Benefit levels and US immigrants welfare receipts

Transcription

1 1 Benefit levels and US immigrants welfare receipts by Joakim Ruist Department of Economics University of Gothenburg Box Gothenburg, Sweden telephone: Keywords: immigration, welfare benefits, welfare magnets hypothesis JEL codes: H24, H31

2 2 Benefit levels and US immigrants welfare receipts Abstract The literature on whether welfare benefits importantly attract international migrants with low labor market attachment yields mixed results. Focusing the empirical analysis not on settlement choices, but on migrants probabilities of actually receiving welfare, has previously resulted in influential evidence in favor of the hypothesis that low skilled migrants are indeed significantly attracted by higher welfare benefits. While previous work has included two time periods, 1980 and 1990, I find that when adding a third period, 1970, no positive correlation is found in any sample or empirical specification, between state welfare benefit levels and native or immigrant households probabilities of receiving welfare, hence no support for the welfare magnets hypothesis. The results indicate that policy makers may have little reason to keep welfare benefits low for fears of attracting low skilled immigrants. 1 Introduction Do generous welfare systems function as magnets attracting immigrants with low abilities to earn their living in the labor market? This is often assumed in public debate in immigration countries, and the subject of a, not very large, academic literature. In this literature, different studies have reached different conclusions on whether US immigrant settlement decisions are importantly determined by state differences in welfare benefit levels, by including different covariates in regression analyses of settlement decisions (Buckley, 1996; Zavodny, 1997; Dodson, 2001). The most cited study on the subject, Immigration and welfare magnets by Borjas (1999), investigates the issue in more detail, by analyzing not the settlement decisions of all immigrants, but the probabilities that immigrants in different states actually receive welfare benefits. The study is quite unique in incorporating differences in immigrants welfare receipts in the empirical analysis. Borjas theoretical model does not predict that US natives migrate to certain states because of their higher welfare benefit levels, since the differences in welfare benefits do not outweigh the costs of migration. For immigrants, on the other hand, the costs of migration are assumed to be equal regardless of in what state they settle, so if they have high probability of being eligible for welfare grants, they will settle in states where these grants are higher. Borjas tests the predictions of the model by investigating correlations between benefit levels in 1980 and 1990 in the program Aid to Families with Dependent Children (AFDC) and the probability

3 3 that a household, native or immigrant, receives a welfare grant. He finds weak support for a more positive correlation between these variables for immigrant-headed than native-headed households, when restricting the sample to female-headed households with children. When including male-headed households, there is no positive correlation for either group. Borjas notes that the empirical indications of excessive clustering of welfare receiving immigrant households are entirely driven by a strongly increasing share between 1980 and 1990 of welfare receiving immigrant households living in California, and an unanswered question is then whether this is actually due to the increasing AFDC benefit levels in California, or to something else that is not included in the empirical specification. The present study aims to provide an answer to this question by adding a third time period, 1970, to the sample, partly motivated by the fact that AFDC benefits in California, relative to the US median state, rose even more between 1970 and 1980 than between 1980 and 1990, while in the former period the share of welfare receiving US immigrant households living in California fell, not rose. In the sample covering , there is no positive correlation between AFDC benefit levels and probabilities of receiving welfare for any group of households, native or immigrant. This result holds across a range of choices of control variables, and sample restrictions. 2 Empirical strategy The program Aid to Families with Dependent Children (AFDC) provided financial assistance, between 1935 and 1996, to low-income households with children. While created at the federal level, benefit levels were set at the state level, resulting in large variation in levels across states. In 1990, Alaska, which offered the highest benefits, offered a maximum level more than seven times as high as that of Alabama, which offered the lowest, to a three-person family. This is the source of variation used by Borjas to investigate correlations between state benefit levels and probabilities of native and immigrant households receiving welfare. The hypothesis that the correlation between benefit levels and probabilities of receiving welfare is stronger for immigrants than for natives follows from Borjas theoretical model, where potentially higher welfare benefits have a positive effect on utility, but there are migration costs. For natives, migration costs outweigh the benefits of moving to a state with higher welfare benefits, while for immigrants, the cost of migration is already sunk, and therefore does not constrain the maximization of utility with respect to welfare benefits.

