THE EFFECTIVENESS OF pH AND ALKALINITY ADJUSThIENTS INREDUCING LEAD AND COPPER LEVELS IN RECHLORINATED AM)CHLORAMINATED TAP WATERbyKenneth C.H. ChanB.A.Sc.(Honours), U.B.C., 1992A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF APPLIED SCIENCEinTHE FACULTY OF GRADUATE STUDIESDepartment of Civil EngineeringWe accept this thesis as conformingto the equired standardTHE UNIVERSITY OF BRITISH COLUMBIASEPTEMBER 1994© Kenneth C.H. Chan, 1994In presenting this thesis in partial fulfillment of therequirements for an advanced degree at the University of BritishColumbia, I agree that the Library shall make it freely availablefor reference and study. I further agree that permission forextensive copying of this thesis for scholarly purposes may begranted by the head of my department or by his or herrepresentatives. It is understood that copying or publication ofthis thesis for financial gain shall not be allowed without mywritten permission.Department ofThe University of British ColumbiaVancouver, CanadaDate_____________(Signature)ABSTRACTThe effectiveness of pH and alkalinity adjustments in reducing copper and leadlevels in rechiorinated and chioraminated tap water was assessed for a study area inGreater Vancouver. Standing cold water, running hot and cold samples were collectedfrom 105 houses that were located in the study areas. The samples were analyzed inthe laboratory for lead and copper concentrations, pH, and alkalinity.After following an complex scheme of data manipulation, sorting, and statisticaltesting, comparisons of the copper and lead levels between the study areas were made.This research study found that pH and alkalinity adjustments were definitelyeffective in reducing the copper levels in rechlorinated and chioraminated tap water.Lead levels were also reduced, but the magnitude of the reduction was statisticallyinsignificant. The effects of pH and alkalinity adjustments on copper and lead levelsdid not appear to be different for rechlorinated and chloraminated tap water.Compared to houses with copper plumbing, houses with plastic plumbing and no-leadsolder had extremely low levels of copper or lead in the tap water. This study alsofound that the age of a house, which was fitted with copper plumbing and was less than15 years old, was not a significant factor in affecting lead concentrations.11TABLE OF CONTENTSABSTRACT iiLIST OF TABLES vfflLIST OF FEGURES ixACKNOWLEDGMENTS x1. INTRODUCTION 11.1 Greater Vancouver’s Drinking Water 11.2 GVRD Initiatives 11.3 Study Area & Preliminary Results 21.4 Objective and Scope of Study 32. BACKGROUND AND LITERATURE SEARCH 42.1 Health Impacts of Lead 42.2 Health Impacts of Copper 42.3 Copper and Lead Regulations 52.4 Principles of Metallic Corrosion 52.4.1 Uniform Corrosion 62.4.2 Galvanic Corrosion 62.4.3 Crevice Corrosion 62.4.4 Pitting Corrosion 72.4.5 Concentration Cell Corrosion 72.4.6 Selective Leaching 72.4.7 Erosion Corrosion 72.4.8 Stress Corrosion 82.4.9 Microbiologically Induced Corrosion 82.4.10 Chemical Corrosion Mitigation 81112.5 Previous Studies in Copper and Lead Levels 92.5.1 Seattle, WA 92.5.2 Boston, MA 102.5.3 Glasgow, Scotland 102.5.4 Portland, ME 102.5.5 Covewood Lodge, NY 112.6 Tap Water Sampling vs. Laboratory Simulations 112.7 Sources of Copper in Tap Water 122.8 Sources of Lead in Tap Water 122.9 Factors Affecting Tap Water Sampling Results 132.10 Factors Affecting Lead Leaching 162.11 Factors Affecting Copper Leaching 192.12 Plastic Pipes vs. Copper Pipes 192.13 Quality Control in Laboratory Tests 202.14 Reporting of Low Level Data 212.15 Considerations in Experimental Design 212.16 Sampling 232.17 Common Assumptions Made in Statistics 252.18 Measuring Central Tendency 262.19 Measuring Spread 272.20 Estimate 282.21 Coefficient of Variation 282.22 Skewness 282.23 Central Limit Theorem 282.24 Hypothesis Testing 292.25 Constructing Power Curves 302.26 Statistics of the Maximum 30iv2.27 Systems of Measurement 312.28 Nonparametric Testing 312.29 Kolmogoroff-Smirnoff Comparison of TwoIndependent Samples 322.30 Median Test 322.31 Kruskal-Wallis H-Test 322.32 Confidence Intervals for the DifferenceBetween 2 Means 342.33 Test for Equal Variance 342.34 Spearman Rank Correlations 352.35 Regression 362.36 One-way ANOVA 362.37 Theory of Causation 383. EXPERIMENTAL METHODS 403.1 Experimental Design 403.1.1 Controllable Factors 413.1.1.1 Direct Control 423.1.1.2 Indirect Control 433.1.2 Uncontrollable Factors 443.2 Arrangements with Governmental Agencies 463.3 Bottle Preparation 473.4 Sampling Package 493.5 Sampling Procedures 493.6 Participation of Homeowners 503.7 Bottle Pick-up 513.8 Lab Testing 513.8.1 Instruments 52V3.8.2 Testing Scheme 533.8.3 Data Recording 534. RESULTS AN]) DISCUSSION 554.1 Summary of Returned Bottles 554.2 Data Cross-check by GVRD 574.3 Reporting of Data 584.4 Data Used for Statistical Analysis 584.5 Non-Parametric Testing 604.6 Other Statistical Problems 614.7 Identifying Flushed Standing Samples 624.7.1 Rejected Samples for n,y,s,w 654.7.2 Rejected Samples for d 654.8 Simple Tests 664.9 Kruskal-Wallis H-Test (Copper) 674.10 Confidence Intervals for the DifferenceBetween 2 Means 684.11 Median Test 684.12 Test for Equal Variance 684.13 Compliance with Regulations 694.14 Plot of Lead vs. Alk and pH 704.14.1 Spearman Rank Correlations 714.14.2 Regression of Alk and pH 714.14.3 Filling in Missing Data Points 734.15 Copper vs. pH 744.l6Leadvs.pH 754.17 ANOVA (Lead) 764.17.1 Using One-Way ANOVA 77vi4.17.2 One-Way ANOVA grouped by area 784.17.3 One-Way ANOVA grouped by pH rangewith age as covariate 784.17.4 Residuals 794.18 Testing of Oldcopper and Plastic 824.19 Comparison of Flushed Cold and Flushed Hot 835. SIJIVIIvIARY AND CONCLUSIONS 865.1 The Effects of pHIAlk Adjustment 865.2 Significance of Other Factors 865.3 Recommendations 876. REFERENCES 88APPENDICESAppendix A Sampling Locations 92Appendix B Written Sampling Instructions 93Appendix C Cartoon Sampling Instructions 94Appendix D Questionnaire 95Appendix E GVRD Data Crosscheck 96Appendix F Raw Data 97viiLIST OF TABLES1 Preliminary Copper and Lead Study Results (From 1988 to 1991)22 Typical Data for a Single-Factor Experiment 373 The ANOVA Table for the Single-Factor, Fixed Effects Model 384 Results of Bottle Treatment Study 485 Summary Information of Returned Bottles 556 Detection Limits of Measurements 587 Summary of Simple Statistical Evaluation 668 Percentage Reduction from Delta (mean values) 679 Difference Between 2 Means 6810 Test for Equal Variance 6911 Regulation Compliance (as per EPA guidelines, 1991) 6912 Regression Analysis of pH versus Alkalinity 7213 Comparison of Oldcopper and Plastic with Main Data 8314 Summary of All Statistical Testings 86viiiLIST OF FIGURES1 Comparison of Data Types 602 Rejection Procedure 643 3D Plot of Lead, Alkalinity, and pH 704 Regression Plot of pH versus Alkalinity 735 Filling-in Missing Data for pH versus Alkalinity Data 746 Copper vs. pH 757 Leadvs.pH 768 Normal Probability Plot of ANOVA of Lead 809 Student t vs. Lead Estimate 8110 Cook’s Distance vs. Lead Estimate 8211 Quantile-Quantile Plot of Cold & Hot Running Copper 8412 Quantile-Quantile Plot of Cold & Hot Running Lead 85ixACKNOWLEDGMENTSThis research project was funded by a Natural Sciences and Engineering ResearchCouncil (NSERC) scholarship, by the UBC Civil Engineering Department, and by theGreater Vancouver Regional District (GVRD). I would like to thank Dr. DonMavinic, and Prof. Jim Atwater of UBC, Mr. Doug Neden, Mr. Mark Ferguson, Ms.Judy Smith, and Mr. Indergit Singh of the GVRD for their help in the planning andanalysis of this project. I want to thank the staff members at the City of Surrey and theCorporation of Delta for their assistance. My sincere appreciation to all thehomeowners in North Delta, Newton, and South Surrey, who participated in this study,for taking the time to collect the water samples. I would also like to thank Ms. SusanHarper, Ms. Paula Parkinson, and Ms. Jufang Zhou of UBC, and Mr. Peter Zadoroznyof the GVRD for their expert help in laboratory sample analysis. Soli Deo Gloria.x1. INTRODUCTION1.1 Greater Vancouver’s Drinking WaterThe water supply of the GVRD comes from three watersheds on the north shoreof the Burrard Inlet. These watersheds, covering a total of 585 squared km of land, arethe Capilano, Seymour, and Coquitlam. Each watershed has its own major storagereservoir. The Cleveland Dam is constructed on the Capilano River, the Seymour FallsDam on the Seymour River, and the Coquitlam Dam on the Coquitlam River.The drinking water supply of the study areas for this research project mostlycomes from the Seymour water source. The water from the Seymour reservoir isuntreated except for the addition of chlorine. In 1992, the chlorinated source waterwas characterized by having an average pH of 6.0, dissolved oxygen content of 10.2ppm, total chlorine residuals of 0.9 ppm as chlorine, total dissolved solids of 16 ppmas CaCO3,and total alkalinity of 2.3 ppm as CaCO3 (GVWD, 1992). The naturalcopper and lead levels of the water in the reservoir was negligible.1.2 GVRD InitiativesA water with these characteristics is considered to have a high corrosionpotential on metals. Since 1984, the GVRD has studied the bacteriologicaldisinfection, and disinfection by-products of the Greater Vancouver’s drinking water(GVWD, 1992). Studies have also been done on the corrosion aspects of the waterquality improvement program of the GVRD. Corrosion can lead to green staining ofporcelain fixtures and basins in bathrooms. It can also increase the copper and leadlevels of the tap water. The economic impact of water corrosion is substantial.Recognizing that the present water supply system is inadequate to meet futuredemands in Greater Vancouver, GVRD is planning a comprehensive program that willincrease storage and transmission capabilities. At the same time, the GVRD’ s DrinkingWater Quality Improvement Plan (DWQIP) is seeking to improve the quality of the1drinking water in order to consistently meet the Guidelines for Canadian DrinkingWater Quality.1.3 Study Area & Preliminary ResultsIn 1988, a monitoring study was initiated in South Surrey, Newton, and Deltato study the effects of secondary disinfectants on the water supply. Newton receivedsecondary chlorine treatment, while South Surrey received chioramination. Theeffectiveness of the secondary treatments on controlling bacterial growth werecompared to the results from the control area in Delta. Besides the disinfectioninitiatives, the study areas were also used to study the dynamics of corrosion control.The pH and alkalinity of the Newton and South Surrey areas were raised in order tomake the water less corrosive. From 1988 onwards, the pH and alkalinity adjustmentshave been in continuous operation. It was hoped that collecting tap water samples fromvarious locations in the three study areas would reveal whether or not pH and alkalinityadjustments were effective in reducing the lead levels of the drinking water.About 15 tap water samples were collected in 1988, just prior to theintroduction of pH and alkalinity adjustments. Three years later, in 1991, about thesame number of samples were again collected. Of the samples, 6 (2 in Newton, 2 inSouth Surrey, and 2 in Delta) were collected from the same houses both in 1988 and in1991. The data almost uniformly supported the conclusion that copper levels in the 1liter standing samples were lowered after the introduction of pH and alkalinityadjustments. However, the lead data yielded inconclusive results (Table 1).Table 1 Preliminary Copper and Lead Study Results (From 1988 to 1991)Sample # - % change in copper - % change in Lead1 -78 -322 -85 +1773 -89 -384 -95 05 +16 +126 -9 +612For half of the 6 houses sampled, the lead level increased in 1991 from 1988, whereasit decreased for the other half of the houses.1.4 Objective and Scope of StudyThe 6 data points that were collected in 1991 and 1988 did not give enough datato do any meaningful statistical analysis on the lead data. The main objective of thisstudy, then, was to collect enough samples from the study areas so that meaningfulstatistical statements could be made on whether or not the pH and alkalinity adjustmentprogram was effective in reducing lead levels in the tap water. Secondly, the copperlevels of the collected samples were also analyzed to confirm the conclusion of theprevious study that the copper level is reduced after the pH and alkalinity of the wateris raised.32. BACKGROUND AND LITERATURE SEARCH2.1 Health Impacts of LeadLead is not a required trace metal for bodily functions. When taken up, itaccumulates in the body. Lead is potentially poisonous to human beings, especiallywhen there is continuous exposure to small amounts of it. Therefore, trace levels oflead in drinking water could become a problem to humans.Lead has the most harmful effects on infants and young children, followed byadult females, and has the least effect, comparatively, on adult males (NationalAcademy of Sciences, 1982). Upon entering the human body, some of the lead isabsorbed in the blood stream and is excreted from the body through the kidneys and theintestinal tracts. Lead can also be deposited in soft tissues, hair and nails. However,most of the lead that stays in the body is found in the skeleton (Department of theEnvironment, 1977). Lead has a half life of 2 to 4 weeks in blood, about 4 weeks insoft tissues, and 27.5 years in bone (World Health Organization, 1984). Lead canenter the body through water, food, or air. It has been shown that the intake and theuptake of lead in drinking water can be a significant percentage of the total intake anduptake (Drill, 1979). When the amount of lead that has accumulated in the body hasreached acute levels, the victim could suffer tiredness, abdominal discomfort,irritability, and anaemia (World Health Organization, 1984).2.2 Health Impacts of CopperIn contrast to lead, copper is an essential trace metal that is required by thehuman for normal physiological functions. Long term exposure to copper at low levelsdoes not have documented toxic effects on humans. But, the intake of very largedosages of copper could lead to severe mucosal irritation and breakdown, capillarydamage, liver and renal damage, and central nervous system irritation, followed bydepression (World Health Organization, 1984). When the body detects that it has anexcessive amount of copper in its system, the person is involuntarily induced to4vomittmg. This action