The Effect of Ethnic Residential Segregation on Wages of Migrant Workers in Australia

Transcription

1 The Effect of Ethnic Residential Segregation on Wages of Migrant Workers in Australia Mathias G. Sinning Australian National University and IZA Bonn Matthias Vorell RWI Essen March 2009 PRELIMINARY DO NOT QUOTE Abstract. This paper investigates the effects of ethnic residential segregation on wages of migrant workers in Australia. We estimate a control function for neighborhood unobservables and account for the non-random sorting of individuals into neighborhoods by instrumenting for neighborhood attributes. The empirical analysis uses data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey in combination with regional information from Census data. Our findings suggest that ethnic residential segregation has a significantly negative effect on wages of skilled migrant workers, while the effect on wages of unskilled migrant workers is insignificant. This result is robust across several empirical approaches. JEL-Classification: F22, R21, R23 Keywords: International migration, segregation, regional labor markets, neighborhood effects This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (MIAESR). The findings and views reported in this paper, however, are those of the author and should not be attributed to either FaHCSIA or the MIAESR. We gratefully acknowledge the support of the Australian Group of Eight (Go8) and the German Academic Exchange Service (DAAD). All correspondence to Mathias Sinning, Social Policy Evaluation, Analysis and Research Centre (SPEAR), Research School of Social Sciences (RSSS), Australian National University, Fellows Road, Coombs Building (Building 9), Canberra ACT 0200, Australia, mathias.sinning@anu.edu.au, Tel: , Fax:

2 1 Introduction A better understanding of the factors that are responsible for a successful integration of current and future immigration cohorts is a prerequisite for effective immigration and integration policies. One factor that has been studied extensively over the last years is ethnic residential segregation (i.e. the clustering of foreign-born populations into ethnic enclaves ), which can help explain why different groups of immigrants are more or less well integrated. Existing studies of ethnic or racial residential segregation have produced rather mixed results. Cutler and Glaeser (1997), for example, find negative associations between segregation and outcomes for young African- Americans. Edin et al. (2003) provide quasi-experimental evidence on refugees in Sweden and find that living in enclaves improves labor market outcomes for less skilled refugees. In this paper, we investigate the effect of ethnic residential segregation on wages of skilled and unskilled migrants. While empirical evidence on the effects of ethnic residential segregation is available for the US and some European countries, one cannot simply assume that findings for the US and Europe are applicable in other immigration countries. This paper addresses this issue by analyzing the effect of ethnic residential segregation on wages of migrant workers in Australia. We take advantage of the opportunity to combine household-based data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey with regional information from the 2001 and 2006 Censuses of the Australian Bureau of Statistics. 1

3 When analyzing the effects of ethnic residential segregation, the sorting of individuals into neighborhoods has to be taken into account. To address this identification problem, existing studies have usually focused on subgroups of foreign-born populations, such as refugees or young immigrants. This paper provides evidence on a representative sample of foreign-born workers without imposing such restrictions. In our empirical analysis, we employ a hedonic house price model to estimate a control function for neighborhood unobservables. This approach reduces the complex correlation structure induced by non-random sorting to a selection problem, which can be addressed by instrumenting for neighborhood attributes. Australia is one of the traditional immigration countries and immigration has been a cornerstone of its economic, social and cultural development. Historically, Australia favored immigration from Europe, leaving little scope for immigration from other continents. The immigration policy moved away from selecting immigrants on the basis of national origin in 1973 (Antecol et al., 2003). In recent decades, Australia has placed a high weight on accepting economic migrants using numerical testing to judge the admissibility of skilled immigrants (Birrell, 1990). In response to an increased demand for high-skilled labor, Australia has adopted initiatives to expand the opportunities for skilled workers to immigrate. Like Canada and New Zealand, Australia has increased the number of visas for permanent migrants selected under the points system leading the number of skill-based immigrants to Australia to triple between 1995 and 2005 (Australian Government Department of Immigration and Citizenship, Research and Statistics Section (AGDIC), 2006). 2

