Hassles and Environmental Health Screenings

A. Lead Screening Background

Federal guidelines mandate that all children on Medicaid must be screened for lead poisoning at ages one and two. In addition, Illinois requires screening for all children living in high‐risk zip codes, defined by housing age and demographic characteristics. Even in these high‐risk zip codes, which include the whole city of Chicago, less than two‐thirds of children born in 2014 were screened (Figure 1, Panel B). This low compliance with the screening guidelines raises the questions of who gets screening and what barriers hinder screening.

Online Appendix Table A.1 shows that children with elevated blood lead levels are more likely to be of low socioeconomic status and to live in old housing than average. These correlations might originate because disadvantaged children are more likely to live in poorly maintained homes. Moreover, parents of low socioeconomic status might be less able to acquire information on the lead status of their residence and to remediate known lead hazards. By requiring screening for children on Medicaid and children living in areas with a high prevalence of old housing, federal and Illinois screening guidelines recognize these higher risks. Indeed, Online Appendix Table A.1 shows that children of low socioeconomic status and those who live in old housing are more likely to get screened. However, the question remains of whether these policies optimally target screening.

I hypothesize that provider access constitutes a barrier to screening, since house visits do not include blood lead screening in Illinois. For comparison, around 80 percent of Illinois children eligible for Family Case Management had three or more well‐child visits in the period 2005–2010, even though the American Academy of Pediatrics recommends six visits during the first year of life, to occur at one, two, four, six, nine, and 12 months of age (Illinois Department of Human Services 2010). As the likelihood that a child has a well‐child visit decreases with age (Caldwell and Berdahl 2013), missed well‐child visits after nine months of age may explain the gap between well‐child visits and lead screening rates.⁵ Section IV.D investigates the alternative explanation that providers’ discretion drives the low compliance with the screening guidelines. My empirical findings support the hypothesis that both travel costs and providers’ quality affect screening rates.

B. The Household’s Screening Take‐Up Decision

Household i perceives benefit b_i from screening their child only if the child is found to be lead‐poisoned: for example, benefits accrue from assignment to case management aimed at reducing damages.⁶ This modeling choice ignores potential screening benefits accruing from negative tests, such as learning that a home is lead‐safe. Parents’ perceived screening benefits depend on several factors, including information about exposure risk, degree of risk aversion, degree of altruism towards the child,⁷ beliefs about treatment costs and feasibility (which may correlate with homeownership) as well as recovery probability,⁸ and additional benefits from visiting the doctor, such as having a physical examination.⁹ My model does not require assumptions on these parameters; the revealed‐preference approach in Section V allows me to compare willingness‐to‐pay (WTP) estimates to estimates of screening benefits computed for different parameter values.

The screening cost, c_i, depends on the nominal screening price, p, and the opportunity cost in terms of the parents’ wage, w_i, and travel time, t_i, which is proportional to distance from the doctor, d_i. I abstract from heterogeneity in p for simplicity, although the cost of a blood lead test in Illinois varies with a child’s insurance coverage.¹⁰ Then, child i is screened if and only if 1

Because t_i ∝ d_i, this inequality yields a cutoff above which a child is not screened: 2

I assume that benefits increase with risk; that is, bʹ(r_i) > 0—the higher the potential exposure, the more likely that screening leads to crucial intervention. Then, riskier children have a higher willingness‐to‐travel for screening. As in the classic ordeals model (Nichols and Zeckhauser 1982), the cutoff increases with risk: 3

Figure 2 illustrates how risk affects the relationship between screening and distance. High‐risk households are less sensitive to distance than low‐risk households: their screening rates decline less sharply with distance (left panel). Therefore, the share of screened children that is high‐risk increases with distance (right panel).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Relationship between Distance and Screening Rates, by Risk

Notes: The figure illustrates the screening predictions from the ordeals model. The left panel plots hypothetical screening rates by distance for low‐risk (L) and high‐risk (H) households. The right panel plots the share of screened children who are high risk by distance as implied by the relationships plotted in the left panel.