helps to reduce the amount of copper in the body. Hence,copper poisoning is actually rarely found in humans. Copper is also of concernbecause it is toxic to fish at relatively low levels. In addition, at high levels, theformation of copper salts can cause blue green staining of plumbing fixtures.2.3 Copper and Lead RegulationsThe USEPA Lead and Copper National Interim Primary Drinking WaterRegulations (1991) specified that lead levels in 1 liter standing samples should notexceed 15 ppb in 10% of the samples taken at the tap. Copper levels should not exceed1.3 ppm in 10% of the samples. The samples are to be taken from the kitchen taps of anumber of targeted residences. The public water utilities must show that the watersupply does not exceed the prescribed levels for both copper and lead. Depending onthe size of the distribution system, a number of samples have to be collected andanalyzed in order to determine whether the supplied water meets the copper and leadregulations. When the regulations are shown to be exceeded, then the water utilitymust implement a corrosion control program. Alternatively, the utility mustdemonstrate that the existing treatment program has already maximized the level oftreatment possible.In Canada, the Guidelines for Canadian Drinking Water Quality proposedifferent regulations for lead levels. The maximum acceptable concentration (MAC)for lead is 10 ppb (Health and Welfare Canada, 1989). However, this guideline appliesto a thoroughly flushed sample. There is no MAC for copper, but it is recommendedthat copper level be kept below 1 ppm. It is not specified whether this aestheticobjective applies to standing samples or flushed samples.2.4 Principles of Metaffic CorrosionCorrosion of metals means that the metal is gradually destroyed by chemical orelectrochemical reactions with the environment. In almost all cases, metal corrosion inan aqueous environment is caused by electrochemical processes (Obrecht and Pourbaix,51967). An electrochemical cell must have four important elements in order for thecorrosion reaction to take place. First, there must be an anode; this is the place wherethe metal is oxidized. Electrons are generated from this place and are passed unto thecathode, which receives the electrons. At the cathode, corrosive substances, such asdissolved oxygen, chlorine, and hydrogen ions are reduced. Between the anode and thecathode, a conductor must exist. This is usually the metal pipe, which allows theelectron to move from the anode to the cathode. Fourthly, an electrolyte must bepresent to provide a medium for moving the various ions involved in the oxidizing andreducing half reactions.2.4.1 Uniform CorrosionUniform corrosion usually occurs in copper and lead pipes. In this form ofcorrosion, the corrosion uniformly penetrates the entire metal surface for a certaindepth. Any given site on the metal surface could be anodic at one moment and becathodic at another (Snoeyink and Kuch, 1985). This phenomenon is promoted whenthe metal being corroded is immersed in acid solutions or in water with high totaldissolved solids (TDS) and with high electrical conductivity.2.4.2 GalvanIc CorrosionGalvanic corrosion takes place when two metals of different electrode potentialscome into contact with each other. The metal with the more positive electrodepotential is the sacrificial anode and the metal with the more negative electrodepotential becomes the cathode. Galvanic corrosion is a problem when copper pipeconnections are soldered by lead solders. The lead solder is the anode, where the leadis converted into soluble ionic species (Oliphant, 1983).2.4.3 Crevice CorrosionThreaded junctions, screwed joints and inverted seams are places where crevicecorrosion could occur. There is poor circulation and oxygen depletion at thesecrevices. Halides and sulfates can migrate into these crevices and combine with6dissolved metals to form strong acids. After the local corrosion process is initiated, thecrevice becomes bigger, and the rate of corrosion accelerates.2.4.4 Pitting CorrosionPitting corrosion or localized corrosion is a process where the anode remains ata fixed location. This type of corrosion generally occurs on ferrous surfaces. Thefixed anodes could be imperfections in the metal or its oxide film. Regions of morehighly stressed metal can also serve as permanent anodes. Because the anode isconfined to a fixed spot, localized corrosion at that spot will eventually lead to a pinhole leak in the pipe. Pitting corrosion also occurs where sediment has built up onbottom of copper pipe.2.4.5 Concentration Cell CorrosionIn a concentration cell corrosion, different parts of the same metal are subject todifferent environmental conditions and have, therefore, different electrode potentials.The difference of electrode potential between different parts of the metal could becaused by minute differences in pH or concentration of dissolved oxygen or hydrogenions.2.4.6 Selective LeachingWhere the supply water is soft and aggressive, the lead component in brass canbe selectively leached away. This leaves the copper with more pores and makes it softand brittle. A similar phenomenon happens when tin is selectively leached from bronzein soft and aggressive water.2.4.7 Erosion CorrosionErosion corrosion occurs when there is cavitation or impingement attacks. Atthe entrance to pipes, and sharp bends, such as joints and elbows of pipes, the velocityof the travelling water changes according to Bernoulli’s Law. The resulting changes inwater pressure at these locations may cause cavitation, where gas bubbles trapped in thewater collapse and produces high localized pressure on the pipe surface. Impingement7attacks occur when the gas bubbles directly strike the metal surface. The energy that isreleased is sometimes sufficient to breakup protection films that have built up on thepipe surfaces.2.4.8 Stress CorrosionDuring the threading of pipe ends, the cold working of the metal often causes ithave to dissimilar stress on different parts of the metal. This could lead to localizedcorrosion. The threaded ends of galvanized steel pipes are especially susceptible to thiskind of corrosion mechanism.2.4.9 Microbiologically Induced CorrosionMicrobiologically induced corrosion is, as yet, a poorly understood, butimportant form of corrosion. Bacteria can live on the surface of pipes. Nitrifiers, forexample, use the ammonia in the water as a source of energy. During biochemcialreactions, oxygen is used to oxidize the ammonia. Hydrogen ions are produced inthese reactions. The resulting acidic environment that is produced encourages thereaction rate of the other forms of corrosion mentioned. Where chloramination is thechoice of disinfectant, the availability of ammonia in the water system clearlyencourages the activities of nitrifying bacteria. Microbiological corrosion can also beinduced by iron and sulfur reducing bacteria.2.4.10 Chemical Corrosion MitigationThe majority of the corrosion mechanisms are influenced by such parameters aspH, dissolved oxygen, standing time, the buffering capacity of water, and the mineralcontent of the water. In general, a soft and aggressive water promotes corrosion, and aharder, less aggressive water decreases the potential for corrosion in the water supplysystem. Controlling corrosion, by making the water harder, is termed “neutralization”.Alternatively, a passivation approach may be employed, where inhibitors, such asphosphate and silicate inhibitors, help to form a protective ifim on the pipe surface toblock the contacts between the electrolytes, the anodes, and the cathodes. The8passivation approach is more expensive to operate than the neutralization approach.Using Zinc orthophosphate inhibitor appears to decrease copper corrosion but mayactually increase lead corrosion (MacQuarrie, 1993).The neutralization approach primarily involves adjusting the pH and alkalinityof the water to make it less aggressive. pH adjustment should always be accompaniedby alkalinity adjustment, in order to provide enough buffering capacity to limit pHfluctuations. Due to the lowering of pH from carbon dioxide absorption, the pH of thewater can drop after the neutralization step. The side effect of the neutralizationapproach are that raising the pH also increases the rate of THM formation anddecreases the effectiveness of chlorine disinfection.2.5 Previous Studies in Copper and Lead LevelsIn previous years, other cities have tried to solve their copper and lead corrosionproblems with pH and alkalinity adjustment programs. Based on the following studies,it has been shown that pH and alkalinity adjustments can reduce the lead levels at thetap by 35% to 75%.2.5.1 Seattle, WAThe Seattle Water Department (SWD) provides drinking water to over onemillion people living in the Seattle Metropolitan Area. The water supplies come fromthe Cedar and the Tolt Rivers, which are characterized by low pH and alkalinity. In1982, a corrosion control program was initiated to reduce corrosion and relatedaesthetic and economic problems. To the Cedar and the Tolt water supplies, 2 mg/L ofcalcium oxide was added. In addition, 9 mg/L of sodium carbonate was added to theTolt supply (AWWARF, 1990). The effectiveness of the treatment program wasextensively monitored throughout the areas serviced by the water supplies. Of thesampling sites, about half were randomly chosen, while the other half comprised ofhouses that made complaints about the water concerning rust stains, yellow water, and9metallic taste. Both standing and flushed samples were taken. The results of themonitoring program showed a lead reduction of 61 % to 68%.2.5.2 Boston, MABoston purchases its drinking water from the Metropolitan District Commission,which operates a surface water supply system. The water delivered to Boston had a pHbetween 5.9 to 6.8, alkalinity of 8 mg/L as calcium carbonate, and hardness of 12mg/L as calcium carbonate (Karalekas et. al., 1983). In 1977, after unsuccessfullytrying to control the lead corrosion problem by adding inhibitors, the pH of the waterwas raised to 8.5 by adding 14 mg/L of sodium hydroxide. Residences from areas thathad lead service lines were chosen for monitoring the lead levels at the tap. Some ofthese residences also had interior lead plumbing. The average lead levels were reducedby 73%, and the variability of the lead measurements also decreased significantly afterthe pH adjustment. The flushed samples exhibited highly skewed concentrationfrequency distributions.2.5.3 Glasgow, ScotlandThe water supply of Glasgow comes from Loch Katrine. The water ischaracterized by low alkalinity and a pH of 6.2. Due to the fact that there wereextensive lead domestic plumbings, 50% of random tap samples registered leadconcentrations in excess of 100 ppb (Richards et. al., 1984). Beginning in 1978, limewas added to the Milngavie Treatment Works. The pH of the water supply wasincreased from 6.3 to 7.8. Random tap samples collected after the pH adjustmentshowed a decrease in lead concentrations to the point where more than 80% of thesamples had less than 100 ppb of lead. In another part of the distribution system,Kings Park (which is close to the end of the distribution system), the pH was increasedto 9.0. The pH adjustments resulted in more than 83% of the samples from this areahaving lead levels less than 100 ppb.2.5.4 Portland, ME10The major source of water supply to the Portland Water District is the SebagoLake. In addition, three well systems supplement this water supply. In 1986, the pHof the water was raised to 8.3 through the addition of sodium hydroxide (AWWARF,1990). The reduction of copper levels in standing samples was 61% to 85%. Thereduction of lead levels was about 69%.2.5.5 Covewood Lodge, NYCovewood Lodge is a resort in upstate New York, and contains several cabins.A spring that has low pH and low alkalinity serves as the major water supply for thisresort. In 1981, a baffled limestone contactor was constructed to treat the water of afew cabins at this resort. The dissolved calcium carbonate raises the pH of the water to7.3 and the alkalinity to 28.5 mg/L (AWWARF, 1990). Over the next 2 years, 23samples were taken and analyzed. It was found that the average lead level for theuntreated cabins was 46 ppb and the average for the treated cabins fell to 18 ppb.2.6 Tap Water Sampling vs. Laboratory SimulationsIn order to assess whether a corrosion control strategy will improve the qualityof supplied water so that it complies with the USEPA copper and lead rule, a fieldstudy or a lab simulation can be conducted. A properly designed tap water samplingsurvey can determine the level of lead and copper in the distribution system, and theeffectiveness of a corrosion control program in reducing the level of lead and copper atthe tap. The significant disadvantage to this approach is that any corrosion controlexperiment could actually adversely affect the quality of the supply water. A saferapproach of determining the effectiveness of a corrosion control program is to do alaboratory simulation of the various corrosion control options. Pipe loops, and coupontesting have been used for a long time to assess corrosion rates and the relative impactsof various water treatment options. The major problem with using a laboratorysimulation is that we do not know enough about the sources of lead, and the11mechanisms for lead mobilization, in order to accurately simulate an actual waterdistribution system.2.7 Sources of Copper in Tap WaterCopper in tap water comes mostly from the leaching of copper from the watersupply system. Leaching of copper can occur in the transmission or the distributionsystem. However, most of the copper leaching takes place in the household plumbings.The copper can be leached from the copper pipes and faucets. Copper leaching frombrass faucets is the major source of trace copper in the tap water.Copper might also be found in significant concentrations in the source water.Copper can be introduced into water from natural and anthropogenic sources. Throughnatural erosion, copper particles can be picked up and deposited in water sources bywind. Infrequent natural phenomena, such as volcanic eruptions, can also extrude agreat deal of copper dust into the environment. Of the anthropogenic sources, copperis introduced into the environment in metal production, wood and fossil fuelcombustion, and waste incineration activities (Environment Canada, 1981).2.8 Sources of Lead in Tap WaterThe