4 These immigration policies have shaped the foreign-born population in Australia in terms of their skills and settlement intentions. Although empirical evidence on the earnings assimilation of immigrants to Australia suggests that the earnings disadvantage disappears over the settlement process (Chiswick and Miller, 1985; Miller and Neo, 2003), less is known about the earnings disadvantage within skill and age groups. Specifically, average income levels of native and immigrant households in Australia are about the same, although immigrants are on average older and better educated than natives. 1 Given this relative educational and demographic advantage, one would expect that average immigrants earn more than average natives. However, such an earnings advantage is not observed, suggesting that immigrants seem to be less integrated into the Australian labor market than is often recognized. Given the existing evidence for other immigration countries, it seems likely that ethnic residential segregation has an effect on wages of immigrant workers in Australia, although the direction of this effect is quite unclear (Borjas, 2000). On one hand, geographic clustering may reduce labor market discrimination and therefore improve employment prospects of immigrant workers. On the other hand, ethnic enclaves may also reduce incentives to acquire country-specific human capital that may be essential for better-paying jobs in the larger national labor market. As a 1 Weekly gross wages of both native- and foreign-born workers are on average about $800, although 49% of the foreign-born and only 39% of the native-born workers have received higher education and average foreign-born workers are three years older than average native-born workers (see Table 1). 3

5 consequence, it seems reasonable to suspect that ethnic residential segregation is less beneficial for skilled than for unskilled workers. Against this background, we investigate whether ethnic residential segregation affects skilled and unskilled migrant workers differently. The results of our empirical analysis reveal a significantly negative effect of ethnic residential segregation on wages of migrant workers in Australia. Differentiating between skilled and unskilled migrants suggests that geographic clustering has a negative effect on wages of skilled migrants especially those who received secondary or vocational education or training. At the same time, the effect of ethnic residential segregation on wages of unskilled migrants is insignificant. These results are robust across several empirical models. They are also in line with theoretical considerations of costs and benefits of geographic clustering. However, the findings reject the hypothesis of positive segregation effects on wages of unskilled migrants, suggesting that estimates for a representative sample of the migrant workforce population may be different from the results obtained for selective subgroups. The paper proceeds as follows. Section 2 provides a discussion of the empirical strategy and describes the data used for the empirical analysis. Section 3 presents the estimation results. Section 4 concludes. 4

6 2 Empirical Strategy and Data 2.1 Empirical Strategy The identification of neighborhood effects is empirically demanding, because observed neighborhood characteristics may be correlated with unobserved individual and neighborhood characteristics. Several empirical strategies have been applied in recent studies. Edin et al. (2003) investigate the effect of living in an enclave on labor market outcomes of refugees, who were distributed by government authorities across locations. Although the experimental design of their study permits an identification of treatment effects, a generalization of the results is rather limited. Since refugees represent a highly selected group of migrants, the causal effect of ethnic residential segregation on their labor market outcomes may be different from that of labor migrants. An alternative approach addresses the complex correlation structure by aggregating to a higher level of geography (Cutler and Glaeser, 1997; Ross, 1998; Weinberg, 2000; Ananat, 2007; Cutler et al., 2000). Although aggregation eliminates the problem of correlation in unobservables among neighbors, Bayer and Ross (2006) note that the identified effects do not only include average neighborhood effects, but also broader metropolitan area effects. As a consequence, the resulting estimates may only be interpreted as neighborhood effects if one assumes that the level of segregation in a metropolitan area does not affect individual outcomes directly. To address the correlation of neighborhood characteristics with both individual 5

