US Permanent Residency, Job Mobility, and Earnings

Transcription

1 US Permanent Residency, Job Mobility, and Earnings Xuening Wang Department of Economics University of Illinois at Chicago November 2017 Job Market Paper Abstract: One concern regarding current immigration policies is that skilled immigrant workers on temporary work visas may be bound to their sponsoring firms in indentured servitude, with weakened bargaining power and lower wages. Using the National Survey of College Graduates, I address this concern by estimating the effect of acquiring US permanent residency on the voluntary job mobility and earnings of skilled temporary professionals. Within an individual fixed effects framework, I find a substantial increase in workers voluntary job mobility following receipt of permanent residency. A decomposition analysis indicates that at least 60 percent of the spike in mobility is driven by voluntary moving being discouraged during the employer-sponsored green card (permanent residency) application process, as opposed to barriers that are induced by temporary visa regulations. In addition, I find that upon acquiring a green card, only male applicants experience an earnings gain for them, the green card premium translates to an 8 percent increase in annual earnings. JEL Codes: J61, J62, J63 Acknowledgments: I thank Darren Lubotsky, Benjamin Feigenberg, Ben Ost, Steven Rivkin, Ce Shang, and participants in presentations at the University of Illinois at Chicago, the Illinois Economics Association annual meeting, and the APPAM 2017 Fall Conference for their valuable feedback. Any remaining errors are my own.

2 1 1. Introduction There has been considerable public debate on the role of high-skilled foreign-born workers in the US economy, concentrating on a tradeoff between the benefits of increased high-skilled labor and the potential negative consequences on the welfare of native-born workers. While the discussion regarding skilled guest worker visa programs, such as the H-1B program, has mostly focused on the optimal number of foreign-born workers to be employed in the skilled labor market, an important issue that has received scant research attention is how the design of these programs and related immigration policies may hinder the job mobility of workers on temporary visas. Specifically, skilled foreign workers are often locked to their sponsoring employers because of regulatory hurdles or the additional risks and costs associated with the process for changing employers. Such distortions in job mobility have led to concerns that guest workers may be exploited by their employers through indentured servitude with weakened bargaining power and wages below their marginal productivity (Dorning and Fanning 2012; Matloff 2003). This paper examines whether high-skilled workers on temporary visas do, in fact, face constraints in their ability to change employers and whether their earnings consequently suffer. Foreign citizens can be legally employed in the United States either as permanent residents (holding a green card) or as workers holding temporary work visas. I refer to the latter as temporary workers or guest workers. A permanent resident is permitted to work in the United States indefinitely and may freely switch employers. 1 Temporary workers, however, face restrictions both on their length of stay and on their ability to move across employers. For example, specialty worker visas (H-1B) and intra-company transferee visas (L-1) are the two most common types of temporary work visas designed for skilled foreign workers. L-1 workers may not change employers under their current visa. H1-B workers are allowed to change jobs if the new employer also sponsors them for a work visa. However, guest workers who have already started the application for permanent residency under the sponsorship of their current employers usually hesitate to move because in most cases, changing to a new job will cause them to lose their position 1 Permanent residents can work indefinitely in the United States as long as they maintain the validity of their green cards. Green cards are valid for 10 years for permanent residents and two years for conditional permanent residents. The card must be renewed or replaced prior to expiration. Cards issued between January 1977 and August 1989 are valid indefinitely. Permanent residents can take any job positions except some in federal, state, and local governments.

3 2 on the long waiting list for a green card. I refer to this distortion in worker mobility decisions as the green card processing constraint. The employer sponsorship requirement of work visas also generates barriers to mobility. Because of the sponsorship costs, all else equal, prospective employers are more likely to hire permanent residents in lieu of guest workers. I refer to this situation as the visa sponsorship constraint. An important goal of my work is to separately identify the contributions of the two mobility restraints, since each has its own policy implication. Institutional restraints imposed on guest worker mobility could generate beneficial effects, such as reducing the cost of labor inputs and incentivizing firms to increase general investment in workers skills because firms recognize that temporary workers are less likely to switch to a new employer. These benefits of labor immobility, however, must be weighed against the adverse consequences created within the high-skilled labor market as a whole. If temporary professionals and native-born workers are substitutable with regard to their skills, the immobility-induced lower wages of temporary workers could spread through the entire market native workers in the same occupations may also suffer from downward pressure on wages or may even be crowded out of the skilled labor market (Doran, Gelber, and Isen 2014). Additionally, restricted labor mobility impedes wage growth and the possible improvement of worker-employer matches, thereby reducing the overall efficiency of the labor market and preventing the immigrant labor force from contributing to the US economy to its full potential (Jovanovic 1979; Madrian 1994). As such, understanding the effects of regulatory hurdles to worker mobility may offer more general insights into worker and firm behaviors. To test for the presence of mobility constraints and immobility-induced lower earnings, I examine three research questions. First, what is the effect of obtaining permanent residency on the probability of voluntarily changing employers? A positive effect would indicate that workers mobility is restrained when they are applying for a green card while on a temporary work visa. Second, what are the dynamic effects of green card receipt on mobility? Based on these two questions, I also aim to separate the relative importance of the green card processing constraint and the visa sponsorship constraint. Lastly, what is the effect of permanent residency on earnings? Direct evidence on earnings can shed light on whether workers pay and promotion are negatively affected by their temporary visa status and limited mobility prior to receiving permanent residency. In answering these questions, I use the National Survey of College Graduates (NSCG) 2003, 2006, 2008, 2010, 2013, and 2015 waves, which has three advantages. First, the dataset