However, the model’s predictions become ambiguous if we consider travel mode, following Alatas et al. (2016). Assume that family assets a_i are negatively correlated with risk, aʹ(r_i) < 0, and that travel time is negatively correlated with assets. For example, traveling by car is faster than walking or using public transit: . Then, 4 5

In a model with assets, cutoffs may either increase or decrease in risk. While the second term in Equation 5 is still positive, the first term is negative; riskier households face higher travel times conditional on distance. Thus, the effect of reducing distance to providers on the average riskiness of screened children is an empirical question. In Section IV.B, I exploit providers’ openings and closings to answer this question.

C. The Planner’s Problem

The socially optimal level and targeting of screening may not coincide with the individual optimum due to externalities. First, lead‐poisoned children negatively affect their classroom peers (Gazze, Persico, and Spirovska 2024) and are more likely to engage in risky and criminal behavior (Aizer and Currie 2019; Feigenbaum and Muller 2016; Reyes 2007, 2015a). Second, detecting lead hazards may prevent exposure of future residents.

I model the social benefits of screening a child as the sum of three components.¹¹ First, I consider the private benefit, b_i – c_i. Second, I add the averted externality i would have imposed on society if they had not been screened, e_i. Third, I add the discounted value of the avoided externalities from preventing exposure among children j ∈ J who will live in i’s building in the future.¹² Summing over the set of screened children S, this yields 6

Thus, some households with low private benefits may have a high social value of screening if they have a large externality or prevention value.

The planner cannot optimally target screening without knowing e_i and e_j. Housing age may proxy for exposure risk at each home. Then, targeting screening based on observable risk may improve upon self‐selection on private benefits. I estimate both the average prevention value of screening (Section IV.C) and the societal benefits of different screening policies (Section V.B).

A. Childhood Lead Screening Measures

The Illinois Department of Public Health (IDPH) provided birth and death certificates for almost 4.5 million children born in Illinois between 1991 and 2016. These records include each child’s name and birth date, allowing me to link these data to the universe of 5.4 million blood lead tests performed in Illinois between 1997 and 2016, with a match rate of 86 percent (Online Appendix Figure A.1). Matched and unmatched tests have similar observable characteristics (Online Appendix Table A.2). Because lead test records are incomplete prior to 2001, I limit my analysis to children born after 2000. I also limit the analysis to children born before 2015 to ensure I observe each child’s outcome by age two. I classify nondeceased children not linked to any tests as not screened. Online Appendix Tables A.3 and A.4 show the number of tests and unique children in my original sample and the number remaining after each data cleaning and linkage step.

The Illinois Department of Public Health collects children’s blood lead records from physicians and laboratories. These records include test date, blood lead level (BLL), test type (capillary or venous), provider and laboratory identifiers, and Medicaid status (albeit incomplete). Capillary tests are prone to false positives. Thus, capillary tests that show elevated blood lead levels (EBLLs), defined as blood lead levels above 9 micrograms per deciliter (μg/dL) of blood, need to be confirmed by another test. For each child, I keep the highest venous test when available or the highest confirmed capillary test when available. My sample includes 70,000 confirmed EBLLs from more than 22,000 children (Online Appendix Table A.5). Some laboratories have minimum reporting limits, meaning BLLs are bottom‐censored; I correct for these limits to obtain correct population estimates of lead exposure.¹³

Birth records also include family characteristics, such as mother’s marital status, age, education, and race, as well as child’s address at birth. I geocode these addresses to link the blood lead data to housing age information (see Section III.C) and census block group median income from the 2015 American Communities Survey. After geocoding, I obtain a sample of more than two million children and more than 2.9 million tests linked to these children. I use birth address rather than address at testing time because I only observe subsequent addresses conditional on a child being screened for lead. Online Appendix Table A.6 shows that even if a third of households in my sample move within a two‐year period, most households remain in homes and zip codes with the same exposure risk.

