Exploring a migration explanation for the EKC

and MSAs are on the downward slope of the estimated EKC. After controlling for the effects of the other explanatory variables, Fig. 1 presents the fitted EKCs using the NPL sites data for both the County sample and the MSA sample. Estimated coefficients on other significant ex- planatory variables also have expected signs. Of note, the variable WHITE proportion of the population that is White is estimated with a statistically significant negative coefficient in the County sample. This is consistent with the evidence found in Berrens et al. 1997 for hazardous waste generation, and Wang et al. 1998 for the assessed risk of NPL sites. The last two columns of Table 2 report estimates from the total counts of all hazardous waste sites ALL-SITES. Since the total data have far fewer zeros than the NPL data, the models fit the cross-sectional data well. Specifically, the total site count data is denser it has few zeroes; 3 of the MSA sample and 15 of the County sample have no sites. The Maddala’s R 2 values are 0.56 for the County sample and 0.63 for the MSA sample, respectively. Again the signs and statistical signifi- cance of estimates on per capita income and its squared term PC-INC and PC-INC 2 , bear out the inverted-U shape of the EKC relationship. Also, the WHITE variable is again estimated with a significant negative coefficient in the County sam- ple. In the County sample, the EKC turning point occurs at a per capita income level of 17 670, which is 2.15 sample S.D. above the sample mean. In our sample, 3.02 or 95 counties lie on the downward part of the EKC. In the MSA sample, there is only weak evidence of the EKC from the ALL-SITES data, since both the linear and quadratic PC-INC terms are imprecisely measured. Regardless, estimates indicate an income turning point for the EKC of 20 300, which is 1.93 sample S.D. above the sample mean. In our sample, 4.68 or 35 MSAs lie on the downward part of the EKC. Again, given the high income turning point, only small percentages of US counties and MSAs are on the downward slope of the estimated EKC. 6 In summary, the count modeling evidence from all four samples County and MSA crossed with NPL-SITES and ALL-SITES indicates the pres- ence of the EKC relationship for hazardous waste sites, with similar per capita income turning points ranging from 17 670 to 20 300.

3. Exploring a migration explanation for the EKC

What is surprising about the existence of the EKC for US hazardous waste sites is that no formal theory advanced so far in the literature is able to explain it. Theories based on shifting of a negative externality e.g. moving the production of pollu- tion-intensive goods abroad do not apply to cur- rent hazardous waste sites, which are effectively active until cleaned. The pace of clean-up has been extremely slow. Despite 13 billion in spending through 1992, only 149 NPL sites had completed construction related to their clean-up remedies and only 40 of those sites have been fully cleaned up CBO, 1992. 7 Arguments based on abatement and incentives to invest in environmentally-friendly technologies Selden and Song, 1994, 1995 are also unsatisfactory for hazardous waste sites. Informal theories have been advanced that public coalitions can exert political influence concerning hazardous waste sites; this is supported by evidence on expansion decisions for treatment, storage and disposal TSD facilities Hamilton, 1993, which are relatively few in number. In contrast, there is no such evidence of collective political influence on speeding the pace of clean-ups Hird, 1990. 3 . 1 . Migration hypotheses This study advances and tests a line of argument in which internal migration plays a central explana- tory role behind the observed EKC for hazardous waste sites. The idea that economic mobility lies behind the EKC for hazardous waste sites is also motivated by the work of Mueser and Graves 1995, who examined internal migration 6 Although not presented here, the fitted curves using the data for ALL SITES produce inverted-U shaped EKCs similar to those shown for the NPL data Fig. 1. 