California school lead testing results and the new lead and copper rule

In 2014 in Flint, Michigan, a water crisis involving lead occurred. The Public Water Systems (PWS) for the area switched water supplies. The local surface water was highly corrosive to plumbing materials in general and lead in particular. While lead service lines (LSL) had been in use for decades in the area, there had been no problems with lead leaching as the PWS added corrosion inhibitors. In 2014 a new water supply was introduced, which was just as corrosive as the old water supply, but no corrosion inhibitors were added. The new water source was corroding the LSL, releasing lead into the water, and causing the increase in blood lead levels (BLL). This event, and other similar ones involving a change in water chemistry, set off a nationwide reaction. In California, the State Water Resources Control Board – Division of Drinking Water (DDW) created a requirement that PWS conduct lead testing in the water at schools covering grades K to 12.

In January of 2017, DDW issued a Permit Amendment (PA) to all Community Water Systems (CWS) and all Non-Transient, Non-Community Water Systems (NTNCWS), which required them to make available all K-12 schools in their service area sampling and analysis services for lead in drinking water. Within 90 days of receiving the request from the school, the PWS had to have completed the initial round of lead monitoring under the final lead sampling plan. The samples had to be analyzed by a laboratory accredited by DDW's Environmental Laboratory Accreditation Program (ELAP) to analyze lead in drinking water. Between 2017 and 2020, 43,803 samples were collected from 7,058 schools in 864 school districts.

DDW used the methods and procedures found in the lead and copper rule (LCR). The PA requires that PWS collect samples from schools. The LCR required that PWS to collect samples from the taps of customers who lived in houses assessed to be at high risk of having lead-bearing plumbing materials. An Action Level (AL) was established at 15 μg/L as a corrosion optimization target, it was not a health-based value. DDW appears to have adopted this approach not because it is optimized for the detection of lead corrosion at any point in the plumbing, but because PWS staff are familiar with it. There was, it would appear, a desire to initiate and complete this effort in a quick and efficient manner. Using LCR procedures allowed for the quickest and least confusing approach.

Each water source is given a ten-digit Federal Reporting Database System (FRDS) number, consisting of the PWS identification number plus a three-digit extension. The first two digits designate the county in which the PWS is located (e.g. 19 = Los Angeles, 30 = Orange, 33 = Riverside, 36 = San Bernardino, 37 = San Diego). The next digit is either 0, 1, or 9 depending on the type of PWS (0 = Non-Community Water System, 1 = Community Water System (PWS), and 9 = Reclaimed Water System). The sample ID consisted of the PWS FRDS number combined with a three-letter ID for each school and a single letter to identify the sample location (e.g., XX10001-AAA-A). Samples collected from reclaimed water systems were excluded.

Additionally, in San Diego County, there were a large number of samples that were from the influent to a seawater desalination plant. Samples from this influent to this plant were excluded as they were raw water and not treated and thus could have no impact on lead corrosion.

The samples consisted of 1 L of first draw water from either kitchens or bathrooms in residential buildings, preferably single-family, detached dwellings. If the 90th percentile of the lead concentration of the samples collected was above the AL, the PWS had to take corrective action to reduce that concentration, such as removing lead service lines (LSL) or adopting corrosion control technology. USEPA had recommended a first draw sample size of 250 mL in their 3 T, however DDW opted to use 1 L in this study as it was a well established procedure among PWS staff and had the fewest unknowns.

The only documented source of lead in LCR monitoring in California is lead from brass corrosion. For some time, it has been known that most, if not all, lead found in LCR samples is present because of the dezincification of brass plumbing materials in fittings, even in houses with LSL (Lytle & Schock 1996). Lead service lines are very rare in California. This being the case, the vast majority of lead and other corrosion by-products would be found in the first 250 mL and the additional 750 mL would tend to dilute the concentrations of those metals. This pattern of most lead leached from brass in household plumbing has been shown in several studies (Kimbrough 2001; Kimbrough 2007; Triantafyllidou et al. 2021).