lead found in tap water could be from the water supply source, the watertreatment process, the transmission and the distribution system, or the service andhousehold plumbings. The following is a list of potential lead sources (adapted fromAWWARF, 1990).Water source and water treatment:• lead containing air pollutions, emitted by industries that are located near raw watersources• deposits of lead-bearing materials that are naturally found in the watershed• point sources of domestic wastewater or industrial discharges, located upstreamfrom the water intake, that might contain leadTransmission and distribution system:12• lead pipes in the distribution system• large water meters or flow detector checks that have a lead counter weight inside• lead-caulking compounds used to seal joints in the water mains• lead gaskets used as flanges to join large valvesService piping and household plumbing:• resetters for meters that used 50:50 lead/tin solder as the joining material• lead or lead-lined iron service lines or premise piping• lead goosenecks or lead pigtails• lead solder used to join copper service lines or copper premise piping• brass fixtures and fittings or pipes with high lead contents• water coolers that have lead components2.9 Factors Affecting Tap Water Sampling ResultsThe variation of tap water results is caused by a number of analytical, chemical,and physical factors. Uncertainties of instrument calibration, and random instrumentresponse errors are two of the most important analytical errors that could be made. Inaddition, there could be procedural errors, such as errors in dilution, and samplemanipulations. The presence of potential interferences in the samples could alsoincrease the analytical variability of data.The lead levels at the tap are affected by the alkalinity, pH, and dissolvedinorganic carbonate (DIC) of the supply water. For example, the equilibrium leadconcentration in the pH range of 6 to 8 could vary by a factor of 5 to 10 per pH unit(Schock, 1980, 1985). The actual amount by which the equilibrium concentrationvaries is dependent on the type of solid that forms on the pipe surface.Changes in any of the chemical characteristics of the water or in the chlorinationpractice, such as the chlorination dosage or the relative proportion of free andcombined chlorine, could cause a subsequent change in the corrosivity of the water.13The initiation of corrosion treatment to the water supply could result in changes to thesolubility and the adherability of the corrosion products to the pipe walls.Likewise, the standing time of a sample is important. The amount of time thewater is left standing in the pipe can affect the pH, chlorine residuals, dissolved oxygenlevel, temperature, calcium and magnesium hardness, and total and carbonate hardnessof the water.Physical factors can also influence the outcome of the lead levels at the tap.There are normally interconnecting lines within a plumbing system in a house. Thewater in the kitchen faucet is connected in some way to the faucets in the bathroom,utility room, and the exterior of the house. Any water usage from any of these faucetswill cause some mixing of the water in the plumbing system.When the faucet, from which the sample is taken, is turned on, a plug flowcondition is created. As the water is drawn through the pipe, the shape and the lengthof the original plug of water flowing through the pipe is altered by the friction of thewater against the pipe wall and by mixing due to turbulent eddies generated in thewater. In almost all houses, the pipe diameter changes from the point where theservice line connects with the house plumbing to where the interior plumbing connectswith the faucet. At these junctures, there could be alterations to the shape of theoriginal plug flow. The extent to which this occurs depends on the flow rate, thedistance between joints, and the size and interior condition of the plumbing system(AWWARF, 1990). The amount of a plug of standing water that can be recovered insample collection depends on the volume of samples taken relative to the diameter ofthe pipe, the size of the plug of standing water, and the degree of water mixing in thesystem. Therefore, even if the faucet from which the samples are taken has not beentouched, the sampled water could still be effected by the mixed water.The contribution of metals from the various parts of the plumbing system can beisolated by varying the volume of water sampled. The precision in identifying the14contribution from the faucets, for example, increases as the volume of water collectedis decreased. The internal volume of the faucets differs according to the faucet design.Most kitchen faucets are usually about 90 to 120 mL (AWWARF, 1990). Bathroomfaucets are smaller. Since standing samples are usually 250 to 1000 mL, measuring thecontribution of the metals from the faucets requires collecting samples that have lessvolume than the volume for the standing samples. In comparing data from onesampling study to another, it is important to note the size of the samples collected.It is important to consider the physical state of the lead which we want tomeasure. The popular mass transfer model of Kuch and Wagner (1983) deals with leadin its various aqueous forms of free ion, ionic complexes, and uncharged complexes.One study examined the size distribution of lead in tap water (AWWARF, 1990). Thestudy found that 65% to 84% of the lead was less than or equal to 0.4jim. Thedissolved lead species are of most concern to sampling studies because they are easilytaken up into the human body after being ingested.The Kuch and Wagner model predicts the amount of lead picked up duringsteady-state turbulent flow through lead pipe. It can also calculate the concentration oflead in the water under no flow conditions in the water. To use the model forpredicting lead concentrations in steady flow conditions, the experimenter must knowthe diameter of the pipe, the pipe length, the water temperature, and the volume rate ofwater flow. As well, an observed equilibrium lead level must be ascertained ahead oftime by analyzing field samples. An estimate for the mass transfer coefficient and thediffusion coefficient of the Pb + + must also made. However, the mass transfercoefficient is only important if the pipes have thick coats of scales that inhibit thediffusion of lead into the water.In addition to the dissolved forms of lead, particulate lead is present in waterdistribution systems as well. Lead can be adsorbed onto foreign particles, such as ironoxides, corrosion products, or calcium carbonate particles. There can also be15adsorption or ion-exchange with sediment materials, colloidal hydrous ferric, andmanganese oxides (AWWARF, 1990). The complicated reactions involving lead withother substances is probably regulated and influenced by the chemical characteristics ofthe supply water.2.10 Factors Affecting Lead LeachingThe contact of water with lead soldered joints makes the water nearby the solderenriched in lead. The contact between the solder and the water allows the lead in anylead containing solder to migrate into the water by the process of simple diffusion.Depending on the amount of time the water is in contact with the solder, and theamount of lead containing solder that is present in the plumbing system, variableamounts of lead can be leached into the water.Small differences in the amount of time water stands in a pipe containing leadcan contribute to considerable differences in the degree of lead leaching. Differencesof 10% to 30% in lead concentration is achieved in standing times that differ only 30 to60 minutes. Lead leaching occurs at a faster rate in pipes of small diameters (USEPAProject Report).Brass faucets contain lead that can leach out into the water. The amount of leadleaching from brass faucets depend on the stagnation (standing) time of the water in thefaucet. The leachability of lead from brass probably decreases with the age of thefaucet (Neff, 1987). The phenomenon is due to the fact that the leachable zone at themetal surface is depleted of lead, or because a passivating film is deposited on themetal surface (Sharrett, 1982; Britton et. aL, 1981). A study that showed very highlead concentrations from standing samples implicated the brass faucet as an importantlead source (Murrell, 1985).The liberation of lead from brass faucets is due to dezincification. In areaswhere the free energy of the alloy is above the average value, dezincification can occur(Oliphant, 1978). These abnormal areas are the result of crystallographic dislocations,16distortion of the normal atomic array due to casting or drawing of the alloy,inhomogeneities in the alloy, or the presence of impurities. At these places, the zincselectively dissolves and leaves the copper and any lead impurities behind. The contactbetween the lead impurities and the copper matrix can result in galvanic coupling andsubsequent lead oxidation.Numerous studies (Moore, 1973; Schaut, 1942; Gregory et. al., 1984) haveshown that there is an approximately 2 to 3 fold increase in lead solubility when thetemperature of the water is increased from 5 to 25 °C. Because temperature affectsvarious dissociation, solubility, and complexation reactions, the actual amount ofincrease in lead solubility will also depend upon the pH of the water and its carbonatecontent.Particulate lead can be formed through the precipitation of lead solids, whichmay deposit on the pipe surface as passivation film. However, if the velocity of thewater flow through the pipe is too high, these lead deposits might slough off.Relatively large pieces of solder can also be dislodged from the solder mass. Thesolder will travel down the pipe until it is stopped by a bend, elbow, restriction, or thescreen in the faucet. Through the mechanism of leaching, the continuous exposure ofthese particulate lead in the plumbing system can elevate the lead levels of tap watersamples.Many plumbing systems use components that are made of a variety of differentmaterials. The service lines could be made of lead, the interior plumbing of copper,and the solder of lead and tin. In these situations, when the different materials comeinto contact with each other, such as when copper pipes are soldered together withlead/tin solder, a galvanic corrosion current is produced (Lyon et. al., 1977). Thisleads to the dissolution of the metals. Solder is shown to be anodic relative to copperpipes. Water acts as a bridge between the solder and the copper, the two poles of thecorrosion cell. Galvanic corrosion at the solder joints can be a problem even if the17capillary joints are well made (Oliphant, 1983). The presence of chloride, and nitratecan increase the galvanic corrosion rates of soldered joints. The chloride penetratesand breaks down the protective ifims on the pipe surface. Nitrate stimulates corrosionactivities at places where the protective film is exposed. The shift in pH changes thesolubiity constants, hence, the degree of protection, of the protective films.Furthermore, the overall corrosion rate at a soldered joint is determined by theagressivity of the water supply.Not all galvanic corrosions are due to the plumbing system itself. At least onestudy has shown that copper in the water supply can deposit on lead pipes and create alocalized galvanic electro-potential cell (Britton et. al., 1981).A newer house, with predominantly newer copper plumbing systems, gives riseto higher lead levels than do older houses. Age has a marginal effect on cold first-flushlead concentrations. However, hot water lead levels appear to be unaffected by age.Cold first-flush lead levels appear to be the same for copper and plastic plumbingsystems. This result is the same for running hot water samples (Singh, 1990).Lead exists in its elemental form in plumbing materials. In alloys, such asbrass, the lead component is spread throughout the alloy matrix in particle form. Theelemental lead is oxidized to the 2 + valence state when it comes into contact withwater. This oxidation process enables the lead to become mobile and transportable intothe water. These oxidation processes usually take place at the anodic areas of corrosioncells.In potable water systems, dissolved oxygen and various chlorine speciesintroduced through disinfection are the most common kinds of oxidizing agents forlead. The oxidation reaction is promoted by increases in the dissolved oxygen content,by decreases in pH, and by the complexation of free lead ions by ligands such ascarbonate, hydroxyl, sulfate, and chloride (AWWARF, 1990). The effects of chlorinespecies on lead oxidation depends on the activities of hydrochiorous acid, hypochiorite18ion, chloramine species and chloride ion. One study showed that, under somecircumstances, chioramination can solubilize more lead than chlorination with freechlorine. However, the rate of corrosion due to chloramination is slower thanchlorination (Treweek, 1985). Also, the complexation of lead by hydroxyl andcarbonate ions predominate in normal situations (Schock, 1985).The rate of oxidation reactions goes up with increasing temperature; however,at the same time, the solubility of many film-forming solids goes down. As a result,the rate of lead oxidation might increase, while the diffusion of the oxidation productsto the surface of the pipe scales could face inhibitions. The scouring action of water onthe pipe scales can expose certain parts of the pipe to oxidation processes. This servesas a new source of solubilized lead and film-forming components.2.11 Factors Affecting Copper LeachingA newer house, with predominantly newer copper plumbing systems, gives riseto higher copper levels than do older houses. Age strongly influences cold first-flushcopper concentrations; however, hot water copper levels appears to be unaffected byage. Cold first-flush has low copper concentrations for plastic pipe compared to thatfor copper pipes. This result also holds true for running hot water samples (Singh,1990).2.12 Plastic Pipes vs. Copper PipesIn contrast to copper pipes, plastic pipes should have less problems withcorrosion and metal leaching. For this reason, plastic pipes are now commonlyinstalled in new homes as the preferred plumbing material (Economic and EngineeringServices Inc., 1990).Plastic pipe is made primarily of polymerized organic compounds. Someresidual unpolymerized monomers may be present. PVC pipes are made by extrudingthermoplastic PVC at temperatures