4 3 contains information on major visa categories, such as green cards, temporary work visas, and temporary visas for study and training, although it does not provide the exact visas within each major category (i.e., H1-B, L-1). Second, the dataset allows me to measure job mobility through responses to a retrospective survey question that asks respondents whether they are currently working for the same employer as on a prior reference date (usually the previous survey wave). Finally, the six waves of NSCG constitute three panels, allowing for longitudinal analysis of changes in immigrant status, job turnover, and earnings. Estimating the causal effect of obtaining permanent residency is complicated by the empirical challenge that the choice and timing of getting a green card is endogenously determined by workers. To tackle this problem, I adopt individual fixed effects (FE) models using a panel of immigrant workers and compare the timing of green card receipt with the timing of job and earnings changes. My within-person estimates complement the recent work by Hunt (2017), who compares the job mobility of new green card recipients with earlier ones under the assumption that there are no other sources of heterogeneity in mobility across cohorts. My work is also related to that of Mukhopadhyay and Oxborrow (2012), who implement a difference-in-difference matching method using immigrants who came to the United States on temporary visas as the counterfactual for those who arrived with a green card, and compare the differences between workers income in the USA and in their home countries. I present four main findings regarding the job mobility and earnings of skilled temporary workers. First, skilled temporary professionals are heavily restricted in their voluntary job mobility, as evidenced by a substantial spike in the mobility rate upon receipt of US permanent residency. Second, at least 60 percent of the spike in voluntary job change behaviors is due to the green card processing effect. Third, half of the mobility effect is driven by Indian and Chinese workers who have particularly long waiting lists for a green card. Finally, obtaining a green card generates a wage gain of 8 percent, but only among male applicants. Together, these pieces of evidence confirm that regulations on employer-sponsored green card applications induce employment lock among temporary workers, but it only decreases the earnings of male immigrants.

5 4 2. Temporary Work Visas, Permanent Residency, and Job Mobility US permanent residents possess broadly similar employment rights as US citizens. Individuals with a temporary work visa, however, often face labor market frictions specific to their status and their legal residence in the USA requires employment with a sponsoring firm. Table 1 presents the annual average numbers and shares of major work visas approved between 2000 and 2015 that are typically awarded to workers with a college degree. 2 H-1B and L-1 visas are the two most common skilled worker visas, jointly representing 80 percent, on average, of the total temporary work visas issued annually. The standard deviations of the annual average percentages indicate limited variations in the shares over time. 3 The dataset for my research does not distinguish between different types of work visas, but since a majority of the workers are likely to hold H-1Bs or L-1s, I describe below the regulations that apply to these workers. The institutional background on other types of visas is presented in Appendix A. H-1B and L-1 visa programs are often referred to as skilled guest worker programs because both require the visa holders to be college graduates (Dorning and Fanning 2012). The H-1B visas are awarded to foreign nationals primarily to fill perceived labor shortages in professional occupations. These visas require the sponsorship of an employer, so H-1B workers may not be self-employed and must leave the USA if non-employed. Each year, a limited number of H-1B visas are allotted to foreign nationals hired by for-profit employers (in most cases through lotteries) while those working for non-profit employers are exempt from the caps. An H-1B visa is issued for an initial period of three years and may be renewed for another three years, implying a maximum stay of six years. If a worker has started the application for permanent residency under the sponsorship by an employer, the H-1B visa remains valid until a green card is either approved or denied. The L-1 visa program is designed for intra-company transferees who have been continuously working abroad for an employer for at least one year and will be employed at US branches of the same employer. With extensions, L-1 visas are valid for a maximum of seven years. 2 The dataset for this study (NSCG) does not provide the exact temporary work visas, and no administrative data are available on the stock of visa holders in each category at any given time. The only administrative data available are the number of visas issued annually. 3 Given that workers with H1-B and L-1 visas tend to stay in the United States longer than workers on other visas, the share of the stock of these two types of workers is likely to be even higher.