The brass found in plumbing fittings is an alloy of copper and zinc with small amounts of lead and nickel. There are many different types of brass used in plumbing fittings. A key difference is how much zinc is used. Brasses with high zinc content, generally yellow and called ‘yellow brass,’ have different mechanical properties than those with high copper content, called ‘red brass.’ There are differences in tensile strength, machinability, elongation, and price, but this study's critical difference is that yellow brasses are much more likely to undergo dezincification than red brasses. Dezincification is the selective leaching of zinc from brass. Lead has been generally found in samples with low copper/high zinc concentrations, indicating dezincification is the cause of the release of lead.

There is little research on water quality parameters that directly impact lead leaching from brass, while there is much research on dezincification. If it is true that lead is primarily leached out early in the process of dezincification, then factors that promote dezincification should likewise encourage the leaching of lead from brass. Many workers report that pH, chloride concentration, bicarbonate concentration, and the ratio of bicarbonate to chloride are variables that impact dezincification (Schock & Gardels 1983).

The PA required the PWS to sample from at least five (5) locations where children would routinely use the fixture for drinking. Food preparation areas could also be sampled. Samples could also be collected from the inlet school where the water leaves the distribution system of the PWS and enters the school's distribution system, such as at the water meter or backflow device. There was a field in the reporting form to describe the sample location. Sampling was to be the same as occurs in the LCR. A one liter (1 L) polypropylene bottle was to be used without any acid preservative (the preservative, nitric acid, would be added in the laboratory prior to analysis).

The sample would be collected first thing in the morning after the water in the pipes had laid quiescent for at least six hours. The samples could only be collected when school was in session between Tuesday and Friday. In the past, the LCR had allowed PWS to flush the lines to be sampled the night before. This practice was not allowed during this program. Likewise, any fixtures' aerators were not removed or altered prior to sampling. Sampling technique and planning are key to study success (Triantafyllidou et al. 2021).

There were two thresholds for assessing laboratory results; (1) The Detection Limit for Reporting (DLR) was 5 ppb, any result below that concentration was reported as <DLR and any result above the DLR is assigned a numeric value. (2) The Action Limit (AL) was 15 ppb. PWSs were required to notify the school within two 2 days of receiving a laboratory result above the AL from the laboratory. The sample location was supposed to be resampled within 10 days of receiving the laboratory results. If the second result was below the AL, a third sample is required to settle the issue. If two of the three sample results were above the AL, the school had to decide what remedial actions to take, while if two of the three sample results were below the AL, the PWS has no additional requirements. However, if two sample results were found to be above 15 ppb at the same sample location, and if the school wanted to continue the investigation, the PWS was required to continue to provide sampling and laboratory services. Many schools did not resample when a result above the AL. Other schools resample more frequently than required, a few cases as many as six times.

The laboratory results were characterized based on the description of the location sampled. They were grouped into six categories: (a) Drinking fountains, (b) Sinks, (c) Food Preparation, (d) Bottle Fillers, (e) Distribution System, and (f) Unidentified. In some cases, the nature of the sample location was quite clear (e.g., ‘drinking fountain by room 3’), and in others, it was entirely unclear (e.g., ‘ABC123’); others were somewhere in between.

The overwhelming majority of lead detected in LCR monitoring in houses comes from the dezincification of brass. The population of brass fixtures covers a wide range of types, ages, and geographical locations. It is argued that the rates of dezincification are a function of water quality. There are significant differences in water quality throughout California, including pH, chloride, and bicarbonate. Differences in dezincification and thus lead leaching should be observable based on geographical areas and types of brass fixtures sampled.

