Dominion Energy & Environmental Racism: a case study in how to lie with maps
Yes, the title is provocative, but its not entirely mine. I simply and liberally borrow from the classic Mark Monmonier primer entitled How to Lie with Maps. But the reality of this ‘case study’ is indeed provocative, and it amounts to nothing less than outright environmental racism under the direction of ‘one of the nation’s largest producers and transporters of energy’, Dominion Energy.
In the following maps, charts and discussion, I detail the mechanisms behind Dominion Energy’s use of a ‘product’ from ESRI based in California; the ‘skills’ of one international environmental firm; ‘validated’ by an academic institution in Virginia and delivered to decision makers that may unwittingly (or not) participate in the statistical erasure of a local, historic, minority-majority community in Buckingham County, Virginia.
The (very) Abbreviated Backstory:
In 2013–14, The Atlantic Coast Pipeline (ACP), at 42 inches and 600 miles, was proposed to cross West Virginia, Virginia and North Carolina with a massive 54,000 hp compressor station slated for ‘area known as Union Hill ’. Shortly thereafter, Dr. Lakshmi Fjord conducted a door-to-door survey of residents in close proximity to the proposed compressor, finding the community to be 83% minoirty-majority, many of whom are descendants of slaves.
Mapping # 1: the (very) Abbreviated Disaggregated Mapping Process:
The results of the survey were mapped, a confidentiality agreement was signed by the cartographer (me) so that the attribute data per survey point would not be shared to the public and potentially expose individual survey participant data. The summary results per .5, .5–1, 1–2 and 2–3 mile proximity distances are shown in the the survey map and table below, demonstrating that the immediate geography is minority-majority in racial composition. This mapping process is generally known as a ‘disaggregated’ process; that is, while the results in the table are summarized for ease of use, the data behind the summary is collected at the actual survey location — the resident address. This is by far the most ‘precise’ data gathering approach for large scale geographies involving human populations:
Mapping # 2: the (not so much) Abbreviated Aggregated Mapping Process conducted by the The Federal Energy Regulatory Commission, Dominion Energy and The Virginia Department of Environmental Quality:
As part of any large infrastructure project, both federal and state, there is a regulatory process that has to include, by law, consideration of Environmental Justice (EJ) issues, communities and importantly local populations. The issue at hand is that certain populations are much more vulnerable and likely to experience an adverse environmental impact for a variety of reasons, two of which are critical threshold criteria: first, race and second, income. More precisely, % minority and % poverty of a local population.
At 600 miles, the ACP is certainly a large project. And the responsibility at the federal level (FERC) is to make sure the applicant (Dominion) is not adversely impacting EJ local populations. In reality what Dominion gave to FERC, and then FERC regurgitated — which, by the way, is pro forma at FERC and a huge problem on its own — is a simple statistical chart in black and white and highlighted in light grey census block groups (CBG) that meet at least one EJ threshold. And lo and behold, there’s the intersecting tracts in Buckingham county, ground zero for the compressor station, highlighted in light grey. Given this inconvenient data point, FERC and Dominion did what every respectable agency and company would do — ignore it and hope no one sees it.
What FERC and Dominion should have done is immediately paused, developed a community based data collection process designed to produce accurate, disaggregated data and resulting summary statistics about the local population at issue. In fact that is exactly what did happen in Mapping #1. The only problem is that FERC and Dominion and further Virginia’s Department of Environmental Quality (DEQ), chose to ignore it.
Unfortunately for FERC, Dominion and the DEQ, people started asking questions. And what happened next is the heart of the lie…
Given mounting pressure from both the local Union Hill community and across the state of Virginia, Dominion and DEQ both started to produce subsequent mappings designed to assuage the public that indeed what they think they know about Union Hill is not really what is going on in Union Hill. An ‘alternative facts’ operation, one might say. In the case of DEQ, they adopted the EJSCREEN tool and presented those details in public without discussion of the explicit warnings that EPA issues when utilizing an aggregated product like EJSCREEN. Regardless, it didn’t seem to go very well:
Dominion, on the other hand, was more sophisticated… maybe calculated is a better term. Their ploy involves 4 components:
- Produce seemingly impressive ‘demographic and income profile’ via an ESRI business product to demonstrate that Union Hill is the exact opposite of what people think it is. No, Union Hill is very white, in fact its at least 70% white:
- Use a less meaningful, purposefully misleading, variable that seems like its an appropriate measure of poverty, but alone, its really not. Hope that no one notices that its not a very good measure of poverty. Disregard the fact that you told FERC already that the geography in question crosses poverty thresholds criteria and indeed is an EJ eligible geography:
- Start showering the local community with grand promises of great things to come and loads upon loads of $$$:
- Get a local academic institution to state they ‘validated’ the… (actually its not quite clear what exactly was validated — the input census data, the analysis, its fit with local population, the temporal extrapolation… no one knows):
If you want to learn about all the promised perks that a 54,000 hp might bring to your local community, you can review that further HERE.