6 5 H-1B workers have been permitted to change employers since the American Competitiveness in the Twenty-First Century Act (AC21) went into effect in 2000 the portability of H-1B visas was one of the policy revisions. Specifically, H-1B workers can change to a job in the same or similar occupational classification as soon as a new employer has filed a petition to US Citizenship and Immigration Services to have the visa transferred. The new employer, however, must also offer H-1B sponsorship. The only difference between a brand-new H-1B visa and a visa transfer is that the latter is not subject to the annual visa cap, and thus job changers do not need to participate in the H-1B lottery again. In contrast, workers holding L-1 visas typically cannot transfer their visas to a new firm. Even though they may find another employer to sponsor them for an H-1B visa, the employment time on the L-1 visa counts toward the H1-B maximum. In spite of AC21 reducing hurdles to mobility, researchers claim that temporary workers still experience a curtailed ability to move because visa regulations distort employer recruitment decisions (Matloff 2003). Since the H-1B visas always hinge on employer sponsorship, temporary professionals, even with the skill qualifications required by a job, may not be employed unless the prospective employer is also willing to undertake the lengthy process of visa application and the sponsorship that induce marked extra costs. The total additional pecuniary cost from hiring a temporary worker ranges from $4,000 to $7,225, which implies that the search cost for a skilled worker is increased by 29 to 75 percent (Depew, Norlander, and Sorensen, 2017). The primary non-monetary cost arises from the firm having to submit a Labor Condition Application to certify the non-displacement of native workers. In addition to these regulations, visa sponsorship requests may influence hiring decisions through other channels that are not necessarily related to worker productivity. For example, the temporary visa status can be a signal to employers that workers lack US-specific human capital (e.g. language and communication skills), or firms may be unsure if workers are likely to stay for long. These costs associated with sponsoring skilled temporary workers tend to decrease the overall market demand for them. Thus, contemporarily employed foreign workers with either H-1B or L-1 visas may have difficulty finding a new sponsoring firm even if they are willing to move. I refer to these sponsorship-related barriers to mobility as the visa sponsorship effect, or simply sponsorship effect. H-1B and L-1 workers can obtain legal permanent residency (LPR) either under the sponsorship of an employer or through other channels such as being sponsored by a family member

7 6 or by making investments. 4 Appendix Table B1 lists the number of LPRs by category and the number of H-1B and L-1 visas granted to skilled workers. The fraction of H-1B and L-1 workers that successfully became permanent residents remains undocumented. Using the statistics listed in the table, however, I provide a crude upper-bound approximation of 32 percent, with the share of workers that obtained employment-based (EB) green cards being roughly 23 percent. 5 Only a minority of temporary professionals are sponsored because LPR petitions incur noticeable costs; theoretically, employers only sponsor workers whose value to the firm outweighs these costs. 6 Moreover, firms are unlikely to sponsor temporary workers for LPR shortly after they are awarded work visas because firms need time to observe and evaluate worker productivity. The importability feature of EB green card applications generates another source of job mobility restraint. Although applicants are allowed to change jobs by law, switching to another employer will force them to restart the lengthy application process. 7 The process to obtain LPR lasts four years on average and may extend to decades in some cases, depending on the category and countries of origin (National Foundation for American Policy, 2011). 8 Labor market distortions arise if workers choose to stay with employers they would prefer to leave, in order to maintain their positions in the visa queue. I refer to this source of employment frictions as the green card processing effect, or simply processing effect. While the sponsorship effect has an impact on all temporary work visa holders, only workers applying for EB green cards are subject to the processing effect. The NSCG data do not provide information on the green card types, but Appendix Table B1 shows that on average, 70 percent of LPRs held by skilled workers are acquired though employment. 9 This implies that my estimate for the processing effect is a lower bound. 4 The EB-5 visa allows eligible immigrant investors to become permanent residents by making the necessary investments to start a new commercial enterprise in the United States that will employ at least ten American workers. 5 The estimation method can be found in Appendix A3. 6 The application and attorney fees for green card petitions are typically $10,000 or more, plus the indirect costs of time. 7 Starting Oct. 1, 2015, workers are able to change employers without jeopardizing their prospect of getting a green card if they have reached the last step of the application process. But the lengthiest period of the process is the second step in which workers wait for a permanent visa to become available. 8 Indian and Chinese applicants face particularly long visa queues. 9 But the fraction of EB green card holders that transitioned from H1-B and L-1 visas is slightly smaller in theory, between 62 and 70 percent because some researchers estimate that 62% of employment based permanent immigrants began as H-1B temporary workers (Hira 2010b). Little is known about the share of EB green cards awarded to L-1 workers.