B. Provider Access Measures

The Illinois Department of Public Health collects the name and address of providers who perform lead tests. A quarter of providers are individuals, while the rest include small groups of doctors and hospitals. I code a provider as entering or exiting the sample the first or last year that I observe them ordering tests, respectively. On average, 4.5 percent of providers enter each year and 4.8 percent exit (4.1 and 4.4 percent, respectively, when excluding providers performing fewer than ten tests per year).¹⁴ These numbers are consistent with statistics on physicians’ churning. For example, 9–13 percent of physicians report plans to retire within three years, and about 2.5 percent new students graduate from medical school each year (Walker et al. 2016; Young et al. 2015). Providers who enter or exit throughout my sample are generally similar to the average provider (Online Appendix Table A.7).¹⁵

To construct a measure of travel costs, I calculate the distance “as the crow flies” between a child’s birth residence and the closest provider open during the child’s birth year.¹⁶ The median child has a provider within 1.2 kilometers (Online Appendix Figure A.4). More than 90 percent of screened children did not get tested at their closest provider (Online Appendix Figure A.5), likely due to preference for continued care after a move (Raval and Rosenbaum 2021; Sabety 2023) or insurance network constraints. Section IV.A investigates the relationship between distance to closest provider and distance traveled.

The impact of access to providers may depend on their propensity to screen, which might be correlated with quality (Vivier et al. 2001). I use the 2019 U.S. News ranking of the medical school the provider attended as one measure of quality (Schnell and Currie 2018). I obtain medical school attended by linking providers to the 2019 Medicare Physician Compare File (MPCF) through name, address, and practice name.^17,18 I also consider measures of quality that directly capture a provider’s lead screening behavior. I define providers as higher quality if they screen more children and/or screen them at the right times according to federal and state guidelines as follows. Because I do not observe a child’s provider if the child is not screened, I calculate a provider’s screening rate as the screening rate for children born within the median‐distance households travel to see that provider, and I weigh unscreened children by the inverse of their distance.¹⁹ Because federal guidelines mandate that all children on Medicaid must be screened for lead poisoning at ages one and two, I compute the share of Medicaid children a provider screened by age one who have a second test by age two. I also compute the share of EBLLs detected by each provider with a required follow‐up within 90 days.²⁰ I then aggregate these screening‐based measures into a summary quality index. Because these screening‐based quality measures might reflect demand‐side preferences, I consider an indicator for performing capillary tests a more objective measure of providers’ propensity to screen because capillary testing requires a machine and may reduce the barrier to screening if households are averse to venous blood draws.²¹

Providers’ screening‐based quality measures and providers’ medical schools capture different provider’s characteristics and are indeed imperfectly correlated (Online Appendix Figure A.8). My empirical analysis is robust to using different quality measures.

C. Childhood Lead Exposure Pathways and Lead Hazard Remediations

Children in homes built prior to 1930 have the highest BLLs in Illinois, after controlling for demographic characteristics and zip code fixed effects (Abbasi, Pals, and Gazze 2020). Indeed, HUD estimates that 87 percent of houses built before 1940 in the United States have lead paint, compared to 69 percent of houses built between 1940 and 1959 and 24 percent of houses built between 1960 and 1977 (U.S. Department of Housing and Urban Development 2011). Thus, I define children born in homes built before 1930 as high‐risk, using parcel‐level data on construction year in the Zillow Transaction and Assessment Dataset.²²

To measure lead hazard abatement following EBLL detection, I use data on addresses that receive remediation funding under HUD’s lead hazard control programs.²³ Because these funds are targeted to low‐income property owners, these data do not cover the universe of lead hazard remediations. Yet, they provide a useful picture of case management following EBLL detection in the absence of more complete data.

A. Does Distance to Providers Decrease Screening and Why?

While the relationship between screening rates and distance to providers in Illinois is U‐shaped in the raw data, it becomes closer to linear after controlling for neighborhood fixed effects (Figure 4). In my main analysis, I drop the 31,178 children who are farther than 20 kilometers from a provider (2.6 percent of the sample), as they are very different from the rest of the sample.²⁴ Indeed, these outliers have a lower elasticity of screening with respect to travel costs (Online Appendix Table A.10).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Determinants of Screening: Distance to Providers

Notes: The figure plots the average likelihood of a child being screened by age two by distance to closest open provider. The bars in the left panel show the number of children in each distance bin on the left y‐axis, and the lines represent their screening rates on the right y‐axis: the solid line plots raw means, while the dashed line plots residualized screening rates after controlling for census block fixed effects. The right panel plots screening rates for each distance bin relative to children born 10–20 kilometers away from open providers controlling for tract fixed effects (dashed line) and block fixed effects (solid line), with vertical bars indicating 95 percent confidence intervals based on standard errors clustered at the zip code level.

