inequality. 8 Conclusion Previous studies such as Gelbach (2004) and Borjas (1999) have found that di erentials in earnings, whether manifested through wages or welfare bene ts, can in uence the migration decision. A higher expected wage would appear to imply that more low-skilled workers would migrate into that state. Under a 24

search framework, Kennan and Walker (2008) model how interstate migration decisions are driven by di erences in income prospects. My paper shows that this is the case for a narrow group of immigrants. Using the minimum wage s close linkage with a low-skilled worker s expected income, I nd that in addition to being an important component of wages, a minimum wage increase can induce migration. I nd signi cant results for immigrants who have been in the US between two and four years. In this window, a dollar increase in the minimum wage is associated with a 26% increase in a state s in ow of immigrants. The results are robust to di erent speci cations of arrival intervals, but I nd insigni cant results for the most recent arrivals and those who have been in the US over six years. To put this number into perspective, we must consider two points. First, though the percentage is large relative to the number of new immigrant arrivals, when converted into person counts the increases in ows represent very small fractions of state populations. Other estimates reported reveal smaller coe cients. Secondly, rarely do we see minimum wage increases of one dollar over the sample period, meaning there is even less reason to expect sharp spikes in migrant ows. These points indicate that the short-term magnet e ect is small, similar to conclusions found in studies by Meyer (2000), Kennan and Walker (2009), and Levine and Zimmerman (1999). In light of my conclusions, the policy implications of raising minimum wages are twofold. The magnet e ect for immigrants is relatively unimportant immediately following a minimum wage increase. Concerning natives, this result implies that the induced immigration 25

is unlikely in uence the e ect a minimum wage increase would have on natives, or to cause displacement, corroborating recent work by Peri and Sparber (2009). Furthermore, if we abstract from disemployment e ects, raising the minimum wage increases immigrant wages by a lower bound of 15 cents for every dollar. These results warrant additional research investigating the potential employment e ects on immigrants. 26

Figure 1 Hourly Wages of the Bottom 10th Percentile of Immigrants in California Note- Wage values are chained to 1994 dollars. The vertical lines represent state minimum wage changes. Some minimum changes occur midyear but wage data is yearly, so the e ect of an increase is not observed until the beginning of the next year. 27

Figure 2 Hourly Wages of the Bottom 10th Percentile of Immigrants in Texas Note- Wages are chained to 1994 dollars. The vertical lines represent state minimum wage changes. Some minimum changes occur midyear but wage data is yearly, so the e ect of an increase is not observed until the beginning of the next year. 28

Figure 3 Hourly Real Wage Distributions for Low-Educated Immigrants density 0.5 1 0 1 2 3 4 hourly wages / previous year's minimum Hourly wage distributions Without increases After increases Note - Real hourly wages are normalized by the previous year s minimum wage to show the growth of wages relative to the previous minimum. The rst distribution groups all observations in which the minimum has not increased, while the second distribution aggregates all observations where the e ective minima has increased. Wages at parity with the minimum are at 1 on the x-axis, while wages which are greater are to the right. 29

Table 1 Comparison of Observable Characteristics CPS (1994-2007) ACS (2000-2007) Average age (yrs) 36 35 Age 30 or less (%) 38 39 Male (%) 64 64 Live in metro area (%) 92 90 Married (%) 54 48 Never married (%) 30 32 Spouse absent (%) 7 9 No children (%) 70 51 One child (%) 11 15 12th grade (%) 40 43 9th grade (%) 9 9 5th or 6th grade (%) 18 17 Latino (%) 63 74 Born in Mexico (%) 46 56 Construction* (%) 11 15 Food service (%) 12 14 Private sector work 98 93 Obs. 96,776 484,917 Note *ACS uses NAICS industry codes while CPS uses SIC Detailed Industry Classi - cations. In the CPS, 21% worked in retail and 7% in personal services. 30