7 While the pace has increased somewhat since the mid1990s, these statistics are relevant to our data. within the US over the three decades between 1950 and 1980. In a cross-sectional study of net migration into 520 county aggregates they found strong evidence that amenities are probably as strong a factor in location decisions as employ- ment opportunities. The Mueser and Graves model describes the dynamics of labor movements as the economy moves towards the steady state spatial equilibrium of the earlier static model of Roback 1982. Mueser and Graves’s model does not differenti- ate by labor quality or social groupings and their migration equations are estimated for the total amount of net migration separately for each decade. Given the significant negative effect of the WHITE variable in our estimated EKC re- sults, we estimate disaggregated net outmigration rate equations since sites are disamenities for two groups: i Whites and ii a minority group- ing composed of Blacks and Hispanics. A wide variety of factors can affect relative migration rates of different racial and ethnic groups. One possible factor is differences in education and human capital, and hence ex ante economic mo- bility. Another possible factor is the ability to get a conventional home loan mortgage. Two specific hypotheses are the focus of the analysis. The first specific hypothesis is: H1: The net outmigration of workers is an increasing function of hazardous waste sites be- yond a threshold level of per capita income. If H1 is valid empirically, then the build-up of hazardous waste sites is not a disamenity source of outmigration until some threshold level of per capita income is crossed. This hypothesis is suffi- cient to produce the inverted-U relationship be- tween income and the number of hazardous waste sites of the EKC. As a simple hypothetical exam- ple, assuming income increases over the life cycle of workers Mueser and Graves, 1995 and that skill levels are homogeneous across age groups, the proposition implies that beyond a threshold level of income younger workers would drive the left upward sloping part of the EKC and older workers would drive the right downward sloping. The second specific hypothesis is: H2: The threshold level of income at which net outmigration is influenced by the count of haz- ardous waste sites is the same as the threshold level of income at which the EKC for hazardous waste sites turns downwards. If H2 holds empirically, it provides a consis- tency check on the hypothesis that migration is a contributing factor to the observed EKC. If out- migration is influenced by the number of haz- ardous waste sites, then the out-flux of persons above a certain threshold of income to cleaner areas should not be statistically different from the observed EKC income turning point. To test H1 and H2, we turn to the estimation of a migration model. 3 . 2 . Migration model In the econometric estimation of the migration equations, net outmigration data across regions county or MSA is used for the half-decade 1985 – 1990 preceding the 1992 EKC data. 8 The basic linear model using cross-sectional data on location i either county or MSA is: NETOMIGR I = b + b 1 SITES i + b 2 PC-INC i · SITES i + Z i L + o i 2 where: NETOMIGR is the net outmigration rate over 1985 – 1990 [outmigration − immigra- tionpopulation] × 100; SITES is the sites vari- able, which is evaluated separately for ALL-SITES and NPL-SITES; and similarly in the interaction term, PC-INC i ·SITES i , with per capita income; Z is a vector of socio-economic variables; b ; b 1 ; b 2 and L are coefficients to be estimated; and o is an error term. We differentiate our find- ings for two groups by separately estimating mod- 8 Migration effects of hazardous waste sites are also possible within a county or MSA and will not be captured at our scale. There is an extensive hedonic property value literature examin- ing the effects of proximity to hazardous waste sites Farber, 1998. Table 3 Variable definitions and descriptive statistics for migration models a Variable County sample Definition MSA sample Mean S.D. Mean S.D. NETOMIGR-W NOUTMIGR, White only 0.44 9.36 − 0.80 2.02 − 0.16 3.86 − 2.69 6.50 NETOMIGR-BH NOUTMIGR, Black and Hispanic only 0.07 0.03 Unemployment rate, 1991 fraction of labor force 0.06 UNEMP 0.02 0.22 1.43 DENSITY 0.78 Population density, 1992 divided by 1000 2.85 0.10 0.09 Farm employment, 1990 million persons 0.03 FARMEMP 0.03 Serious crimes per million persons, 1989 divided by 1000 CRIME 2.81 2.28 4.55 2.77 1.64 5.02 PC-INC×ALL-SITES 5.11 PC-INC·ALL-SITES 9.36 PC-INC×NPL-SITES PC-INC·NPL-SITES 