8 7 In sum, the sponsorship effect stems from the organizational sponsorship mandated by the guest worker programs, whereas the processing effect is a consequence of regulations on EB green card applications. By disentangling these effects, this paper aims to assist policy-makers in setting the target for reforms. 3. Theoretical Framework 3a. The Effect of Permanent Residency on Job Mobility In a static framework, the acquisition of permanent residency should raise labor mobility as workers extricate themselves from the legislative restrictions extrapolated above. Understanding the dynamics of job moves prior to and after green card receipt, however, is key to building an appropriate empirical model. Figure 1 plots the hypothetical dynamic effects of visa status and the EB green card application process on the rate of voluntary job change for an immigrant worker i, who initially works under a temporary work visa and eventually becomes a permanent resident under the sponsorship of his/her employer. The horizontal axis denotes in years the time relative to green card receipt, and the vertical axis tracks the probability of voluntary employer change in year t. The hypothetical trajectory in job mobility is graphed abstracting away from any time trends in job turnover, or age and experience effects. The long-run steady-state mobility rates of temporary workers and permanent residents are P1 and P4, respectively. The sponsorship effect suggests P4 is greater than P1. Based on the institutional background described in the previous section, when worker i submits an EB green card application at time t1, the probability of job move is expected to decline from P1 to P2 because changing jobs implies restarting the green card application process. Mobility decreases continuously between t1 and t2 since opportunity costs are rising as the worker gets closer to receiving a green card. Once the green card is awarded at t2, the immigrant worker likely experiences a sharp spike in mobility not only because he or she has escaped the visa-queue trap, but also because more firms may be willing to hire him or her. If compressed labor mobility is assumed to surge by the same degree, then given that workers are typically on temporary visas for multiple years (t2 t0 > 1), the rise in voluntary moving in year t2 should exceed the difference between P4 and P2. The spike in mobility in year t2 should eventually diminish and converge to the steady-state mobility rate. However, the trend in mobility between t2 and t3 is less predictable. Even though workers are free of institutional barriers to mobility, searching and relocating may

9 8 cause a delay in job transitions, implying that the probability to move is increasing between t2 and t3. On the other hand, if the timing of green card receipt is known beforehand and workers can engage in searching prior to t2, then mobility is likely to be downward-sloping because further job changes are less likely to happen following a recent move. Due to the ambiguity, I simply assume that the probability to move is constant between t2 and t3. In this graph, the instantaneous effect of acquiring LPR is P3 P2, which can be decomposed into three segments. The decrease in voluntary moving from P1 to P2 is due to the green card processing effect. Permanent residents persistent higher mobility rate relative to that of temporary workers (P4 P1), ceteris paribus, reflects mobility frictions driven by the visa sponsorship effect. The spike in mobility beyond the steady-state rate of permanent residents (P3 P4), however, can be a consequence of either sponsoring or processing effects, and is referred to as combination effects. In this study, I empirically investigate all three effects, estimate the instantaneous effect, and calculate a lower bound of the processing effect. My investigation of job mobility complements two contemporaneous papers. Using a sample of immigrants with initial visas for temporary work, Hunt (2017) compares job turnover rates among workers who obtained a green card in the past four years to the rates among earlier green card recipients. My analysis indicates that this comparison is only valid in providing general evidence for restrictions on mobility. In addition to the total mobility constraint, my paper also investigates the dynamics in job mobility and identifies the component effects, which have distinct policy implications. Another limitation of Hunt (2017) is that its sample excludes a substantial share of temporary workers whose initial visa was for study or training. 10 Using data from six large Indian IT firms, Depew, Norlander, and Sorensen (2017) examine inter-firm moves among H1-B workers within the United States and find that these workers exhibit a significant amount of mobility after six to nine years, only 14 percent of the workers remain in the same firm. Although their data have the advantages of determining temporary visa types and tracking short-term job changes, my data better represent the skilled guest worker population. 10 Descriptive statistics based on NSCG show that, among current temporary workers, the proportion of individuals arriving on visas for study/training outweighs that of individuals on any other types of visas.

10 9 3b. The Effect of Permanent Residency on Earnings In a perfectly competitive labor market free of frictions, a temporary visa holder should receive the same equilibrium wage as a comparable permanent resident. Employment frictions imposed by the immigration rules, however, potentially decrease the wage rate of temporary immigrants. One source of friction is the explicit costs of visa sponsorship, which are presumably capitalized into worker wages. Since these costs are only 4 to 7 percent, on average, of a skilled worker s annual salary, their influence on earnings is small. 11 The main friction is the institutional hurdles to job mobility, rendering workers unable to effectively respond to adverse employment situations, such as below-equilibrium wage rates. Job-to-job transitions are documented to be a crucial pathway for wage growth (Topel and Ward 1992). Once these mobility restraints are removed by receipt of a green card, workers can pursue a higher wage by either directly switching to a higher-paying job or threatening to leave their current employers. Therefore, the acquisition of permanent residency is expected to increase worker earnings, or in other words, generate a green card premium. The dynamic effects of a green card on earnings can differ from those on job mobility. Although workers have a curtailed ability to move during the lengthy application period, employers are unlikely to cut their wages because the decision to sponsor workers for LPR implies firms appreciation of worker productivity and anticipation of long-term employment relationships. Quite the contrary, employers may still allow wages to grow (but presumably at a rate lower than the case of full mobility) for fear that workers would change jobs after receiving permanent visas. However, green card sponsorship costs may be transferred to workers, implying an absolute decrease in earnings when the application process starts. In either case, the wage growth rate is predicted to rise abruptly upon green card reception and then fall to a more steady rate. One caveat is that the preceding hypotheses only apply to successful EB green card recipients; the earnings effects on the majority of temporary workers who are never sponsored for green cards are likely to be different. Unsponsored workers should suffer a larger wage penalty because they are more likely to work under short-term contracts. Yet, my empirical design only allows me to estimate the effects on the successful transferees, which should be smaller than the 11 Appendix Table B3 shows that the mean salary of skilled workers is roughly $100,000, and the H1-B sponsorship costs between $4,000 and $7,000.