In order to detect any differences, the results were grouped by the year in which they were collected. The data were also grouped by the county in which they were collected. If the county where the school was located (some water districts provided water to schools in different counties than where they were located) had at least 250 individual results, it was assessed by itself. If there were fewer than 250 results, they were grouped with counties with fewer than 250 samples, which are also geographically proximate. Del Norte, Humboldt, Lassen, Marin, Mendocino, Modoc, Shasta, Siskiyou, and Trinity counties were one group along the Northern tier of California (NC). Amador, Butte, Colusa, El Dorado, Glenn, Lake, Napa, Nevada, Placer, Plumas, Sierra, Solano, Sutter, Tehama, Yolo, and Yuba counties were the second group in the Northern Inland (NI) area. Monterey, San Benito, Santa Cruz, San Luis Obispo, and Santa Barbara counties were a third group of the Central Coast area (CC). Calaveras, Kings, Madera, Mariposa, and Merced were a fourth group of the Central Inland (CI) area (Kings County was not contiguous with the other counties). Imperial, Inyo, and Mono counties made up the last group along the Eastern tier of California (EC) area (Imperial County was not contiguous with the other two). Sample results were also grouped by the type of sample location (i.e., fountain, sink, kitchen, bottle filler, system, etc.) tested. The results from the drinking fountain were also sub-grouped by county, as were the results of sink and food processing locations. The results from resampling efforts were grouped by the order in which they were collected.

The lead data were assessed four ways:

• (a)

  The frequency of the lead detected above the DLR or frequency of detection.

• (b)

  The frequency of lead exceeding the Action Level or frequency of exceedance.

• (c)

  The median lead concentration among samples that had concentrations of lead greater than the DLR.

• (d)

  The median lead concentration of lead among samples that had results greater than the AL.

These data were compared with the median concentration of pH, chloride, bicarbonate, and the ratio of bicarbonate to chloride concentrations for each county. DDW has an online Microsoft Access database for the year 2015 through 2021.

• (a)

  Significance – Several statistical tests were used in this study. In each case, differences between test populations were considered significant if the probability of random variation was less than 5% (a < 0.05).

• (b)

  Normality – The distribution of each data set was assessed using the Shapiro–Wilk Test (SWT), and the skewness and kurtosis were assessed. All data in this study was non-normally distributed for either skewness or kurtosis, or both. Kurtosis is the measure of how sharp the peak of a frequency–distribution curve as compared to the width of the tails. Kurtotic distributions may have wider than normal tails or narrower than normal peaks.

• (c)

  Median Concentration of Two Populations – When two populations were compared for median concentration, the Mann–Whitney Rank Sum Test (MW), the non-parametric equivalent to the Student's t-test, was used for non-normally distributed data.

• (d)

  Median Concentration of Three or More Populations – When three or more populations were compared for concentration, the Kruskal–Wallis One Way Analysis of Variance on Ranks (KW) was used. The KW test produces the Kruskal–Wallis Statistic (H).

• (e)

  When a statistically significant difference was determined to be present, the median of each group was compared with each other using Dunn's method. Dunn's method produces a value for the Difference of Ranks (DOR), a Q-value, and the probability, p. If a pairwise comparison had a probability of 5% or less, the DOR was among the highest in the study, and the Q-value is greater than four, then the difference was considered significant. The lower the DOR and the Q-value, the less likely a difference was truly significant.

• (f)

  If two populations were being compared pair-wise, if their variances are not equal, this indicates that even if the p-value was less than 5%, it is less likely that the difference was truly significant.

• (g)

  When the variance of two populations was compared for concentration, an F-test was used. The standard deviation (SD) of each population was calculated, the square was taken (the variance), and the ratio of the two variances was determined (F-value). The F-value was compared to an F table against the degrees of freedom.

• (h)

  Frequency – When the frequency at which lead was detected, either above 5 ppb or 15 ppb, was compared between two or more populations, the chi-squared test is used. In this case, an expected frequency was determined based on state-wide statistics. The observed number was determined based on state-wide statistics.