Panel A of Table 1 estimates that being 1 kilometer farther away from a lead‐screening provider, a 30 percent increase over the mean distance, decreases the likelihood that a child is screened by age two by 0.4 percentage points, or 0.9 percent relative to the mean. These estimates imply an elasticity of screening with respect to distance to the closest provider of −0.03. Einav, Finkelstein, and Williams (2016) estimate an elasticity of −0.02 for take‐up of cancer treatment, while Herbst and Tekin (2012) estimate an elasticity of −0.13 for take‐up of childcare subsidy. Nonlinear estimates in the right panel of Figure 4 imply that the screening rate in Illinois would have been two percentage points higher (4.35 percent) if every child in my sample had a provider within 1 kilometer.

Because most households do not visit their closest provider, I also estimate the effect of distance from the provider of choice using a two‐sample two‐stage‐least‐squares (2SLS) model. Using the sample of screened children, Panel B of Table 1 estimates that being 1,000 meters farther from the closest provider translates into an extra 75–280 meters traveled to get a child’s first lead test, providing a strong first stage in all but the most stringent specification with house fixed effects. Bootstrapping this first‐stage relationship to predict distance from provider of choice for all children, Panel C estimates that an increase in distance to the provider of choice of 1,000 meters (13.4 percent) reduces screening by 1.3–7.2 percentage points, yielding an elasticity of −0.21 to −1.01. Importantly, 2SLS scales the reduced form estimates by the share of compliers, that is, the households who change provider following openings and closings.²⁵

Interpreting the magnitude of the effect of distance on screening take‐up requires data on households’ transportation mode, which I do not observe. Thus, I use car travel times, at 1–1.5 minutes per kilometer in Illinois (Agbodo and Nuss 2017). The estimates in Table 1 imply that a $6.25 increase in travel costs (a 15‐minute two‐way trip to the doctor at 7.5 kilometers each way and $25 hourly wage) decreases screening take‐up by 9 percent.²⁶

Proximity to healthcare providers might affect screening not just by enabling more timely visits to the doctor, but also by changing the information set available to families. Indeed, a provider opening (or closing) affects an entire neighborhood—more families now have more direct access to a doctor and might learn about lead screening, potentially spreading information and making lead exposure more salient. To examine the role of information and salience, I first focus on families with multiple offsprings where at least one child is screened, and I estimate the effect of that child having a BLL of 10–14 μg/dL relative to 5–9 μg/dL on the siblings’ screening decision. This exercise isolates the effect of the information shock that one child has an EBLL on the decision to screen other children within the family who might also be at risk of lead exposure. Second, I explore how the effect of distance to providers changes with proxies for lead risk information at the neighborhood and family levels.

Table 2 shows that siblings of a child with EBLL are 27 percent more likely to be screened in the 30 days immediately after the EBLL is diagnosed, over a mean of 3 percent, and are still 5 percent more likely to be screened after two years than siblings of children with BLLs 5–9 μg/dL. Moreover, these effects appear to be symmetric for older and younger siblings, albeit larger for older siblings who are less likely to be screened as they age. These findings highlight two facts. First, there is ample variation in screening decisions even within families, suggesting that parents re‐optimize based on information shocks. Second, screening costs are nonzero, or else parents would not respond to information changing the expected benefits of screening. Section V.A estimates parents’ willingness to pay for screening.

Table 3 investigates whether providers’ openings and closings increase screening primarily by increasing its salience. Columns 1–3 estimate Equation 8 controlling for screening rates at the neighborhood–cohort level. These estimates are virtually indistinguishable from those in Panel A of Table 1, suggesting that information cannot fully explain the effect of travel costs. Next, I study how the travel elasticity of screening changes with information. Column 5 focuses on siblings and interacts distance to provider with indicators for younger siblings of tested children, as well as younger siblings of children with EBLLs.²⁷ Travel costs have larger effects for younger siblings with screened older siblings, unless the older sibling has an EBLL. I interpret these coefficients as suggesting that a negative test for one child decreases the expected benefits from screening their sibling, thus making parents more cost‐sensitive. Vice versa, a positive test reduces the sensitivity to travel distance. Column 6 repeats this exercise but interacting distance to providers with an indicator for a child living in the same building having an EBLL and finds similar patterns.

