Several new U.S. laws and additional funding for border enforcement made crossing more difficult: the Illegal Immigration Reform and Immigrant Responsibility Act of 1996, the Homeland Security Act of 2002, the USA Patriot Act of 2002, the Enhanced Border Security and Visa Entry Reform Act of 2002, the Intelligence Reform and Terrorism Prevention Act of 2004, the REAL ID Act of 2005, and the Secure Fence Act of 2006. According to a survey of migrants, the cost of crossing the border with the help of smugglers, or “coyotes,” rose substantially since mid-1990s .Cornelius notes that increasing coyote costs are associated with decreases in the probability of returning to a country of origin and with increases in deaths along the border. Pena shows that border enforcement is negatively associated with agricultural worker migration specifically. Newspaper articles indicate that the U.S. government substantially increased U.S.-Mexican border enforcement since the mid-2000s. In addition, changes in U.S.-Mexican foreign relations and in Mexican public policy reduced incentives for its citizens to move to the United States in the second half of our sample period . Mexican farm laborers were less like to migrate to the United States because of increased economic growth in Mexico, rising productivity, and decreased birth rates . The 1997 anti-poverty Programa de Educación, Salud y Alimentaciónin Mexico increased welfare in Mexico through education, health, and conditional cash transfer initiatives, which decreased the incentive for workers to cross the border . Oportunidades also increased agricultural production in Mexico.For example, the 1986 Immigration Reform and Control Act conferred legal status on many previously unauthorized workers, which provided a path to a legal permanent residence status and citizenship. By so doing, IRCA reduced the share of unauthorized workers during the 1990s. Over time, many of these workers left agriculture. Together,vertical farm these factors reduced the number of undocumented workers from Mexico in the United States.

Martin reviews the history of immigration legislation and domestic enforcement and concludes that the e-verify program had little impact during the period immediately after IRCA went into effect. In contrast, Kostandini, Mykerezi and Escalante show that after 2002, counties participating in the Department of Homeland Security’s 287 enforcement program had fewer foreign-born workers, reduced labor usage, and experienced changes in cropping patterns among producers. In our empirical analysis, we investigate whether the willingness of a worker to migrate within the United States depends crucially on legal status. A variety of other structural factors also affected the supply and demand for U.S. farm labor. In recent years, increased consumer demand for fresh fruits and vegetables and expanded exports of agricultural commodities led to greater production of labor-intensive crops . Agricultural producers have responded to higher labor costs by improving productivity through increased mechanization and more efficient cultivation practices . These changes, by altering the value of the marginal products of labor across areas, affected the incentives to migrate within the United States.Ideally, we would like to model and test the effects of each of these various shocks. However, the number of institutional and policy changes are large relative to the number of years in our data set, 1989 to 2009. Thus, it is not feasible to test and model these shocks individually. Rather, we estimate a migration model for each of the large, annual cross sections, and allow the coefficients in each year to change, so as to reflect the structural change over time stemming from all these individual shocks.These trends hold for various sub-samples. Figure 2 plots the proportion of migrant farm workers from 1989 through 2009 by legal status, region, and age.

The solid line in each sub-panel indicates the proportion of migrants in the full sample. sub-panel 2A shows how the proportion varies by legal status over time: citizens, legal permanent residents, and unauthorized workers. On average over the entire period, 18% of citizens migrate compared to 39% for legal permanent residents, 60% for those with other work authorization, and 48% for those who are unauthorized.5 Thus, a higher share of unauthorized and other authorized workers migrated than did citizens and legal permanent residents in the sample overall.6 The figure illustrates how the migration rates for authorized versus unauthorized workers both fall over the last decade of our sample. sub-panel 2B presents the proportion of migrants by geographic migration patterns. Traditional networks of migrants follow typical U.S. harvest patterns by starting in the south and moving north as the season progresses.The NAWS classifies workers into three north-south streams based on their work location at the time of interview and therefore includes both workers who follow-the-crop and those who work in a single location. As the figure shows, migration rates were generally higher for Eastern and Midwestern stream workers than for Western stream workers. The migration rate declines over time for all streams. sub-panel 2C shows that workers younger than 35 are slightly more likely to migrate than are older workers. Again, both groups show a decline in the rate of migration in the recent period. These results also hold for other demographic variables that are correlated with age, such as education and experience levels. Our definition of a migrant includes both of the NAWS’s sub-categories of migrants: follow-the-crop migrants and shuttle migrants. Follow-the-crop migrants are workers who move between U.S. farms as the agricultural season progresses. Shuttle migrants move between their homes and a single distant work site.Figure 3 shows how the share of farm workers who follow-the-crop or are shuttle migrants varies over time. The migration rate for both groups fell over our sample period.