7 and neighborhood unobservables, we employ a two-part estimation strategy proposed by Bayer and Ross (2006). Specifically, we use a hedonic house price model to estimate a control function for neighborhood unobservables. The intuition behind this approach is that house prices indicate the attractiveness of the neighborhood for unobserved reasons. Consequently, a function of house prices may be used as a proxy for unobserved neighborhood attributes. To illustrate the empirical approach, we assume that a worker s wage rate depends on his or her own characteristics, the characteristics of the neighborhood and an unobserved error term. Specifically, we consider the following model: ln y ij = X iα + Z jβ + ε ij, i = 1,..., N, j = 1,..., J, (1) where y ij denotes the wage rate of individual i in neighborhood j, X i is a vector of individual characteristics and Z j is a vector of neighborhood attributes. The parameter vectors α and β capture individual and neighborhood effects, respectively. We can decompose the error term in equation (1) into three components: ε ij = u i + v j + w ij, where u i represents unobserved individual characteristics, v j includes unobserved neighborhood characteristics and w ij is an error term with the usual properties, i.e. E(w ij ) = 0 and E(w ij Z i, X j ) = 0. Any non-randomness in the sorting of individuals across neighborhoods produces a correlation of unobserved individual and neighborhood characteristics with observed variables in equation (1), which will bias OLS estimates of α and β. Following Bayer and Ross (2006), we eliminate 6

8 the bias caused by unobserved neighborhood characteristics by estimating a control function for v j as the average residual for each neighborhood, using a hedonic house price model. Specifically, house prices, p ij, may be described by ln p ij = H iγ + Z jδ + λ j + η ij, (2) where H i represents a vector of housing unit attributes and λ j is a control function for the unobservable part of the neighborhood quality. Given the estimate of the control function, λ j, we can rewrite equation (1): ln y ij = X iα + Z jβ + θ λ j + ε ij, i = 1,..., N, j = 1,..., J, (3) where ε ij = u i + w ij. Estimating equation (3) reduces the complex correlation structure to a selection problem attributable to the correlation between (observed and unobserved) neighborhood characteristics and unobserved individual characteristics. To address this correlation, we use a lag variable of ethnic residential segregation as an instrument. 2.2 Data We utilize the opportunity to combine household-based panel data with economic and social information of the neighborhood. Specifically, data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey will be combined with regional information from the 2001 and 2006 Censuses at a statistical local area level (Australian Bureau of Statistics, 2006). 2 2 Statistical local areas have about the same size as post code areas. They represent the smallest geographical region available in both data sets. 7

9 The HILDA Survey is a broad social and economic longitudinal survey which began in 2001, with particular attention paid to economic and subjective well-being, labour market dynamics and family dynamics. The panel includes about 20,000 individuals in about 8,000 households. Interviews are conducted annually with all adult members of each household. As the HILDA Survey has a longitudinal design, most questions are repeated each year. In addition, specific questionnaire modules are included each wave, focusing on questions that will not be covered every year (such as family background and personal history, household wealth, retirement and plans for retirement, etc.). The In-Confidence Release of the HILDA survey includes information about post code and statistical local areas (SLAs) of the household s residence. 3 Since the Australian Census of Population and Housing contains information about the whole population of Australia, it is our main resource for reliable regional characteristics on the statistical local area and statistical subdivision (SSD) level. The Basic Community Profiles of the Censuses include a variety of neighborhood characteristics, such as age distributions, migrants (including country of birth and year of arrival), children and child care, dwelling structure, earnings, education, employment, family formation and dissolution, home ownership, etc. 3 The data used in this paper were extracted using the Add-On package PanelWhiz v2.0 (Nov 2007) for Stata. PanelWhiz was written by Dr. John P. Haisken-DeNew (john@panelwhiz.eu). The PanelWhiz generated DO file to retrieve the HILDA data used here and any Panelwhiz Plugins are available upon request. Any data or computational errors in this paper are our own. Haisken-DeNew and Hahn (2006) describe PanelWhiz in detail. 8

