THE HEALTH AND ECONOMIC EFFECTS OF DRINKING WATER
David McCormick and Cristine Candela
Bendix Corporation
Bendbr
Applied Science
& Technology
Division
-------
THE HEALTH AND ECONOMIC EFFECTS OF
DRINKING WATER
By
David McCormick
Cristine Candela
Contract No. 68-01-1867
Project No. WA73-A368
Project Officer
Dennis Tihansky
Washington Environmental Research Center
Washington, D. C.
Prepared for
Office of Research and Development
U.S. Environmental Protection Agency
Washington, D. C. 20460
-------
EPA Review Notice
This report has been reviewed by the Environmental
Protection Agency and approved for publication.
Approval does not signify that the contents neces-
sarily reflect the views and policies of the Envi-
ronmental Protection Agency, nor does mention of
trade names or commercial products constitute
endorsement or recommendation of use.
-------
ABSTRACT
This study examines the relationship between sub-clinical common
illness such as upset stomach and other low level diseases, and quality of
drinking water and water treatment plant characteristics. Such illnesses
account for the greatest occurrence of health problems in the country.
For example, in 1969 the National Health Survey found that between 67
and 72 percent of disability days arising from illness fell into the sub-
clinical category.
The basic hypothesis tested was that if one accounts for all macro -
systemic influences on health, the occurrence of sub-clinical illnesses
will be highest in those communities with the worst water quality. The
measures of drinking water quality and water treatment plant character-
istics were obtained from a Public Health Survey study entitled, "Commu-
nity Water Supply Study". Elementary school absenteeism was used as a
surrogate measure for sub-clinical disease. The primary statistical tool
employed was stepwise multiple regression.
A linear regression was performed of elementary school
absenteeism onto total plate counts, measures of water treatment facilities,
and a series of control variables. The resulting regression was essentially
a damage function from which percentage absences resulting from incre-
mental changes in water quality were predicted. Under the assumption
that low level illness in the adult population is affected by a constant
proportion to childrens1 illness, the number of drinking water-related
absences and their associated dollar value were estimated.
111
-------
CONTENTS
Abstract
Sections
I Conclusions 1
II Recommendations 2
III Problem and Approach 3
Background 3
Community Water Supply Study, 1970 3
This Study 4
IV Literature Review 7
Human Health Effects from Water Pollution 7
Waterborne Disease Outbreaks 8
Viruses in Water Supply 12
Health and Mineral Characteristics of Water
Supplies 16
Coliform and Pathogenic Organisms 17
Bathing and Recreational Water 18
Summertime Gastrointestinal Illness 19
Benefits from Reduction of Waterborne
Disease Outbreaks 20
Summary 22
V Research Design 25
The Problem 25
Data 26
Absenteeism Data 26
Water Quality Data 30
Air Quality Data 32
Socio-Economic Data 33
Climate 34
Hypotheses 35
Analytical Methodology 40
Missing Data 41
Case Selection and Aggregation 43
The Test 45
A General Model 45
Regional Analyses 47
Six Area Grouping - Rural 50
Six Area Grouping - Urban/Suburban 51
Effects 54
Conclusions 64
VI Bibliography 66
Appendices 71
A. California Results
B. The Profile of Medical Practice
iv
-------
SECTION I
CONCLUSIONS
A consistent relationship was found between operator salary and
elementary school absenteeism, controlling for community wealth and
urbanization. The frequency of testing finished water, times the salary
of the operator(s), results in a useful index of the operation of a water
treatment facility.
Regression analyses allowed us to predict the percentage of
elementary school absences which result from incremental changes in
water quality. The results broadly indicated that a . 2 to . 5 percent
increase in school attendance can be expected for each $1000 increase
in operator salary. Nationwide and extrapolating to the adult population,
improvements of this magnitude would produce savings (in wages and
medical expenses) of approximately $250 million per year.
Conclusively, our findings show that water treatment plant
operators' salaries appear to be a useful predictor of elementary
school absenteeism. However, one cannot assume that pay raises
will immediately improve community health. The very low salaries
which are now prevalent have resulted in the recruitment of poorly
educated and undertrained operators. While, in time, higher salaries
will make possible the selection of higher caliber operators, in the
short term, the salaries would have to be supplemented by vigorous
operator improvement programs if any significant result is to be
expected.
Therefore, we computed benefits on the basis of increments in
improvements in the various measures of water quality. The various
increments predict reductions of 3% to 10% of the digestive and non-
respiratory illnesses. In the specific areas studied, this translated to
reductions in elementary school absences on the order of 300, 000 to
600,000 absences per year per incremental improvements. Converting
this to worker absenteeism using a constant of proportionality of . 62,
this becomes a savings in the order of 23 million dollars per year per
increment. For the nationwide sample, we would cautiously estimate
that elementary school absences would be reduced on the order of
magnitude of 11-12 million absences per year per incremental im-
provement. The wages and medical expenses saved would range in
the order of 250 million dollars per year per incremental improvement.
-------
SECTION II
REG OMMENDATIONS
In carrying out the study, every attempt was made to be conser-
vative in our estimates. The medical cost figures, for example, assume
that there are no respiratory diseases resulting from drinking water. If
they were transmitted in a manner similar to gastrointestinal illness,
the medical cost figures would be more than twice what we have estimated.
At all other points in the research program, when we were faced with two
alternatives, one of which would tend to inflate the results and the other
of which would tend to deflate them, we systematically chose the more
conservative.
We believe that the results indicate that if improvements are made
in the quality of drinking water, substantial savings will be realized. As
the CWSS report suggested and our study has confirmed, the problems are
most severe in the rural non-farm communities. There the economic
savings alone would be most dramatic both because the effects are rela-
tively large, and the costs of resolving the problem are relatively small.
Inmost cases, upgrading the personnel and procedures appears to be all
that is needed.