11 10 overall effects. The impact of a permanent visa on earnings can also be heterogeneous across the categories of green card receptions, although the direction of the heterogeneity is less predictable. Family-sponsored applicants, for example, may suffer smaller wage penalties than EB applicants because the application period does not tie them to employers. In contrast, the sponsoring decision in itself is an indication of employer satisfaction of worker performance, so their wages may be less affected compared with the other green card types. Since my data do not include LPR types, my estimates are the average effects weighted by the number of EB and non-eb permanent immigrants. 4. Data My empirical analysis draws on the 2003, 2006, 2008, 2010, 2013, and 2015 waves of the National Survey of College Graduates (NSCG). Conducted by the National Science Foundation, the survey samples individuals with at least a bachelor s degree, who are residing in the United States during the survey reference period and are seventy-six years old or younger. The NSCG data are publicly available and best suited for my study because the surveys have a large sample size with information on both immigrant visa status and job mobility. The four base survey waves draw stratified random samples from the college-educated respondents to either the US Census (2003 wave) or the American Community Survey (2010, 2013, and 2015 waves). The survey design generates a gap, ranging from one to three years, between the sample frame and the survey year, implying that short-term migrants are absent from the samples. The 2003 survey, for example, was administered three years after the 2000 Census, and therefore only immigrants who arrived by 2000 and lived in the United States for at least three years are present in the data. The design of the surveys generates three short panels, allowing for longitudinal analysis. The National Science Foundation conducted follow-up surveys in 2006 and 2008 on a cohort selected from the 2003 wave with education or employment in science and engineering or related fields, which provided the first panel. Beginning in the 2010 cycle, NSCG implemented a revised sampling method, the rotating panel design, in which a new panel of respondents is selected in each survey year; these respondents participate in three biennial follow-up interviews before rotating out of the survey. This process leads to a second three-wave panel. Columns (2) and (3) of Appendix Table B2 present the sample sources for each survey wave, which consist of both a new sample and a returning sample from the previous survey cycle. Since the 2010 wave also

12 11 draws individuals from the 2008 NSCG, a third two-wave panel is available. These panels are the basis for my individual FE models, as discussed in Section 5. The immigrant visa status of foreign-born respondents can be identified through responses to several survey questions. Each participant in the survey is first asked whether he or she was a US citizen during the survey reference period, and if so, whether he or she was native-born or became a citizen through naturalization. Foreign nationals are further classified as either permanent residents or temporary visa holders. Permanent residents are also asked to provide the year when their green card (permanent residency) was awarded, while visas held by temporary residents are divided into four major categories: for temporary work (e.g., H-1B, L-1A, L-1B), for study or training (e.g., F-1, J-1, H-3), as the dependent of another person (e.g., F-2, H-4, J-2, K-2, L-2), and for any other reason. These non-us citizens are also asked what their visa status was when they first came to the United States for six months or longer, and the options are permanent US resident visa (green card) and the temporary visa categories listed above. The measurement of job mobility rests on three retrospective questions regarding an individual s past employment. First, respondents are asked whether they were working for pay at both the current survey reference date and an earlier reference time (roughly two years ago); if so, the survey asks if they were working for the same employer and in the same type of job. Appendix Table B2 lists these reference dates, which are roughly two years apart. Individuals who changed jobs, employers, or both during the two-year window are further asked to select the reasons for the change from several non-mutually exclusive options, such as job location, change in career interests, pay and promotion, family reasons, retirement, and layoffs. To conduct my empirical analysis on visa status, mobility, and earnings, I restrict the sample to non-disabled foreign-born workers aged who are employed in both periods referenced in the job mobility questions. Since my study only considers direct job-to-job transitions within the United States, I exclude immigrants who entered the country or received new full-time education after the previous mobility reference date. Temporary residents holding visas other than those for work are also dropped. Further operations in constructing the sample are described in Appendix A2. My sample contains nearly twenty thousand individuals whose characteristics are summarized in Column (1) of Appendix Table B3.