10 In the empirical analysis, a cross-section of native- and foreign-born workers observed in 2006 is employed. We restrict our sample to persons between 18 and 65 years residing in statistical subdivisions of major urban areas. We use the logarithm of weekly gross wages as dependent variable in our wage regression. We further consider a set of individual socioeconomic and demographic characteristics in our empirical analysis. Specifically, the vector X i of equation (3) includes a quadratic function of age and indicator variables for migrant workers, gender, part-time employment, levels of educational attainment and metropolitan areas. After excluding all observations with missing values on one of the variables used in the analysis, our sample contains 3,018 native- and 943 foreign-born workers. Table 1 includes descriptive statistics of all variables used in the empirical analysis. There are many ways to measure the different dimensions of segregation (Massey and Denton, 1988; Echenique and Fryer, 2007; Simpson, 2007). The Australian Census provides information about the size of the foreign- and native-born population within statistical local areas. We use this information to define the most straightforward measure of segregation, i.e. the share of migrants in the neighborhood. The vector Z j of equation (3) includes the measure of ethnic residential segregation and an interaction term between a variable indicating foreign-born workers and the segregation measure. Given the empirical findings provided by Cutler and Glaeser (1997), we expect that the effect of segregation on wages of native workers is not significant. Since the Australian Census data allow us to create a measure of ethnic residen- 9

11 tial segregation for 2001 and 2006, we are able to use a lag variable of segregation as an instrument to address the possible correlation between segregation and unobserved individual characteristics. Moreover, we estimate a hedonic house price model for the year 2006, using housing attributes, metropolitan area characteristics and segregation as explanatory variables to derive a control function for neighborhood unobservables. To account for correlation between this control function and unobserved individual characteristics, we estimate a second house price model for 2005 and employ the resulting control function as an instrument. The estimates of the house price models are available from the authors upon request. 3 Estimation Results Table 2 includes the estimates of linear regression models (with and without statistical subdivision fixed effects) and instrument variable models of equation (3). In all cases, the estimates provide evidence for an inverted U-shaped wage-age relationship. After controlling for other observable characteristics, wages of female workers are about 17% lower than wages of male workers. Moreover, gross wages of parttime employed workers are about 90% less than those of full-time employed workers. The returns to education are significantly positive, with the reference group being Year 11 and below. As expected, the wage differential between native- and foreign-born workers is insignificant. The neighborhood unobservables are positively associated with the 10

12 wage rate and highly significant. While the share of migrants in the neighborhood does not affect wages of native-born workers, the effect of segregation on wages of foreign-born workers is significantly negative. Differences between the linear regression models and the instrumental variable approach are very small, suggesting that the correlation between segregation and unobservable individual characteristics can be neglected. At the same time, the F-statistics of the first-stage regression of the instrumental variable model suggest that our instruments are strong predictors of the neighborhood attributes. Table 3 include the estimates of regression models that were fully interacted with the indicator variable of migration status. Since differences in the returns to observed characteristics between native and migrant workers are small, the estimates in Table 3 do not differ substantially from those in Table 2. The values of the R- squared measure also suggest that including interaction terms between migration status and observed characteristics does not change our results. Since estimating a fully interacted model does not seem to be necessary, we are able to estimate a modified version of the model presented in Table 2, where we split up the interaction term between the migrant indicator and the share of migrants into different levels of education. The estimates obtained from this exercise are presented in Table 4. The numbers point to substantial variation in the effect of ethnic residential segregation on wages of migrants with respect to educational attainment. Specifically, wages of low-skilled migrants seem to be not affected by segregation, while wages of migrants who received secondary or vocational education 11