-------
SECTION -IK
PROBLEM AND APPROACH
Background
Conventional rehtoric used to be that U.S. drinking water was of the
highest quality and purity. In fact, most Americans took-water purity for
granted, considering it a fact of life. Recent studies, however, have served
to dispel such inaccuracies. Besides the alarming discoveries of carcino-
genic agents in drinking water, there have been analyses such as Craun and
McCabe's findings that I) water-related epidemics affecting an average of 100
persons per outbreak occur at an average rate of one per month; " 2)
McDermott's estimates that of annual occurrences of 500, 000 infections of
hepatitis and two million of aseptic meningitis, both can be transmitted by
water^»; and 3) Mosley's 1967 survey of waterborne disease viruses, which
opened a wide range of possible links to major health problems. )
Community Water Supply Study, 1970
In July 1970, the Department of Health, Education, and Welfare
published results of their own nation-wide evaluation of drinking water
quality and water treatment plants. This Community Water Supply Study
(CWSS) shows that the quality of drinking water is far below general ex-
pectations.4' Significant results were:
Forty-one percent of the water systems, delivered inferior
water to 14% of the study population.
More than one-third of the tap water samples contained con-
taminants in excess of Public Health Service standards.
Over half of the systems had physical or operational de-
ficiencies.
1) G. Craun, L. McCabe, "Waterborne Disease Outbreaks, 1961-1970".
Annual Meeting of the American Water Works Association, 1971.
2) J. H. McDermott, "Virus and Water Quality: Occurrence and Control -
. Conference Summary" (13th Water Quality Conference, February 1971),
p. 213.
3) Mosley, James, "Transmission of Viral Diseases by Drinking Water"
from Bery (eds.) Transmission of Viral Diseases by the Water Route,
New York, 1965.
4) McCabe, Leland J. , J. Symons, R. Lee, and G. G. Robeck, "Survey
of Community Water Supply System", Journal of the American Water -
works Association, Vol. 62, No. 11, November 1970, p. 670-687.
-------
• More than three-fourths of the plant operators had inadequate
training in fundamental water microbiology; nearly half in
application chemistry.
• Control, inspection, surveillance, and testing programs were
entirely lacking, infrequently performed, or well below stand-
ards in the vast majority of systems.
Surprising as they may be, the CWSS results are probably optimistic,
since water quality measurements and standards suffer from a lack of virus
data. Direct methods for measuring viruses in large quantities of water are
under development but not yet available. Although present water purification
technology can handle all significant bacterial problems, in practice, availa-
ble technology is not always applied. And it is far from certain that viruses
and harmful chemicals can be easily removed with existing technology.
In recent years, the American press and popular literature have re-
flected a growing public awareness of environmental pollution. At the same
time, public attention has focused on the cost in higher taxes and consumer
prices to clean it up. Confronted with this price tag, taxpayers and the
decision-makers who represent them naturally ask what they're buying.
In the area of water pollution control, there are numerous payoffs, including
direct economic benefits such as commercial fishing, measured real estate
values, increased recreational use, and generally improved aesthetic value,
all of which contribute to improving quality of life for Americans.
Among the costs incurred by our degraded water supplies are those of
purifying it to make it safe to drink. Alternatively, of course, failure to purify
the water sufficiently can be argued to increase the possibility of illness.
However, the question arises whether and to what extent incremental
degradation of the drinking water has negative effects on community health.
If such effects exist, there are associated, definitive economic implications
both on the cost of operating the treatment facilities and on the costs of im-
proving the quality of the intake water quality.
This Study
Numerous studies have already been undertaken to examine the effects'
of water degradation on serious communicable disease. In view of this fact,
this study proposed an examination of data relating to sub-clinical illnesses
such as gastroenteritis and other non-specific low level diseases. Such
illnesses account for high occurrence of health problems in the United States.
In 1969, for example, the National Health Survey found that between 67 and
72 percent of disability days arising from illness fell into the sub-clinical
illness category.
Based on the assumption that the economic costs of sub-clinical ill-
nesses affect a far wider distribution of people than do major illnesses,
lost productivity alone due to sub-clinical absences in 1969 totaled in the
-------
general range of $4 1/2 billion dollars.* Consequently, the costs of sub-
clinical illness are large and, in fact, are more than twice the total EPA
1974 estimated budget outlays.
This study was, therefore, undertaken with the basic hypothesis
that any significant health effects from water degradation would result pri-
marily in sub-clinical illnesses. To perform the study, some reliable
measure was needed for community health. Worker sick time data, the
most direct measure are not usually reported for a community. Even where
they are available, sick time data are notoriously unreliable and can tend to
follow a form of Parkinson's Law: sick days off expand to absorb sick days
allowable. Therefore, a widely reported indicator of community health had
to be chosen. We selected absenteeism from elementary schools as our
indicator for the following reasons:
• Records in most school districts are reasonably accurate
and complete.
Elementary school children represent a wide spectrum of
socio-economic classes in a community.
Truancy is less serious problem for early grades, than for
high school.
• Both children and adults are susceptible to low level illnesses;
therefore, school absenteeism can be linked to work absenteeism.
• Biases in absenteeism due to non-water related phenomena
can be partly explained by measurable variations in socio-
economic community characteristics.
For the communities studied, we performed a linear regression of
elementary school absenteeism onto total plate counts, measures of the
treatment facilities, and a series of control variables.
The resulting regression is essentially a damage function from which
we were able to predict the percentage absences resulting from incremental
changes in water quality. Under the assumption that low level illness in the
adult population is affected by a constant proportion to childrens1 illness,
the number of drinking water related absences and their associated dollar
value were estimated. This was done for the nine areas used in the CWSS
study: the State of Vermont, and eight standard metropolitan statistical
areas (New York, N. Y. ; Charleston, W. Va. ; Charleston, S. C. ;
Cincinnatti, Ohio; Kansas City, Missouri-Kansas; New Orleans, Louisiana;
Pueblo, Colorado; and San Bernadino-Riverside-Ontario, California. In
Based on 368 million disability days to persons 14 and over due to acute
illnesses, 67% of which are sub-clinical, a median, daily wage of $21.00,
and 58% employment.
-------
addition, extremely general estimates were made for the total nation.
The ensuing sections of this report will review:
1.) the findings from a literature review covering re-
search on water quality relationships to low level
illness;
2.) the research design and methodology, and
3.) description of analysis and findings.
Five appendices list data resources. A bibliography
provides references relevant to the research.