• (i)

  Correlation – the Spearman Rank Order Correlation (SROC) was used to compare two variables to determine if there was a correlation between them. This is the non-parametric equivalent of the Pearson Product Moment Correlation.

All of the data collected in this study were non-normally distributed by the SWT (p < 0.001). The results were both skewed and kurtotic. This outcome was expected for the lead data as it was ‘left censored’ through the use of the DLR. Quite aside from that, there were many individual results above the upper 95% percentile in almost every case, which also contributed to the results' highly skewed and kurtotic distribution.

In examining Table 1, state-wide, 43,803 samples were collected, of which 2,338 were detected above the DLR, which is 5.3% and 474 exceeded the AL, which is 1.1%. This same pattern was seen in all three years between 2017 and 2019. The median concentration of all samples above the DLR was around 8 ppb and the median concentration of the samples above the AL, which was approximately 26 ppb. The results for each year between 2017 and 2019 were very similar. The mean results were considerably higher than the median results, consistent with the right-skewed population. The data would suggest that the source of the lead was consistent throughout the period of the study.

Table 2 shows the distribution of lead results within the school population based on counties and groups of counties as described above for chi-squared analysis.

• a.

  Frequency of Detection – Every county had at least one sample with a lead concentration above the DLR. At one extreme, Kern County had 113 samples with lead concentrations above the DLR out of 1,077 total samples or 10% and at the other extreme was Sonoma County, which only had six (6) samples with lead concentrations above the DLR out of 651 samples, or 0.9%. The calculated chi-squared value was 156, which is greater than the critical value for a = 0.05 with 23 degrees of freedom which was 35. This result means that the observed differences in frequency of lead detection between counties were significant. The CI Group, Kern, NC Group, San Bernardino County, and Sonoma County appear to be the counties that show the largest difference from the state-wide population as a whole based on the chi-squared values. All counties had between 1, and 10% of their samples had lead concentrations above the DLR, 16 of the 24 counties and groups had a detection frequency of between 3 and 7%. With a few exceptions, the detection frequency was largely uniform throughout the state.

• b.

  Frequency of Exceedance –Every county had at least one sample with a lead concentration above the AL except Sonoma County, which had none. The calculated chi-squared value was 103, which was greater than the critical value for a = 0.05 with 23 degrees of freedom, which was 35. This result means that the observed differences in frequency of lead detection between counties are significant. Kern County, NC Group, Orange County, San Diego, and Sonoma County appear to be the counties that show the largest difference from the state-wide population as a whole based on their chi-squared values. All counties except Sonoma County were between 0.5% and 2.7%.

Table 4 shows the distribution of lead results arranged by county groups based on the type of sample location. This table shows considerable differences among the different counties as to what type of sample location was tested. State-wide, 64% of all samples were collected from drinking fountains. However, in Santa Clara County, only 33% of all samples were from drinking fountains, while in San Francisco County, it was 81%, with Los Angeles County having 62%. In Stanislaus County, 47% of all samples came from sinks, while in San Diego County, the figure was only 3.7%. Similar situations exist for food preparation areas, bottle fillers, and distribution system samples, to say nothing of the unidentified samples. At least some of the differences between the median concentration of lead and the frequency of detected and exceeded samples are due to the samplers' very different sampling practices in different counties.

Table 5 shows the distribution of lead results arranged by county groups but for only drinking fountains.

• a.

  Frequency of Detection for Drinking Fountains – Every county had at least one sample with a lead concentration above the DLR in drinking fountains. The calculated chi-squared value is 112, greater than the critical value for a = 0.05. This result means that the observed differences in frequency of lead detection between counties are significant.

• b.

  Frequency of Exceedance for Drinking Fountains – Every county had at least one sample with a lead concentration above the AL except Sonoma County, which had none. The calculated chi-squared value is 100, greater than the critical value for a = 0.05 with 23 degrees of freedom. This result means that the observed differences in the frequency of lead detection between counties are significant.