0.57 2.27 1.91 4.30 MNFEMP Manufacturing employment, 1990 million persons 0.17 0.10 0.17 0.08 0.84 0.17 Proportion of White racial background, 1990 0.82 WHITE 0.16 0.12 0.15 BLKHISP 0.14 Proportion of Black and Hispanic background, 1990 0.13 a The proportion of the US population that was White non-Hispanic origin in 1996 was 73.3. The proportion Black non-Hispanic origin was 12.1. The proportion of Hispanic origin regardless of current race was 10.5. For the county sample, the population-weighted averages of WHITE and BLKHISP are, respectively, 76.2 and 19.38. Here WHITE includes persons of Hispanic origin, while BLKHISP excludes Hispanics who are White. els for NETOMIGR-W and NETOMIGR-BH, which denote the net outmigration for Whites W, and Blacks and Hispanics BH, respectively. Definitions and descriptive statistics for the vari- ables used in the migration analysis are presented in Table 3. The test of H1 is based on the estimates for NETOMIG-WNPL-SITES and NETOMIG- WALL-SITES. In reference to the generic Eq. 2, we have, NETOMIG-WSITES = b 1 + b 2 · PC-INC, which is a function of per capita income. This response turns positive when PC- INC \ − b 1 b 2 . Thus, we can test H1 by esti- mating the threshold of per capita income where the response in net outmigration to hazardous waste sites i.e. NETOMIG-WNPL-SITES turns positive if at all. The second hypothesis H2 is based on a test of the difference between the level of per capita income at which the EKC turns downwards, with the level of per capita income at which the response NETOMIGSITES turns positive. Explanatory variables Z in the outmigration equations are motivated by the four broad factors underlying the Mueser and Graves 1995 specifi- cations. These four factors are: i amenities; ii employment-related factors; iii demographic fac- tors, and iv other regional factors. Amenities are measured by hazardous waste site counts ALL- SITES or NPL-SITES and rents RENT. The reason for including RENT as a measure of amenities is that we do not have data on the full variety of amenities such as climate, proximity to recreational locations, air quality, and other qual- ity of life variables. A pragmatic solution in this exploratory context is provided by Graves 1983, who found that rents serve as an excellent proxy for amenities. Hence RENT acts as the ‘com- posite amenity’ following Graves 1983. Employment-related factors are measured by manufacturing employment MFGEMP and farm employment FARMEMP, and the unem- ployment rate UNEMP. In the County sample, since a majority of counties are non-urban, agri- cultural employment, as captured by FARMEMP should be important. UNEMP proxies the de- mand for the product composition of an area; regions with declining industries should experi- ence an attrition through outmigration. Demo- graphic factors include the proportion of the population with at least a high school degree HSGRAD, and proportion of Whites WHITE in the region’s population, and proportion of Blacks and Hispanics BLKHISP. HSGRAD Table 4 White net outmigration rate model NETOMIGR-W a Scale NPL-SITES ALL-SITES Variable COUNTY MSA COUNTY MSA 10 − 4 – PC-INC·ALL-SITES – Turning point 2.146 3.243 4.395 4.168 – FIT PC-INC·NPL-SITES 10 − 4 1.127 2.261 1.343 2.903 – – – FIT 1 – – Amenity − 3.254 ALL-SITES − 6.909 −4.131 −3.176 NPL-SITES 1 − 1.855 − 2.274 – – −2.798 −2.204 10 − 3 − 114.893 − 84.413 − 85.534 − 97.072 RENT −8.707 −4.524 −7.922 −5.098 1 47.590 6.000 Employment 61.580 3.428 UNEMP 41.157 5.550 73.072 3.306 1 − 19.431 MNFEMP − 18.062 − 22.038 − 15.226 −2.632 −4.399 −2.564 −5.358 1 29.142 6.104 − 10.563 − 23.779 20.609 6.722 FARMEMP −0.780 −1.321 10 − 4 41.457 6.173 Demography 30.797 3.673 PC-INC –FIT 27.947 4.008 13.058 1.359 HSGRAD 1 − 22.706 − 35.247 − 19.844 − 10.694 −1.442 −1.609 −2.989 −3.591 1 − 11.731 − 15.265 − 17.396 − 9.409 WHITE −6.278 −7.093 −4.319 −5.619 Other 1 CRIME 0.423 4.410 0.416 3.608 0.580 4.981 0.813 4.337 1 − 0.378 DENSITY − 0.449 − 0.140 − 0.147 − −0.646 0.856 −2.400 −2.108 WEST 1 5.814 6.601 5.155 3.859 4.592 5.276 Regional dum- 4.350 2.609 mies 1 4.887 7.559 5.809 6.622 MIDWEST 4..079 6.4255 6.401 5.601 1 7.089 7.459 NORTHEAST 6.984 7.460 6.171 6.832 6.407 5.416 1 11.291 4.104 27.402 4.264 Constant 10.225 3.966 37.076 4.321 3141 748 3141 n 748 0.246 0.288 R 2 0.252 0.314 a Variables scaled to be uniform in size. To interpret