4 4 The identifying equation is P ijt =X ijt β+δw jt +αy jt +s j +d t +ε ijt (1) where P is the probability that any individual in household i in state j in period t receives a welfare grant, X is a vector of socioeconomic characteristics of the household and the household head, w is the maximum AFDC benefit level for a family of three persons, y is a vector of logged state mean disposable income per capita and the unemployment rate, s and d are state and time fixed effects. The vector X includes the numbers of persons in total, persons under 18, and persons above 65 in the household, the household head s age, gender, years of education, and dummies for being black, Hispanic, or Asian, and for immigrants also dummies for year of immigration in five-year intervals. With the state fixed effects s j, the equation identifies how variation in probabilities of receiving welfare within each state depends on variation in benefit levels within the same state over time. The coefficient of interest is δ, and Borjas finds it to be positive and significant for all female headed households, with higher point estimates for immigrant households than for native households, but the differences are not significant. Borjas notes that the results are strongly driven by California, where both the maximum benefit level and the share of welfare receiving immigrant households increased substantially between 1980 and 1990, as seen in Table 1. While it is not possible to verify that this correlation is causal, it is not necessarily a problem that the results are driven by the state with the highest benefits in 1990 (apart from Alaska, to where migration costs are arguably higher also for immigrants), as according to the model there is no reason for an immigrant to settle in the state with the second highest benefits. Hence the main contribution of the present study is to provide more robust results by adding a third time period and thus decreasing the dependence of the results on particular Californian circumstances between 1980 and Figure 1 confirms that this strategy adds important additional variation in the explanatory variable, by plotting the correlation between the changes in the first and second decades covered by the study. The correlation shown in the graph is positive and marginally significant. The observations with the largest positive changes are California in the first decade, and Alaska, followed by California, in the second. Insert Table 1 here Insert Figure 1 here

5 5 Individual and household level data is obtained from the Integrated Public Use Microdata Series (ipums, Ruggles et al, 2010) of the 1970, 1980 and 1990 censuses. The 1980 and 1990 samples contain 5% of the immigrant population in each year, while the 1970 sample contains 1%. Weights are used in the analysis to adjust for this. Data on benefit levels is from the US House of Representatives (1996). Data on income and unemployment levels is from the US Bureau of the Census (various issues). Data on average disposable household income per state in 1970 was unavailable, and was thus replaced by data on average personal income. 3 - Results Estimates of the parameter of interest, i.e. the correlation between AFDC benefit levels and probabilities of receiving welfare benefits, are shown in Table 2. Standard errors, which are clustered at the state level, are shown in parentheses, and numbers of observations, in thousands, are shown in brackets. Results obtained using the main specification, including all observations and no additional controls, are shown in column (1). As there may be differences in behavior of migrants from higher and lower income countries, column (2) limits the immigrant samples to immigrants from Latin America, Asia (except Japan) and Africa. Column (3) includes controls for ethnic concentration across states. These measures are set to zero for countries of origin with less than 1,000 observations in the total sample (all three sample years together), where ethnic concentration is deemed irrelevant. For the 72 origins 1 with at least 1,000 observations, two measures of ethnic concentration are included, one being the fraction of all immigrants from country c in period t that lives in state j, the other being the fraction of all inhabitants in state j in period t that is made up of immigrants from country j. As differences in welfare benefits may be of more interest for people with less education, columns (4) and (5) limit the sample to households where the head has not obtained a high school degree, column (4) without and column (5) with the ethnic concentration controls. Insert Table 2 here Table 2 reports twenty point estimates for the coefficient of interest referring to immigrant groups, and eighteen out of these are negative, whereas positive signs were predicted by the model of Borjas. The two positive point estimates are for recently immigrated (i.e. last 5 1 The United Kingdom is merged into one country of origin. The Azores and Portugal are kept separate, and have >1,000 observations each.