Table 2 Minimum Wage Changes State Changes Year Wages Alaska 1 2003 7.15 California 2 2001-2002 6.25, 6.75 Connecticut 6 2000-04, 2006 6.15, 6.40, 6.70, 6.90, 7.10, 7.40 Delaware 2 2000-01 5.65, 6.15 District of Columbia 2 2005-06 6.50, 6.90 Florida 1 2006 6.40 Hawaii 3 2002-03, 2006 5.75, 6.25, 6.75 Illinois* 2 2004-05 5.50, 6.50 Maine 4 2002-03, 2005-06 5.75, 6.25, 6.35, 6.50 Massachusetts 2 2000-01 6.00, 6.75 Minnesota 1 2006 5.25 New Jersey 1 2006 6.15 New York 2 200-06 6.00, 6.75 Oregon 5 2000, 2003-06 6.50, 6.90, 7.05, 7.25, 7.50 Rhode Island 3 2000-01, 2004 5.65, 6.15, 6.75 Vermont 5 2000-01, 2004-06 5.75, 6.25, 6.75, 7.00, 7.25 Washington 7 2000-06 6.50, 6.72, 6.90, 7.01, 7.16, 7.35, 7.63 Wisconsin 1 2006 5.70 Note *Illinois minimum wageis not re ected in my estimation. 31

Table 3 E ect of Minimum Wages on the Hourly Wages of Immigrants (1) (2) (3) (4) (5) Minimum wage (full sample) 0.141** 0.137** 0.143** 0.147** 0.226** (0.064) (0.030) (0.073) (0.073) (0.088) Minimum wage (Latino sample) 0.254*** 0.241*** 0.231*** 0.238*** 0.282*** (0.052) (0.055) (0.062) (0.063) (0.092) Demographic covariates X X X X X Family covariates X X X X Employment covariates X X X Age polynomials X X State and year FE X X X X X State time trends X Obs. in full sample 85,776 85,776 81,602 81,602 81,602 Obs. in Latino sample 67,299 67,299 63,934 63,934 63,934 Note Standard errors are reported in parenthesis below the coe cients. * indicates signi cance at the 10 percent level, ** at 5 percent and *** at the 1 percent level. Standard errors are clustered by state and observations are adjusted using CPS person weights. Di erences in sample sizes are due missing values of ethnicity, race, or usual hours worked. 32

Table 4 ACS Data Means & Standard Errors States with States without All states changes changes Di erence Immigrants 130,388 220,278 85,444 134,834*** (292,120) (440,326) (159,658) Minimum wage 5.44 6.02 5.15 0.87*** (0.60) (0.75) (0) State population 5.63 6.77 5.06 1.71** (in millions) (6.30) (8.85) (4.43) Unemp. rate 4.75 4.88 4.68 0.20 (1.15) (1.27) (1.08) Manufacturing inc. 43,220 47,648 41,006 6,642*** (8,559) (10,323) (6,495) GDP 35,476 42,747 31,840 10,907*** (12,386) (18,550) (4,379) Obs. 408 136 272 Note Standard errors are reported in parenthesis below means. ** indicates di erence between states with and without changes are signi cance at 5% level and *** at the 1% level. Immigrant counts are the sum of ACS weighted observations, collapsed on state and year. 33

Table 5 E ect of Minimum Wages on Immigrant Flows by Year of Arrival Year of arrival yrs<2 2<yrs<4 4<yrs<6 yrs>10 Minimum wage t 1.1648.2600**.0562 -.0528 (.1422) (.1084) (.0921) (.0596) GDP per capita t 1 (x 1,000).0603.0995*.0744.0167 (.0632) (.0549) (.0462) (.0286) Unemployment t 1.0484.1203*.0284.0347 (.0791) (.0714) (.0656) (.0379) Manufacturing t 1 (x 1,000) -8.94e-04 9.37e-03 -.013 2.43e-04 (.0206) (.0187) (.0127) (6.88e-03) State & year FE X X X X State trends X X X X R 2.94.96.97.99 Obs. 404 405 401 408 Note Standard errors are reported in parenthesis below the coe cients. * indicates signi cance at the 10 percent level, ** at 5 percent and *** at the 1 percent level. Standard errors are clustered by state. Sample sizes vary slightly due to not observing immigration in all states and years for each interval. 34

