Wherefore Art Thou: Emerging Characteristics in California’s Hispanic Out-Migration

The loss of California population to domestic migration has undergone significant changes over the last fifteen years. While traditionally the outflow has been seen as "white or middle class flight," recently a new underlying dynamic has begun to assert itself: Hispanic out-migration. Having spent some time in California, both newly-arrived and later generation Hispanics are subsequently migrating to other states in increasing numbers. This movement is notable for the change in destination, particularly to more rural states, as well as education levels; this migration appears to be most prevalent among the least/lesser educated. We examine the underlying forces behind this Urban to Rural out-migration shift using a nested logit migration model augmented by a spatial correlation component. Using IRS and ACS data, we examine the relative effects of education, same-race, and amenities on this migration decision. We show that race concentrations and urbanization are less important than cost-of-living and wage/employment considerations--elements that tend to favor rural areas. The growing prominence of the Hispanic ethnicity in the US population, and its changing distribution and dispersal pattern, motivates specific attention to Hispanics changing roles in models of internal migration. The history of such models has shown increasing elaboration over past decades. Early migration researchers conceptualized the migration process as a largely labor-market phenomenon where migration responds mainly to the spatial disparities in economic opportunities. The typical model explaining origin-to-destination specific flows of migrants included such factors as wage level and unemployment rate, together with distance and origin and destination populations (Lowry, 1966). Later, ‘quality of life’ factors, particularly those related to climate, became introduced into these models as movement from the snowbelt to the sunbelt became more widespread (Graves, 1976; Greenwood 1981; Liaw and Ledent, 1987). More recently, the wide disparity in housing costs across states has entered into the calculus of movers (Turner, 2000; Coy, 2002) such that it should also be considered in models of internal migration within the United States. While US migration patterns adhere to a standard set of determinants largely dependent on personal characteristics such as age and education, it is increasingly important to incorporate the differences that race and ethnic background may create. In examining the migration patterns of California Hispanics we will address the role of raceethnicity in two ways. First, we will reassess the role of so called ‘cultural constraints’ as they affect departures and destination choices for different race-ethnic groups (Liaw and Frey, 1996). These constraints shape migration patterns for these groups due to their needs for social support networks, kinship ties, and access to informal employment opportunities that tend to be available in areas that house large concentrations of coethnics. We argue that Hispanics as a group are acting more like average migrants with a consequent reduction in their reliance on cultural support networks. Secondly, we examine the locational choices of these Hispanic migrants and attempt to parameterize a choice theoretic using a nested logit model that incorporates the rising number of urban to rural migration decisions. Theoretical Considerations Theories of internal migration recognize the multiplicity of choices that individuals face when making a migration decision. The most accepted perspective argues that individuals decide where to live based on a comparison of the expected long-term costs and benefits of living in different places, including the place they currently reside in. However, the literature is not clear about how exactly the migratory decision is made. Do individuals decide to leave their place of origin first, and then decide where to move to? Or do they decide that they want to migrate only after they have compared the conditions in all the destinations they are aware of and know where they will be moving to? If so, do they value the conditions in their place of origin more than they value the conditions in other destinations, or would they be willing to move to any place that has relatively better living conditions? Different approaches to these questions lead to different methods and statistical models of the migration process. According to Han (2000), migration is a complex process, which, as far as its emergence and development is concerned, is continually determined through a multiplicity of causes and factors. As a rule the causes triggering this process are a mixture of objectively compelling exogenous factors (e.g. company contacts or attraction through employment scouts) and subjectively justified decisions (e.g. better employment opportunities, starting a family). A classic approach to explaining the complex and multicausal determinants of migration can be found in the theory of so-called push and pull factors. Push factors (migration factors) comprise all those conditions of the migrants’ area of origin that induce them to migrate or temporarily migrate, such as political pressure, economic outlook and housing costs. Pull factors (factors that attract migrants) are those circumstances in the host area that motivate and encourage them to migrate. Factors that may attract migrants are, for example, a more receptive social structure, economic prosperity, better