13 or training are significantly negative. The estimates further suggest that wages of migrants with university degree are negatively affected by residential segregation as well, although these effects are only significant at a 10%-level. In sum, these findings reveal a significantly negative effect of ethnic residential segregation on wages of migrant workers. At the same time, the numbers point to substantial variation in the effect of segregation across education levels. 4 Conclusions This paper investigates the effect of ethnic residential segregation on wages of skilled and unskilled migrants. While empirical evidence on the effects of ethnic residential segregation is available for the US and some European countries, one cannot simply assume that findings for these countries are applicable in other immigration countries. This paper addresses this issue by analyzing the effect of ethnic residential segregation on wages of migrant workers in Australia. Existing studies have usually focused on subgroups of foreign-born populations such as refugees or young immigrants to address identification problems induced by the sorting of individuals into neighborhoods. This paper provides evidence on a representative sample of foreign-born workers without imposing such restrictions. In our empirical analysis, we estimate a control function for unobserved neighborhood characteristics, using a hedonic house price model. This approach reduces the complex correlation structure induced by non-random sorting to a selection problem, which we address by 12

14 instrumenting for neighborhood attributes. The empirical findings reveal a significantly negative effect of ethnic residential segregation on wages of migrant workers in Australia. Differentiating between skilled and unskilled migrants suggests that geographic clustering has a negative effect on wages of skilled migrants especially those who received secondary or vocational education or training. At the same time, the effect of ethnic residential segregation on wages of unskilled migrants is insignificant. The results are robust across several empirical approaches. They are also in line with theoretical considerations of costs and benefits of geographic clustering. However, the findings reject the hypothesis of positive segregation effects on wages of unskilled migrants, suggesting that estimates for a representative sample of the migrant workforce population may be different from the results obtained for selective subgroups. 13

15 Tables Table 1: Descriptive statistics Natives Migrants Mean SD Mean SD Socioeconomic characteristics: Weekly gross wage Age Female Part-time employed Educational attainment: Graduate certificate or higher Diploma or bachelor Certificate I-IV Year Year 11 and below Metropolitan area: Sydney Melbourne Brisbane Adelaide Perth Canberra Rest of Australia Share of migrants: Share of migrants (2006) Share of migrants (2001) Control function: λ (2006) λ (2001) N 3, Note. Weighted numbers based on weights provided by HILDA.

16 Table 2: Results from model without interaction terms (1) (2) (3) OLS OLS FE IV Age 0.078*** 0.077*** 0.078*** (0.006) (0.006) (0.006) Age squared *** *** *** (<0.001) (<0.001) (<0.001) Female *** *** *** (0.022) (0.021) (0.022) Part-time employed *** *** *** (0.027) (0.027) (0.028) Graduate certificate or higher 0.472*** 0.467*** 0.472*** (0.040) (0.040) (0.040) Bachelor or diploma 0.313*** 0.312*** 0.313*** (0.033) (0.033) (0.033) Certificate I-IV 0.114*** 0.111*** 0.114*** (0.034) (0.033) (0.035) Year *** 0.114*** 0.119*** (0.036) (0.036) (0.036) Migrant (0.067) (0.066) (0.068) λ 0.145*** 0.226*** 0.147*** (0.035) (0.045) (0.041) Share of migrants (0.147) (0.212) (0.148) Migrant share of migrant ** ** ** (0.194) (0.193) (0.198) Constant 4.945*** 5.007*** 4.951*** (0.122) (0.133) (0.122) R-squared F-statistics of first-stage regressions: Share of migrants Migrant share of migrants λ Note. Number of observations: 3,961. Weighted numbers based on weights provided by HILDA. Standard errors, which are reported in parentheses, were adjusted to take clustering within statistical local areas into account. (1) and (3) include metropolitan area fixed effects; (2) includes statistical subdivision (SSD) fixed effects. p<.1; ** p<.05; *** p<.01.