-------
SECTION IV
LITERATURE REVIEW
Human Health Effects from Water Pollution
Method - To ascertain what is already known about the costs of water
pollution with relation to low level illness, a survey of the relevant
literature was conducted. Although several articles are discussed
and quoted with regard to water quality and illness, the major studies
reported were reviewed critically for the utility of their objectives,
fulfillment of these objectives, and the analytical methods used.
Sources are mentioned wherever given. Emphasis is given to con-
sistent findings between water quality and illness, with attention
to geographical aspects involved.
Related Research
There is still a paucity of research in the area of water and human
health, particularly with respect to low level illnesses. Also, there
is a dearth of records and statistics dealing with minor illnesses.
In this literature review, the goal is to complement our research
objective to measure the economic costs to society in lost wages and
medical expenses arising from low level illness associated with water
quality. The inadequacy of background data is significant. Krishnaswami
has pointed out, "that man's direct and indirect contact with water and
the aquatic..environment varies considerably, and that the total impact of all
environmental factors - air, water and food - cannot be compart-
mentalized in assessing their significance to health and disease".5)
In a quote from Dubos, it is stated that: "Experimental and
epidemiological studies (e.g., common cold and infectious hepatitis,
respectively), have shown that even in the transmission of pathogenic
organisms and the production of overt disease ... a simple relation-
ship of causal specificity does not always exist. The presence of a
pathogen or a toxicant is only a necessary condition, but not a
sufficient condition for the causation of a disease. " °'
5) S. Krishnaswami, "Health Aspects of Water Quality" American
Journal of Public Health, 61:2260 (November 1971). For
a literature search of published and unpublished estimates of
the cost of air pollution damage, see Larry B. Barrett and
Thomas C. Waddell, Cost of Air Pollution Damage: A Status
Report, EPA, February 1973. Also see literature review
contained in Lester B. Lave and Eugene P. Seskin "Air
Pollution and Human Health", Science 169, 21 August 1970,
723-733.
6) Dubos, Rene1. Man, Medicine and Environment. New York:
The New American Library, Mentor MY 942, 160 pp. 1969.
-------
Krishnaswami adds that, "The overt clinical manifestation of severe
chemical deficiency or of acute toxicity can be evaluated and their
causes discussed relatively easily. However, the role of water con-
stituents (relatively significant concentrations of various synthetic
organics, heavy metals, and other chemical toxicants) in the probable
multifactorial causation or contribution to chronic ailments and other
bodily stresses have not received serious consideration. "?)
Thus, in conducting studies in the area of water quality and minor
illness one must deal with the fact that their occurrence is negligibly
reported. Many individuals will forego any physician consultation
during low level illnesses, but additionally, records that do exist
are incomplete as to causation, circumstances, length of illness, etc.
Waterborne Disease Outbreaks
Weibel. et al.8)
Despite such limitations, Weibel, et al, were able to compile diseases
by identity finding that the number of reported epidemics and cases
of waterborne disease outbreaks in the United States between 1946 and
I960 were primarily classified as "gastroenteritis". (See Table 1.)
"This category and 'diarrhea' (16 outbreaks and 5,160 cases) are not
specific diseases, because the etiologic agent has not been determined
through laboratory analysis. Even with laboratory analysis, the
causative agent cannot always be determined.")
7) S. Krishnaswami, p. 2261.
8) S. R. Weibel, F. R. Dixon, R. B. Weidner, and L. J. McCabe,
"Waterborne Disease Outbreaks, 1946-60", Journal of the
American Waterworks Association, 56:-949 (August 1964).
9) S. Krishnaswami, p. 948.
8
-------
TABLE 1
10)
Waterborne-Disease Outbreaks in the United States,
1946-60, by Type of Illness and System
nine s s
Private or Semi-
public Systems
Outbreaks
Cases
Public Utilities
Outbreaks
Cases
All Systems
Outbreaks
Cases
Gastroenteritis
Typhoid
Infectious hepatitis
Diarrhea
Shigellosis
Salmonellosis
Amebiasis
Other
Total
92
33
14
7
4
3
2
3
158
4,233
403
430
320
596
22
36
16
6, 056
3411J
6
9
9
7
1
0
4
70
9, 397
103
500
4,840
5,057
2
0
29
19,928
126
39
23
16
11
4
2
7
228
13,630
506
930
5, 160
5,653
24
36
45
25,984
10) Weibel, et al.
11) One gastroenteritis outbreak also included a typhoid case.
-------
Weibel, et al also report on a review of waterborne-disease outbreaks in
the United States and Canada by Gorman and Wolman/^'covering the period
from 1920 to 1936, in which the case distribution (116, 000) by type of ill-
ness was: 88 percent diarrhea, 11 percent typhoid fever, and 1 percent
dysentery. A second review by Eliassen and Cummings 'covered water-
borne outbreaks in the United States between 1938 and 1945 with a case
illness distribution (110,000) of: 91 percent gastroenteritis (which prob-
ably corresponds to the "diarrhea" classification of the earlier study); 8
percent dysentery? and, 1 percent typhoid.
The Eliassen and Cummings report used the term "outbreak" to refer to
two or more cases; the Gorman and Wolman report defined it as consist-
ing of five or more cases.
Weibel, et al also show a distribution of outbreaks by month for the years
1946-60^ which indicate a peaking of outbreaks due to private supplies in
the summer months. (See Table 2. )14) This is of extrinsic interest to. our
research in that our stress is rather on usage of school absenteeism, data, covering
the time periods of September to June. It is a finding which is given
corroboration by more recent analyses done by Petersen and Hines, which
will be reported on in a later section of this report.
Unfortunately, the o~nly explanation offered with regard to causal factors of such
occurrence is either greater pollution during the summer months, or pos-
sibly increased usage by more susceptible individuals (tourists, campers,
travelers, etc. ).
G. Craun, L. McCabe15)
During 1961-1970, Craun and McCabe reviewed 128 known outbreaks of
disease or poisoning attributed to drinking water, with 46, 369 illnesses
and 20 deaths. Their definition of "outbreak" reflected at least two cases
of infectious disease, associated with water used for drinking or domestic
12) Gorman, A. C. & Wolman, A. , "Waterborne Outbreaks in the
United States and Their Significance. Journal of the American
Water Works Association 31:222 (February 1939).