However, its the actual mapping and data behind the operation that is critical to uncovering the lie.
First, the product. ESRI (actual name: Environmental Systems Research Institute) is an industry standard mapping platform. Privately held, proprietary, a long, storied history with a charismatic leader that just gave away a huge piece of California the Nature Conservancy. I use ESRI products day in, day out as some/many/most GIS analysts do also. By no means is ESRI a bad company, a bad software or producer of bad products. However, a product used for incorrect purposes ends up a bad product. And that is what has happened here.
Dominion contracted with the environmental firm Environmental Resource Management (ERM) to do the larger ACP mapping as well as the secondary run at EJ analysis for the Buckingham Compressor SNAFU. Important to keep in mind — this analysis was not part of the primary analysis submitted to FERC; and in fact contradicts that primary analysis that found EJ eligibility at the intersection with the proposed compressor site. In the following excerpt, Dominion claims more recently, based on the results of the ESRI run, that EJ eligibility is no longer an issue because the median income is sufficiently high as to not warrant consideration:
But median income is not the EJ criteria that is typically used for EJ analysis. It was not used in the FERC analysis to get the certificate in order to proceed with the project, but somehow its now a criteria to permit and build the project. Regardless of this subterfuge, % Poverty is indeed triggered by a significant margin, and it remains the correct variable to use for EJ analysis. It is described well via the US Forest Service statement on Environmental Justice; and can been seen in the project area overlay for Census Tract Groups utilizing ACS 2013–27 data, mapped by author:
Low-income status is determined by comparing annual income to a set of dollar values called poverty thresholds that differ by family size, number of children, and age of householder. If a family’s before-tax monetary income is less than the dollar value of their threshold, then that family and every individual in it are considered to be living in poverty. For people not living in families, poverty status is determined by comparing the individual’s income to his or her poverty threshold.
Even as this obfuscation fails, its only amounts to the ‘little lie’. The BIG lie, the one that has Union Hill, Virginia and watchers outside the state, outraged — and rightly so — is the attempt at statistical erasure of race. This is particularly outrageous given that the Union Hill community is largely African American and Native American- African American, with a profound connection to an unique history rooted deep in this local geography. The Union Hill community was demonstrated and validated via the disaggregated community study to be diametrically different than the Dominion’s following analysis results.
To follow, Dominion’s utilization of an ESRI Demographic and Income Profile report to summarize local population as 70%+ White accompanied by a sample of the ESRI report for population counts at .5 mile proximity:
Since the ‘data’ (actually its not the data that’s at issue, its the analysis method, but everyone confuses the two to bad ends) carries the ESRI imprimatur, ‘validated’ by the L. Douglas Wilder School of Government and Public Affairs at Virginia Commonwealth, the natural conclusion is that ESRI is not only right but unquestionably right. In fact, ESRI is not wrong, not at all. Technically what is wrong is the use of the this particular tool to utilize aggregated census data through a process of apportionment to assign population quantities (population counts) and categorical qualities (race) to a particular large scale local population (Union Hill).
To understand this we need to unpack it step by step:
- To start, we need to recognize that the data that is being utilized by EJSCREEN, by ESRI, by regulators, by Dominion, by myself is census data… period. While ESRI does indeed utilize other proprietary data sources for a myriad of variables, the ones at issue here — population counts and race — come directly from the census. And importantly there are several vintages being utilized. ESRI uses 2010 decennial data as the baseline for the profiles; FERC used 2014 ACS census data; I have been using the latest vintage 2013–2017 ACS census data. But regardless of the vintage, the trends are always going to be similar because its essentially the same dataset measurements, different vintages.
- Second, census data is always an aggregated product, meaning it is a collection of statistics about a population across geography that typifies that population, not individuals per se. In other words, across distances of census block groups, tracts, counties, states, ect. the census data is organized in such a way as to mask individuals in order to get to a summary of all the individuals in the particular census geography. In the end, census data results in approximations of the population; it doesn’t claim locational precision at the household or individual level. And importantly, in more rural locations such as Buckingham County generally, it spans across relatively large geographies. This can, and does indeed, cause issues when typifying large scale local geographies because the summation at the aggregated census unit stands in for the local dynamics of a large scale local geography.
- Third, in using a census input for a custom point — in this case the proposed compressor — a unique geographical unit must be created — a .5 mile, 1 mile and 2 mile radius. This unique geography then has to be overlaid to the census geography and the data from the census has to be ‘extracted’ proportionately into the unique geography — .5 mile, 1 mile and 2 miles in this case. Its this ‘extraction’ process — known technically in GIS as apportioning or apportionment that is at play. This is usually acceptable in dense urban areas where the census geographies are ‘tight’ but in more rural locations the aggregation can really produce disastrous results, as has happened here. And this is the BIG lie. That Dominion hasn’t acknowledged this and rather utilizes the implied authority of ESRI and a local university to ‘validate’ inappropriate analysis — that is bad and its literally traumatizing people. This is not an exaggeration; people in Union Hill are very upset and deeply offended by the results of the ESRI profile, in particular.