education and wage/salary opportunities relative to those in the original area. It is generally assumed that with modern information, communication, and transportation capabilities push and pull factors are becoming ever more important to individual migration decisions. Gatzweiler (1975) pointed out that in the end every migration decision is the result of push factors from the source area and pull factors from the target area working together. An array of approaches in the migration-theory literature aim to identify and explicate important determining factors for an individual’s willingness to migrate or for aggregated migration flows. The starting point of most theoretical models attempting to explain individual migration decisions is the neo-classical approach. The majority of microeconomic models are based on this approach, which views migration as a form of investment that is worthwhile or “profitable” for some individuals, but not so for others. To this a human capital factor is added so that migration takes place when the cost directly incurred through it will be reimbursed or will “pay for itself” through higher income in the future. Because of unemployment and other economic and non-economic aspects, migration is often connected with financial and social risks. According to neoclassical models, migrants must overcome a strong preference for one’s present environment, high migration costs, potentially poor labor market chances, great uncertainty and the hope that conditions in the origin region might unexpectedly turn for the better. There are considerations on the macro-economic level as well, which in the end can be traced back to micro-economic foundations. Among these are demographic trends, self-selection of migrants, self-sustaining migration and institutional restrictions on migration. Demographic trends are quite important: higher population figures in the sending area lead per se to greater migration flows. With regard to the causes of selfsustaining migration, so called network effects command the greatest attention. These result from the fact that, apart from the contacts amongst themselves, migrants above all, maintain good contact with their originating area. Through this exchange of information, the information and migration costs go down for all future migrants. People who have migrated in the past help the next ones with assimilation in the receiving area and also help reduce psychological costs that may arise through separation from one’s original area (Bauer, Epstein and Gang, 2000). Bartel (1989) studied the migration behavior of different groups of migrants (Asians, Europeans and Hispanics) to the United States in 1980. His research shows that the network effects are very strong. Regions with a high number of residents belonging to a particular ethnic group are the preferred destination of migrants of that respective group. In both their microeconomic and macro-economic studies, Bauer and Zimmermann (1995) found a high level of significance for network effects on migration. In a recent study Bauer, Epstein and Gang (2000) examined the influence of a migration network on migrants’ decisions based on location. They observed that the size of the Mexican network within the United States has a positive effect on the likelihood of migration. This is in contrast to the native or assimilated population. This group of individuals tends to treat locational amenities and economic prospects as the primary determinants of locational choice when migrating. While relative social connectivity may have strong influence, this influence is at a familial level and is usually independent of greater community considerations and ties. Nested Logit Models Models using the nested logit specification have been increasingly prevalent in the literature of internal migration during recent years, and good examples of their application are Liang and White's study of the impact of market transition in interprovincial migration in China in 1983-1988 (Liang, 1997), and Frey et al. study of the factors affecting population redistribution in the U.S. (Frey et al., 1996). We justify the nested formulation of the model over multinomial conditional logit (MNCL) mainly, because we suspect that the expected utilities associated with the unobserved destinations’ choices can be correlated, which is a violation of the independence of irrelevant alternatives (IIV) assumption required for the correct us of MNCL. If an individual perceives the destination alternatives as close substitutes, for example, if migrants perceive the attributes of Chicago and Dallas as similar, then the unobserved factors that affect one destination may also affect another, because the utility is no longer stochastically independent but is correlated through the error terms across those alternatives. See Greene (2003) and Knapp, et. al. (2001) for a complete discussion. Secondly, we also think that some of the factors (push and pull) that influence stay home or migration decisions are distinct and by being jointly modeling the two can be revealed. For example, mobility for some individuals may occur because the destination attributes that maximize the utility (higher salary, established social networks, incentives for more employment etc.) are less attractive for the home destination and are more attractive for a out of state destination, i.e. the individuals may choose another utility maximizing destination type of move, which can be thought of as distinctly different. But the destination decision can’t be made independently of the features