17 Table 3: Results from model with interaction terms (1) (2) (3) OLS OLS FE IV Age 0.079*** 0.078*** 0.079*** (0.007) (0.007) (0.007) Age squared *** *** *** (0.000) (0.000) (0.000) Female *** *** *** (0.026) (0.024) (0.025) Part-time employed *** *** *** (0.031) (0.031) (0.032) Graduate certificate or higher 0.466*** 0.461*** 0.465*** (0.046) (0.046) (0.046) Bachelor or diploma 0.319*** 0.317*** 0.317*** (0.039) (0.038) (0.039) Certificate I-IV 0.151*** 0.146*** 0.151*** (0.038) (0.036) (0.038) Year *** 0.147*** 0.154*** (0.038) (0.037) (0.038) Migrant (0.308) (0.312) (0.309) λ 0.148*** 0.228*** 0.157*** (0.039) (0.046) (0.048) Migrant λ (0.063) (0.063) (0.077) Share of migrants (0.150) (0.210) (0.151) Migrant share of migrants *** ** ** (0.195) (0.196) (0.199) Constant 4.892*** 4.960*** 4.894*** (0.129) (0.142) (0.128) R-squared F-statistics of first-stage regressions: Share of migrants Migrant share of migrants λ Migrant λ Note. See Note to Table 2.

18 Table 4: Results from Interactions with Educational Attainment OLS OLS FE IV Share of migrants Migrant share of migrants Graduate certificate or higher * * Bachelor or diploma * * Certificate I-IV *** *** Year *** *** Year 11 and below λ 0.144*** 0.227*** 0.147*** Constant 4.906*** 4.976*** 4.912*** Control variables Yes Yes Yes Interacted model No No No Note. Estimates based on model specification (2) of Tables 2 4. See Note to Table 2.

19 References Ananat, E. O. (2007, August). The wrong side(s) of the tracks estimating the causal effects of racial segregation on city outcomes. NBER Working Papers 13343, National Bureau of Economic Research, Inc. Antecol, H., D. A. Cobb-Clark, and S. J. Trejo (2003). Immigration Policy and the Skills of Immigrants to Australia, Canada, and the United States. 38 (1), Australian Bureau of Statistics (2006) census of population and housing - data dictionary. Reference guide, Australian Bureau of Statistics. Australian Government Department of Immigration and Citizenship, Research and Statistics Section (AGDIC) (2006). Settler Arrivals Technical report, Canberra. Australian Government Printing Service. Bayer, P. and S. L. Ross (2006, April). Identifying individual and group effects in the presence of sorting: A neighborhood effects application. Working papers , University of Connecticut, Department of Economics. Birrell, R. (1990). The Chains that Bind: Family Reunion Migration to Australia in the 1980s. Canberra: Australian National Printing Service. Borjas, G. J. (2000). Ethnic enclaves and assimilation. Swedish Economic Policy Review (7), Chiswick, B. and P. W. Miller (1985). Immigration Generation and Income in Australia. Economic Record 61 (73), Cutler, D. M. and E. L. Glaeser (1997). Are ghettos good or bad? The Quarterly Journal of Economics 112 (3), Cutler, D. M., E. L. Glaeser, and J. L. Vigdor (2000). When are Ghettos Bad? Lessons from Immigrant Segregation in the United States. Journal of Urban Economics 63, Echenique, F. and R. G. Fryer (2007). A Measure of Segregation based on Social Interactions. Quarterly Journal of Economics 122, Edin, P.-A., P. Fredriksson, and O. Aslund (2003). Ethnic enclaves and the economic success of immigrants evidence from a natural experiment. The Quarterly Journal of Economics 118 (1),

20 Haisken-DeNew, J. P. and M. Hahn (2006). PanelWhiz: A Menu-Driven Stata/SE Interface for Accessing Panel Data. mimeo, Massey, D. S. and N. A. Denton (1988). The Dimensions of Residential Segregation. Social Forces 67, Miller, P. and L. Neo (2003). Labour Market Flexibility and Immigrant Adjustment. Economic Record 79 (246), Ross, S. L. (1998). Racial Differences in Residential and Job Mobility. Journal of Urban Economics 43, Simpson, L. (2007). Ghettos of the Mind: The Empirical Behaviour of Indices of Segregation and Diversity. Journal Of The Royal Statistical Society Series A 170 (2), Weinberg, B. (2000). Black Residential Centralization and the Spatial Mismatch Hypothesis. Journal of Urban Economics 48,