13) Eliassen, R. 8t Cummings, R. H. "Analysis of Waterborne Out-
breaks, 1938-45" Journal AWWA, 40:509 (May 1948).
14) Weibel, et al, p. 954.
15) G. Craun and L. McCabe, Ibid.
10
-------
TABLE 2
16)
Seasonal Distribution of Waterborne-Disease
Outbreaks in the United States, 1946-60, by
Type of System and Affected Population
Month
January
February.
March
April
May
June
July
August
September
October
November
December
Total
Month
unknown
Grand
total
Public
Utilities
%
7.6
10.6
7.6
7.6
6.1
3.0
12.1
9.1
9.1
6.1
9.1
12.0
100.0 .
Private or Semipublic
Systems
Percent-
age of
Total
3.9
3.9
2.6
5.9
6.6
12.5
29.0
15.8
8.6
6.6
2.0
2.6
100.0
Affected Population
1*
3
4
3
4
3
7
10
4
4
6
3
1
52
0
52
2f
1
0
0
1
2
2
0
1
1
3
0
1
12
3
15
3+
2
1
1
4
5
10
34
19
8
1
0
2
87
2
89
4§
1
1
1
2
*Habitual users of supply.
•^School children.
^Visitors.
§ Unclassified
16) S. R. Weibel, et al
11
-------
purposes. Data sources included state health departments,
medical and engineering literature, newspaper clippings, polls of state
sanitary engineers and epidemiologists.
Their findings concluded that on the average, one waterborne outbreak
that is known about occurred per month with something over 100 persons
becoming ill. Some of the illness was quite severe and about two deaths
per year occurred from waterborne outbreaks.
There were 94 outbreaks in private supplies and 34 outbreaks in public
supplies. Most of the outbreaks were classified as gastroenteritis; these
include 38 outbreaks and 25, 800 cases.
Viruses in Water Supply
Dr. Luther L. Terry, U. S. Surgeon General, was quoted in '62 as saying,
"We are by no means sure that at least some viruses are not slipping
through our present water purification and disinfection processes and en-
tering our water mains". '
Jh reporting on the 13th Water Quality Conference, McDermott 'summa-
rized the reports on virus and water quality as follows:
1. Most speakers agree that there is a growing concern with
the problem of viral pollution of water in this country.
2. It was generally agreed that, because the basic properties
of viruses are poorly understood, their transmission by the
water route has not been fully appreciated. However, sufficient
evidence has been presented to support the thesis that virus can
be transmitted by water and result in human infection and disease.
3. The speakers unanimously agreed that there is insufficient
technical data now at hand to document conclusively the idea that
virus-free water can be attained with our present wastewater
treatment and domestic water supply treatment processes and
practices.
17) Quoted in Report of Committee on Recreation Use of Sources of
Public Water Supply, N. Eng. Water Works Association, Journal
New England Water Works Association, 85:1:92 (1971).
18) J. H. McDermott, Ibid
12
-------
4. Thus, immediate attention should be given to additional studies
and investigations in the following subject areas: viral detection
and enumeration methodology, basic properties of enteric virus,
transmission through the aquatic environment, viral disease of
man and associated epidemiological studies, and unit process re-
search and development.
In a discussion of viral disease and epidemiology McDermott reports that
we can reasonably be assured that the transmission of human enteric
diseases are largely dependent upon water as the vehicle.
Metcalf and Melnick presented papers at the conference indicating that one
strain of virus may produce illnesses with widely variable incubation periods
and manifestations, e. g., a coxsackievirus may produce meningitis in one
person, myocarditis in another, and diarrhea in a third, etc. ''
S. F. B. Poynter20)
S. F. B. Poynter, Senior Bacteriologist, Metropolitan Water Board,
addresse s this same area of concern: Infection may be brought about by
ingestion of virus contaminated food or water or by downward spread from
the respiratory tract. The prolonged excretion of the virus in the faeces
which follows infection causes faecal contamination to be a major factor in
their spread, and probably most enterovirus epidemics are started by
symptomless excreters.
Poynter also specifically indicated that Adenoviruses are aetiologically
associated with upper respiratory tract infections.
James W. Mosley21)
Mosley attempted to classify how often water is responsible for transmission
of viruses responsible for infectious disease. He reported that infectious
19) T. G. Metcalf, "Biologic Parameters in Water Transmission of
Viruses;" J. L. Melnick, "Detection of Virus Spread by the Water
Route, " Proceedings of the 13th Water Quality Conference, Univer-
sity of Illinois at Urbana-Champaign (1971).
20) S. F. B. Poynter, "The Problem of Viruses in Water", Water
Treatment and Examination, 17:187 (1968).
21) Mosley, James, Ibi d.
•13
-------
hepatitis is the only disease caused by an agent having the characteristics
of a virus, for which evidence of waterborne transmission has been accept-
ed by all workers in the field. Nonetheless, he considers relatively
neglected the entities known as viral gastroenteritis and diarrhea, although
numerous waterborne epidemics are documented.
He points out that the limitations of using the term "viral" to apply to gas-
troenteritis or diarrhea are that it provides a wastebasket for those episodes
due to agents other than viruses not yet identified as pathogenic, as well as
for those instances in which the search for known pathogens failed because
of delay in the investigation or technical inadequacies in the laboratory.
"Despite these limitations, experimental studies have in-
dicated that forms of gastroenteritis and diarrhea trans-
missible from person-to-person are due to agents with
characteristics of viruses (Gordon et al, 1949; Jordan
et al, 1953; Reimann, 1963). It seems plausible, there-
fore, that under some circumstances such agents could be
waterborne. In addition, it has been possible to associate
epidemics of gastroenteritis and diarrheal disease in new-
borns with infection by enteroviruses (Eichenwald et al»
1958; Lepine et al, I960). The extent to which enteroviruses
are responsible for these syndromes in older children and
adults is less clear. Echovirus 11, however, has been re-
ported to have caused gastroenteritis in adults working with
this agent in the laboratory (Cramblett et_al» 1962; Klein et
al, I960). No association of enteroviruses with waterborne
episodes has been achieved, but it is uncertain whether any
serious attempts have been made to do so. "
Finally, Mosley indicates that evidence for waterborne transmission of
viral disease has been based primarily upon epidemics in which cases
occurred within a sufficiently short period of time to make person-to-
person transmission an unlikely explanation.
