Equity as Afterthought: Lead Service Line Replacement in Pittsburgh,PA

Regulating Pb

In 1991, the U.S. Environmental Protection Agency issued the Lead and Copper Rule (LCCR), ²⁴ which regulates levels of Pb and copper in drinking water. Since its introduction, the rule has been revised and strengthened several times. As of this writing, the most current version—the Lead and Copper Rule Improvements (LCRI) released in October 2024—mandates that all drinking water systems across the U.S. must identify and replace lead service lines by 2037. ²⁵ In 2021, after years of public outcry and media focus on the issue of lead-contaminated water in Flint, MI and other cities, the Biden Administration announced its Lead Pipe and Paint Action Plan (LPPAP), a federal plan allocating billions of dollars to the removal and replacement of lead services lines across the United States. ²⁶ As the Biden administration elevated environmental justice across federal policy, equity and environmental justice became defining elements of the discourse on LSLR programs. Such framing marked a significant shift that positioned LSL removal not only as a technical and public health imperative, but as a central plank of the Biden administration’s (2020–2024) broader environmental justice agenda.

LSLR and equity

What constitutes equity in the context of lead service line replacement programs? According to the LSLR Collaborative, a national coalition of 28 organizations representing public health, water utilities, environmental groups, labor, consumer advocacy, housing, and various levels of government, “an equitable program will recognize that not everyone has the same societal and economic advantages, and provide support, not equally across the population, but rather as appropriate according to an individual’s circumstance.” ²⁷ The LSLR Collaborative defines an equitable program as one that recognizes historic inequities, addresses income disparities, prioritizes tenants, and engages communities genuinely. ²⁸

Although existing scholarship has focused heavily on the unequal distribution of lead exposure and its effects, relatively little attention has been paid to the question of equity within the context of LSLR. One exception is the work of Baehler et al. (2021), ²⁹ which evaluates an LSLR program in Washington, D.C., and demonstrates that lower-income households and households of color were systematically less likely to obtain full service line replacements (the safer intervention) and more likely to receive unsafe partial replacements. In this study, we turn our focus to the sequencing and prioritization dimensions of equity in LSLR programs. Per the LSLR Collaborative, because each community has unique patterns of lead service line occurrence and social disparities, replacement planning offers a chance to prioritize replacements in ways that address environmental justice concerns (e.g., prioritizing households with multiple sources of lead exposure). Thus, in a context where widespread environmental remediation is required, questions of equity and environmental justice are closely intertwined with sequencing (i.e., which households in which geographical areas receive their service line replacement first).

The case of Pittsburgh, PA

In 2016, Pittsburgh residents learned that samples of the city’s water tested positive for elevated levels of lead, an exceedance of the EPA action level (15 ppb) at that time. ³⁰ Subsequent investigation revealed that lead levels had been steadily rising for decades, a result of deteriorating water infrastructure, poor system maintenance, and controversial cost-cutting measures—including a change in water treatment chemicals—which increased water line corrosion. ³¹ As a result of this discovery, PWSA began the process of LSLR in 2016. In addition to the initial discovery of high lead levels in 2016, the water authority was sued by the state of Pennsylvania, and later entered into a $500k settlement agreement, for failing to notify and warn residents that LSLR work could (and did) cause temporary spikes in water lead levels. ³² Against this backdrop, PWSA (now known as Pittsburgh Water) took public strides to integrate equity into its general operations and LSLR program. In 2019, the water authority formed the Community Lead Response Advisory Committee (CLRAC). That same year, PWSA also convened a “Water Equity Learning Team” which joined the national Water Equity Taskforce, ³³ a network of cities that collaborate to develop more equitable water policies and practices. In 2021, the Water Equity Learning Team (comprised of a number of community groups) published a report that outlined a water equity roadmap for Pittsburgh. The report discussed income and racial disparities and recommended that, “PWSA follow a replacement prioritization plan suggested by the Community Lead Response Advisory Committee. The plan ranks at-risk populations as a priority for LSLR.” ³⁴ Since the beginning of the LSLR process in 2016, and as of this writing, PWSA has replaced over 14,000 lead service lines ³⁵ within its water service area, as shown in Figure 1 and Table 1.

OPEN IN VIEWER

FIG. 1.

Map of PWSA Service Areas. PWSA, Pittsburgh Water and Sewer Authority.

OPEN IN VIEWER

Table 1.