6 6 years) female headed households, in columns (4) and (5), but the standard errors for these two estimates are several times higher than the point estimates. Hence, with three time periods, there is no support for the prediction that higher welfare benefits attract higher numbers of welfare eligible immigrants. The estimated coefficients on the controls for ethnic concentration show that immigrants living in states with higher concentration of their own ethnicity are less likely to receive welfare benefits. When both measures are included simultaneously, as in the specifications behind the results in columns (3) and (5), the coefficient on the share of one s own ethnicity living in the same state is sometimes positive, but when only one of the two measures is included, it is almost always negative and significant at the 1% level. Figure 1 identifies some possibly important outliers in the explanatory variable, being Alaska in the top of the plot, California in the upper right corner, and Texas furthest to the left. Excluding the far outlier Alaska does not affect the results, as few people live there. Yet California and Texas are populous states, and excluding them changes the sign on some of the estimates in the main specification, but they all still have high p values (at least 0.18). In the original results reported by Borjas, the estimates on the coefficient of interest were positive and significant for all female-headed household groups, natives and immigrants alike, while the point estimates were larger, but not significantly larger, for immigrants. In one instance, not all of the difference between the results reported in Table 2, and those of Borjas, can be attributed to the additional sample period, as I have not managed to exactly replicate the original results for recently immigrated female headed households. The original twoperiod results, and my replications, are shown in Table 3. The point estimates are close for all groups except for recently immigrated female headed households with children, although the 95% confidence intervals of the two estimates do intersect. Insert Table 3 here The reason for the failure to replicate the point estimate for this group has not been found. After communication on the issue with the original author, I have used the documentation behind the original results that was still available, to make the data treatments as similar as possible. Yet not all of the original documentation was still available, and the differences shown in Table 3 thus remain. It is not surprising that the difference is largest for the group of recently immigrated female headed households with children, as this group is quite small, and

7 7 in particular as there are many states with very few observations on such households that actually receive benefits. In 1980, the median state has 22 such households, while in 1990 it has 26. Results obtained in this sample are thus highly sensitive to variation in data treatment. Due to the sensitivity of the results for recently immigrated female headed households, it is probably more appropriate to focus the reading of both the two-period and the three-period results on those referring to the group consisting all female-headed immigrant households with children, which is seven times as large, and has an equally significant positive estimate in Borjas original results. For this group, the additional sample period fully explains the difference from Borjas results with two periods, as my replication of that estimate in Table 3 even has a slightly larger positive point estimate than that reported by Borjas, while the estimate in Table 2 is negative, and significantly different from that in Table 3. Also for the group of native female-headed households with children, my point estimate with two periods is slightly larger than Borjas, and hence the difference between the latter and my three-period estimate, which is positive but not significant, is fully explained by the addition of the third period. 4 - Conclusions In the well-cited paper Immigration and welfare magnets, Borjas concludes that higher state benefit levels in the program Aid to Families with Dependent Children (AFDC) have attracted immigrant households, at least those headed by females, with high probability of receiving welfare. Yet he acknowledges that the empirical evidence relied on to support the conclusion is relatively weak, and that the results are strongly driven by California. In the present paper, I provide a more robust analysis, by extending the data set to cover three US censuses instead of two, including 1970 in addition to 1980 and The results show no indication of a positive correlation between benefit levels and probabilities of receiving welfare, neither for natives nor for immigrants, across a range of different empirical specifications. In the behavioral model outlined by Borjas, state differences in welfare benefits need not affect native location patterns, as the costs of relocation may easily outweigh the increase in expected income. Still, one could expect a positive correlation also for natives, if higher benefits attracted more workers already present in the state into the system. My results do not support that prediction, which may indicate that the AFDC system has been functioning well enough to deter such behavior.