The estimated effect of screening costs is robust to different specifications, samples, travel costs measures, functional forms, and outcome definitions. Table 1 shows robustness to controlling for different location fixed effects, suggesting that spurious correlation does not drive my findings. Moreover, I estimate similar elasticities for children at different distances from providers, although absolute changes appear to have larger effects at smaller distances (Online Appendix Table A.13).

Online Appendix Table A.10 explores different specifications and distance measures. Column 3 controls for census block group trends to assuage concerns that neighborhood changes, such as gentrification, drive the estimated relationship between screening rates and distance to providers. Column 4 uses average distance from the closest five providers, as households do not always visit the closest provider. While attenuated with respect to my preferred estimate, the coefficient on this variable is negative and significant. Column 5 uses distance from the census block centroid to remove variation in travel costs due to children living in different buildings within the same block, yielding estimates that are not statistically distinguishable from my preferred estimate. Related, Online Appendix Table A.14 includes both distance to the closest provider and distance to the five closest providers. Because distance to closest provider has a higher explanatory power, provider density does not appear to drive my findings.

Logistic and ordinary least square (OLS) regressions that include regressors’ block‐level means but omit block fixed effects yield similar results to my preferred linear probability model (Online Appendix Table A.15). This approach avoids the incidental parameters problem (Neyman and Scott 1948) and is equivalent to the linear fixed effects model if there is no correlation between the relevant regressors and the fixed effects (Mundlak 1978; Chamberlain 1984; Bafumi and Gelman 2016). This equivalence is important because Section V.A uses the logit framework to estimate households’ willingness‐to‐pay for screening. This table also shows robustness to measuring screening at different ages, consistent with most children being screened by age two (Online Appendix Figure A.10).

B. Does Distance to Providers Affect Selection into Screening?

Distance to providers decreases screening take‐up, but for whom is theoretically ambiguous. On one hand, families with low exposure risk will not be willing to travel farther. On the other hand, children facing high travel costs, who may also be at high risk, may forego screening.

To assess how the composition of screened children changes with distance, I estimate Equation 8 on the sample of screened children, with children’s characteristics as outcomes. I examine ex ante observable and unobservable exposure risk, as measured by housing age and lead levels. Consider two children, one in an old house and one in an adjacent new house. There is a clinic 250 meters away, and both get screened. Years later, two new children move in, but the clinic is closed, and the closest provider is now a kilometer away. Only the child in the old house gets screened. In this example, the probability that a screened child lives in an old home increases with distance: it is 0.5 at 250 meters and one at 1 kilometer. Data from this example suggest that hassles improve targeting based on observable risk, as illustrated in Figure 2.

In contrast to this example, Table 4 does not support the hypothesis that children screened at farther distances have higher observable or unobservable risk—they are less likely to live in a home built prior to 1930 and have slightly lower BLLs (only significant when controlling for census tract fixed effects). Children screened at higher distances are also slightly less likely to be Black or Hispanic, with significant estimates only when controlling for tract fixed effects. These findings are largely robust to including time‐varying neighborhood controls (Online Appendix Table A.16).²⁸

C. Does Proximity to Providers Improve Children’s Outcomes?

Because distance decreases screening for high‐ and low‐risk children alike, decreased provider access may hinder detection of lead poisoning. If lower detection rates lead to fewer remediations, future residents may face increased risk. I investigate how distance affects prevention outcomes by estimating Equation 8 with the following outcome variables: an indicator for EBLL detection (equal to zero if the child is either not screened or has a BLL lower than 10 μg/dL), age at first and highest test, as well as indicators for remediations and subsequent EBLLs at the same location.