Timeline of Major Events

Year	Event
2016	Tap water samples in Pittsburgh reveal an exceedance of the EPA action level
	PWSA begins the process of LSLR
	PWSA is sued by the state of Pennsylvania for failing to notify and warn residents that lead service line replacement work may cause temporary spikes in water lead levels
2019	The Citizens Lead Response Advisory Committee (CLRAC) is formed
	Pittsburgh Water Equity Learning Team is formed
	PWSA joins the Water Equity Taskforce
2021	Biden-Harris Lead Pipe and Paint Initiative Announced (LPPAP) funded by the Bipartisan Infrastructure Law (BIL)
2023	PWSA wins the Association of Metropolitan Water Agencies’ (AMWA) Environmental Justice and Equity Utility Management Award
2024	EPA unveils the final Lead and Copper Rule Improvements (LCRI)
2025	PWSA announces replacement of over 14,000 LSLs

Quantitative analysis

This analysis draws on primary data covering the replacement of more than 12,000 LSLs undertaken by PWSA between 2016 and 2024, acquired via a right-to-know request filed with PWSA. These records included essential information such as the address of the replaced lines, the date of replacement, and whether the replacement occurred on the public or private side of the property. To assess equity, this LSLR data was integrated with census-tract-level demographic indicators from the U.S. Census Bureau’s 2024 American Community Survey (ACS). Key indicators of vulnerability included median household income ³⁶ (to differentiate affluent from less-affluent census tracts), percentage of BIPOC (particularly relevant given the documented racial disparities in environmental exposure), and the percent of children in a census tract with Elevated Blood Lead Levels (EBLL), data made available by the Allegheny County Health Department. ³⁷ This comprehensive data collection strategy aligns with the principles highlighted by the LSLR Collaborative, which emphasizes the use of demographic, economic, and public health data to identify communities disproportionately at risk and to ensure equitable program implementation.

Spatio-temporal mapping

Because environmental inequality is inherently spatial, ³⁸ we employed spatio-temporal mapping techniques to analyze the data. ³⁹ All analyses were conducted at the census tract level to balance geographic precision with statistical power. Replacement records were geocoded and aggregated to tracts, and we calculated the average replacement date (expressed as a fractional year) for each tract, as well as the total number of replacements. To account for tract size, we normalized replacement counts by total population, creating a “replacements per capita” measure. Our focus on the sequencing dimension of equity helps us understand environmental justice in remediation efforts, as emphasized by the LSLR Collaborative, which notes the priorities of agencies conducting replacements.

We began with multivariate ordinary least squares (OLS) regression to model the average replacement date as a function of median income, percent of BIPOC residents, and percent of children with an EBLL, while controlling for population. Recognizing that a simple average can obscure temporal dynamics, we also constructed a binary “early” indicator (tracts replaced before versus after the median date) and fitted a logistic regression to examine whether demographic characteristics predicted being in the early-replacement group.

To capture how the influence of each factor might have shifted over the course of the program, we implemented a time-varying logistic regression on a tract-month panel dataset. This model interacted each demographic variable with a smooth spline function of time. This allowed, for example, the effect of the percent BIPOC variable on the probability of a replacement occurring in a given month to change continuously. Moreover, this approach directly addresses the question of whether initial prioritization patterns later changed over time.

We complemented these regression-based approaches with distributional and nonparametric methods. Lorenz curves and concentration indices were constructed by ranking tracts by each demographic variable and plotting the cumulative share of replacements against the cumulative population share. The concentration index summarizes whether replacements were disproportionately concentrated among tracts with higher or lower values of a given characteristic. We computed these separately for the pre-2019 and 2019-onward periods (using January 2019 as the cutoff) to visualize how the distribution shifted after the committee’s formation and membership in the Water Equity Task Force.

Spearman’s rank correlation provided a simple, robust measure of the monotonic relationship between each demographic variable and the average replacement timing, complementing the regression results.

Finally, to assess whether spatial clustering might be influencing our inferences, we computed Moran’s I on the residuals of our regression models and on the raw average replacement timing. We also produced Local Indicators of Spatial Association (LISA) cluster maps to identify tracts where replacement timing was significantly higher or lower than neighboring tracts, after accounting for demographics. Together, these methods allow us to examine equity not only as a final outcome, but as a dynamic process, revealing whether and how prioritization changed over the life of the program.