13 12 5. Empirical Methods In this section, I empirically answer the question of whether temporary workers face institutional constraints on their job mobility by measuring the total effects of obtaining permanent resident status on temporary workers job changing rates as well as the three component effects mentioned in Section 3. To examine the presence of mobility-related lower earnings, I estimate the effects of having a green card on worker annual earnings. Two endogeneity problems are inherent in gauging these effects. First, acquiring LPR is a voluntary choice, either solely made by individuals, or jointly made by workers and firms. If workers sponsored for LPR and those without sponsorship have differential propensities to change jobs or different earnings even in the absence of green card applications, comparisons between temporary workers and green card holders would produce biases due to selection. Hunt (2017) approaches this issue by comparing the job mobility of recent green card recipients with that of earlier ones. Yet, this approach solves the selection problem only if there is no unobserved differences across cohorts of green card holders. To address the possible presence of cross-cohort heterogeneity, I employ individual FE models, accounting for any permanent individual-specific component to mobility. I first present my empirical specifications for the job mobility effects. The estimation of earnings effects largely follows the same framework. My analysis starts with a simple individual FE model that compares job mobility within the same worker before and after a green card is awarded, controlling for year effects: Pr (Firmchange it, t 2 ) = α i + β 1 PR it + β 2 NC it + δ t + ε it (1) where i is an immigrant worker, and t denotes the survey year. The dependent variable is an indicator that worker i changes employers voluntarily at least once between the prior reference date and the current survey date, which are roughly two years apart. Voluntary employer changes are defined as those not caused by layoffs or retirement. The variable α i is a permanent component of worker i s probability to move across firms. PR it is a dummy equal to one if worker i is a permanent resident in year t; NC it is a dummy equal to one if worker i is a naturalized citizen in year t; the omitted group are temporary workers; and δ t indicates survey year FE. Since this

14 13 approach only exploits variations within the same individual over time, the survey year dummies simultaneously control for age and experience effects in mobility. 12 I do not control for time-varying variables endogenous to a worker s employer changes, such as marriage, number of children, and work-related characteristics, that are likely to change concurrently with or following a job move. Education is a time-invariant variable by construction because only workers who had completed their most recent degrees prior to the previous mobility reference date are included in the sample. As the primary parameter of interest, β 1 is the long-run average effect of a change in visa status from temporary workers to permanent residents on the changes in the probability to switch employers. Theoretically, it is the difference in average mobility rate on either side of t 2 in Figure 1, although in practice, because the NSCG panels are fairly short, my estimates tend to reflect more of the dip and spike in job moving around t 2. On the other hand, β 2 is the effect of transitioning from a temporary visa to citizenship. Next, I implement an FE model with dynamics to trace out the time-varying effects of permanent residency on job mobility. This model allows a direct examination of mobility patterns relative to the time of green card receipt so that the hypothetical trends presented in Figure 1 can be tested. The model is specified as follows: n Pr (Firmchange it, t 2 ) = α i + j= m β j GC i,t+j + δ t + ε it (2) where, in comparison with the previous model, the dummies for visa status are replaced by a series of time dummies indicating the year relative to green card receipt. The m lagged dummies show if mobility is depressed prior to acquiring green card, and the n post-treatment dummies capture the dynamics in job mobility after LPR is awarded. In practice, this model includes single year indicators from five years prior to green card receipt to nine years after it. The omitted group are those who received their green card ten years prior or more Specifications controlling for age or experience yield similar estimates. 13 In this regression, the omitted group should ideally be the temporary workers who are going to receive green cards in four years or more so that trends in mobility are estimated relative to these initial values. Unfortunately, because the panel data are quite short, leading to an especially small sample size of this group, omitting it would cause large standard errors in the estimation of all the time dummies. Therefore, the omitted group is selected from the other end of the time line.

15 14 Because of the short nature of the NSCG panel data, Equation (2) cannot capture the trajectory of job mobility prior to three years ahead of green card receptions. This implies that P1 in Figure 1 cannot be measured by the model, suggesting that the processing effect (P1 P2) cannot be disentangled from the sponsorship effect (P4 P1). Given the data limitation, the best possible approach to separately identify the decomposed effects is to adopt OLS regressions utilizing the variations in job moves and visa status across immigrant workers. Specifically, pooling the six waves of data together, I estimate the following regression model: Pr (Firmchange it, t 2 ) = α + β 1 PR it,gc 6 + β 2 PR it,gc 5 + β 3 TWGC it + β 4 X it + δ t + ε it (3) where PR it,gc 6 is a dummy equal to one if worker i has been holding LPR for at least six years or if worker i is a naturalized citizen in year t whose year of green card receipt is unobserved (referred to as old permanent residents). PR it,gc 5 indicates that permanent resident i in year t acquired LPR five years ago or less (referred to as new permanent residents). The year six is used as a dividing line because the FE with dynamics model indicates that permanent immigrants converge to the long-run steady-state mobility rate in the sixth year following green card reception. TWGC it is a dummy equal to one if temporary worker i is receiving LPR in four years or less (observed from later survey cycles), implying that the worker is already in the application process and is referred to as an ongoing applicant. Obviously, not all temporary workers in the application process are captured by the dummy, so the omitted group includes the guest workers who are either not applying for green cards or at least not observed applying for one; they are referred to as nonapplicants. X it are controls including age and its square; working experience and its square; years in the United States; gender, marital status, and their interaction; presence of children and its interaction with gender; respondent location; the highest degree type; majors of the most recent degree; employer sector; and occupation. To minimize the possibility of controlling for changes endogenous to job moves, the time-variant variables all refer to the survey year of a respondent s first appearance in the data. As the parameter of interest, β 1 measures the difference in mobility between long-term permanent residents and temporary workers not observed applying for LPR conditional on observable characteristics. It is an upper bound of the sponsorship effect noted in Figure 1 because