There is "feeling" at this time that the role of water transmission of viruses
is, therefore, especially related to municipal systems using surface water.
14
-------
O. C. Liu*
In a personal communication, Dr. Liu addressed the question of dollar
costs associated with occurrence of viral illnesses. He believes that from
available epidemiologic data, virus is perhaps the most important water
pollutant which could produce harmful effects in man.
Pointing out that water may be polluted by biologic, chemical and radio-
active pollutants, Liu suggests that by eliminating these pollutants,
human health may be benefitted and these benefits may be converted into
dollar values. Health benefits are enumerated as either dollars saved
for medical expenses as if the patient had not been sick and/or the
dollars which could be earned in one's lifetime if the patient dies.
Liu groups diseases produced by enteric viruses into two major
categories:
a) Acute clinical illness
b) Health effects from subclinical infections
The subclinical infections are of primary concern to us in this study.
Liu reports that subclinical infections of enteric viruses in man, repre-
senting 100-1000 times the incidence of clinical infections, are far from
harmless. Certain numbers of enteric viruses have been associated with
serious delayed effects.
He states that the whole group of enteric viruses may potentially be trans-
mitted by water through: a) drinking and culinary use; b) recreation; c)
agriculture; and d) food.
* - O. C. Liu, M. D. Personal Communication, Office of
Research and Monitoring, Environmental Protection
Agency, Narragansett, Rhode Island (1972).
15
-------
With the present knowledge, therefore, he estimates that health benefits
could be of two extremes:
(a) One extreme is represented by use of the published data oij waterborne
disease incidence as a basis for the health benefit estimate. Practically,
these include only two diseases: infectious hepatitis, and gastroenteritis or
diarrheal disease. But he personally feels that the reported figures (Mosley,
Wiebel, et al, etc.), probably represent only a fraction of those which
actually occurred. "If we estimate health benefits according to the above,
the total value will be too low to be realistic or true. "
(b) The other extreme is to estimate the benefits by assuming that eradica-
tion of all enteric virus diseases incidence may be achieved if the water
pollution from virus is under control. Liu feels there is circumstantial
evidence to indicate that the latter might be the case. "Intensive research,
however, is badly needed in this area in order to definitely identify this
problem. The long and insidious route of transmission by water is not at
all understood at present. Water, however, may serve as an essential
link in the chain of events in perpetuating these viruses. Once this link
is broken, the virus finally may stop to exist among human hosts thereby
eradicating these diseases. "
Liu gives his estimated dollar value which may be saved per annum, if
indeed eradication of enteric virus diseases in this country can be
accomplished by control of water pollution. For acute gastroenteritis
and diarrheal diseases (2,000, 000 work days lost/year at $30/day) the
rough estimated figure is $60 million.
Health fa Mineral Characteristics of Water Supplies
Neri, Hewitt and Schreiber^Z)
In a recent review for the Journal of Epidemiology, Neri, et al present
a synopsis of findings on literature investigating relationships between
health and mineral characteristics of local water supplies. They
remark on the general impression resulting from the literature reviewed,
(53 journal articles and papers) "that some 'water factor' does exist, but
no agreement is in sight concerning the identity of this factor, its likely
mode of action or even the pathological effects produced by it".
22) Luciano C. Neri, David Hewitt and George B. Schreiber, "Can
Epidemiology Elucidate the Water Story?" Journal of Epidemiology
99:2, February 1974, 75-88.
23) A comprehensive review of the literature is contained in Masironi, R. ,
Miesch, A. T., Crawford, M. D., et al, "Geochemical Environments,
Trace Elements & Cardiovascular Diseases, " (forthcoming). Another
aspect of water mineralization damages can be found in Tihansky,
' Dennis, "Economic Damages from Residential Use of Mineralized
Water Supply", Water Resources Research, April 1974.
16
-------
An overview of the findings from this literature review indicates that water
quality relates to cardiovascular disease, or some subdivision of it. But as
Neri, et al point out, the support for this finding is not yet even very strong.
In fact, "'it has been progressively weakened by the admission of bronchitis
'effect1, ' by the suggestion that there is also an infant mortality 'effect1
and by-the estimate based on Canadian statistics that more than half the ex-
cess mortality in soft water areas is certified to noncardiovascular cause
of death.26)
In their review, Neri, et al emphasize the need for an increase in epidemi-
ological studies conducted in a systematic fashion, with due regard to the
need for specifying and testing explicit hypotheses. They recommend studies
which utilize a method such as partial correlation analysis. Their stress on
more rigorous analyses in the field of epidemiology is illustrative of a weak-
ness inherent in research - no matter what the discipline - which accents
descriptive and anecdotal criteria in the achievement of results.
Coliform and Pathogenic Organisms
T. Virarghavan2 " indicates that in studies of water quality, "The coliform
organisms, because of their large numbers in fecal matter, offer a far more
satisfactory approach to the detection of fecal contamination than do the
pathogenic organisms. A search for pathogenic organisms is very difficult
and time consuming and may prove fruitless when fecal contamination is
present but specific pathogens are absent". He further points out that the
water utility profession relies on coliform examination to determine whether
water is free from disease producing microorganisms.
24) Crawford, M. D. , Garner, M. J. , Morris, J. N. "Mortality &:
Hardness of Local Water Supplies, " Lancet 1:827-831, 1968.
25) Lowe, C. R., Roberts, C. J. , Lloyd S., "Malformations of
Central Nervous System & Softness of Local Water Supplies, "
British Medical Journal, 2:357-361, 1971; and Crawford, M. D. ,
Gardner, M. J., Sedgwick, P. A. , "Infant Mortality and Hardness
of Local Water Supplies". Lancet, 1:988-992, 1972.
