This week, I’m featuring a guest post from Christopher Timmins, Associate Professor of Economics at Duke University.
—
Prompted by a number of headline cases of hazardous waste contamination in the late 1970s, the US Congress established the Superfund program under the Comprehensive Environmental Response, Compensation, and Liability Act (CERCLA) of 1980. Under that program, the Environmental Protection Agency has placed some of the most contaminated waste sites in the country on the National Priorities List, where federal funds are directed towards cleanup when those funds cannot be recovered from a “responsible party”. Since that time, the EPA has identified more than 47,000 hazardous waste sites that potentially require some sort of cleanup action. By 2007, more than 1,500 sites had been placed on the NPL.
Superfund site cleanups are not cheap. Even before the budget crisis in Washington, there were many calls for an evaluation of the costs and the benefits of the program – the former requiring a careful accounting of all the resources that go into a cleanup, and the latter necessitating a non-market valuation exercise. Superfund cleanups do not trade in the marketplace, but the houses surrounding a remediated site do – this provides one of the most simple and straightforward ways in which to look for estimates of the benefits of a Superfund cleanup – property value hedonics. Economic theory tells us that homebuyers’ marginal willingnesses to pay for the remediation of a site can be deduced from what they are willing to additionally pay for a house that is near a remediated site versus a house that is near a site that has not been cleaned up.[1] This seems like it should be a simple empirical problem. Take two houses – one in a neighborhood surrounding a site that has been cleaned up and one in a neighborhood surrounding a site that has not. How do their prices differ? The problem here (and in most hedonic analysis) is that these two neighborhoods may be different in ways that are not observed by (and hence, not controlled for) by the researcher. These unobservables can confound the effect we are trying to measure.[2]
Empirical techniques have been developed to try to deal with these confounding unobservables. In the parlance of applied microeconomists, researchers have looked for “quasi-experimental” data variation. True “experimental” variation would require the researcher to randomize over which sites get cleaned up – something that the real world isn’t quite ready for yet. Quasi-experimental variation is variation in which sites are cleaned that mimics experimental variation in one way or another. One of the cleanest forms of variation used to date in the literature was suggested by Michael Greenstone and and Justin Gallagher.[3] They demonstrated that a twist in how the first wave of sites proposed to the NPL in 1982 were funded led to two batches of sites that were very similar in terms of how the EPA evaluated the hazard they posed, but only one of which received funding. Comparing similar neighborhoods across these two groups allows the researcher to safely assume that unobservables are not the driving force in any differences in housing prices. Rather, those differences must arise from the effect of the treatment being administered by EPA.
The Greenstone and Gallagher analysis, while relying on a clean source of quasi-experimental variation in data, suffered from another flaw. In particular, it relied on publicly available data describing housing prices at the level of the Census tract. Census tracts are areas containing several thousand individuals and can have a radius of several kilometers. In their hedonic analysis, Greenstone and Gallagher used the median housing price in each census tract as their housing price measure. While this may seem like an innocuous choice, it can have big impacts. In particular, the externalities associated with many Superfund sites are highly localized – unless contamination works its way into groundwater or leaks into the air, it may not affect people outside a narrow geographical range. Still it could have a big impact on the people it does affect. Given the size of a census tract, it is therefore possible that a tract median price may not be affected at all by the cleanup of a Superfund site, while other parts of the house price distribution could be dramatically affected. Indeed, Greenstone and Gallagher found no effect of Superfund cleanup on median house prices. This result has had big effects on Superfund program policy decisions over the last few years.
In our work,[4] we address this problem of localized externalities on several levels. First, using publicly available census housing price data, we take an extra step to construct other parts of the house price distribution (e.g., the 10th, 20th, etc… percentiles of the distribution of house prices in each census tract) and look to see how these vary with site cleanup. Where effects at or above the median become small or disappear altogether, we find big effects of cleanup (e.g., 18.2%) at the bottom end of the of the house price distribution. Within a census tract, it is the cheaper houses that are benefitting from the Superfund program.
Next, we verify this result using restricted-access census block data. Census blocks are much smaller than tracts (think about a typical city block). Data on census blocks can only be used on-site in a Census Bureau Research Data Center, and only after going through a long series of security checks. These data confirmed once again the presence of big cleanup effects on nearby houses (e.g., 19.4% on houses within one kilometer).
Based on these results, we feel confident that Superfund cleanups do, in fact, provide benefits to nearby homeowners. Whether those benefits are big enough to exceed cleanup costs is another question that we are currently studying. Passing a simple cost-benefit test will be made more difficult by the fact that benefits are localized, so a small set of houses will see price increases from cleanup. Whether the benefits to those houses are sufficient to justify the costs is an empirical question. Note that this sort of analysis ignores benefits that might arise from improved labor market conditions if a site cleanup leads to economic revitalization of a neighborhood. It may also ignore health benefits if people aren’t aware of them, and hence don’t build them into their home-buying decisions.
[1] Note to reader – this is a delicately worded sentence. Current research in non-market valuation theory (hedonic theory in particular) is aimed at rectifying shortcomings in the simple hedonic analysis that prevent the researcher from recovering preferences from compensating differentials in housing prices. For example, when a change in a neighborhood is non-marginal (i.e., big relative to the size of the housing market), simple welfare measures are invalid. Moreover, we may expect people to move around (i.e., re-optimize) in response to the change. Simple hedonic techniques are not able to handle either of these possibilities.
[2] Note that research on this question has seldom even bothered trying to compare neighborhoods containing Superfund sites (cleaned or uncleaned) with neighborhoods that contain no sites at all – the thought being that the determinants of housing prices that we can’t see are likely to be very different across these two groups – we should at least restrict those unobservables to be the sort we would find in the vicinity of a site of some type.
[3] Greenstone, Michael, and Justin Gallagher, “Does Hazardous Waste Matter? Evidence from the Housing Market and the Superfund Program,” The Quarterly Journal of Economics, 123 (2008), 951-1003.
[4] Gamper-Rabindran and Timmins (2011). “Does Cleanup of Hazardous Waste Sites Raise Housing Values? Evidence of Spatially Localized Benefits.” http://econ.duke.edu/~timmins/Gamper_Rabindran_Timmins.pdf