Is the Water Safe? Health-Based Drinking Water Violations in New Jersey

Background

Under the federal Safe Drinking Water Act, public water systems must meet health-based standards for contaminants and treatment. When they don’t, it counts as a health-based violation, for example when bacteria are found, lead levels exceed the “action level,” or a chemical goes above its legal limit. A national study by the Natural Resources Defense Council and partners found that water systems that stayed out of compliance the longest were more likely to serve communities with larger shares of people of color and residents with limited English (NRDC, 2019).

Earlier posts looked at specific contaminants such as PFAS and lead. This post takes a broader view: across all health-based standards, are violations more common in water systems serving New Jersey’s overburdened communities?

Data

EPA publishes every drinking water violation reported by states through its ECHO database, including the date each violation was first reported (U.S. EPA, n.d.-a). That field lets me avoid hindsight. I kept only health-based violations whose non-compliance began between April 2021 and March 2026 (five years) and that had been reported by May 7, 2026.

I focused on New Jersey’s 555 active community water systems, the ones that serve people where they live, and calculated the share of each system’s service-area residents living in an overburdened community (OBC).

Findings

What the violations are

There were 966 health-based violations at 551 systems in this period, across all system types. Most involved two rules:

• Revised Total Coliform Rule (439): bacterial contamination or failure to fix sanitary defects.
• Lead and Copper Rule (334): failures related to lead control and treatment.

Smaller numbers involved nitrate, disinfection byproducts (TTHM), and naturally occurring radioactive elements.

Figure 2. Coliform and lead and copper rule violations account for most health-based drinking water violations.

Systems serving overburdened communities had more violations

Share of service-area residents in an OBC	Systems	% with ≥ 1 health-based violation	Mean violations per system
Under 25%	359	18%	0.40
25–75%	74	19%	0.49
Over 75%	114	32%	0.76

Figure 1. Water systems serving mostly overburdened communities are more likely to have a health-based violation.

Testing it with a count model

Violation counts are small whole numbers, mostly zeros, so I modeled them with count regression. A first Poisson model showed overdispersion: the counts varied about 3.7 times more than the model assumes (variance 1.9 vs. mean 0.5). When that happens, Poisson standard errors are too small, so I also fit a negative binomial model, which allows extra variation. It fit much better (AIC 919 vs. 1,262).

Predictor (negative binomial)	Rate ratio (95% CI)	p-value
+10 points OBC share	1.07 (1.01–1.14)	0.028
System size (10× population)	0.58 (0.39–0.87)	0.009
Surface-water source	1.77 (0.95–3.32)	0.075
Privately owned	0.46 (0.22–0.94)	0.034

Holding system size, source, and ownership constant, each 10-point increase in OBC share is associated with about 7% more health-based violations. Going from a service area with no OBC residents to one that is entirely OBC corresponds to roughly twice the expected violations (1.07¹⁰ ≈ 2.0). Larger systems have fewer violations per system, consistent with economies of scale in staff, testing, and treatment.

Why this matters

Many violations are fixed quickly. A single coliform detection may lead to flushing and retesting within days. But the pattern matters: if systems serving overburdened communities violate standards more often, residents there are more likely to receive boil-water notices, to drink water that briefly failed a standard, and to lose trust in their tap water. Small and older systems often lack money and trained staff. Programs that help systems consolidate, hire operators, and replace aging infrastructure, especially in overburdened communities, could address the root causes.

Limitations

• Counts, not severity or duration. One brief coliform detection counts the same as a long lead exceedance.
• Reporting depends on monitoring. Systems that monitor more may find more violations, and missed monitoring is a separate (non-health-based) category.
• Service-area demographics are approximate. For small systems whose service areas contain no block-group center, I used the block group at the service area’s own point.
• Association, not cause. OBC share may stand in for system age, finances, or other unmeasured factors.

Code project (Level 4): count models and overdispersion

Goal: Test whether health-based violations are more frequent in systems serving overburdened communities, using an appropriate model for count data.

Data: EPA ECHO SDWA bulk download (violations, water system details, reference codes), filtered to New Jersey; NJDEP water service areas; NJDEP EJ data (Mar 31, 2026 version).

Method:

1. Stream the 4 GB national violations file and keep New Jersey rows.
2. Keep health-based violations with non-compliance beginning April 2021 – March 2026 and first reported by the cutoff date (the anti-hindsight filter).
3. Count violations per active community water system and attach OBC share, size, source, and ownership.
4. Fit Poisson and negative binomial models, check dispersion, and compare AIC.

pois = smf.glm(f, data=d, family=sm.families.Poisson()).fit()
print(pois.pearson_chi2 / pois.df_resid)          # ≈ 3.7 → overdispersed
nb = smf.negativebinomial(f, data=d).fit()
print(np.exp(nb.params))                          # rate ratios

New skills: handling very large files, diagnosing overdispersion, and choosing between count models.