between 150 to 200 °C. In order to make theprocess more stable, lead or tin compounds are commonly added. The presence of this19lead in the PVC pipes could contribute, but not necessarily, to minor levels of leadleaching. As well, there are other problems with using plastic pipes, such as the risk ofleaching carcinogenic and other organic compounds.2.13 Quality Control in Laboratory TestsA sample validation process should be in place to make sure that a measurementis correctly reported for the sample from which it came. Where the samples are takenfrom a targeted population with known water characteristics, the samples should beanalyzed for those characteristics to verify that they correspond to the expected values.If a sample does not meet the criteria for a good sample, it should be excluded from thedatabase.For analyses involving metal ions, samples should be acidified to pH of lessthan 2. The acidification of the samples minimizes the possibility of the metal ionsprecipitating or adsorbing onto the walls of the containers (Mancy, 1971).The quality of sampling data depends heavily on the precision and the bias ofthe measurement methods and instruments used. If the samples are tested by severallaboratories, the variability of measurements between the laboratories will be affectedby the type of instruments and reagents used, the sampling handling techniques, thediffering abilities of analysts, and the quality of laboratory support facilities.Both quality control and quality assurance programs should be employed.Quality control refers to those activities, such as spiking, and calibration, that are usedto assess the quality of the measurements. Quality assurance is the larger, overallmanagement system that ensures that the quality control program is working effectively(Keith, 1991). Normally, quality control charts are drawn to measure the stability ofthe measurement instruments. Standards are periodically tested. The measurementprocess is out of control when a measurement of the standard is above the upper (UCL)and lower than the lower (LCL) control limits; these are defmed to be plus and minus 320sigma around the sample mean. When the process is out of control, the instrument hasto be recalibrated and the samples reanalyzed.2.14 Reporting of Low Level DataZero or negative values in measurements are usually considered to be outliers.This presents a problem when most of the measurements in a sampling program, suchas found in this study, have true values that are expected to be close to zero.Every measurement instrument has its detection limits. There are three basicexpressions of the detection limit. Firstly, there is the limit of detection (LOD). Thisis the lowest concentration level that can be statistically determined to be different froma blank at a specified level of confidence. Secondly, there is the reliable detection level(RDL). This is the concentration at which we can say it is extremely likely that thereis detection. Thirdly, there is the limit of quantitation (LOQ). This is defmed as thelevel above which concentrations can be specified with a certain degree of confidence(Keith, 1991). The LOD is usually set at three times the standard deviation of theinstrument precision (3a). This ensures that we encounter “false” positives only 0.1%of the time, i.e., 99.9% level of confidence. If RDL is chosen to be 6cy, as is theconvention, then the chances of having false negatives is also 0.1 %. The LOQ isusually recommended to be set at lOa. Values at the LOQ have an uncertainty of plusand minus 30% at the 99% confidence level.A measurement that is lower than the LOD is sometimes not included in thedata analysis because we can not be sure about the actual values of these very low levelmeasurements. These measurements theoretically have finite, and positive values.Ignoring all data less than LOD might result in a left-censored data set. This meansthat the resultant database might be biased to the right. Some people prefer to retain allof the data as is, including all the values that are less than LOD (ASTM, 1984).2.15 Considerations in Experimental Design21Most of the previous studies that looked at the metal levels in buildings orhouses only reported the mean values of the metal levels. However, almost none ofthese same studies examined the variability of the metal levels in the system. It hasbeen well documented (Bailey, 1986) that the mean lead level in one house can besignificantly different from the mean lead level in another house of the same area. Thisresult shows that the distribution of lead in the water supply system is a highly variableprocess. Not only is there a great deal of variability between the sampling sites, thereis also a high degree of variability within the same site. Published field and laboratorystudies of lead, solder, and brass corrosion (AWWARF, 1990) indicate that theequilibrium condition is usually not achieved in most samples taken. According to theKuch and Wagner model of the lead stagnation curve, there is a sharp rate of changefor lead before the equilibrium level is reached. This means that at concentrationsmuch below the equilibrium, the level of lead in a standing sample could vary by awide margin. In this case, it is clear that sampling should be repeated several times,for each house, in order to reflect the variability within each site.To ascertain the effectiveness of a corrosion control strategy on reducing metallevels at the tap, detailed statistical analyzes should be performed. Some problems arecommon to statistical reporting (Study Group on Environmental Monitoring, 1977).These include: lack of statistical sophistication, no calculations of the precision ofestimates, no statement of the test hypothesis, insufficient sample size, no descriptionof the method of sample collection, no non-response mechanism, and no respondentbias mechanism.Much effort should be paid to the selection of sampling sites to ensure that thesampling program yields the desired results. In general, the more sampling sites areincluded in the study, the more accurate will be the results. However, it is not easy topersuade some homeowners to provide standing samples. Due to budgets, anymonitoring program will have constraints on how many samples can be collected and22analyzed. A crucial question to be considered at the start of the experimental design iswhat kind of gain in accuracy is there for every marginal increase in the samplingeffort (Gilbert, 1987).2.16 SamplingSampling means to select a few houses and to measure their metal levels,instead of doing the measurements for all the houses in the study area. The results weobtain from the few selected houses can also tell us something about the metal levels ofthe entire study area, through the process of statistical inference.In a sampling survey, attention must be paid to all aspects and phases of thesurvey. Poor work in just one phase of the survey may ruin the results even if all theother phases have been done well.Sampling survey theories have been developed extensively for normallydistributed distributions. A large part of these theories is concerned with finding aformula for the means and variances of the distributions. The sample survey theorydiffers from the classical theory of sampling in that a population group in survey workcontains a finite number of units, whereas the classical theory assumes an infinitelylarge population. For practical purposes, the difference between the two theories areseldom important.Nonprobability sampling refers to those sampling methods that are not amenableto sampling survey theories because the selection of samples is not random. Forexample, the samples might be selected haphazardly, or the selection process mightinvolve human judgment. In judgment, or purposive, selection, the sampler inspects aheterogeneous population and selects a typical unit, which the sampler deems to beclose to the average of the population. Nonprobability sampling can yield useful resultsif good judgment is employed.If the population has an underlying normal distribution, good sampling tends tomake the sample distribution more normal. Bad sampling practice usually results in the23presence of many outliers, which potentially not only skews the sample distribution,but also increases the sample variance and decreases the precision. When possible, thecause of the outliers should be identified and the design of the sampling survey shouldbe adjusted accordingly.Nonresponse refers to not being able to measure some of the units in a selectedsample group. As a consequence of nonresponse, the statistical estimates may bebiased, since the nonresponse part of the population may be different from the part thatdid respond. Evidences from previous studies suggest that the magnitude ofnonresponse bias varies widely from survey to survey (Cochran, 1977). Because thesample actually obtained is smaller than the size of the targeted sample, the variancesof the estimates are increased.Stratification sampling is an efficient way to sample a heterogenous population,especially when it is possible to divide the population into subpopulations, each ofwhich is internally homogeneous. Each subpopulation is called a “strata”. Since eachstrata is homogeneous, an estimate of any stratum mean can be obtained from a smallsample size. The estimates for the strata can be combined to form an estimate for thewhole population group. When used properly, stratification almost always results in asmaller variance for the estimates than is given by a simple random sampling.Quota sampling is basically stratified sampling with nonrandom selection ofunits within a stratum. Because the sample selection is not strictly random, the usualstatistical formulae do not apply. Sampling continues until a targeted number ofsamples is attained.Single-stage cluster sampling is type of stratified sampling. Here, instead ofrandomly selecting units within each stratum, only one cluster is randomly selected.All the units that are defined to be within the cluster will be sampled. Cluster samplingis the most economical way to do sampling if it is expensive to travel between theunits, and the distance between the units is long.24Before any of the stratification sampling methods can be employed, thefrequency distributions of the population must be known. This kind of information iseither available from previous studies done on the same population group, or if not, canbe obtained by doing a preliminary sampling. This technique is known as the doublesampling or two-phase sampling. This method is useful only if the behaviour of theparameter to be measured in the population group does not change with respect to time;otherwise, the results of the preliminary sampling can not be used to define the stratafor the later sampling stage. The accuracy of multiple-stage sampling improves as weincrease the number of samplings. However, respondents who are repeatedly asked forthe same information may not be willing to cooperate after a certain time.2.17 Common Assumptions Made in StatisticsMost of the standard statistical tests that have engineering applications aredeveloped based on certain assumptions. First, the underlying distribution is usuallyassumed to be normally distributed. This assumption is often made because theanalysis of non-normal distributions is highly difficult to compute. In cases where theunderlying distribution is not normal, transformation routines can be applied totransform the original distribution into a normal one. Sometimes, non-parametricstatistics have to be employed if the transformation routines are unsuccessful.In theoretical statistics associated with the normal distribution, the population isconsidered to be infinite. In sampling statistics, the population is almost always finite.As long as the sampling population is large in comparison with the size of the samples,there is not much difference between theoretical and sampling statistics.The selection of the samples from the population is usually assumed to beindependent and random. In other words, the probability of selecting a certain sampleis equal to, and not influenced by, the selection of another sample. To ensure that thisassumption holds true, the population size must be large. Also, the selection ofsamples must follow a random scheme. In real life, the “random scheme” is actually25computed by a random number generator that gives numbers, within a predeterminedrange, with a equal degree of probability.2.18 Measuring Central TendencyThe measurement error is composed of the random error plus the systematicerror. The random error is related to the precision of the measurement process and thesystematic error reflects the accuracy of the measurement. The systematic error is alsocalled the bias, which is the difference between the expected value of the distributionand the population mean.Of the systematic errors, blunders is one type. A blunder is a technical termmeaning that a wrong sample is measured, or there is a misreading of the measurementscale, or a mistake in transcribing or transposing measured values. If large, a blundermay show up as an outlier. But once a blunder is made, it is next to impossible torecover the true value.Due to the central limit theorem, we know that the magnitude of the randomerror decreases as the number of measurement increases. As n gets larger, the sampledistribution also approaches the normal distribution. There is no safe rule available topredict the sample size needed to approximate the normal distribution. However, forsamples that primarily deviate from normality due to positive skewness, a crude rule isn>25G12where G1 is Fisher’s measure of skewness, and is equal toG1This rule is designed so that a 95 % confidence probability statement will be correct94% of the time (Fischer, 1932).For statistically independent variables, the expected value, E, has the property:E(ax+by+cz+ . ..) =aE(x)+bE(y)+cE(z)+Also, if U is defined as26U=ax+by+cz, then the bias of U isB(u) =aB(x) +bB(y) + cB(z) +... (Mandel, 1964).In statistical language, the central tendency is measured by a parameter, themost common of which are the mean, median, and the mode. The mean, or arithmeticmean, is defined as the sum of all the observations divided by the number ofobservations, i.e.The median, in a set of observations that is ranked or arranged in order of magnitude,is the middle observation. If the number of observations is even, then the median is theaverage of the two middle observations. As a measure of the central tendency, themedian is not as prone to be affected by outliers as the mean measurement. Themedian is also easier to calculate than the mean, but, the mean has an advantage that itis always an unbiased estimate of the population mean (Kennedy, 1986). Mode is thevalue that occurs most frequently in a set of observations, or, in a continuousdistribution, the value with the highest frequency.2.19 Measuring SpreadFor statistically independent variables, the variance, V, has the property:V(ax +by + cz + .. .)