• c.

  Median Concentration of Lead in Detected Samples in Drinking Fountains – Orange County had the lowest median lead concentration at 6.3 mg/L, and San Joaquin County had the highest median concentration at 10 mg/L. There are 253 comparisons among the 23 county groups used in this study. The KW test was performed, and the H value was 62.5 with 22 degrees of freedom; the probability was much less than 0.05 (p < 0.001). There is a significant difference in the median lead concentration between the groups of counties. However, only eight of these comparisons showed a significant difference using Dunn's method. Comparing San Joaquin County and Orange County, the DOR was 353; the Q-value was 3.8, and p = 0.037. Similarly, the comparison between San Joaquin County and Riverside County showed a DOR of 353, a Q-value of 3.8, and p = 0.037. When compared to Orange County and Riverside County, San Mateo County had a DOR of 303 and a Q-value of 4.0, and p = 0.02. Similarly, the DOR between Alameda and Orange Counties was 341, Q = 4.9, and p <0.001.

• d.

  Median Concentration of Lead in Samples Exceeding the AL for Drinking Fountains – Ventura and Tulare Counties had the lowest median concentration of lead and samples with lead above the AL in drinking fountains at 20 μg/L, again ignoring Sonoma County, which had no samples with concentrations of lead above the AL, and Stanislaus County had the highest median concentration at 39 μg/L. It is worth noting that Sonoma County had the second-highest median lead concentration even though there were six samples with detectable amounts of lead. There are 253 comparisons among the 23 county groups used in this study. The KW test was performed, and the H value was 19.2 with 21 degrees of freedom; the probability was much higher than 0.05 (p = 0.57). So median lead concentration in samples with concentrations above the AL was effectively uniform in all counties, except Sonoma.

• e.

  Correlation between bicarbonate, chloride, ratio, and pH against Median Concentrations of Lead and the Frequency of Detection and Exceedance. The median concentration of lead for samples with a lead concentration above the DLR for each county group was matched with the median value for bicarbonate, chloride, the ratio of bicarbonate to chloride and pH using the SROC. No significant correlations were found. The correlation coefficient (R2) was less than 0.2 in all cases and the p-value was more than 0.05, although the correlation coefficient for the median value for chloride and the frequency of results above the DLR was almost significant (R2 = −0.41, p = 0.053).

• f.

  Similarly, when the same test was performed using the median concentration of the samples with lead with concentrations greater than the AL, there were also no significant correlations. The correlation coefficient (R2) was less than 0.2 in all cases, and the p-value was more than 0.05.

The data shown in Tables 4 and 5 suggest that there might be an important difference between samples collected from drinking fountains and other sample locations. Table 6 shows the distribution of lead results arranged by county groups but only samples collected from sinks and food preparation areas. For this part of the study, the results from sinks and food preparation areas were combined as they appear to have very similar characteristics in terms of median lead concentrations and frequency of detection and exceedance. These results will also allow a larger number of samples which will increase the power of the statistical tests.

• a.

  Frequency of Detection – Every county had at least one sample with a lead concentration above the DLR. The calculated chi-squared value is 178, greater than the critical value for a = 0.05. This result means that the observed differences in frequency of lead detection in samples with results above the DLR between counties are significant.

• b.

  Frequency of Exceedance – Every county had at least one sample with a lead concentration above the AL except Sonoma County, which had none. The calculated chi-squared value is 182, greater than the critical value for a = 0.05. This result means that the observed differences in the frequency of lead exceedance between counties are significant.

• c.

  Median Concentration of Lead in Detected Samples – The median lead concentration in samples collected from sinks and food preparation areas with detected quantities of lead was 8.8 mg/L. Kern County had the lowest median lead concentration at 5.5 μg/L, and Santa Clara County had the highest median concentration at 14 μg/L. There are 253 comparisons among the 23 county groups used in this study. The KW test was performed, and the H value was 36.5 with 22 degrees of freedom; the probability was much less than 0.05 (p < 0.027). There is a significant difference in the median lead concentration between the groups of counties. However, no two counties showed a significant difference using Dunn's method.