26) Neri, L. C. , Mandel, J. S. , Hewitt, D., "Relation Between Mortality
& Water Hardness in Canada", Lancet 1:931-934, 1972.
27) T. Viraraghavan, "Water Quality & Health" Journal of the American
Water Works Association (October 1973) pp. 647^o50.
17
-------
Virarghavan makes the statement that although pathogenic organisms may
survive water treatment that apparently removes all coliform bacteria,
the evidence available from outbreaks of waterborne disease from treated
public water supplies shows that the coliform examination has been a satis-
factory measure of the microbial safety of water.
In this same regard, Krishnaswamir states, "There is no evidence that
bacterial, enteroviral and other microbial diseases are frequently trans-
mitted to man by water that meets the relatively stringent bacterial, dis-
infection, and chlorine residual standards for drinking water".
Bathing and Recreational Water
It is of interest to note that while uniformity exists for bacteriological
standards of drinking water in the United States, most of the states have a
variety of bacteriological standards for bathing water.
The most pertinent study completed in the bathing water quality and health
area was carried out by Stevenson prior to 1952.^°' The goals of the r.esearch
were to determine what frequency of swimming might be expected from the
population groups chosen; to determine what relative increase in illness in-
cidence might be expected in the cleanest waters; and to determine what
differences in illness incidence might be expected from swimming in water
containing various degrees of bacterial pollution.
The studies were conducted over two months with families living near 1)
a great lake; 2) an inland river and a fresh water recirculating pool, and
3) a tidal water. Stevenson's overall findings were the following:
a) An appreciably higher overall illness incidence may be ex-
pected in the swimming group over that in the non-swimming
group.
b) When the total illnesses among swimmers and non-swimmers
were compared, there appeared to be no significant correlation
between the illness incidence and the quality of water in the
areas studied.
c) Some 100 percent higher illness in the group under 10 years of
age was shown than for those over 10 years of age.
d) Among swimmers, eye, ear, nose and throat ailments represented
more than 50 percent of the illnesses reported; gastrointestinal
disturbances about 20 percent and skin irritations the remainder.
28) Krishnaswami, p. 2262
29) Stevenson, A. H. , "Studies of Bathing Water Quality & Health",
American Journal of Public Health, 43:5, 1953, 529-538.
18
-------
e) Specific correlation between illness incidence and bathing
in waters of a particular bacterial quality was observed in
two instances. Illness frequency was significantly higher
among swimmers when the average total coliform density
was up to 2700 per MPN rather than below 43 percent MPN.
Stevenson was particularly cautious about the latter finding as only three days
were on each side of the comparison; plus, thorough analysis of the effect
that weather might have had on the result was not permitted.
McKee indicated findings which showed that sewage-polluted sea water
carried only negligible risk to health unless the water was "highly" polluted.
What remains unknown is the generalizability of these findings across time
and place. Research in water recreation and health has received particu-
larly little attention, due to the difficulties inherent in carrying out a valid
study (i.e., data collection problems, costs involved in a study of statis-
tically significant proportions, difficulties with data multi-collinearity,
etc.). Nonetheless, the findings described above, as well as other studies
•which reveal rising illness in the summer months, indicate the need for
program emphasis in this research area.
Summertime Gastrointestinal Illness
Petersen and Hines carried out a study on the relationships between summertime
gastrointestinal illness and the sanitary quality of water supplies in six rocky
mountain communities. 31) The research performed is notable for its employment
of rigorous statistical analyses. Additionally, Petersen and Hines identify their
analytical methods and data sources.
Their query examines existence of a correlation between unreported gastro-
intestinal illness and the populations using the water supplies. Approximately
100 families in six Rocky Mountain communities were surveyed. No schools
were in session during the period.
Variables included: family experience with gastrointestinal illness during June,
July and August of 1957. (Gastrointestinal illness was defined as diarrhea with
or without nausea and vomiting, persisting for at least one day); and bacterio-
logical water analysis for each town over the past 3 years.
Illness experienced was obtained by means of survey on a house-to-house basis
by 2 investigators, using a standardized form, directed to one responsible
household person, usually the mother. Data were recorded on McBee cards
specifically designed for gastrointestinal illness. Table compilations were
made up of the observations.
30) McKee, J. C. and Wolf, H. W. "Water Quality Criteria", California
State Water Quality Board Publication No. 3-A, 1963.
31) "The Relation of Summertime Gastrointestinal Illness to the Sanitary
Quality of the Water Supplies in Six Rocky Mountain Communities",
American Journal of Hygiene, Vol. 71:314-3ZO, November 1959.
19
-------
Bacteriological water analysis was taken from the office records of the State
Health Department. Numerical values were assigned based on a demon-
strated consistency of quality over a three year period.
Data manipulations were conducted by means of simple chi square tests of
epidemiological factors associated with the illnesses showing significant
differences between the group of communities with water of low sanitary
quality and those in the group having high sanitary quality.
Their findings concluded that no patterns could be found from analysis of
chronological arrangement of cases of illness in each community by date
of onset, which could be interpreted as being the result of a common source,
single exposure type of infection. Only 20 of the 206 cases showed a history
of eating away from home during the week preceding illness. No correlation
was found between incidence of illness and particular brands of milk.
No correlation was found to environmental sanitation deficiencies, because,
of 570 premises surveyed, all had running water inside the house, and only
three had sewage disposal of a type other than an indoor, water-flush system.
Communities with water supplies of low sanitary quality were found to have
an overall attack rate of 13. 9 percent, while communities with water of high
sanitary quality had an overall attack rate of 8. 8 percent. This correlation
between water quality and attack rate was found consistently when computed
separately for each community.
The attack rate for males was higher than the rate for females in both groups
of communities with no significant difference. A concentration of illness
among the young appeared to be greater in communities with water supplies
of high sanitary quality. Attack rates computed by family size indicated no
consistent gradation. The percentage of cases occurring in July and August
indicated a definite increase as the summer progressed, for both groups of
c ommuniti e s.
Interestingly, an overall attack rate among people who had lived in the
communities with low sanitary quality water supplies for less than two years,
•was twice that of the people who had lived in the communities more than two
years.