Aggregate patterns

Over the study period, all 85 census tracts within the PWSA service area received at least one LSLR. The average replacement date (expressed as a fractional year) ranged from 2016.1 to 2022.6. ⁴⁰ Spearman rank correlations (Table 2) show that, over the entire program, tracts with higher median income tended to be replaced earlier (ρ = –0.27, p = 0.01), although tracts with higher percentages of BIPOC residents tended to be replaced later (ρ = 0.26, p = 0.02). ⁴¹

OPEN IN VIEWER

Table 2.

Spearman Rank Correlations with the Timing_Avg Variable

Variable	Spearman rho	p value	N
Median Income	−0.27	0.01	85
Pct_BIPOC	0.26	0.01	85
Pct_EBLL	0.14	0.20	85

To understand how the influence of race shifted over time, we estimated a time-varying logistic regression on a tract-month panel (8,667 observations). At the beginning of the program (early 2016), tracts with higher percent BIPOC were significantly less likely to receive a replacement (β = –0.31, p = 0.007). A significant positive interaction with the second spline term (β = 0.39, p = 0.002) indicates that this disadvantage diminished over time and eventually reversed. By late 2022, higher percent BIPOC tracts had become more likely to be replaced than whiter tracts. In contrast, the positive effect of EBLL on replacement probability remained constant throughout (β = 0.35, p < 0.001), with no evidence of change over time.

In the Lorenz curves in Figure 2, note that the closer a Lorenz curve lies to the diagonal, the more proportional the distribution. The Robin Hood (RH) (Hoover) index indicates the share of replacements needing reallocation to achieve equality. Before 2019, %BIPOC showed a modest bias toward whiter neighborhoods (CI = –0.16, RH = 0.02), requiring just 2% of replacements to be redistributed. After 2019, the distribution became nearly neutral (CI = –0.02, RH = 0.07). Income was a far stronger predictor early on as the curve sat well below the diagonal (CI = 0.13, RH = 0.25), meaning 25% of replacements would need to move from higher to lower-income tracts. This inequality was almost entirely corrected after 2019 (RH = 0.04). For child lead risk (EBLL), both periods showed positive concentration (CI = 0.08 pre, 0.39 post), with the Hoover index rising from 0.20 to 0.31 showing a deliberate and growing prioritization of high-risk areas after the CLRAC. Thus, class-based inequity was initially larger than racial inequity, yet both came close to resolution after 2019. Following Tsai’s caution ⁴² we note that the small number of census tracts (pre-CLRAC n = 77) may bias the RH and CI downward, leading to an overly optimistic impression of early racial equity. The more robust, statistically significant time-varying logistic model therefore serves as our primary evidence for early racial disadvantage, although the inequality metrics should be treated as suggestive.

OPEN IN VIEWER

FIG. 2.

Lorenz curves for % BIPOC, Median Income, and % EBLL, pre and post CLRAC. Curves below the diagonal indicate greater share to tracts with higher characteristic values; CI and RH quantify inequality.

The time-varying coefficient trajectories (Fig. 3) mirror these indices. The coefficient for percent BIPOC began negative (β = –0.31, p = 0.007) and, driven by a significant positive interaction (β = 0.39, p = 0.002), steadily increased, crossing zero around 2022. In contrast, the coefficient for EBLL remained positive and flat throughout (β = 0.35, p < 0.001), with no significant change over time. Together, these complementary analyses show that early racial inequity was corrected only after sustained community oversight, although health risk was consistently prioritized. ⁴³

OPEN IN VIEWER

FIG. 3.

Results of Time-Varying Logistic Regression (Spline vs. Demographics).

Spatial clustering

A LISA analysis identified isolated clusters of tracts with significantly early or late replacement timing (Fig. 4). The low-low (early) cluster includes tracts in neighborhoods like Stanton Heights, Perry North, and Upper Lawrenceville, all with median incomes above $80,000. The high-high (late) cluster includes several Bloomfield tracts and one Shadyside tract, with median incomes between $47,000 and $68,000. No significant global spatial autocorrelation was detected (Moran’s I = 0.048, p = 0.199); therefore, these local patterns should be interpreted with caution. Our primary conclusions are drawn from the time-varying logistic regression and concentration indices, which do not rely on spatial assumptions.

OPEN IN VIEWER

FIG. 4.

LISA Cluster map of average replacement timing by census tract.