of alternative locations (destinations). Thirdly, the nested model offers researchers a method of linking the choice of destination with the type of move (Stay-Home or Migrate) and captures any feedback between the two simultaneously determined decisions. Furthermore, decomposition of direct and cross elasticities into branch (upper-level) and choice (lower-level) between destination effects can be estimated. Consider now the problem of choice of area location. Suppose that an individual faces such a problem with a choice of Stay-Home or Move indexed as i = 1,2 (upper level) and location/area choice mode j = 1,2,...,Ni (lower level) Our formal model of the two-level nested logit model as demonstrated by Greene (2003) and modified to our specification is Pr [choice j, branch i] = Pj|i * Pi , where i refers to the type of move (Stay-Home or Move) and j refers to the destination choices (Urban area, Rural area, etc.) Thus nested logit models divide the migration process in two stages, however, the overall probability of out-migration in nested logit models does not depend exclusively on the characteristics of the place of origin but also on the conditions in all other areas in the country. This is because the probability of out-migration is conceptualized as a logistic process that includes a summary measure of the conditions in all the other areas. This summary measure is known as the “inclusive value” (IV) and represents how attractive is the option of migration for each state. In addition, nested logit models are flexible enough to allow the assumption that individuals value the conditions in their place of origin differently than the conditions in other states (The β coefficients are different for sending and receiving states.); and also because they allow the possibility that individuals do not consider the same characteristics of origin and destination when making their migratory decisions (The variables included to describe the characteristics in the state of origin need not be the same then the variables included to describe the characteristics of the potential destinations.) The unconditional probability of migrating to any particular state j is the product of the overall probability of leaving the state of origin times the probability of going to j given that a migration occurs. When a nested logit model is fitted, one needs to calculate first the the probability of moving from origin i to destination j, given that individuals are moving out of i. This is given by: ! !!" !h!"! ! ≠ ! = ! !!!!!!!!!!" ! !!!!!!!!!!" !!! eq (1) Secondly, one calculates the inclusive value for each state of origin, which is given by: !"! = ! !!!!!!!!!!" ! !!! !"# ! ≠ ! eq (2) And thirdly, one calculates the overall probability of out-migrating from state i, as given by: ! !! = ! !!!!!!!!"! !!! !!!!!!!!"! eq(3) The inclusive value and the parameter φ is a measure of the correlation among the random error terms due to unobserved attributes of destination choices and is used as a test for random utility maximization in a nested logit model. The expression φ IVi captures the feedback between the lower level (destination choice) and upper level (type of move) of the model. The parameters α, β, and φ are then estimated using FIML simultaneous estimation. The model is consistent with utility maximization if the conditions, 0 < IVi ≤ 1, are satisfied. If IVi = 1, the nested model collapses to the multinomial conditional logit model (MNCL). The full information maximum likelihood (FIML) estimation method for nested logit models estimates all of parameters simultaneously by maximizing the unconditional log-likelihood; Log-L = Σ log P(j|i) + log P(i) eq(4) Nested logit models can also be interpreted in terms of a utility maximizing decision process (Hunt 2000; Heiss 2002). One can think that the utility that individuals living in state I would get from moving to state j would be given by the function: !(!!") = !! + !!! + !!!" !"# ! ≠ ! eq(5) And that the utility that individuals would get from staying in their place of origin would be given by: !(!!!) = ! + !!! eq(6) Individuals decide whether to leave the area they live in or not by comparing their expected utility there (E(Uii)) with the maximum utility they could get in any other area (max(E(Uij)). If they decide to migrate, they will move to the place that maximizes their utility The Data The covariates used in the analyses were chosen according to the neoclassical, social capital, and to the empirical literature on the determinants of internal migration in developed countries. One of the most common assumptions in the literature of internal migration is that migrants go from areas with low wages and high unemployment to areas with high wages and low unemployment (Todaro, 1980). The probability of migration decreases with the distance that separates the state of origin from the state of destination (Lucas, 1997). Independent of the labor market conditions and the distance separating them from different destinations, individuals are more likely to migrate to places they know more about. Some factors that have been found to increase the knowledge of a particular destination are the presence of individuals who have migrated there before and infrastructure, trade and administrative links between the state of origin and the destination. While an extensive literature demonstrates that migration is cumulative and that the probability of migrating from one place to another increases with the number of individuals who have migrated before (Massey, 1994), the strength of this