Benefits from Reduction of Waterborne Disease Outbreaks
Jack Lackner did the only cost-benefit study we found at all attributable to
•water quality effects on some low level illness. ' He queried the extent
to which benefits from anticipated reduction in waterborne disease outbreaks
and health damage can be predicted! His point of reference •was the Safe
Drinking Water Act of 1973.
32) "Safe Drinking Water Act of 1973: Estimated Benefits and Costs:,
EPA Office of Planning and Evaluation.
20
-------
He examined disease data in the United States, 1961-1970; flourida-
tion practices in the United States (1967 capital investment data con-
sidered in relation to 160 million people served); and, future water
supply system upgrading (costs for 1600 supplies).
Data and sources were: 1) Disease incidence from Morbidity and
Mortality, annual supplement, Summary 1970, USDHEW, Center for
Disease Control; 2) Bacteriological data, effectiveness of flourida-
tion, and lead excess data from the "Community Water Supply Study ";
3) Capital investment data in public water supply facilities from 1967
Department of Commerce data and; 4) Department of Commerce and
the American Water Works Association estimates on future costs of
upgrading water supply systems to comply to 1962 PHS Drinking Water
Standards.
Data manipulations included the following: 1) Judgmental estimates
were made by Public Health Scientists that 1/3 or approximately
245, 000 of reported disease cases are in fact waterborne. An assump-
tion that the recording of cases of waterborne'disease is about 10 per-
cent of actual number of cases of waterborne disease, increased the
number to 2,450, 000 in the 10 year period. 2) Estimation of in-
come at $25 per day and hospitalization at $200 per day. 3) Assump-
tion of 1,000,000 cases of drinking water gastroenteritis per year at
comparable costs of $50 per case. 4) Correction of poor flouridation
. using "Community Water Supply Study" indicator of 38 percent of the
86, 136,000 people receiving ineffective levels of flouridation as it
pertains to dental disease. 5) Capital investment data in public water
supply facilities considered in relation to the 160 million people served
at $312 per capita. 6) Assumption of three chlorinators needed per
system for $24, 000 systems at a cost of $3, 000 each. 7) Community
water supply figures indicating that 1,000 supplies need defluoridation
equipment, and 600 supplies need correction of failure to meet constituent
standard of lead.
Lackner's findings included a total for estimated annual benefits of
$810, 000, 000. This figure was derived from the $306 million attribut-
able to reductions in communicable diseases; $50 million attributable
to reductions in the incidence of gastroenteritis; $454 million attribut-
able to reductions in dental decay.
DOC and AWWA data estimated that $6 billion should be spent at a rate
of $500 million per year over a 15 year period for improvement of water
supply facilities. The yearly cost of program to correct inadequate
design, construction, operation and maintenance of water supply facili-
ties, was estimated at $2. 5 billion. The direct Federal cost of imple-
menting legislation on water quality was estimated at $27 million in
FY '74, increasing to $63 million in FY '78; estimated cost to states
was $50 million each year; estimated costs to communities was $165
million annually. The additional future costs required to upgrade all
water supply systems was estimated at $6 billion, to be spent at the rate
of $500 million per year.
-------
Summary
Overall, the Petersen and Hines morbidity study revealed a signifi-
cant relationship between the sanitary quality of a community's water
supply and the incidence of gastrointestinal illness in that community.
Communities served by contaminated water suffered an attack rate
of 14 percent during the summer months as opposed to the 9 percent
attack rate for people using water supplies showing no contamination.
The latter finding corresponds to the Weibel, et al, peaking of water-
borne disease outbreaks due to private supplies, in the summer
months.
Lackner's analysis resulted from studies showing that 30 million
people, served by 8,000 community water supply systems, were re-
ceiving inadequate flouridation in their water for maximum benefit
to their dental health. Furthermore, 64. 4 million people, served by
approximately 16, 000 systems, were receiving water containing
dentally insignificant flouride ion levels. His cost benefit study gives
rough estimates of monetary and health savings derivative from effec-
tive flouridation levels, which are calculated to be in the millions. No
other researcher to our knowledge has published findings comparably
achieved.
Of the literature reviewed relevant to water quality and low level ill-
ness, the above two studies were the most analytically rigorous, and
empirically definitive.
More precise analyses have been undertaken in the area of air pollu-
tion and health, several of which are quantifiably verifiable. Partic-
ularly notable is work done by Ridker 33) on morbidity, mortality
and respiratory disease, and the subsequent studies by Lave and Seskin
34) which expanded the diseases covered by Ridker. Ridker estimates
that 18 to 20 percent of the approximately $2 billion in national health
costs results from air pollution. Lave and Seskin, in a conclusion
which they consider to be conservative, estimate that air pollution
damage amounts to 4. 5 percent of all the economic damage associated
with morbidity and mortality.
Similar findings are being reported which specifically indicate relation-
ships between air pollution and serious illness. For example, a recent
report by Henderson, Menck and Casagrande of the University of
Southern California indicates that residents of South Central Los Angeles,
who have for years lived near oil refineries and chemical plants, are
more likely to die of lung cancer than people who reside elsewhere.
33) Ridker, Ronald G., Economic Costs of Air Pollution.
New York, Frederick A. Praeger, 1967.
34) Lave, L. B. and E. P. Seskin. "Air Pollution & Human
Health", Science 169(3947): 723-733, August 21, 1970.
22
-------
In a study which specifically addressed air pollution vis-a-vis low
level illness, residence data again proved of consequence: Chapman,
et al, queried to what extent air pollution exposures related to the
prevalence of chronic respiratory disease. 35) Using data from
New York, findings suggested that moving from a polluted area to
a clean one can promote substantial improvement in symptoms.
Other evidence from the same study suggested that sulphur oxide
pollution exerts an effect on chronic respiratory disease which may
rival the effect of cigarette smoking, particularly in areas where
high exposures to sulphur oxide are coupled with suspended
particulates.
A final air pollution/low level illness study by Thompson, et al, in-
vestigated the effects of air pollution on the common cold. 3°)Using
multiple regression it was found that statistical analyses differed
markedly by season of the year and by whether incidence or common
cold was being examined. Meteorologic variables used appeared to
be more related to the common cold rates than the pollutant variables,
but examination of the data failed to "explain" the reasons underlying
this finding.