To understand how bad this is, we can recreate the process utilized by the ESRI demographic profile and compare it to the disaggregated method. The Dominion analysis was likely performed using an ESRI extension entitled Business Analyst. Its important to pause and realize this is NOT a planning tool for siting point-source toxic shale gas infrastructure. The repercussions of a bad analysis run on a business, while bad, are going to be resolved because a business has the resources to resolve them. A bad analysis run on a local community in a existential face-off against with one of the most powerful extraction companies in America — that is a whole other universe of badness.
So, to recreate, the following assumptions are made:
- The input data is from the US Census, ACS data 2013–17. It is paired with the appropriate geographies at the census block group level for intersecting tracts at the emission site. The exact datatable is B03002.
- A quick sidenote: the analysis point that was utilized is incorrect. I’ve utilized the incorrect location (approx. 650 feet southeast in the middle of the road) as the analysis point for consistency.
- The analysis geometries are recreated at .5, 1 and 2 miles.
- Geographical apportionment is conducted using the census data as the attribute input data. In the process, the correct proportions of the analysis geometries relative to the whole of the census geometries are determined, and then by that same factor the attribute data is proportioned to the analysis geometries. In effect, the aggregated census data is being ‘weighted’ — just like teachers weight grades — but here the ‘weight’ is the area of the analysis geometry as it overlays to the various census geometries.
- The attribute data is then collected into a new table and mapped and labeled on top of each analysis geometry.
The purpose of recreating this process is simply to show that there is no ‘magic’; there’s no special ‘authority’ or ‘formula’ that ESRI possesses and Dominion delivers from on high.
The following insinuation by Dominion’s spokesperson that they somehow ‘own’ the data and that its ‘better’ that other data is incorrect. Yes, census data is good (and importantly Dominion doesn’t ‘own’ it); but its really bad when used incorrectly to typify a large scale, local population.
Dominion contends that its own data is “the best available . . . because it is unbiased,” company spokesman Karl Neddenian said via email.
Importantly, the GIS geoprocessing steps which Dominion adopted, either knowingly or not — in their contract with ERM using the ESRI demographic profile product — can be partially recreated and then compared to the disaggregated local survey result for the core variables that are important, namely population count and race. And through that comparison, the ‘ESRI approach’ is the not only a bad approach in this particular case, but the worst possible approach because it ends up statistically erasing actual populations and their identities in this large scale, local geography.
The results of the first .5 mile analysis run show the results misrepresent this particular local geography. As was shown in the ESRI profile, the population count was 10, here it is 12. The total % White in the ESRI profile was 77.8%, while here its is 75%. While the numbers are indeed slightly different, these are two different vintages of census data, and also slightly different collection processes from decennial census to the ACS 5-year estimate. But the larger data trend remains the same across the % White variable from the larger census block groups into the smaller unit, misrepresenting the high concentration of minority-majority population that actually lives in close proximity to this proposed compressor.
This trend continues at the 1 mile proximity where in the ESRI profile the population count was 95, here it is 54. The total % White in the ESRI profile was 70.7%, while here its is 64%:
At the 2 mile proximity where in the ESRI profile the population count was 303, here it is 239. The total % White in the ESRI profile was 71.5%, while here its is 61%:
On January 8th, 2018 The Virginia DEQ air board voted to permit this compressor station. Undoubtedly this is a terrible political and environmental decision not only for Virginia but the country. Inevitably the case will go to court. When it does, the mapping and the data will be on the side of those who would bear the burden of a 54,000 HP shale gas compressor designed to run 40+ years, 24–7, emitting thousands of tons of both damaging Methane (CH4) globally, and Hazardous Air Pollutants (HAPS) locally over its life cycle.
Based on the data and the preceding recreation of the ESRI analysis method employed by Dominion to utterly misrepresent and ignore disaggregated data collection methods — seemingly in coordination with regulators who’ve given such a spirited defense of the faulty methods — its clear that Dominion is involved in grave malfeasance that is nothing short of textbook Environmental Racism:
Environmental racism refers to any environmental policy, practice or directive that differentially affects or disadvantages (whether intended or unintended) individuals, groups or communities based on race or colour.
Big company, big dollars, small community: Dominion deal sparks dissent in community facing gas project by Gregory S. Schneider, Washington Post, 12/9/2018. Accessed 01/06/2019.
BREDL: DEQ attempting to erase history of Union Hill, Augusta Free Press, 01/06/2019. Accessed 01/06/2019.
Dismantling Environmental Racism in the USA, Robert Bullard, 1999. Accessed 01/07/2019.
Stephen Metts is a GIS analyst and instructor based in New York State. His research interests covered in this article include participatory methods of GIS; shale gas development and infrastructure; climate justice and environmental justice.