8 8 For immigrants, the results imply a clearer refutation of the predictions of Borjas model, that migrants with high probabilities of receiving welfare would be attracted to settle in states with higher benefits. The crucial assumption for this prediction is that for international migrants, migration costs do not depend on in which state they settle in the US. This simplifying assumption finds little support in the data, as is perhaps most easily seen from a glance at the geographical distribution of immigrants from Latin America and the Caribbean, which make up a fourth of the total immigrant stock in 1980, and a third in In both years, the concentration of these immigrants is highest in the four states that share a border with Mexico, i.e. those that are closest to Latin America, plus Florida, which Caribbean migrants most easily reach by sea. While one of these five states, California, also has one of the highest AFDC benefit levels in both census years, the other four have below median values in both years. This simple observation indicates that migration costs are not independent of state of settlement even for international migrants, and may hence explain why the results reported in this paper are not in line with those predicted by Borjas model. Acknowledgements I am grateful to Arne Bigsten and Lennart Flood for their useful comments on earlier drafts of this paper, and to George Borjas for his assistance in my replications of his earlier results.

9 9 References Borjas, George, 1999, Immigration and welfare magnets, Journal of Labor Economics, 17, Buckley, F. H., 1996, The political economy of immigration policies, International Review of Law and Economics, 16, Dodson, Marvin, 2001, Welfare generosity and location choices among new United States immigrants, International Review of Law and Economics, 21, Steven Ruggles, Trent Alexander, Katie Genadek, Ronald Goeken, Matthew Schroeder, and Matthew Sobek, Integrated Public Use Microdata Series, Version 5.0 [Machinereadable database], Minneapolis, MN: Minnesota Population Center US Bureau of the Census, (various issues), Statistical Abstract of the United States, Washington DC: US Government Printing Office US House of Representatives, 1996, Background material and data on programs within the jurisdiction of the committee on ways and means (Green Book), Washington DC: US Government Printing Office Zavodny, Madeleine, 1997, Welfare and the location choices of new immigrants, Economics Review, Federal reserve bank of Dallas, 2-10

10 10 Table 1. AFDC maximum levels and welfare receiving households in California over time Year Maximum AFDC Benefit level in California relative to US median state Share of welfare receiving immigrant households living in California (%) Share of welfare receiving female headed immigrant households with children living in California (%)

11 11 Table 2. Results in full sample, (1) (2) Main Only Latin America, Asia, Africa native (0.019) [82] all immigrant (0.045) [57] recent imm (0.111) [8.1] native (0.011) [1,079) all immigrant (0.015) [732] recent imm (0.023) [92] 0.108* (0.053) [39] (0.137) [6.5] 0.051* (0.014) [363] (0.032) [65] (3) Control for ethnic concentration Female head with children (0.038) [57] (0.122) [8.1] All households 0.026* (0.011) [732] (0.030) [92] (4) No high school degree (0.024) [28] (0.067) [26] (0.165) [4.0] (0.009) [356] (0.020) [314] (0.034) [31] (5) (3) & (4) combined (0.053) [26] (0.178) [4.0] 0.023* (0.010) [314] (0.050) [31] Notes: Table shows the estimated partial effects of logged AFDC benefit levels on probabilities of receiving welfare. Standard errors clustered at the state level are shown in parentheses, and numbers of observations, in thousands, are shown in brackets. A (*) denotes significance at the 5% level.

12 12 Table 3. Borjas results for , and my replications (1) Borjas (2) Replication Female head with children native 0.088* (0.031) [67] 0.099* (0.047) [66] all immigrant 0.112* (0.054) [53] recent imm * (0.103) [7.8] native (0.009) [796] all immigrant (0.015) [655] recent imm (0.026) [87] All households 0.126* (0.060) [55] (0.126) [7.8] (0.013) [787] (0.015) [686] (0.030) [87] Notes: Table shows the estimated partial effects of logged AFDC benefit levels on probabilities of receiving welfare. Standard errors clustered at the state level are shown in parentheses, and numbers of observations, in thousands, are shown in brackets. A (*) denotes significance at the 5% level.

13 Figure 1. Correlation between decadal changes in AFDC maximum benefit levels ($)