Table 5 shows that children who live 1 kilometer closer to a provider are 3.3 percent more likely to be diagnosed with an EBLL (Column 1). Because screening increases by 0.9 percent per kilometer (Table 1), the higher EBLL detection rate is likely due to both the extensive margin and selection on the intensive margin. Moreover, these children are screened six days earlier and are younger when their highest BLL is recorded (Columns 2–3). Early detection may improve long‐term outcomes by reducing exposure and enabling access to treatments, which Billings and Schnepel (2018) show can improve outcomes for lead‐exposed children.

Next, I ask whether distance affects prevention and outcomes beyond the first lead‐exposed child in a house. Columns 4 and 5 find no evidence that proximity to providers is associated with higher HUD‐funded remediation activity at a child’s home within three years of birth or with lower future EBLL rates at that home.²⁹ Column 5 rules out effects in the same order of magnitude as the direct detection effects. Summing up, these findings suggest that proximity to providers improves detection of lead exposure, potentially improving access to treatment for directly exposed children but without beneficial externalities on future residents through increased remediations. Potential explanations include further barriers to remediations, which might be stronger for high‐risk families, who tend to be of low socioeconomic status, as well as temporary remediations whose effectiveness might fade over time.

D. Does Providers’ Quality Affect Screening?

The disparity between well‐child visits and screening rates discussed in Section II.A suggests that providers may exercise discretion in screening. As providers’ practices differ greatly (Mullainathan and Obermeyer 2019; Kwok 2019; Fadlon and Van Parys 2020; Silver 2021; Currie, MacLeod, and Parys 2016; Van Parys 2016; Fletcher, Horwitz, and Bradley 2014; Epstein and Nicholson 2009), I ask how access to providers of different quality affects screening take‐up.

I regress a child’s screening indicator on indicators for the presence of any provider and of a high‐quality provider within given distances of a child’s birth address. Screening quality measures include whether providers offer less‐invasive capillary tests, adherence to screening guidelines, and screening rates. Importantly, while a provider’s screening rate might reflect demand‐side preferences, I consider the indicator for providers offering capillary testing as more directly capturing a provider’s propensity to screen. Moreover, I test for sorting by investigating whether proximity to a provider with a good screening record has additional explanatory power over proximity to a provider who attended a top‐20 medical school because parents may more easily observe and select on a provider’s alma mater than screening record. I estimate 9

where k ∈ {<1 km, 2–5 km, 5–10 km, 10–20 km}, and X_igy are individual‐ and neighborhood‐level characteristics that may correlate with parents’ and providers’ screening propensity. Variation in proximity to providers of different quality in this equation is given by providers entries and exits, as well as differences in children’s addresses within a geography, as in Equation 8.

Figure 5 shows that proximity to providers increases screening, and more so if the providers are of high quality, using any quality variable. For example, screening increases by 2.5 percentage points when there is a provider within 1 kilometer, and further by 6.5 percentage points when that provider offers capillary testing. Figure 4 estimates that having a provider within 1 kilometer increases screening by four percentage points on average. The effect of general proximity is 29–58 percent the effect of supply‐side providers’ quality, depending on the quality measure. Because I estimate similar effects of proximity to high‐quality providers when using capillary testing ability and screening‐based quality, I interpret my findings as suggestive that providers’ discretion matters beyond demand‐side preferences. Moreover, screening‐based quality measures have additional predictive power beyond a provider’s alma mater, suggesting that these results are not driven by households with a higher propensity to screen choosing providers with better education.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

Determinants of Screening: Provider Quality

Notes: Each panel plots coefficients from a regression estimating the effect of having any provider (dark gray/blue bars), a high‐quality provider based on the definition in each panel (light gray/orange bars), and a provider who attended a top‐20 medical school (medium gray/maroon bars) within each concentric buffer indicated on the x‐axis on screening take‐up. The quality index includes screening rates in a provider’s catchment area, as well as a provider’s rate of follow‐up within 90 days on cases of EBLLs and a provider’s rate of adherence to Medicaid guidelines, that is, the rate at which children on Medicaid screened by that provider at age one have a second test at age two. Providers’ catchment areas are computed based on the median distance of children to their screening providers in my sample. Within catchment areas, I compute provider‐level screening rates by weighting unscreened children by the inverse of their distance to the provider. The sample includes all geocoded children born 2001–2014 whose closest provider is within 20 kilometers. Each regression includes child‐level controls, as well as birth year and block fixed effects. Vertical bars display 95 percent confidence intervals based on standard errors clustered at the zip code level.