=a2V(x)+b2V(y) +c2V(z) +In statistics, the spread of a distribution is measured by the variance, which isdefined asJ2=_______Because the variance has units of the square of the units of the variate, the standarddeviation is often used in place of the variance. The standard deviation, a, is definedas the square root of the variance, and has units the same as those of the variate.27When the mean value, ji, of the population is not known, an estimator can beused to estimate the standard deviation. The estimator is defined as=V n—iThe estimator is almost the same as the standard deviation, except that the denominatoris n-i instead of n.2.20 EstimateThe statistic that estimates a parameter of a population is called the estimate. Agood estimate should be unbiased, consistent, efficient, and sufficient. Unbiased meansthat the estimate is the same value as the true value for the population. The estimate isconsistent if it approaches the population value as the sample size increases. Efficiencyrefers to the variance of the estimate. A highly efficient estimate has a low estimatevariance. Sufficient means the estimate has used all the information a sample containsabout the parameter to be estimated.2.21 Coefficient of VariationSupposing that x can only be positive, the ratio of the standard deviation to themean value is called the coefficient of variation, or coefficient of variability, c.v.forx>O.xThe convenience of this quantity is that it is a dimensionless measure of dispersion,with the mean value as the measuring unit.2.22 SkewnessThere are many ways to measure skewness. In general, skewness is a measureof the asymmetry of the distribution, as opposed to the perfect symmetry of the normaldistribution.2.23 Central Limit Theorem28The central limit theorem proves that if we take independent samples from apopulation with a finite variance, all of size N, then the averages of these samples willresult in a sample distribution that tends toward normal. This result is true regardlessof the original population from which the samples were taken. Also, the larger the N,the greater will be this tendency towards normality.2.24 Hypothesis TestingHypothesis testing has to do with making inferences about the population, basedon the information from the samples. The hypothesis that a certain parameter of twopopulation distributions agree with each other is called the null hypothesis, whichassumes that the difference between the parameters is zero. The null hypothesis istested against an alternative hypothesis. If the null hypothesis is tested to be not true,then the alternative hypothesis is said to be true, provided that the null hypothesis plusthe alternative hypothesis cover all the possible outcomes.One problem with almost all hypothesis testing is that we do not know thepopulation distributions. Instead, we only have sample distributions that approximatethe populations. The samples from any one population will have sampling variations.Therefore, hypothesis testing of two sample distributions will practically always yield adifference.Hypothesis testing is meaningless unless we know the accuracy and theconfidence intervals of the conclusions from the hypothesis tests. To determinewhether or not the differences are due to the differences of the populations or thesampling variations, the probability of the conclusions being right must be stated. Thenull hypothesis is rejected when the testing indicates that there is only a small chancethat the populations are the same.The level of significance, x, of a hypothesis test is the maximum probability ofrejecting a true null hypothesis. For most statistical tests, a. is chosen to be 0.05. The29power, 1-f3, of a hypothesis test is defined as the probability of rejecting a false nullhypothesis.A test is unbiased if the probability of rejecting the null hypothesis, Ho, whenHo is false, is always greater than or equal to the probability of rejecting Ho when Hois true, i.e., f3cL (Conover, 1980).A test is conservative if the actual level of significance of the hypothesis test issmaller than the stated level of significance.Conclusions that are drawn from hypothesis testing are never sure. At best,they give us an indication of what might be true about the population. The conclusionsfrom one hypothesis test can give us clues as to what other kinds of hypothesis testsought to be performed. By properly structuring the hypothesis testings, we canhopefully find out what we want to know from the data that have been collected.2.25 Constructing Power CurvesA power curve is a plot between the power of the test, 1-3, and the sample size.It is very useful to be able to know the power of the test associated with the sampledata being tested. However, the construction of power curves requires prior knowledgeof the population means, which are usually not available for most studies (Yamane,1964).From the theoretical results of power curve analyzes, we know that, for thesame level of significance, a one-tail test is always more powerful than a two-tail test.2.26 Statistics of the MaximumWhen we are sampling a population for the maximum values obtained over adefined time interval, we end up with a set of extreme values. The class of distributionfunctions that can describe this kind of distribution is called the “double exponentialdistribution function”. Theoretically, it can be shown that the double exponentialdistribution is the limiting distribution of extreme values of large samples taken frompopulations such as Gaussian (Kinnison, 1985). Many types of environmental pollution30problems, including tap water sampling, can be viewed as an extreme value problem.The “maximum value” statistics should be used if we want to know the most likelymaximum values to be obtained in a sampling program that spans across various timesand locations.2.27 Systems of MeasurementWe normally measure things using the system of real numbers. For example,we say that a certain pen is eight inches long or a chair across the room weighs fivepounds. This type of measurement is sometimes also called the ratio scale ofmeasurement. The main characteristic of this measurement system is that there is atrue zero point. In systems where the zero point is only arbitrarily set, such as in athermometer measurement, the measurement is defined in terms of an interval scale.In such a system, the intervals between the numbers have an empirical meaning, but theratio between numbers do not. Certain other measurements are not as easilyquantifiable. Licence plate numbers and zip codes of addresses, for example, aredefmed rather arbitrarily. These measurements are said to be nominal. In a taste test,a panel of judges are asked to rank the flavor of chlorine in drinking water samples.On a scale from 1 to 5, one being no chlorine detectable and five being strong chlorinetaste detected, only the relative order or position of the parameter of interest isimportant. This system of measurement is called “ordinal”.2.28 Nonparametric TestingIt is hard to handle sample distributions that are not normal or log normalbecause most statistical tests are designed on the basis of the normal distribution.Whenever a nonnormal distribution is encountered, it can potentially be transformedinto a normal distribution by the log function or the functionf(x) = where q belongs to the set of real numbers. The most commontransforming functions of this form are the square root, the negative reciprocal, and thenegative reciprocal square root (Tukey, 1977).31However, in situations where even the transformation procedure can nottransform the actual distribution into one that is normal or log normal, thenonparametric testing method should be used. Nonparametric statistical methods arethose that can be used on data based on almost any measurement system, such asnominal, ordinal, interval, and ratio data. It can also be employed when thedistribution function of the random variable producing the data is either unspecified, oris specified but has a large number of unknown parameters.2.29 Kohuogoroff-Smirnoff Comparison of Two Independent SamplesThe Kolmogoroff-Smimoff test can tell us if two independent samples frompopulations with continuous or discrete distributions, but both of the same type, aredrawn from the same population. This test is robust against differences in the shape ofthe distribution, especially differences in the mean, median, dispersion, and skewness.The maximum difference between the cumulative disthbution functions, F, ofthe two populations serves as a test statistic, D.D = max (— — —)fl‘2The critical D value, for an aggregate sample size of 35, can be approximated byDa = Ka/ni +n2V i2If ni =2, Ka= 1.36 when a =0.05 (Sachs, 1982). If D> Da then there is asignificant difference between the distribution of the two populations.2.30 Median TestThe median test examines whether the sample sets being tested come frompopulations that have the same median. However, the underlying populations need notbe identical when the Ho is true.2.31 Kruskal-WaHis H-Test32The Kruskal-Wallis H-Test is the extension of the Mann-Whitney Test for TwoIndependent Samples. Where there are k random samples, each of which is possiblyfrom a different population, the H-Test tests the null hypothesis of whether or not all ofthe populations are identical. The alternative hypothesis is that one or more of thepopulations tend to give larger values than the other populations, i.e., not all thepopulations have identical means.The H-Test is similar to the median test. However, the H-test uses moreinformation contained in the observations than the median test. Therefore, the H-test isusually more powerful than the median test. The disadvantage of using the H-test isthat all the observations have to be ranked in the combined sample. For a large dataset, the H-test involves more work than the median test.The assumptions of this test are that all samples are random samples from theirrespective populations. In addition to independence in each sample set, the varioussample sets must be mutually independent from each other. Also, the measurementscale of the data must be at least ordinal.The test statistic, T, is defined asT=-1-(Y—— N(N+l)2)S2 ‘=‘n 4S is defined asS2= N_lR3S N_l(j)N4)N is equal to the total number of samples of all the sample groups. And R is theaggregate rank, across all the sample groups, of each observation. The null hypothesisis rejected for T> Ta. For k=3, and n >5 for each sample group, Ta can be found in atable (Conover, 1980).If, and only if, the null hypothesis is rejected, we can determine each pair of thecompared populations differ from each other. The test statistic is33jR1 2 N—l—T 1/2 1/2>t1_(/2)(S ) ( )n. n1 N—k i n1Population i is said to be different from population j if the inequality holds true.2.32 Confidence Intervals for the Difference Between 2 MeansThe confidence intervals of the difference between 2 means can be computed bythe non-parametric method that bears the same name. In this method, it is assumedthat the two distributions being compared are identical except for a difference in thelocation of the mean. If n denotes the number of samples in distribution A, and mdenotes the number of samples in distribution B, then we calculate an intermediate,k = wa,2 — n(n + 1)/2, where Wa12 is a function of n and m (Conover, 1980). For allpossible pairs of (X,J.), the kth largest difference, U, and the kth smallest difference,L, are the respective upper and lower limits of the confidence interval for thedifference between 2 means, i.e.,P[LE(X)-E(Y)U]l-a2.33 Test for Equal VarianceThe test for equal variance is designed to test whether or not population X hasthe same variance as another population, Y. If the populations of X and Y have normaldistributions, the F test should be used in place of the nonparametric test for equalvariance. However, the F test is extremely sensitive to the assumption of normality.Even if the true underlying distribution is a double exponential distribution, which canresemble the normal distribution, the true level of significance may be two or threetimes as large as it is supposed to be. Therefore, the F test is not safe to use unless weare sure that the populations are normal. The A.R.E. of the Squared Ranks Test forEqual Variance is 0.76 (Conover, 1980) if it is used instead of the F test, when thepopulations are actually normal. But, for other distributions, the A.R.E. increases andapproaches unity.34In addition to the usual assumptions of independence, and randomness, the testfor equal variance also assumes that the measurement scale is at least interval. For atwo-tail test, the null hypothesis is Var(X) =Var(Y).Let n denote the number of samples in distribution X, and m denote the numberof samples in distribution Y. The absolute deviation of each observation from themean is U =X—p, i=1,...,n; and J’.The ranks 1 to n+m are assigned to the combined sample of Us and Vs. Insituations where several Us and Vs are exactly equal to each other, the average of theranks, if there are no ties, are assigned to each value. The rank of each observation isdenoted by R(U) and R(.).The test statistic is-T-nR21r ,1/2tim p4 nm[N(N_i)i=’ N—i’ ‘where N=n+m, and,= ±{$[R(u1)]2+[R(v)]2};= $[R(U)] +[R(v,.)]4The null hypothesis for a two-tail test is rejected if i’ is less than ctI2 or greater than 1-cd2 (Conover, 1980).2.34 Spearman Rank CorrelationsWhere the data consists of pairs of numbers (x,y), a measure of correlationbetween the two numbers can be calculated. Correlation estimates the degree ofdependence between x and y. The Spearman measure of correlation in ranked data isdefined as- iJ][R(Y) - LiJ]n(n2—1)11235The correlation measure only assumes values between -1 to +1. Thecorrelation is closer to +1, if the larger values of X tend to be paired with the largervalues of Y. If the larger values of X tend to pair with the smaller values of Y, themeasure