relationship diminishes with the increasing assimilation of the individual with in the general culture. To address this issue a spatial correlation measure will be used, based on the proportion of similar ethnicity conditioned by education level present within a radius of 5 miles for urban areas, 15 miles for rural. Numerous ways exist of defining the levels of urban hierarchy and the migratory flows between them. We choose to use the Rural-Urban Continuum Codes (RUCC) developed by the Economic Research Service of the USDA. We define RUCC codes 1-4 & 6 to be Urban, and the remaining 5,7,8,9 as Rural counties. This allows us to create a DiffR variable that encompasses the 4 possible origin-destination pairings (Rural-toRural, Rural-to-Urban, Urban-to-Rural, Urban-to-Urban). For the model fitted in this paper, the probability of out-migrating from any area (P(Mi)) is determined by its unemployment rate, the proportion of the labor force earning less than twice the minimum wage (Wi), Median Household Income (Yi), and Rental housing cost (RCi). Among those who migrate, the probability of going to a specific state (P(mij)) modeled by DiffR, is determined by the population size in the state of destination (Pj), the unemployment rate in j (URj), wages in j (Wj), Median Household Income in j (Yj), the Rental costs in j (RCj), distance and contiguity between i and j (Dij and Cij, respectively), and he proportion of individuals spatial correlated within the given radii (PM5j, PM15j). Data is taken from the IRS county-to-county migration data files for 2006-2008 for the base rates and conditioning ratios, and the 5-year American Community Survey (ACS) data for 2005-2009 for demographic, social and economic values. Results Table 1 shows the estimates of the nested logit model of the destination mode choice example. Our dependent variable is a destination choice and includes the 5 alternatives in DiffR plus the No-move reference case; No-Move, Rural-to-Rural, Rural-to-Urban, Urban-to-Rural, and Urban-to-Urban. The alternative-specific constant for an alternative has been included in the model to capture the average effect on utility of all factors that are not included in the model. (Train 2002). As can be seen the only two that are significant are the moves to rural areas, suggesting that the utility (or probability) of the individual for all alternatives relative to No-Move, increases. As we can see the wages, housing costs and employment are significant drivers of migration. Interestingly, it does not appear that the spatial correlation coefficient is significant, suggesting that these migrants are less interested in a core of self-similar individuals by ethnicity and more interested in raw economic opportunity. While the signs and significance of a coefficient in nested logit models have standard interpretation, the magnitude of the coefficients do not, thus Table 2 reports the attribute’s direct and cross elasticity effects between the choice theoretics. Table 1: Estimated Results of NLM Model Explanatory Variables Coefficient t-­‐stat Rural-­‐Rural_Const 0.784 4.299** Rural-­‐Urban_Const -­‐0.1884 -­‐1.716 Urban_Rural_Const 2.1071 14.218** Urban_Urban_Const -­‐0.11 -­‐0.492 Unemployment -­‐0.2125 -­‐4.698** Wage 0.2892 4.463** Household Income 0.1415 2.794** Rental Costs -­‐0.0289 -­‐2.67** Spatial_Corr 0.0384 0.663 Distance_Dummy -­‐1.0651 -­‐1.109 Contiguity_Dummy 0.0037 1.152 Inclusive Value Parameter Stay Home 0.3608 1.956** Move 0.7304 3.944** R-­‐Adj. 0.715 Log-­‐Likelihood -­‐1739.01 Table 2 somewhere here, along with relevant discussion. Initial results suggest that the magnitude higher elasticity values for a Rural choice reveal a strong attraction for these areas over more traditional urban environments. This attraction would seem to be overriding any attractive ability from the increased accessibility and familiarity provided by similar ethnicity enclaves. Conclusion We used a discrete choice random utility model to investigate individual choices among a discrete number of alternatives, taking into consideration the characteristics of each alternative by means of a nested logit model. Concentrating specifically on individuals of Hispanic ethnicity we examined their relative migration choice destination based on economic and spatial nearness and relative ruralness. We may infer from our results that personal factors and cultural connectedness were much less important than the socio-economic variables in accounting for the decision choice behaviors of outward migration. We argue that the simultaneous estimate of the Stay vs. Move decision and choice of location provided by the nested logit framework is confirmed as the most correct framework given the highly significant coefficients on the inclusive value components. A comparison of the branch components between the destinations Urban and Rural, reveal a magnitude higher elasticity values for the Rural areas over Urban. This suggests a greater attraction for such areas as a destination for Hispanic migration. A comparison of cross-elasticities shows that employment, housing cost and income tend to have stronger impact for rural destinations over urban ones. This implies that while racial and ethnic social ties may once have played an important part in determining the preferred destination local of Hispanic migrants, recent trends and changes have altered this influence. Economic opportunity and relative living costs now serve as the primary attractant for this group of individuals, much like they are for a majority of the overall population.