A. Exposure Risk and Willingness‐to‐Pay for Screening

I derive the WTP for screening by defining household i’s utility from screening as 10

where d_i is distance from provider, θ_i is opportunity cost of travel time, p is the price of a test, and α_i and β_i are preference parameters (Einav, Finkelstein, and Williams 2016). Assuming that α_i = δ^αX_i + ϵ_i, β_i = δ^βX_i and that ϵ_i follows a Type I extreme value distribution, i’s WTP for screening is .

Table 6 estimates the marginal effect of distance on screening take‐up of households with different characteristics using both the linear probability model in Equation 8 and a logit model. To recover α_i while avoiding the incidental parameters problem (Neyman and Scott 1948), I include block‐level means of controls. I then compute WTP using the logit estimates. Column 1 reports estimates for the whole sample, while other columns report estimates for subsamples, obtained by interacting distance to the closest provider with indicators for household characteristics.

The average household has a negative WTP for screening, yet households in the riskiest homes, those built prior to 1930, are willing to pay $7.81 for screening on average (Table 6). Similarly, households with low socioeconomic status have a higher WTP for screening than better off households, consistent with their heightened risk even after controlling for housing age. Because Panels A and B of Table 6 do not show large differences in the elasticity to travel costs, β_i, across housing vintage, race, and ethnicity, these different WTPs suggest households have different valuations of screening benefits, α_i.

Because I do not observe the price each household pays for a test, which could vary with insurance status, I perform a bounding exercise. If all households face the same price, Table 6 implies that households in pre‐1930 homes are willing to pay up to $31.44 more than households in newer homes, as the test price would cancel out of the difference. If, instead, households in pre‐1930 homes have no co‐pay, while low‐risk households pay the maximum full price indicated in outreach materials in Illinois ($43, as discussed in Section II.B), the difference in WTP between high‐ and low‐risk households becomes negative. Conversely, the difference widens to $74.44 if riskier households pay full price ($43) due to lack of insurance, while low‐risk households do not pay. Because households often do not visit their closest provider, I can further divide the WTP estimates by the average ratio between minimum and actual distance, 75–281 meters per kilometer (Table 1), yielding a difference in WTP of $111.89–419.20. Still, my definition of travel costs likely overestimates WTP, as high‐risk households are less likely to drive, meaning they need more time to travel a given distance.³⁰

To interpret the magnitude of these WTP estimates, I need a measure of screening benefits. Under risk neutrality and perfect information, benefits are the converse of the expected costs of lead poisoning, as screening enables detection (Table 5) and potential treatment. Yet, the literature lacks rigorous and comprehensive estimates of the cost of an EBLL. The correlation between IQ losses and BLLs implies an expected lifetime cost of living in a pre‐1930 home relative to a new home of $910 (Schwartz 1994). While, this estimate does not account for unobserved innate ability correlated with lead exposure, it also omits the opportunity cost of the additional time parents spend caring for a poisoned child and additional damages not measured by test scores.

B. Policy Counterfactuals

This section simulates the impact on EBLL detection of four screening policies and one remediation policy in the 2014 cohort as modeled in Equation 6. First, I look at incentives for households and providers. Then, I look at a policy opening screening locations in each zip code. Finally, I evaluate a 100 percent screening requirement for children in homes built prior to 1930. Moreover, I compare these policies to subsidizing full remediation for addresses with EBLLs.

Table 7 reports the number of additional children screened and additional poisoning cases detected under each policy. I compute additional detection rates assuming that marginal children have the average poisoning rate in the 2014 cohort, based on my finding that hassles do not improve targeting (Section IV.B). When evaluating the screening mandate for old homes, I use the poisoning probability among children living in old homes. I compute the private benefits of each policy by summing the WTP for screening of the marginal households, b_i – c_i, estimated in Section V.A.³¹ I assume no prevention benefits from screening because Section IV.C finds no evidence that proximity to providers reduces exposure of future residents. Examining the opportunity cost of using public funds for these policies is outside the scope of this paper.