• d.

  Median Concentration of Lead in Samples Exceeding the AL – Ventura and Tulare Counties had the lowest median lead concentration at 20 mg/L, again ignoring Sonoma County, with no samples with lead concentrations above the AL, and Stanislaus County had the highest median concentration at 39 mg/L. It is worth noting that Sonoma County had the second-highest median lead concentration even though there were six samples with detectable amounts of lead. There are 253 comparisons among the 23 county groups used in this study. The KW test was performed, and the H value was 13.8 with 21 degrees of freedom; the probability is not higher than 0.05 (p = 0.88). So median lead concentration in samples with concentrations above the AL was effectively uniform in all counties, except for Sonoma County.

• e.

  The median concentrations for bicarbonate, chloride, the ratio of bicarbonate to chloride, and pH were compared with the median lead concentration for samples with results above the DLR and the AL against the median concentrations of the four water quality parameters listed above. No significant correlations were found for these combinations for sinks and food preparation. The correlation coefficient (R2) was less than 0.2 in all cases, and the p-value was more than 0.05.

• f.

  The median concentrations for bicarbonate, chloride, the ratio of bicarbonate to chloride, and pH were compared with four variables associated with results from sinks and food preparation areas using the SROC. The median lead concentration for samples with results above the DLR, the median lead concentrations for samples results above the AL, the frequency of results above the DLR, and the frequency of results above the AL were all compared pair-wise.

• g.

  In contrast, when the frequency of results above the DLR for sinks and food preparation areas was compared to these four values using the SROC, two significant correlations were found. The correlation coefficient for chloride and the percent measured above the DLR was −0.55, which was significant (p = 0.006), while the correlation coefficient for the ratio of bicarbonate to chloride with the percent detected above the DLR was 0.45 (p = 0.032).

• h.

  Three significant correlations were found when the four chemical variables were compared to the frequency of results above the AL for sinks and food preparation areas. The correlation coefficient for chloride and the percent exceeded was −0.5, which was significant (p = 0.014); for pH and percent exceeded, the correlation coefficient was 0.50, which was significant (p = 0.017), and for the ratio of bicarbonate to chloride, it was −0.44 p = 0.04.

The data showing the samples with lead concentrations above the AL is somewhat misleading. As noted above, samples with lead results above the AL were supposed to be resampled. However, 205 locations that produced samples with concentrations above the AL were not resampled. Of the 178 sample locations resampled, 68 had a second result above the AL, 61 had a result below the DLR, and 94 had a value greater than the DLR but below the AL. 110 samples locations had a second result above the DLR but less than the AL. These should have been resampled but sometimes were not. In the third round, there were 96 samples collected. 48 were <DLR, 20 were above the AL, and 28 were above the DLR but below the AL. The data are summarized in Table 7. While there were only 473 samples in all rounds of sampling which had lead concentrations greater than the AL, there were only 383 samples with results greater than the AL. Two (2) locations had concentrations below the DLR but in later rounds had results above the AL. In other cases, after two positive samples, the water dispenser was replaced and then sampled some more. Suffice it to say, samples above the AL represent a complex set of results that do not correspond to one sample to one location, as is generally the case for results less than the DLR or above the DLR below the AL. However, the number of involved samples is very small compared with the entire study population. As an interesting note, one PWS found elevated lead at a school. Their staff collected three follow-up samples, each of which was 250 mL. Only the first of the three samples had lead above the DLR.

The data in Table 1 show that the frequency at which samples had results above the DLR or the AL and median concentration of lead measured was consistent throughout the period of the study and the same with the median lead concentrations among the samples above the DLR and AL. There appeared to be little difference between detection frequency and exceedance among the various counties.