of correlation tends to -1. If there the pairing of X and Y do not follow anypredictable pattern, then the correlation should be close to zero.2.35 RegressionAs opposed to correlation methods, regression methods are used to inspect moreclosely the relationship between x and y in bivariate data sets. One important objectiveof regression methods is to predict a value of y where only the value for x is known.This is done based on the information we obtain from existing (x,y) data sets.The variables used for regression analysis should be carefully selected. Also,we should know beforehand whether the form of the fitting function is linear ornonlinear. A linear regression function is in the form y =a+bx, where a is theintercept, and b is the slope of the regression line. Both the a and b regressioncoefficients have physical meaning. Therefore, given that the data can be fitted by alinear regression line, the most important task is to determine the values for the a and bcoefficients.The best empirical fit of the data points may not necessarily be accurate. It ispossible to develop, between variables, relationships that are completely meaninglessin a practical sense (Montgomery, 1991).Regression relationships are only valid for the range of values in the originaldata. Regression models should never be extrapolated to predict values outside of therange of the original data.2.36 One-way ANOVAThe analysis of variance, ANOVA, seeks to determine if the differencesbetween groupings of data are significant. Data is classified into groups based on thedistinguishing features of the data. In the analysis of water samples, the chemical36parameters of the samples can be used to group data. Lead samples with low pH andlow alkalinity can be grouped together, and samples with higher pH and higheralkalinity grouped separately. In this illustration, pH and alkalinity are said to be theANOVA factors of the lead analysis. If the analysis reveals significant differences inthe lead concentrations between the various pH and alkalinity data groupings, we cansay that the pH and alkalinity factors significantly influence the lead data. Thisillustration is also an example of a two-way ANOVA because there are two factors: pHand alkalinity. In an one-way ANOVA, the data is grouped based on only one factor(Montgomery, 1991).The study of ANOVA is an extremely vast and complicated field of study instatistics. For the purposes of this particular research study, it will suffice to explainthe simplest case.A typical data table for a single-factor experiment is presented in Table 2.Table 2 Typical Data for a Single-Factor ExperimentTreatment Data Totals AveragesLevel1 Yii Y12 Yin Yl2 Y22 Y2n Y2a Yal Ya2 Yan Ya YaFrom the data table, we can calculate the total of the treatment level totals, yy = and the average of the averages is, y = yIN.i=1 j=1In addition to these, we can calculate other important quantities, such as:SsT=2(yy); sSE=SSE=(YYl);SST=SSfreat,fleflj+SSE;i=1 j=1 2=1 J=I37Source of Sum of Degrees of Mean Square F0Variation Squares FreedomBetween SStreatments a-itreatments F —° MS,Error (Within SSE N-a MSEtreatments)Total SST N-iThe most important part about this whole table is the F0. If the computed Fisher valueexceeds the critical Fisher value, then we conclude that the between treatment factor issignificant.2.37 Theory of CausationStatistical methods can function to demonstrate the relationships, theinterrelatedness, or the correlations between variables; however, no statistical methodscan prove that there is certain causation. Suppose we find a positive relationshipbetween the number of books a high school student purchases and the grade pointaverage the student receives. We can demonstrate through the use of statistics that astudent who purchases a lot of books also receives a high grade point average.However, it is not valid to, therefore, say that buying a lot of books will ensure thestudent of a high grade point average. There may be other factors involved here. Forexample, the students who purchased a lot of books may also have spent a lot of time inreading those books. As a result of their reading efforts, their increased knowledge ofthe school subject matters has helped them to excel in examinations. We see, then, it isin fact reading books, not purchasing them, that affects grade point averages. A poorstudent who does not have the money to purchase books, but assiduously pours throughMSat,nen ttm; and MSE=. We summarize these important deriveda—i N—avariables in the one-way ANOVA analysis table (Table 3).Table 3 The ANOVA Table for the Single-Factor, Fixed Effects Model38volumes borrowed from the local library, may do better in examinations than studentswho buy a lot of books but do not read them.In certain cases, the presence of correlation between variables has no meaningin terms of causal relationships. For example, we find that there is a strong correlationbetween the length of the right and the left arm, and the height and body weight of aperson (Sachs, 1982).There is another class of noncausal relationship called the inhomogeneitycorrelation. Consider a population that has three subgroups. When we measure twoquantities: A and B, they may not be correlated in each of the subgroups, but acorrelation may be found when all the three subgroups are analyzed together as onepopulation.Although we do not know for certain if there is a causal relationship betweencorrelated variables, but we can make this conclusion once we have excluded otherpossibilities. This method of proving causality is only useful if we have an exhaustivelist of all the possible types of noncausal correlation relationships. Furthermore, wemust have a way of detecting these relationships.393. EXPERIMENTAL METHODS3.1 Experimental DesignThe study areas of this research project are located in North Delta, Newton, andSouth Surrey of the Greater Vancouver Regional District. North Delta is the controlarea. Due to bacteriological and chemical chlorine demand within the distributionsystem, the supply water from the Seymour Reservoir is stripped of chlorine residualsby the time the water reaches North Delta. In North Delta, no chemicals are added tothe water to adjust for pH and alkalinity. The Seymour water supply also feedsNewton and South Surrey. In Newton, the water is rechlorinated and adjusted for pHand alkalinity. In South Surrey, the water is chioraminated and also adjusted for pHand alkalinity levels.The major water feeder main for Newton comes into the distribution grid at theintersection of 128th Street and 64th Avenue. Similarly, 144th Street and 32nd Avenueis the feed point for South Surrey. As the water enters into the distribution grid, thechemical characteristics of the feed water, such as residual chlorine levels, pH, andalkaiinity, change according to the physical, chemical, and biological conditions of thedistribution grid. In general, as the water travels farther away from the feed point, theresidual chlorine levels drop. The change for pH and alkalinity is harder to predictbecause the interactions between the supply water and the distribution pipes could leadeither to increases or decreases of pH and alkalinity.By collecting tap water samples from North Delta, Newton, and South Surrey,we can compare the copper and lead levels in Newton and South Surrey against those inthe control area. The comparison tells us if pH and alkalinity adjustments are keyfactors in lowering copper and lead levels in tap water. Because pH, alkalinity, andresidual chlorine levels vary in Newton and South Surrey, each area has to be studiedinternally to see if there are significant internal differences.40Samples from the control area should have fairly homogenous pH, alkalinity,and residual chlorine measurements. Therefore, random aerial sampling (takingsamples from locations randomly distributed throughout the control area) would be thebest way to assess the copper and lead levels. In contrast to North Delta, Newton andSouth Surrey will have pH, alkalinity, and residual chlorine measurements that span awide range. Simply taking random aerial samples from each of these two districtswould not reveal significant differences, if there were any, in copper and lead levelswithin each district. Assuming that water with higher residual chlorine levels also hashigher copper and lead levels, areas closer to the feed points should have the highestmetal levels in Newton and South Surrey and areas farther in distance from the feedpoints should have the lowest metal levels. By getting a representative set of samplesfrom an area closest to and another set of samples farthest from the feed points, wehave an excellent chance of finding any significant differences in metal levels betweenthe two areas. Due to the spatial variations in water quality parameters, we cannottravel too far away from one sampiing site to another before we find substantiallydifferent pH, alkalinity, and residual chlorine levels. Hence, cluster sampling is thebest way to get the two sets of representative samples from Newton and two sets fromSouth Surrey (Appendix A).3.1.1 Controllable FactorsBesides pH, alkalinity, and residual chlorine levels, there are other factors thatcould influence the copper and lead levels in tap water. Most of these factors can becontrolled directly or indirectly. By direct control, we mean that we can control thequantity or quality of the parameter of interest. For example, we can control the pH,alkalinity, and residual chlorine levels inside the rechiorination station. By indirectcontrol, we mean that we cannot control the parameter of interest, but we can use othertechniques to influence the outcome of a variable. Outside of the rechlorination station,the pH, alkalinity, and residual chlorine levels begin to vary in an unquantifiable way41because we do not have precise information on every part of the entire distributiongrid. But, if our objective were to collect tap water samples that have a similar level ofresidual chlorine, we can indirectly achieve this objective by taking our water samplesfrom houses that are very close to each other. This way, although we cannot say “apriori” that the levels of residual chlorine in all the samples will be exactly the same,the chances are that they will be similar.3.1.1.1 Direct ControlThe age of the house, from which the water sample is collected, is a factor thatcan be controlled directly. This study investigated the worst potential copper and leadproblems, which previous studies have shown to be houses plumbed with copper pipingthat are less than ten years old. Therefore, the houses selected for this study had to beless than ten years old. Some very old houses may have been renovated in the past tenyears with new copper plumbing. But, since it is nearly impossible to find out whichold house has new plumbing, it was much easier to target for houses built in the pastten years.A problem arises with using this strategy because some of these very newhouses had plastic plumbing. These houses are not likely to have a problem with metalleaching into the tap water, and these houses were not included in this study. Theresidents of the houses were asked whether their plumbings are plastic or copper. Ifthey were not sure, we could check the water service pipes underneath the kitchen sink.Most houses that have plastic plumbing inside the house will also have plastic servicepipes. However, this method of determining the predominant type of plumbingmaterial used in the house is at best tenuous. On the other hand, this was probably theonly method available, if the residents are not sure what type of plumbing is in thehouse.Samples were collected both from houses that have plastic and houses that havecopper plumbing. The data from houses plumbed with plastic pipes had to be analyzed42separately from those with copper pipes. Results from the two separate data sets werecompared with each other to confirm that houses with plastic pipes indeed did not havea metal leaching problem.All tap water samples were collected from the kitchen tap of houses (singlefamily dwellings). The metal levels in the tap water from the same house will vary,depending on from which tap the water is collected. This is due to the fact that the tapfaucets contribute to the metal leaching problem. Not only are different faucets in ahouse of different sizes, but they may be of different makes and have a differentpercentage mixture of metallic components that are susceptible to leaching. Collectingwater samples only from the kitchen tap minimizes the variation in tap water metallevels due to the variation in the faucets.This study targeted houses, rather than apartments or buildings, because thedifferent types of dwelling cannot be studied together. Previous studies have shownthat the type of building will influence lead levels in tap water. Apartments andbuildings have, in general, higher levels of lead than houses. Apartments or buildingshave much longer internal plumbing loops than houses. The average residence time ofwater in the plumbing system of a house is much less than in an apartment. Therefore,there is less contact time between water and the leachable metals in the plumbingsystem of a house (Singh, 1990).For the past few years, on-line water purification devices have become popularin the market. The house residents were asked if they have installed such devices ontheir kitchen taps. In cases where they had, they were instructed to turn off thosedevices during the sampling process. Some of these water purification devices are veryeffective at removing metals in the water. The failure to consider this variable could bedetrimental to this study.3.1.1.2 Indirect Control43The natural level of lead and copper in the supply water could be a problem ifthey were much higher than the levels leached from the distribution and plumbingsystems. Fortunately, GVRD data show that the natural lead and copper levels in theSeymour source water is extremely low. Hence, the background metal levels were notexpected to “mask” the effects of pH and alkalinity adjustments on copper and leadleaching. Furthermore, there could be a problem if the drinking water of North Delta,Newton, and South Surrey was supplied from water sources that have different waterchemistry. Having anticipated this problem, the drinking water of all the study areaswere supplied exclusively from the Seymour reservoir.The type of solder and the lengths and sizes of pipes used in the