16 15 some of the workers in the omitted group may actually have already been sponsored for LPR, thereby subject to the processing effect as well. One caveat is that the employer sponsorship regulation may differentially affect future green card applicants and those never sponsored for green cards. β 1 captures the average effect on these two groups. The fact that in NSCG, employer changes are determined over a two-year interval implies that multiple job changes over the interval are counted as one change and the precise time of job moves is also missing. The date of a green card, on the other hand, is a calendar year, thus the timing of employer change with regard to green card receipt cannot be perfectly determined. To at least partially address the noise in relative timing, I assume that for job movers, the job mobility date is the year when the current job starts. This assumption and the variable of green card reception date allow me to impute observations of job move and visa status for years between the mobility windows, thereby generating panels spanning the years , , and , respectively. Using the expanded panels, the three models presented above are estimated with the dependent variable as the probability of voluntary employer change in year t, rather than over the interval of t and t 2. This approach has the advantages of a larger sample size and producing estimates that more accurately capture the timing of job changes with respect to green card receipt. However, it may introduce attenuation bias due to potentially more measurement errors in the job move date variable. As such, estimations based on both the original sample and the expanded panels are presented for the purpose of comparison. The effects of obtaining permanent residency on earnings are estimated using the same individual fixed effects models described above. LnI it = φ i + γ 1 PR it + γ 2 NC it + δ t + u it (4) n LnI it = φ i + j= m γ j GC i,t+j + δ t + u it (5) In Equations (4) and (5), the dependent variable is the log of annual earnings of worker i in year t. Earnings are adjusted for inflation using the annual average consumer price index collected from the Bureau of Labor Statistics. The worker fixed effect, φ i, captures all time-invariant characteristics affecting earnings. The γ coefficients are the effects of permanent immigration status on earnings.

17 16 My FE design relies on the assumption that conditional on immigrant-specific time trends in mobility (or earnings), no time-variant individual-specific factors are correlated with both the timing of green card receipt and voluntary moving (or earnings). While this assumption is inherently untestable, unreported regressions have addressed some of the confounding circumstances. In cases in which green cards are obtained through making an investment, since starting one s own business implies an individual has switched from a salaried job to selfemployment, a job change and an increase in earnings are directly related to green card receipt. My test on whether recent green card recipients are more likely than other immigrants to become self-employed offers no support for this hypothesis. Another similar possibility, which is more likely among women, is that green cards may be acquired by marrying natives; getting married generates higher probabilities to move geographically and in the meantime change employers. Although information on spouse status is unavailable, I find no evidence that female workers recently being awarded LPR were more likely to get married shortly before its reception. Nonetheless, I cannot account for the cases in which workers receive a green card when their spouse obtains one through employment. In fact, these situations are likely leading to heterogeneous effects by gender, as discussed in the following section. 6. Results 6a. Job Mobility Effects I. Simple Individual FE Estimates Results from estimating Equation (1) are shown in Table 2. Consistent with Hunt (2017), I find that permanent resident status positively affects worker mobility, although my estimates are markedly larger. Column (1) shows that overall, compared with temporary visas, holding a green card renders workers 8.1 percentage points more likely to change employers over a two-year period. The increase in job mobility rate is 6.6 percentage points when the status change is from temporary visa to naturalized citizenship. The evidence that transitioning from a green card status to citizens is associated with a decline in mobility aligns with my hypotheses plotted in Figure 1 in that the temporal spike in job moves declines to a long-run steady-state rate after five years on green cards. The estimates based on a one-year window follow the same pattern workers annual job changing rate rises by 2.7 percentage points due to the receipt of permanent residency. The fact that the effect on one-year mobility is less than half that of the two-year estimates can be driven either by

18 17 individuals probability of moving over a two-year window being more than twice the annual mobility rate or by measurement-error-induced attenuation bias. While it remains untestable, in any case, the estimated one-year mobility effects are lower bounds. The size of these estimates is more sensible in comparison with the average job turnover rates. Based on the sample average job mobility rates reported in the first row in Appendix Table B4, immigrants are 64 to 74 percent more likely to change employers due to green card receptions. Because workers from India and China face particularly long green card queues as a result of the per-country quota, larger post-green-card mobility responses are expected among these immigrants. Table 3 presents the mobility effects separately by top sending countries of highskilled workers. As predicted, Columns (2) and (3) show that Indians and Chinese are the main groups driving these effects. The average increase in two-year job mobility rates following LPR receptions among them are 14.1 and 21.2 percentage points, respectively (Panel A), whereas no significant effects are observed among workers from Canada and the United Kingdom. The oneyear effects are more similar between Chinese and Indians, 5.4 and 5.8 percentage points, correspondingly (Panel B). Column (6) shows percent decreases in the mobility effect when Chinese and Indian workers are excluded from the sample. The heterogeneous effects suggest that a considerable portion of the spike in mobility is driven by the processing effect. The heterogeneity also precludes any time-varying confounders that may have contaminated my FE estimates unless the impact of those confounders on Chinese and Indian workers differs from that on other groups. The effect of permanent visas on job turnover also tends to vary by degree type. Workers with different educational attainments are likely to fall into different EB green card application categories. Those with more advanced degrees generally have shorter waiting lines for green cards and are thus less affected by the processing effect. Workers employed in higher education and research institutions, mostly doctorate holders, are also largely immune from the sponsorship effect because recruitment decisions by these institutions are less likely to be influenced by visa status. Table 4 presents my estimates by respondent highest degree type. Consistent with the institutional setting, doctorate recipients are found to be less restricted in their ability to move, while the effects on bachelor s and master s holders are quite similar.