coefficients in Table 3 units, scale the estimate by the scale factor. Fitted values of PC-INC·ALL-SITES –FIT, PC-INC·NPL-SITES–FIT, and PC-INC–FIT are generated by regressing these variables on all the right-hand side independent variables and their squared terms. Denotes significance at the 0.05 level. measures the upward economic mobility of an area, so that migrants are attracted to areas with high values of HSGRAD. These raceethnicity variables investigate the attractiveness of living in an area with a similar demographic make-up. Other regional factors include population density DENSITY and the crime rate CRIME, both of which are expected to lead to outmigration. Similar to the EKC estimation, dummy variables for census regions West, Midwest, Northeast, with South as the baseline are included. 3 . 3 . Model results Table 4 presents OLS estimates from the White net outmigration equation, after again correcting for the endogeneity in PC-INC term by the Kele- jian 1971 method. The adjusted R 2 values of 0.25 for the County sample and 0.29 for the MSA sample indicate a fair fit. 9 From the estimates presented in the first two columns of Table 4, the derivative of interest is NETOMIG-W NPL-SITES. For the County sample, this response equals − 1.855 + 1.127 × PC-INC × 10 − 4 . Therefore, NETOMIG-WNPL- SITES \ 0 only if PC-INC \ 1.8551.127 × 10 − 4 = 16 460. The inference is that, in the County sample, net outmigration by Whites in- creases in response to the presence of NPL sites after a threshold level of per capita income of 16 460. At lower levels of income we infer that, after controlling for other socio-economic, re- gional, and amenity factors, NPL sites are not important in location decisions. In the MSA sam- ple, NETOMIG-WNPL-SITES \ 0 after per capita income reaches 16 932. Hence, for the NPL sites, hypothesis H1, that sites emerge as a factor in migration decisions after a certain level of income, is confirmed. This is a necessary condi- tion if migration is a contributing factor for the observed EKC. For total sites, H1 is also confirmed. Estimates from the last two columns of Table 4 show that for ALL-SITES the responses are similar. Sites begin to matter beyond a per capita income of 15 163 in the County sample, and 15 720 in the MSA sample. Table 5 presents estimates from the net outmi- gration rates of Blacks and Hispanics taken to- gether. The model fits poorly for the County sample, and the estimates on the coefficients of interest are measured quite imprecisely. This is to be expected because much of the migration in this group is inter-urban migration Nord, 1998. Hence we focus on the estimates from the MSA sample where the equation has a better fit. In the MSA sample, the coefficients of interest are esti- mated precisely, and afford the inference needed for tests of H1. Hazardous waste sites are seen to be a factor in the Black and Hispanic net outmi- gration rate with NETOMIG- BHNPL-SITES turning positive for PC-INC \ 17 537, and NETOMIG-BHALL-SITES turning positive for PC-INC \ 15 656. Hence net outmigration is at levels of income not significantly different from that for Whites. 10 Table 6 collects previous results about income turning points for the EKC, and the threshold level of income at which NETOMIGSITES turns positive for all groups, together with their standard errors. This information is used in test- ing H2 that the income turning points are equal to the level of income at which the build-up of sites begins to affect outmigration. The last three columns of Table 6 report pairwise t-tests. Con- sider the column labeled ‘1 vs. 2’, which tests the equality of the EKC income turning point with the income threshold for White net outmi- gration. The results support H2: the evidence indicates no statistical difference. Notably, the income turning points and the income threshold for White net outmigration are both precisely estimated, and failure to find a difference is not due to high standard errors, but rather that the estimates are statistically similar. Other t-tests from Table 6 also support H2, using the income thresholds from White net outmigration data or from Black and Hispanic net outmigration data although the difference of means test for ALL- SITES data for the MSA sample may be influ- enced by the imprecise measurement of the income turning point.

4. Discussion and conclusions