plumbingsystem were variable in each house. Even though the old type of lead/tin solder hasbeen prohibited in municipal plumbing codes for many years, they are still sometimesused illegally due to the ease of handling lead/tin solder. Houses that have more of thelead/tin type solder, rather than the newer tin/antimony solder, in the house plumbingwill probably also have higher lead levels at the tap. Unfortunately, there is no way toquantify the amount of lead/tin solder present inside the house or in the distributionsystem. The technique of sampling a number of houses instead of a few houses solvesthis problem. Sampling ensures that some houses sampled have more lead/tin solderand some houses have less. The average of the lead concentration of the samplinggroup will then reflect an averaged amount of lead/tin solder.The temperature of the supply water in the various areas of the distributionsystem could also affect the rate of metal leaching from the pipes, solders, and faucets.However, considering the close proximity of the various sample locations, and the factthat solar heating of the supply water is retarded by the soil cover on top of the entiredistribution system, the temperature variation across the sampling sites was notexpected to significantly affect metal levels.3.1.2 Uncontrollable Factors44There is one very important factor of copper and lead leaching that iscompletely uncontrollable in a field study. This factor is the standing time of the watersamples. This study examined standing cold water, flushed cold water, and flushed hotwater; however, the focus of the study was on the standing cold water samples becausethey are likely to have the highest levels of copper and lead. Standing samples are alsoimportant because they are used to determine the compliance of trace copper and leadlevels in USEPA regulations.In a laboratory or coupon study, the standing time of water can be controlled.In a field study, however, the standing time cannot be controlled precisely. Thestanding time is the time between the last use of water in the house before theoccupants go to sleep and the time the sample is collected first thing in the morning,i.e., before any water is used in the morning. There are at least four potentialproblems with the control of standing time. One, not everybody participating in thisstudy would sleep the same amount of time. Two, one or more of the residents mayhave to use the bathroom sometime during the night. When this happens, some of thestanding water is flushed out of the plumbing system. This has the effect of decreasingmetal levels in the house plumbing, and hence, decreasing the effective standing time.Three, when the occupants wake up first thing in the morning, they may use thebathroom or the kitchen before they remember to do the sampling. This is probablythe most serious problem, and this scenario is more likely to happen if there are manyoccupants living in the house. The larger the household, the higher is the chance ofhaving at least one of the occupants forgetting about the sampling study. Four, somehouses have one or more leaky water taps. The faucets may be getting too old orsomeone may forget to shut-tight a tap, which causes continuous slow leakage of thestanding water. Depending on the degree of leakage, the effective standing time coulddrop insignificantly or a great deal.45The participants of this research project were instructed about these potentialproblems. But human behaviour is hard to predict and impossible to control.Therefore, we can predict that the samples come from a wide variation of standingtime. Since we wanted our standing cold water samples to have remained in the pipesfor at least six or seven hours, samples that had an effective standing time less thanthat will significantly bias the copper and lead data toward the low side. We alsoexpected that a lot of the scatter in the copper and lead data could be attributed to thestanding time problem.There are two possible ways to get around this problem. Both are highlyimperfect. Assuming that most people sleep more than six hours and that the waterleakage problem is insignificant, we can ask the participants if they used any waterbetween the time they went to sleep and the time they collected the samples. Someparticipants who know that they have made a mistake may not be willing to admit it.Also, the person who is asked this question may not know whether or not anotherperson in the household has used any water before collecting the water samples. Thismethod of determining if the standing sample from a particular house has been flushedor not is seriously flawed.The second way to solve this problem is to identify the flushed standing samplesby using statistics. This approach, while imperfect, is nevertheless more scientific thenthe first approach described earlier. This research project adopted this approach (referto section 4).3.2 Arrangements with Governmental AgenciesBefore the samples were collected, various levels of government wereconsulted. In particular, the GVRD, the Municipality of Delta, and the City of Surreygave this researcher permission to solicit houses, within their jurisdictions, forcollecting tap water samples. Assistance was also offered by these government46agencies to provide letters of recommendations, in the event that the home owners haddoubts about the legality of this project.From the archives of these government agencies, it was possible to locate theaddresses of about two hundred houses that were newer than ten years old, plus a fewmore houses that were between ten to fifteen years old. These houses, locatedthroughout the five sample areas (one in North Delta, two in Newton, and two in SouthSurrey), were the target houses for collecting water samples. The GVR]) assisted inmailing out letters to these homeowners, informing them that this researcher mightcome to their house to ask for their cooperation in this study and to provide tap watersamples from their kitchens.3.3 Bottle PreparationAll the bottles used in this study were brand new plastic bottles that weredelivered to the UBC laboratory sealed in the original packaging. This was the firststep toward minimizing the potential of metal contamination of the bottles due to airparticulates, solvents, or chemicals in the lab. Because the degree of existingcontamination due to copper or lead particles on the inside surface of the bottles wasunknown, a bench test was performed. Six different kinds of bottle preparationtechniques were tested. Three bottles were prepared for each bottle treatment. Eachcopper measurement was repeated five times, and lead three times.The first set of bottles (NT) were the controls. Only deionized distilled waterwas poured into the bottles, and no pretreatment was given. The second set of bottles(AiD) were pretreated by acid washing with 10% nitric acid for one minute and thenrinsing with deionized distilled water. The third set (A5D) was similar to the secondset, except that the acid wash time was five minutes; the fourth set(A9OD) had a ninetyminute acid wash time; the fifth (AOD) had the acid stay in the bottle for overnight.The sixth set of bottles (SDAOD) were pretreated by the method recommended by theStandard Methods (17th edition). The bottles were soaped, rinsed by deionized47distilled water, acid washed overnight (about eight hours), and then rinsed again bydeionized distilled water. This last method is extremely labor intensive. To wash allthe bottles needed for this study (about 600) by this method would have requiredapproximately 2 months for one person. If any of the first five shorter methods provedto be equally as effective as the method recommended by the Standard Methods, thenthat method was adopted by this study.Table 4 Results of Bottle Treatment StudyTreatment Method Average Copper (ppm) Average Lead (ppb)NT 0.023 0AiD 0.000 0A5D* 0.016 0A9OD 0.016 0AOD 0.016 0SDAOD 0.000 0The results of the bottle treatment study (Table 4) shows that the deionizeddistilled water used for this study was free of lead. Also, the zero level of leaddetection, regardless of the treatment method employed, means that the bottles had nolead contamination.The detected levels of copper were extremely low compared with the range ofcopper concentrations that one would expect to find in the collected tap water samples.Therefore, it did not really matter which treatment method was chosen. The methodrecommended by Standard Methods worked the best; but, for the purposes of thisstudy, is inconsequential. All the treatment methods resulted in a lower level of coppercontamination. A decision could have been made not to treat any of the bottles.However, to avoid the unlikely possibility that excessive amounts of lead or copperparticulates might be found in some of the bottles due to chance, it was decided to acidwash all the bottles used in this study for five minutes, followed by rinsing withdeionized distilled water.483.4 Sampling PackageThe sampling packages that were to be delivered to the homeowners wereprepared beforehand, at the UBC lab. The package was a large ziplock bag that had,sealed within it, a 1L bottle, two 125 mL bottles, a copy of the letter that the GVRDhelped to send to the homeowners beforehand, and two instruction sheets. The oneliter bottle was marked with the number “one”. This bottle was used to collect thestanding cold water sample. The metal levels in this sample came primarily from theinterior home plumbings, soldered joints, and faucets (AWWARF, 1990). One literwas chosen to be the size of the standing sample because the one liter sample size wasrecommended by the USEPA in the May 1991 Lead and Copper National InterimPrimary Drinking Water Regulations.One of the two 125 mL bottles was marked with the number “two” and wasused to collect the flushed cold water sample. The other 125 mL bottle was markedwith the number “three” and was used to collect the flushed hot water sample. Allthree bottles were empty and capped. The flushed samples measured the metal levelsin the water distribution system. The sample volume for the flushed samples was notan important issue, assuming that the kitchen tap had been adequately flushed beforethe sample collection. Since the metal levels in the distribution system have lessvariation compared to those inside the house plumbings (AWWARF, 1990), collectinga large or a small does not make much of a difference. A relatively small sample sizewas chosen, making storage and transportation more convenient.The first page of the instruction sheets was a concise written explanation of thesampling procedures (Appendix B). The second page was a self explanatory cartoonthat helped the homeowner to visualize the sampling procedures, in case theinstructions on the first page was unclear to them (Appendix C).3.5 Sampling Procedures49Two identical ziplock bag packages were delivered to each house thatparticipated in this study. One was marked bag “A” and the other bag “B”.On the first morning after being contacted by the researcher, the homeownerwas to take the three bottles out of bag “A” and place them, in order of theirnumbering, on the counter beside the kitchen sink. First thing in the morning, beforeany water was used in the house, the homeowner was to turn on the cold water tap andfill bottle number 1. The cold water tap was left on, until the water became cold,before filling bottle number 2. When the water from the cold water tap turned cold, itwas a sign that water came from the distribution pipes outside the house plumbing.After the flushed cold water sample was collected, the cold water tap was turned off,and the hot water tap was turned on. Bottle number 3 was filled after the water becamehot. All three bottles were to be capped tightly, and sealed in the ziplock bag forpickup. On the next morning, the second morning after being contacted, the sameprocedure was repeated for bag “B”.3.6 Participation of HomeownersThe participation and cooperation of the homeowners in this study wasabsolutely crucial to its success. First, the homeowner had to be willing to participatein the study. Second, the homeowner had to follow the sampling procedures correctly.In order to obtain a database of sufficient size, there was a target of enlisting 35houses in each of North Delta, Newton, and South Surrey to participate in the study.Having 35 data points for each area allowed the researcher to do statistical comparisonsbetween the study areas with an acceptable degree of confidence (see Section 4), whilekeeping the sampling and lab testing efforts at a manageable level. Quota samplingwas employed for this study. Targeted homeowners, who had been contactedpreviously by mail, were contacted personally by door knocking. They were asked ifthey were willing to participate in the study. Those who were willing were given thesampling packages and were instructed about the sampling procedures. The50homeowners were also asked a series of questions regarding their houses (Appendix D).For a variety of reasons, a portion of the homeowners who were contacted did not wantto participate in the study. Enough homeowners on the list of targeted addresses wereasked until 35 homeowners in each study area agreed to participate.By far the biggest error in this study was due to the homeowners not followingthe sampling instructions. The most common error was the use of water in the houseprior to sampling. Other less common, but equally serious, errors included: notletting the cold water flush enough when collecting bottle number 2, not letting the hotwater get hot before collecting bottle number 3, and mixing the order of collecting thesamples.3.7 Bottle Pick-upThe sample bottles, sealed in the ziplock bags “A” and “B”, were left outsidethe houses after the second morning of sampling. The bottles were picked up in theafternoon and brought back to the UBC lab for analysis. Due to the large number ofsamples that had to be delivered and picked up, the sampling process took place over athree week period between September 20th to October 10th, 1994. Each bottle wastracked so that the bottles from each house could be picked up right after the secondmorning of sampling.3.8 Lab TestingThe pH, alkalinity, and chlorine residuals in tap water deteriorated quickly. Todetermine the values of these parameters for the houses, the standing sample from bag“B” of each house was measured in the lab on the same day that it is picked up. Thecopper and lead metals did not