19 18 II. Individual FE with Dynamics and the Decomposed Effects The coefficient estimates of Equation (2) are plotted in Figure 2. Recall that the omitted group in generating these plots are workers that have held a green card for 10 years or longer. By and large, the point estimates of the dynamic patterns in job mobility effects are in consonance with my conceptual framework. Panel A displays a downward trend in mobility in the years prior to green card reception. Although no such trends are found for annual mobility rates (Panel B), an abrupt spike in voluntary move prevails upon LPR reception in both cases. The spike in Panel A indicates that in the first year following receiving a green card, workers are 10.3 percentage points more likely to change employers during a two-year window relative to one year prior; the jump in voluntary job moves in Panel B shows a prompt increase in yearly mobility rate of 3.2 percentage points during the year when LPR is awarded. These instantaneous effects confirm the existence of institutional restraints to employer change while workers are on temporary visas. As my hypothesis presumes, these effects soon subside after the fifth year on green cards and reach a constant job changing rate. Even without the combination effects component (P3 P4 in Figure 1), the restraints are at least 8 percentage points over a two-year window and 2.2 percentage points over one year. Due to few observations available for temporary workers in the years prior to receiving green cards, especially three years or more prior, estimates from these lagged terms are associated with large standard errors and thus fail to provide evidence on mobility rates before the start of a green card application (P1 in Figure 1). Therefore, the FE model with dynamics cannot separately identify the sponsorship and processing effects. Yet, based on the dynamic plots, the summations of these effects are estimated as 8 (two-year window) and 1.4 (oneyear period) percentage points, respectively. The decomposed effects are estimated using OLS regressions, as presented in Table 5. Recall that the omitted group in these regressions are non-applicants (temporary workers not observed as having started applications for green cards). Job changing rates are compared between these workers and three other groups new permanent residents (on a green card for five years or less), old permanent residents (for six years or more), and ongoing applicants. Conditional on observable characteristics, the difference in mobility between non-applicants and old permanent residents is my upper bound estimate of the sponsorship effect. In Table 5, once the key determinants of job mobility, age and experience, and demographic characteristics are controlled for, the coefficient estimates are fairly stable. Based on the last column, the sponsorship effect is

20 19 estimated as 1.7 percentage points, at most, over a two-year period (Panel A) and 0.6 percentage points for one year (Panel B). Taken these numbers together with the summation of the two effects estimated in the dynamic plots, I calculate lower bounds of the processing effects of 6.3 and 0.8 percentage points, respectively. Since the regression of mobility over two years is less likely to be affected by measurement errors, I prefer this specification for drawing conclusions. Compared with the sample average job changing rates reported in Table B4, my estimates suggest that when green cards are awarded, temporary workers are at least 60 percent more likely to switch employers due to previously being locked by the application process. 6b. Earnings Effects and Heterogeneity by Gender Since men and women tend to obtain green cards from different sources and they also differ in their labor market behaviors, an analysis stratified by gender is conducted for both mobility and earnings. I begin with presenting the estimates on job mobility. Table B5 indicates no significant differences across gender, although female mobility does appear to be slightly more affected by a green card. Figure 3 shows the dynamic effects over a two-year window by gender. While the trajectory in job moves are similar across genders, the mobility propensities relative to the longrun steady-state rates are saliently different. Male workers exhibit broadly the same mobility patterns as the full sample. In contrast, unlike men who are more likely to move than ever in the years following green card receipt, women have job mobility rates that are actually slightly below the long-run rates. In other words, the combination effects illustrated in Figure 1 are not found among women. The differential mobility behaviors can be caused by the fact that women devote themselves more to the family and thus their job search efforts are equally distributed over the years following green card receipt, whereas being more career-focused, male immigrants strive to find better employment as soon as the barriers to mobility are removed by a green card. Results on log annual earnings are presented in Table 6. Column (1) shows the estimates of Equation (4) for the full sample. Overall, workers are found to experience a 5.5 percent growth in earnings upon obtaining permanent residence status and naturalized citizenship leads to another 1.5 percent increase. Columns (2) and (3), however, indicate that these effects are entirely attributed to male immigrants the acquisition of green cards generates 7.8 percent more earnings for male workers, and naturalized citizenship raises their earnings by a further 1.9 percent. The reason that obtaining naturalized citizenship raises earnings is because citizens are able to take