Most had a frequency similar to the state-wide average. There were apparent exceptions; Sonoma County appears to have a much lower frequency of samples with lead above both the DLR and AL. Sonoma County had six (6) samples out of 651 with results above the DLR and no samples above the AL. This result is quite a bit below the expected rates based on the state-wide frequencies and the frequencies of other county groups. The ratio of observed to expected results was near 6:1. The CI county group, the NC county group, and San Bernardino County appeared to have a higher than expected frequency of samples with concentrations of lead above the DLR, although in these cases, the ratio was less than 2:1. For results above the AL, Sonoma County was different from every other county as it had none. The NC group had twice as many results above the AL as it should have based on the state-wide frequency, and San Francisco, San Joaquin, and San Mateo appeared to have a slightly higher frequency of results above the DLR, around 3:2, while Orange and San Diego Counties seemed to have lower frequencies. However, while the differences seem to be statistically significant for these few counties, they are not large in an absolute sense.

Suffice it to say, for any given county or group of counties, there does not appear to be any connection between the frequency at which sample concentrations exceeded either the DLR or AL and the median concentration of lead measured. Where differences did exist, they were not large.

The internal components that delivered water mainly were entirely plastic or stainless steel.

Interestingly, the median concentration is 9.9 mg/L among those seven samples, higher than the other groups except for the distribution system. This result is analogous to Sonoma County's situation, which had the lowest number of samples with lead above the DLR, but those samples had the second-highest median lead concentration among the counties. Probably of greatest significance is that the population of drinking fountains had a detection and exceedance frequency of about half that of the population of sinks and food preparation areas. There are significant differences between the type of brass used in drinking fountains and the brass in sinks and faucets. It is known that brasses with high zinc content (‘yellow brass’) have a much higher tendency to undergo dezincification than brasses with low zinc content (‘red brass’). It has also been shown that lead is released by yellow brasses during dezincification, particularly in the earlier phases. Drinking fountains are designed to withstand much greater mechanical stress than are kitchen and bathroom faucets and so probably used red brass for that reason. It seems clear from the results in Table 4 that there are substantial differences between the counties in terms of how the type of sample locations were sampled. State-wide, 64% of samples were collected from drinking fountains, 12% from sinks, and 11 from food preparation areas. Previous research has shown that the composition of brass has a significant impact on the corrosion by-products released, as does pH (Lytle & Schock 1996).

Notably, only 48% of samples collected in Stanislaus County were drinking fountains, and 47% were collected from sinks, plus 15% from food preparation. In contrast, in San Francisco County, 81% of the samples were collected from drinking fountains, and only 7.5% and 3.7% of samples were collected from sinks and food preparation areas, respectively. The differences between these different types of sample locations are significant. This difference could account for at least some of the differences in the counties.

Counties with a greater frequency of sampling drinking fountains might reasonably be expected to have higher frequencies of samples with lead concentrations above the DLR and AL.

The difference between counties can be partially attributed to the type of sample locations. However, another critical variable appears to be water chemistry. There was a strong and significant correlation between the chloride concentration and the detection frequency of lead in samples. This relationship was also true for the ratio of bicarbonate to chloride and pH. All three of these variables likewise had a significant correlation with exceedance frequency.

However, these correlations were only found with the samples collected from sinks and food preparation sample locations, not from drinking fountains. Samples collected from drinking fountains showed no such correlation. This study's data suggest a significant physical difference between the brass found in drinking fountains, which might be red brass, and the brass found in sinks and food preparation locations, which might be yellow brass. Perhaps it is the case that yellow brass, which is more susceptible to dezincification than red brass, is more influenced by differences in water chemistry. Specifically, different concentrations of chloride, different ratios of bicarbonate to chloride, and different pHs favor dezincification and lead leaching in yellow brass. If this is true, the corollary would be that the rate at which drinking fountains made of red brass dezincify and leach lead are not impacted by the same chemical parameters. They generally leach less lead than the yellow brass fixtures, and the rate is independent of the four water quality parameters measured in this study.

