Cleaning EPA's Dirty Sewers Data
Wheeler, Larry, Smith, Grant, Investigative Reporters and Editors, Inc. The IRE Journal
You're reading a featured story from Uplink, IRE and NICAR's online publication devoted to computer-assisted reporting - available at http://data.nicar.org/uplink.
The Environmental Protection Agency estimates that the nation's aging and overburdened sanitary sewer systems overflow at least 23,000 times a year and discharge between 3 billion and 10 billion gallons of raw sewage into streams, rivers and lakes.
The problem is even bigger - 850 billion gallons bigger - when older sewer systems that discharge both storm water and sewage from the same pipe are included in the tally.
But there is no national database to track when, where and how much.
The best the EPA has to offer is its Enforcement and Compliance History Online database (ECHO; www.epa-echo.gov/echoj.The searchable database contains thousands of records depicting enforcement actions, including fines taken against hundreds of publicly owned sewage treatment plants. Many of those enforcement actions were related to illegal or chronic sewage discharges. The online ECHO database worked well for searching out a few facilities in one region, but it didn't answer the question journalists (and readers) typically want to know: Which facility got fined the most?
To begin answering that question, we downloaded the raw enforcement data from Integrated Data for Enforcement Analysis (IDEA; www.epa.gov/compliance/data/systems/multimedia/idea/ index.html), another EPA database.
IDEA is a sprawling collection of three datasets never designed to talk to each other. To construct a complete picture of sewer enforcement and compliance across the nation, we needed to turn the three IDEA datasets into one all-powerful, all-knowing database.
There was nothing in the EPA data dictionary that warned us this would be a mind-bending task, even with substantial help from our dear friend Microsoft Access database manager.
Using variables strewn across 1 8 tables and two EPA datasets, we first defined our universe as all publicly owned sewage treatment plants with a daily treatment capacity of at least I million gallons.
We came up with about 4,250 municipal sewage plants in 50 states and the District of Columbia.
Next, we threw a five-year time bracket around the data to make our results comparable to ECHO.
Eventually, we created a single database that included the name and location of each facility, plus federal, state and local penalties assessed against each facility since 2003. It was neatly organized with one record for each facility and used an EPA national permit number as a unique ID. This allowed us to sort, rank, and look for trends and extremes.
That's when we noticed Ridiculous Data Problem No. 1: State or local fines and federal fines were recorded in separate fields. In some cases, the two fields could be combined to get an accurate total. Other records recorded the same fine in multiple fields, and combining them would have resulted in inaccurate double counting. Ultimately, we devised a set of criteria to sum the penalties for each facility and compared our results to the ECHO database results to ensure accuracy.
We still had a third IDEA dataset to contend with, which became the root of most of our difficulties. …