Because estimates of the externality of lead exposure e_i are not available, for each policy I compute the per child difference between the policy’s private benefits and its costs. This difference indicates the minimum value of the average externality that would make each policy cost‐effective. All the screening policies I study appear to be cost‐effective for externality values lower than $15,976, the estimated spillover effect that a lead‐poisoned child imposes on their school peers (Gazze, Persico, and Spirovska 2024). As this value omits the crime costs of lead poisoning, it underestimates its total externality, further implying the cost‐effectiveness of the screening policies examined.³²

First, I study the effect of incentivizing households for screening, following a large literature on immunization incentives (Banerjee et al. 2010; Bronchetti, Huffman, and Magenheim 2015). To simulate a travel voucher households could receive upon screening, I assign incentives based on the zip code average realized travel distance, valued at 1.2 minutes per kilometer and $25 per hour ($10.5 on average). I identify the marginal children screened under this policy as those whose WTP turns from negative to positive under the counterfactual policy, weighting by the realized probability of screening for a given WTP. Column 1 of Table 7 shows that this policy’s private benefits are positive but lower than the incentives disbursed as many inframarginal households receive subsidies.

Second, I consider a pay‐for‐performance (PFP) incentive for low‐performing providers, a policy with mixed success (Li et al. 2014; Alexander and Schnell 2019). Under PFP, I assume that providers in high‐risk zip codes with screening rates lower than 50 percent screen an additional 10 percent of random children in their catchment area. Column 2 of Table 7 shows that PFP would lead to screening around four times more children than the household incentive, but achieve a similar private benefit, due to poorer targeting.

Third, I simulate a provider opening at the centroid of each zip code without providers in 2014. In the past, lead screening was offered at the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC), the single largest point of access to health‐related services for low‐income preschool children in the United States (General Accounting Office 1999), and WIC status appears associated with increased screening (Vaidyanathan et al. 2009). Alternatively, pharmacies could be equipped with capillary screening kits at the cost $7.96 per test. While this policy would only screen 882 more children, the benefits for these marginal children appear to outweigh the program’s cost (Column 3 of Table 7).

Fourth, I consider a mandate to screen all children in homes built prior to 1930, which leverages observable exposure risk to target screening. Column 4 of Table 7 shows that, compared to the screening incentive in Column 1, this policy yields fewer additional screenings and lower private benefits, but similar rates of poisoning detection. This result is consistent with the finding in Section IV.B that households do not self‐select into screening based on better information about unobservable risk. Thus, the social planner may be able to target screening based only on observable risk. However, it may be prohibitively costly to screen all children in old homes.

Fifth, I consider a policy that keeps screening constant but assumes perfect remediation after EBLL detection, preventing new lead poisoning cases at homes with previous cases. In the 2014 cohort, 638 homes had an EBLL. Because 10.3 percent of addresses with EBLLs in the 2001–2003 cohorts have another child with EBLLs within ten years, I assume that remediating these 638 homes would prevent 66 new cases. The average remediation cost in the HUD data for the 2010–2016 period is $10,646, suggesting lead poisoning externalities need to be on the order of $100,000 for remediations to be cost‐effective in terms of prevention benefits only. Importantly, I do not have estimates of averted case management costs that would factor in prevention benefits.

This section evaluates the impact on EBLL detection of five screening and remediation policies. Overall, policies increasing screening rates have modest private benefits for marginal children, but may be cost‐effective after taking into account lead poisoning externalities as small as $3,500. Specifically, I consider a screening subsidy, which allows households with the highest WTP at the margin to select into screening, and find that even this policy has small private benefits. Then, I consider supply‐side policies, such as a PFP incentive and an increase in provider locations, and find that while both have worse targeting outcomes than the screening subsidy, PFP leads to higher screening rates and thus higher poisoning detection rates. To better study targeting, I next consider a screening mandate in old homes and find that it leads to similar poisoning detection rates as the subsidy, suggesting that households do not have private information on unobservable risks. Finally, I examine perfect remediation and find it not to be cost‐effective because of the uncertainty in turnover of residents at each address. Importantly, this section does not consider additional welfare implications of changing provider access, for example, stemming from other health outcomes.