deteriorate as quickly as the other measured parameters.Therefore, the measurement on copper and lead levels was deferred for a few weeksafter all the bottles were picked up from the houses. Copper and lead were measuredfor all the bottles from bags marked “A” and bags marked “B”.51After the bottles were brought into the UBC lab, each bottle was coded so thatone could identify later on where each bottle came from. The pH, alkalinity, andchlorine residuals were measured before acid was added to preserve the water samples.Preservation with acid was necessary for measuring lead and copper. Copper ions hada tendency to adsorb onto the bottle surface and, thus, the addition of acid de-adsorbedthe copper ions. For lead analysis, the acid added served as a matrix modifier thatreduced interferences (Standard Methods, 17th edition). Nitric acid at 2.5% wasselected as the preserving acid. Normally, a concentration of 0.3% would have beenused. To make the copper and lead testing feasible in terms of time, the samples werenot digested before measuring the metal levels. A higher concentration of acid wasadded to the samples in order to compensate for this deficiency. This did not mean thattotal metal was recoverable without going through the digestion step; however, at leastall of the dissolved metals could be recovered.In essence, this procedure measured and analyzed the dissolved copper and lead.This study focused on the dissolved metals, since they account for most of the totalmetals in tap water (AWWARF, 1990). Also, dissolved copper and lead have morepotential harmful effects on humans than the particulate counterparts, since the humanbody can uptake the dissolved forms much more easily. In the past, studies that werestrictly concerned with dissolved lead species filtered the samples through a memberfilter of 0.4 or 0.45mm pore size. This procedure was not effective because thedissolved lead tended to adsorb onto and was retained by the filter material (Schock,1983).3.8.1 InstrumentsThe pH of the samples were measured by a Fisher glass electrode and aBechman pH meter. Because tap water has a low ionic strength, an orion buffersolution was added to the water to give more stable readings. The pH meter wascalibrated by pH buffer solutions. Total alkalinity was determined by the titration52method (Standard Methods, 17th edition). The free and total chlorine levels weremeasured by the DPD Colorimetric Method.Copper was analyzed by a Thermo Jarrel Ash Atomic AbsorptionSpectrophotometer, using lean acetylene fuel. Samples were aspirated directly out ofthe sample bottles in order to avoid contamination in the measurement stage. Standardsolutions, in the range of copper concentrations we expected to find, were made. 2.5%nitric acid was added to the standards in order to make them equivalent to the samples.The AAS instrument was recalibrated against these standards for every ten samplesmeasured. All measurements were repeated five times. Only the averages of theserepeated measures were reported.Lead was determined by atomic absorption graphite furnace. Lead sampleswere poured into pre-rinsed sampling cups before the graphite furnace instrument testedthem with the autoanalyzer. The quality control procedure was the same for lead as forcopper.3.8.2 Testing SchemeFive percent of the standing samples were cross-checked by the GVRD waterquality lab. Interlaboratory testing could help to determine whether or not theequipment at the UBC lab was working properly and was in good condition.Because there were roughly six hundred samples that had to be measured forcopper and lead, it took two months to complete the testing for all the samples. Duringthis testing period, the working conditions of the instruments changed according to theenvironmental conditions of the lab, on any given day. The same sample that wasmeasured on one day might have a different reading than on another day. Thevariations of the instruments could have been purely random or may have hadsystematic trends. Besides the technique of constant recalibration, the instrument effectwas taken care of by selecting the sample to be tested in a random sequence.3.8.3 Data Recording53There were inevitably some errors in the process of reading, recording, andtranscribing data. The validity of the databank would be in jeopardy if the rate of thesekinds of error was high. There should not have been any reading and recording errorsfor the lead and copper measurements because the test instruments automatically printedout the data on paper. To assess the rate of transcription error made in the process oftranslating the data into the computer databank, the data in the databank was crossreferenced with the data tapes. The rate of error that could not be corrected was foundto be 0.1%.544. RESULTS AN]) DISCUSSION4.1 Siimmry of Returned BottlesOf the 105 houses that agreed to provide samples for this study, 92 of themreturned bottles. Thirteen houses did not return any bottles. The 88% return rate washigh for a study of this type. Participants of this study were informed about thepotential dangers of consuming water with excessively high levels of copper and lead.They were told that their participation was vital to the success of identifying the safetyof the water supply in their area. Those who wanted to know about the outcome of thisstudy received the conclusions of this study after it was completed. Personalcommunication with the house owners, and linking this study to their health and safety,were the main ingredients of obtaining a substantial number of samples. A summary ofthe information about the returned bottles is provided in Table 5.Table 5 Suinmar Information of Returned BottlesArea code # pH** alk** free chlorine* chloramine* samplingDelta d 19 n n .01 .01 randomNewton 1 n 31 y .28 .05 clusterNewton2 y 10 y .91 .10 clusterS.Surreyl s 19 y .29 .27 clusterS.Surrey2 w 13 y .33 .44 cluster* GVRD data (avg for Sept 20- Oct 10)** The presence of pH or alkalinity adjustmentsThe samples from the 92 houses were collected from the five different samplingareas. The North Delta area had 19 bottle returns. As expected, free chlorine andchloramine residuals were almost zero because all of the chlorines had been oxidized inthe system by the time the supply water from Seymour travelled to North Delta. NorthDelta was given a shortform “d” in sample identification and statistical analysis. Thefirst of the two sample areas of Newton, “n”, registered levels of chlorine residualsmuch higher than the Delta area. Comparing the second of the two sample areas ofNewton, “y”, to “n”, we clearly see that the free chlorine residuals of “y” was much55higher than “n”. This was expected since “y” was closer geographically to the watersupply feedpoint than “n”.Chioramine was added to the two South Surrey study areas. The first of thetwo sample areas of South Surrey, “s”, had higher levels of free chlorine andchloramine residuals than “n” or “y”. The second of the two sample areas of SouthSurrey, “w”, had chloramine levels that were even higher than “s” because “w” wascloser to the feedpoint than”. pH and alkalinity tests on the samples from the fivestudy areas showed the addition of chemicals had substantially raised the pH andalkalinity of the water in Newton and South Surrey. In all, measurements of the fourwater quality parameters: pH, alkalinity, free chlorine residuals, and chloramineresiduals, agreed well with the expected levels.The results of this statistical analysis will be valid only for the range of thewater quality parameters detected in the samples of this study. The conclusions of thisstudy do not apply to ranges of water quality parameters that are above or below thosein the study samples. Conclusions based on statistical statements are also specific to thestudy areas. Lead and copper levels in the other areas of the Greater Vancouver WaterDistrict have to be examined separately, because the water quality characteristics andthe water distribution systems might be significantly different from those in this study.However, the same methods and analysis techniques can be applied when studyingthese other regions of Vancouver.Although the chlorine levels of the samples were measured in the UBC lab,none of these measurements were used in the data analysis. All the measured chlorinevalues were very low in comparison with the GVRD chlorine data.There was a time gap of several hours between the time the homeowners tookthe water samples and the time the chlorine levels were measured in the UBC lab.Because chlorine is a very volatile gas, much of it escapes from the bottles even if thebottles are tightly capped. For any chlorine measurement to be valid, the analysis must56be carried out concurrent with the time of the actual sampling. Fortunately, the GVRDregularly monitors various water quality parameters from many places in the GreaterVancouver. There was at least one of these GVRD monitoring sites in each of the fivestudy areas. The reported chlorine residuals in this study were the time averaged (Sept20 to Oct 10) values of all the monitoring stations in each area.4.2 Data Cross-check by GVRDThe accuracy of the measurement data from the UBC lab was cross-checked bythe GVRD lab (Appendix E). Because the standing samples constituted the focus ofthis research, only standing samples were selected for cross-checking interlaboratoryaccuracy. Of the 177 standing samples from bags “A” and bags “B”, fifteen percent,or 28, of the samples were randomly selected and sent to the GVRD lab. Copper andlead were measured by the GVRD lab using the same principles of measurement asthose in the UBC lab.The GVRD and the UBC lab data for the 28 samples revealed some differences.While the copper values for the two data sets were remarkably similar, 9 out of 28 ofthe lead measurements appeared to be significantly different. If the UBC and theGVRD data sets were significantly different from each other, then either the UBC orthe GVRD lab or both labs had a problem with measurement accuracy. TheKolmogorov-Smirnov test was employed to test for a statistical difference between thetwo thta sets. The UBC data set was randomly drawn from the pool of standingsamples, which was the population of values measured by the UBC instrument. Thenull hypothesis of the test was that the GVRD data set belonged to the same populationof values as the UBC data set. The rejection of the null hypothesis meant that theGVRD data set differed significantly from the UBC data set. If this were the case,more interlaboratory testing had to be conducted to identify the sources of themeasurement inaccuracies.57A two sided test was performed, with n=28, c=0.O5, and Dcritical=0.36.For copper, d=0.07. Because d < 0.CUiC\O00-..C.\O00Uict-CD-00000000000000)t)C00UiOcMUiC00Ot)CUiL’JCO000,UiUL’)00—-‘0.00I-—I————————I-—I-————————)IUiCCUiCCCCOCcMCCCUiUicMUiUiCCUiCCUiCCCUiUiUiCUiUiCCCUiUiCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC CUi.UiUiCDUi‘CO—00-CUi.U00C—D—00000090p090ZUi‘O-CCoUi00CUit-).I-))I-000OCC-0CD-oL’-)pp.00.Oppoo.000o.o00..aoUi.OCCOUiUiUiOUiCCOUiCC0UiCCCCUiCCUitiUiUio,ibboCoUiUiCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCp----p---..c. L’)I-t.JC00\DpCoopCUCCCMQOCCO100000000)10000000000000000000oOCICflr,I2rjCflCICl)Ct)ClCl)Cl)Cl)Cl)Cl)CI)CflCl)Ci)Cl)Cl)Cl)Cl)—.>4>4>4>4>4>4>4>4>4>4>4>4>4>4CUi.L’.)0D00C-00—CUi.t3‘-.-.O00UiU)clQ>4>4>4>4>4>4>4>40pppopppppppc000ccceeoccc00PP0P0PoPo0*OU)L’.)0\OO-I-Ui0Vi-OL’)‘-DUi000C0CCO-%D00->4>4>4>4>4>4>40ppppppppppPPppppP000000000000000Ct000t.)00t.)o00O00000C0t000—It.)U)*>4>4>4>4>4>4>4>40ppppppppppppppppppp000000000000000000*0000U).U).Ui00U)0000U)000U)0Ui0U)U)>4>4>4>4>4>4>4000000000PPP000PPOOOPPO000000———bb.—o—o—0———0—0*O.01’.)ViU)U)40UiVi)O—0-\000ViU)>4>4>4>4>4>4>4>4>4•ppppppppppppppp00000000000000C.0000t’.)0‘00Vi-000—000—0000..——*>4>4>4>4>4>4>4000ooppppppppppppooo0000U)00000P0000000000*•L’)000.Ci-L’.)0.000C.00U)L’.)C\D000U))L’.)—aViU)———-—————>4>4>4>4>4>40 -a -Ut)iiiII—IIIIIII*0O•ViViVi.)\O00001t-.).Vi0U))OC-OC.tU)>4>4>4>4>4>4>4>40 rIIIIIIIIIIIIIII-t-.)0U)0—‘10it)—Ct’t—)t)00—L’)c’-)L’-—*>4>4>4>4>4>4>4>40) wIIIIIIIIIIIIII-vI—0•-I——0—I—I‘)i—0I—t-)—0C‘000)0U)*>4>4>4>4>4>4>4UIIIt—)11IIIiII*0L’.)L’)U)U)o.t’)C01-000—.CU)Ui00..L’-)C‘‘-Vi>4>4>4>4>4>4>4F)-U *00)00-0-’ciia0)C00CC0->4>4>4>4>4>4>4(i)-U *IIIIIIIIIIIIIIIIIU)0t)t)L’)-U)‘t-Jt’)t.)0t)0Q0U)0L’))I-CCl)Cl)Cl)CV)0Cl)Cl)—.CC00->4>4>4>4>4>4CCCCOO00CCCPP00p00ppCCC1’.)PCCCCC0t’.)1Pt.)P—c*0CJi0L’Jt)OO00CCCC00O000Cc.C..WC).C-0J100>4>4>4>4>4>4pppppppppppppppppppp0000000C0CCC0CCCPcCC*tJCCL).1CC000L’)01J.J10—C‘CC)CC-tJL’)CL’J00>4>4>4>4>4>40pppppppppPppppppppppppCc-COCO0CCCCOOCCCCC0000CC*C).C00CuCCI-0O-‘clC).C‘000CJL’JC..C•CC>4>4>4>4>4——CCCCCC00CCCC0CCCCCCCCCCccC)b.*00C)1OOcCC’00000CCOcCt)C\0CC00>4>4>4>4>4MCCCCCCpppppppOcCPC’CCCCCCCCCCCCCCCC*CCCCC.Cc)CL’.)‘CCCCCOCCL’)-CC-00CI-CCCC>4>4>4>4>4(A)oopopppppppppppocoocc-)CCCC0C0CCCCCCCCCCCCCCCC*CCc)cAJiC.lCCc)0L’JCC-‘-CC‘DC.cAo0’CCCC->4>4>4>4>4>4IIIIIIIIIIIIIIII1’.)I-I-CV—Cul-‘...Ct)c)t.).•.cv..0’C...çi.1.*>4>4>4>4>4>4r.A)IIIIIIIIIIIIIIIII-CCc)Ct.)i—CI—11C—.l1.CcC‘—IS)C—)0CCC*>4>4>4>4>4>40 (*)-)CCCC)CCJI-CC-‘a-CCCCC-CI)C)C-CCC-C-t)*>4>4>4>4>4-.IIIIIIIIIIIIIIIIIIII-i—c.c).1.cCIS.)IS)I—’I—’IS).rco0—))ci)CIS)i—0cçç*>4>4>4>4>4F’)IIIC.IIIIIIIIIIIIIIIIIIIIIIIcici)—.1CciIS)-CCCci)IS)ci)-..‘—.l.—C.—.IS.)ci)cicCC*!>4>4>4>4>4Ci)CIS)ti)Ci-IS)-CCCC-C)-CCCCid# alC* a2C* a3C* blC* b2Ct b3C* alP* a2P* a3P* blP* b2P* b3P*dl 0.24 0.15 0.18 0.29 0.11 0.2 21 1 24 5 2 5d2 x x x x x x x x x x x xd3 0.91 0.13 0.23 0.23 0.11 0.13 1 -1 2 -1 0 0d4 1.4 0.5 0.32 1.45 1.22 0.35 3 2 4 29 20 5cL5 0.4 0.2 0.21 0.42 0.18 0.22 -1 1 3 4 5 2d6 0.17 0.11 0.17 0.28 0.36 0.15 -2 3 7 13 3 5d7 0.19 0.08 0.14 0.08 0.21 0.16 6 6 4 -2 3 4d8 1.16 0.84 0.21x x x 10 22 4x x xd9 1.35 0.59 0.27 1.32 0.34 0.27 24 2 3 4 1 3dlO 2.29 0.24 0.37 2.01 0.84 0.38 8 1 5 11 2 4dli x x x x x x x x x x x xd12 1.87 0.11 0.38 1.93 0.14 0.4 20 2 30 18 1 8d13 0.2 0.07 0.04 0.06 0.07 0.11 12 2 3 2 2 18d14 0.29 0.16 0.26 0.23 0.08 0.26 2 0 4 5 1 1diS 0.64 0.43 0.22 0.55 0.29 0.22 3 0 — 5 -1 1d17 0.08 0.09 0.27 0.07 0.52 0.48 -1 2 13 9 4 10d16 2.12 0.09 0.09 1.9 0.07 0.06 3 1 1 3 0 0d18 1.81 0.84 0.23 1.86 0.29 0.29 7 -2 4 14 0 -2d19 0.06 0.01 0.18 0.17 0.02 0.11 3 0 8 0 -2 2d20 1.76 0.42 0.45 2.01 0.97 0.42 9 -2 5 6 15 3d21 1.78 0.87 0.44 1.76 0.2 0.39 25 2 -2 17 3 3* a (sample collected from morning 1)b (sample collected from morning 2)1 (standing sample)2 (running cold sample)3 (running hot sample)C (copper, ppm)P (lead, ppb)105