The results from Sonoma County seem distinct from the other counties. It is the only population of results with not a single result above the AL. There were only six (6) results above the DLR, the next closest was the three counties in the EC group, which had 17, but Sonoma had 651 total samples while EC had 239. This result meant that only 0.9% of all samples had results above the DLR, less than a third of the frequency of the next closest county, which was Sacramento, at 3.2%. This result is the lowest observed frequency of detections among all counties.

This same pattern for Sonoma County is valid for the drinking fountains, where only 0.5% of samples were above the DLR, and among the sinks and food preparation samples, the frequency was 2.1%. While the median concentration of lead for samples with lead concentrations above the DLR was a bit on the high side compared with other counties and the State-wide median, 10.2 mg/L, the mean concentration was by far the lowest at 9.4 mg/L, and it was the only county where the mean concentrations were less than the median. It was also the only county or county group with no lead results above the 95th percentile. The standard deviation was only 2.3 mg/L, less than half of the next closest county, San Diego, at 7.8 mg/L. Sonoma County also had the lowest maximum concentration of lead reported, 11.7 mg/L, less than a quarter of the next closest county, Sacramento, at 60 mg/L.

Sonoma County stands out as having the lowest median pH among all counties at 7.29, the lowest 25th percentile at 7.00, and the lowest 75th percentile at 7.60. The ratio of bicarbonate to chloride is 8.16, which is the fourth-highest among the counties. So it might appear as if the unusual results in Sonoma County could result from beneficial water chemistry and a selection of sample locations that favored lower rates of dezincification and lead leaching. Overall, there were significantly fewer samples with reportable quantities of lead collected in Sonoma County, and the concentration of the lead in those samples was also lower. The PWS that used three 250 mL samples to follow-up a sample showed that a smaller sample size is more effective at identifying lead leaching when brass in the fixture is the source of lead.

It seems reasonable to observe that when schools are being sampled, greater guidance needs to be given to location selection. As can be seen, there are vast differences in how locations were selected, which had a significant impact upon results. More focus needs to be placed on locations where students actually consume water, for example drinking fountains, and less in bathroom sinks. The 1 liter sample size seems a poor choice as the lead (and other metals) are found in the only first 250 mL (Kimbrough 2001; Kimbrough 2007; Triantafyllidou et al. 2021). The greatest risk of lead exposure is from the corrosion of LSL, as was shown in Flint Michigan, which the first draw sampling approach does not capture.

During this study, 43,803 samples were collected from 7,058 schools in 864 school districts from every county in California between 2017 and 2020. Overwhelmingly, the results were uniform, 95% of the samples were less than the DLR, and just under 1% were above the AL. Among the small number of samples that did have reportable lead concentrations, the results were almost entirely uniform. The concentrations were the same from one county to the next. While a few counties might have had higher or lower concentrations of lead that were different to a statistically significant degree, this was only among the small minority of samples that had results above the DLR, and the differences, significant though they were, were slight.

What was clear, however, was that samples collected from drinking fountains had significantly lower numbers of results above the DLR and AL than samples collected from sinks, food preparation areas, and the distribution system of local PWSs but higher frequencies than bottle fillers. It seems possible that this difference results from the use of different types of brass.

It is possible that the drinking fountains were made from red brass, while the faucets in sinks and food preparation areas were made of yellow brass. Yellow brass is more subject to dezincification and may, thus, release lead more frequently than red brass. pH, chloride, and the ratio of bicarbonate to chloride appears, to influence the rates of dezincification and lead leaching, at least in some types of brass fixtures.

All relevant data are available from an online repository or repositories. https://www.waterboards.ca.gov/drinking_water/certlic/drinkingwater/leadsamplinginschools.html

The authors declare there is no conflict.