Limitations

We acknowledge several data limitations. The EBLL data, obtained from the Allegheny County Health Department, reflect only children who were tested and may not capture the full population at risk; moreover, sample sizes at the census tract level can be small, necessitating caution in interpreting tract-level percentages. ⁴⁵ American Community Survey 2024 (ACS) estimates are subject to sampling variability, particularly for smaller tracts, which may introduce measurement error in our demographic predictors. Although the OLS model’s overall fit is marginally non-significant, the individual coefficient for percentage of BIPOC residents remains statistically significant (p = 0.046), as individual variables can reveal valid relationships even in social science models where low r² values are common due to unpredictable institutional behaviors and unobserved local factors in cross-sectional data. ⁴⁶ We address this limitation by corroborating our findings across multiple lines of evidence, including Spearman rank correlations and time-varying logistic regression, both of which independently confirm our findings. Additionally, although replacement records were obtained through a formal right-to-know request and are presumed to be complete for PWSA’s reported work, we cannot rule out the possibility of missing or mis-coded entries. As PWSA was unable to produce records about the distribution of LSLs, we chose to normalize our data by population rather than by existing LSLs, which may be unevenly distributed within the city. Despite these limitations, the consistency of findings across multiple analytical approaches—time-varying logistic regression, concentration indices, spatial clustering, and non-parametric correlations—strengthens confidence in the overall patterns we report.

Conclusion

This study provides empirical evidence that PWSA’s early LSLR implementation (2016–2018) did not prioritize equity. Instead, during this period replacements were disproportionately concentrated in tracts with higher percentages of white residents and higher average median incomes. The time-varying logistic model demonstrates that the disadvantage for BIPOC tracts was statistically significant at the program’s outset and only diminished over time. Concentration indices confirm this pattern: before the CLRAC’s formation and PWSA’s membership in the Water Equity Task Force in 2019, replacements were concentrated in census tract areas with higher percentages of white residents (CI = –0.14); afterward, the distribution became racially neutral. Income showed a similar but weaker trajectory, shifting from favoring higher-income tracts pre-2019 to favoring lower-income tracts 2019-onward.

In Pittsburgh’s case, early inequity was only revealed through independent analysis and was not publicly documented by the utility itself, highlighting the need for transparency and accountability mechanisms. In the case of the ongoing clean water transition, community engagement is of paramount importance as repairs cannot be completed without the public’s cooperation and buy-in. This emphasis on engagement aligns with a robust body of environmental justice literature that identifies meaningful public participation as a central tenet, with scholars underscoring that equitable outcomes depend on inclusive decision-making processes that empower affected communities.^47,48,49

We do not believe Pittsburgh’s LSLR program is an outlier. Rather, it mirrors challenges observed in other cities. Washington D.C.’s cost-sharing model, which linked lead line replacement access to household wealth, illustrates how neoliberal policy design can create disparities. However, as our time-varying analysis shows, the eventual correction of racial disparities from 2019 onward suggests that such failures can be addressed with sustained advocacy and oversight. Federal funding mechanisms anticipating partial homeowner contributions similarly risk cementing racial and class inequities, as seen in both cities. With changing federal priorities, such funding mechanisms themselves are in flux.

Some key takeaways for other municipalities considering LSLR include: First, equity should be embedded in prioritization criteria from a program’s inception, rather than applied later as a corrective measure. Second, transparency in prioritization decisions is essential. Public dashboards showing how sequencing aligns with stated equity goals can build trust and enable accountability. Third, equity and public health objectives need not conflict. Despite early disparities in race and income, PWSA’s eventual correction of the imbalances led to sustained prioritization of high-EBLL areas, thereby demonstrating that improving equity does not come at the expense of protecting children from lead exposure, rather, these goals can be pursued in tandem.

Looking ahead, many unanswered questions remain with regard to how equity has or has not shaped decision-making in the thousands of cities now undertaking LSLRs. To date, no comparative, systematic studies have examined equity outcomes across different cities or regions. One major barrier is the absence of standardized, publicly available data. Future research could critically examine this data gap itself, asking what it reveals about transparency, accountability, and the political will to address environmental justice concerns at scale. Another direction could be efforts directed toward the development of an equity-weighted prioritization tool to assist water systems in the task of strategically sequencing repairs in ways that deliver the greatest benefit to historically marginalized and high-risk communities. Embedding community co-design into the creation of weighting criteria and decision objectives will be essential to ensure that equity-focused algorithmic tools reflect lived experiences, enhance procedural justice, and support more accountable infrastructure planning.