After the first year of the sample, the share of shuttle migrants exceeds that of follow-the-crop migrants. Analyzing these types of migrants separately produces results similar to those reported for the combined group in the following sections. To estimate a migration model, we can use any of the standard binary choice models: logit, probit, and the linear probability model. Because the share of workers who migrate lies between a quarter and a half in most years, all three methods produce nearly identical results in terms of the marginal effects of individual variables, their ability to predict, and our other analyses. For presentational simplicity, we use the linear probability model.9 We estimate separate migration models for each year of the sample. We did so because the coefficients are not constant over time.We tested and rejected that the intercept and slope coefficients are constant across various time-period aggregations such as the two halves of the sample and each pair of successive years. We examine how the probability of migrating varies with three groups of demographic variables: individual characteristics, family attributes and assets, and employment experiences. Our individual demographic variables include age; years of school; a dummy for female; dummies for Hispanic, African American, and American Indian ; dummies for legal permanent resident, unauthorized worker, and other authorized worker ; and a dummy for whether the individual speaks at least some English. Family characteristics include whether the worker is married, lives with a spouse in the United States, and lives with at least one child under 18 years of age. Family wealth and income variables include whether the individual owns or is buying a house in the United States; whether the individual owns or is buying a car or truck in the United States; and the worker’s self-reported real personal income in the previous year . We used lagged personal income to avoid endogeneity. Our employment variables include years of farm experience; a dummy if the employee performs semi-skilled or skilled work or supervises others; and a dummy if the worker was hired by a farm labor contractor . 12 Our dependent variable equals one if the worker is a migrant and zero otherwise. Table 2 shows estimates of our model using data from 1989, the first year of our sample ; for 1998, the end of our stable period ; and for 2009, the last year of the data . The table reports robust standard errors in parentheses. Based on hypotheses tests, we can reject the hypothesis that all the slope coefficients are identical in any two years. We can similarly reject any of the other aggregations over time. The following discussion focuses on the estimates from the 1998 model . Nine out of the 19 slope variables are statistically significantly different from zero. A female is 15 percentage points less likely to migrate than a male, which is a large difference given that the sample average probability of migrating is 53% in 1998.

Hispanics are 15 percentage points more likely to migrate than are non-Hispanics. Skilled workers are 7% percentage points less likely to migrate than unskilled workers. Surprisingly,nft vertical farming age and farm experience have negligible effects. Married workers who do not live with their spouse in the United States are 19 percentage points more likely to migrate . However, married workers who live with their spouse in the United States are 10 percentage points less likely to migrate . Similarly, workers are 11 percentage points less likely to migrate if they live with their children. Presumably, these family-oriented workers see themselves as having a higher opportunity cost of migrating. The probability of migrating falls with lagged personal income. We expected this result because the main purpose of migrating for these workers is to earn a higher income. Workers hired by farm labor contractors are 15 percentage points more likely to migrate than are those who are directly hired by farmers. Farm labor contractors provide labor to many farms and may provide transportation to distant jobs. In contrast, a worker hired by a farmer is likely to work at a single location. We had expected that legal status of workers would play an important role; however, no clear pattern emerged. In the 1989 and 1998 regressions, we cannot reject the hypothesis that the coefficients on unauthorized workers are zero . The coefficient is negative and statistically significant in the 2009 regression. We see the same pattern for other-authorized workers. In contrast, legal permanent residents were 14 percentage points more likely to migrate in 1998, but the difference was not statistically significant in the other two years.Schneider and Ingram argue that elements of policy design are political phenomena amenable to empirical analysis. In particular, “Data can be generated by the study of texts, such as legislative histories, statutes, guidelines, speeches, media coverage, and analysis of symbols contained therein” . The data used to analyze the policy design of SB 700 was gathered from several sources. The first is the legislative record for the bill. Since SB 700 was passed and signed into law by Governor Davis, material from the governor’s chaptered bill files are part of the database. In addition to the official record of the bill, newspaper articles and editorials concerning SB 700 were included for analysis. Searching the ProQuest newspaper database from February 21, 2003 to September 22, 2003 using the keyword “SB 700” yields 81 newspaper articles and editorials for use in the narrative analysis. Articles and editorials after this date are used to examine the dynamic relationship between narratives, policy tools, and agents and implementation structures in the implementation activities of SB 700. A special report entitled “Last Gasp” published in the Fresno Bee on December 15, 2002 is also included in the narrative analysis. While the report predates the introduction of SB 700, it is included because it is continually cited by the bill’s author and its supporters and opponents in the legislative materials. Its importance in shaping the debate over this bill warrants its inclusion. The specific form of discourse analysis used to analyze the texts of SB 700 will be Roe’s narrative policy analysis, which consists of two stages. The first is the disaggregation of the text into discrete problem statements, which contain the simplest assertions of causal relationships or sets of causal relationships that link problems to their source . The second stage requires the aggregation of all the problem statements across the entire “data set” or texts. This allows the researcher to see the pattern of commonly identified problems and causal relationships concerning the policy. It is these aggregated problem statements that are then identified as narratives .