Table 6 E ect of Minimum Wages on Immigrant Flows Without Controls Year of arrival yrs<2 2<yrs<4 4<yrs<6 yrs>10 Minimum wage t 1.1732.2783**.0630 -.0486 (.1436) (.1091) (.0940) (.0565) State & year FE X X X X State trends X X X X R 2.9408.9605.9716.9866 Obs. 404 405 401 408 Note Standard errors are reported in parenthesis below the coe cients. * indicates signi cance at the 10 percent level, ** at 5 percent and *** at the 1 percent level. Standard errors are clustered by state. Sample sizes vary slightly due to not observing immigration in all states and years for each interval. 35

Table 7 E ect of Minimum Wages on Immigrant Flows for Recent Immigrants Year of arrival 1<yrs<5 1<yrs<6 2<yrs<5 2<yrs<6 2<yrs<7 3<yrs<6 Minimum wage t 1.1776**.1098**.2146***.1411***.1344***.1101** (.0853) (.0418) (.0788) (.0427) (.0426) (.0493) Economic controls X X X X X X State & year FE X X X X X X State trends X X X X X X R 2.9655.9722.9714.9781.9821.9767 Obs. 407 407 406 406 406 403 Note Standard errors are reported in parenthesis below the coe cients. * indicates signi cance at the 10 percent level, ** at 5 percent and *** at the 1 percent level. Standard errors are clustered by state 36

References [.] Borjas, George J. 1999. Immigration and welfare magnets. Journal of Economic Literature 17, no. 4, part 1: 607-637. [.] Borjas, George J., and Lynette Hilton. 1996. Immigration and the welfare state: immigrant participation in means-tested entitlement programs. The Quarterly Journal of Economics 111, no. 2: 575-604. [.] Cebula, Richard. 1979. The determinants of human migration. Lexington, MA: Lexington Books. [.] DiNardo, John, Nicole M. Fortin, and Thomas Lemieux. 1996. Labor market institutions and the distribution of wages, 1973-1992: a semiparametric approach. Econometrica 64, no. 5:1001-1044. [.] Dube, Arindrajit, T. William Lester, and Michael Reich. Forthcoming. Minimum wage e ects across state borders: estimates using contiguous counties. Review of Economics and Statistics. [.] Gelbach, Jonah. 2004. Migration, the lifecycle, and state bene ts: how low is the bottom? Journal of Political Economy 112, no. 5: 1091-1131. [.] Kennan, John, Walker, James R., 2008. The e ect of expected income on individual migration decisions. Working Paper no. 9585, National Bureau of Economic Research, Cambridge, MA. 37

[.] Kennan, John, and James R. Walker 2009. Wage, welfare, and individual migration. Working paper, Department of Economics, University of Wisconsin- Madison, Madison. [.] Lemieux, Thomas. 2006. Increasing residual wage inequality: composition e ects, noisy data, or rising demand for skill? American Economic Review 96, no. 3: 461-98. [.] Lemos, Sara. 2005. Political variables as instruments for the minimum wage. Contributions to Economic Analysis & Policy vol. 4, no. 1, article 16. [.] Levine, Phillip B., David J. and Zimmerman. 1999. An empirical analysis of the welfare magnet debate using the NLSY. Journal of Population Economics vol. 12, no. 3: 391-409 [.] Lubotsky, Darren. 2007. Chutes and ladders? A longitudinal analysis of immigrant earnings. Journal of Political Economy vol. 115, no. 5: 820-867. [.] Meyer, Bruce. 2000. Do the poor move to receive higher welfare bene ts? Working Paper, The Harris School of Public Policy, University of Chicago, Chicago. [.] Mo tt, Robert. 1992. Incentive e ects of the U.S. welfare system: a review. Journal of Economic Literature vol. 30, no. 1:1-61. [.] Neal, Derek. 1999. The complexity of job mobility in young men. Journal of Labor Economics 17 (April): 237-261. 38