Overview of Findings
While highly definitive findings examining water pollution relationships to
low level illness are still in the future, of the studies conducted to date, the
evidence points to probable significant relationships. An overview of find-
ings from the literature reviewed can be summarized as follows:
• The role of water as a causal factor in occurrence of low level
illness has not received serious consideration by scientists.
This results partially from the fact that in both water and air
pollution studies, the role of environmental pollutants in
mortality and chronic disease is difficult to quantify because
so many other determinants of death and disease cannot be
adequately measured. A simple relationship of causal specificity
does not always exist.
Studies of waterborne illness covering the years 1920-1936, and
1946-1960 have classified over 80 percent of the cases as gas-
troenteritis or diarrheal illness. While not specific diseases
because no etiologic agent can be identified, these categories
of illness account for the vast majority of American illness due
to waterborne sources.
35) Chapman, Robert S. et al, "Chronic Respiratory Disease Symptom
Frequency in Late Adolescence, In Parents of School Children and
in Elderly Persons", Presented at the American Medical Associa-
tion's Air Pollution Medical Research Conference, Chicago, 111. ,
October 1972.
36) Thompson, et al, "Health and the Urban Environment: Air Pollu-
tion, Weather and the Common Cold", Journal of Public Health,
Vol. 60, No. 4(1970), 731-739.
23
-------
Additional studies are needed in the area of viral detection, enumera-
tion, transmission, and elimination because their basic properties
are poorly understood.
There is sufficient evidence that virus can be transmitted by water
and result in human infection. However, intensified research has
yet to prove conclusively whether bacterial, enteroviral and other
microbial diseases are frequently transmitted to man by water
that meets the bacterial, disinfection and chlorine residual stand-
ards for drinking water.
Past evidence for waterborne transmission of viral disease has
been based primarily upon epidemics in which cases occurred
within a sufficiently short period of time to make person-to-
person transmission an unlikely explanation.
There is evidence that viruses may produce widely variable
incubation periods and manifestations in different individuals.
Some sources indicate that the role of water transmission of viruses
is especially related to municipal systems using surface water.
Although present water purification technology can handle all
significant bacterial problems, in practice, available technology
is not always applied. Many water systems are overage, sub-
standard and unable to meet peak demands.
Some morbidity studies reveal a significant relationship between the
sanitary quality of a community's water supply and the incidence
of gastrointestinal illness.
Research is needed in the area of bathing and recreational water
to study indications that swimmers suffer from eye, ear, nose,
throat ailments, and gastrointestinal disturbances more often
than non-swimmers.
Two studies have shown a peaking of waterborne illness in the
summer months, particularly as this season progresses.
Residence data in recent studies indicate that longevity in a
polluted area contributes to symptoms of serious diseases.
However, an overall illness attack rate among people who have
lived in communities for shorter periods of time can be observed
as being higher than for people who have lived in communities for
longer periods (for example, more than two years).
One estimate states that $60 million can be saved per annum through
eradication of gastroenteritis and diarrheal disease from water
pollution.
24
-------
SECTION V
RESEARCH DESIGN
The Problem
The driving question of this research project developed from the CWSS
study of drinking water around the nation. If, indeed, drinking water is not
uniformly good and water treatment standards are not consistently followed,
does this have any health implications with economic effects?
There is an image that our drinking water is pure and that there are
no serious health problems resulting from it. It is now apparent that the
first half of this image is shattered. Yet there are large numbe rs of
epidemologists who would still argue that there are no serious health effects.
They will point out that there are very few known outbreaks of epidemics
which can, with certainty, be traced to drinking water. They will argue
that we produce anti-bodies to protect us from most bacteria so that even
if our drinking -water is poor, we will develop immunity to most diseases
it could produce. Mexico is indicated as a prime example of a country with
contaminated drinking water which presumably does not affect the health
of the residents.
To a very large extent, these critics are correct. Indeed there have
been isolated known water-related epidemics. We do develop
effective anti-bodies and the water probably could get considerably worse
before epidemics of major serious diseases occur.
Despite this, the critics may be missing a very important point.
People do not have to contract a major disease to get sick. Unreported
sub-clinical illness may not be fatal, but it is unpleasant and causes lost
work and occasional medical costs. And secondly, the absence of epidemics
does not rule out the possibility that the drinking water may carry micro-
organisms which do, in fact, result in illness at a sub-epidemic level.
As we have said above, the over-whelming majority of health related
absenteeism comes not from the "serious" diseases such as cancer, heart
attacks, etc. but from sub-clinical illnesses such as flu, upset stomachs,
and colds.
We believe that if our drinking -water has any appreciable economic
effect on the community, it will be through low level illness. Variations
from community to community in this type of illness could easily be
significant yet go unnoticed.
25
-------
Data
Given the above general hypotheses, it is necessary for us to
operationalize the concepts and to postulate a set of more specific data-
based hypotheses. In doing this it is useful to describe the data which we
would ideally want in order to best test our general hypotheses (not that
we could get it, because we always have to make some concessions to the
reality of availability.) Knowing what we would want, however, enables us
to better understand the characteristics of the results which we obtain
with real data. Ideally, to perform a study with these goals, we would
like a good measure of adult illness which would reasonably be related to
drinking water. Ideally again, this would include only those sub-clinical
illnesses whose virus or bacteria can be transmitted by drinking water.
We would want a good measure of water quality including average coliform
counts for a year, and the maximum coliform count for a period perhaps
as long as a month.
Additionally, we would want measures of community socio-economic
characteristics such as average income, population density, racial
distribution, and housing characteristics. Finally, we would want to hi ve
average and extreme values for particulate counts and sulfur oxides.
Absenteeism Data
Obtaining direct measures of adult sub-clinical illness is virtually
impossible without extensive surveying. The problem is that there is no
clear measure of low level illness by community for those communities
which we also have water quality data. The National Health Survey is an
excellent measure of illness of all types, but it is a nationwide sample
which cannot reasonably be broken down to the small communities in which
we are interested.
People generally do not report low-level illnesses unless they are
particularly severe or prolonged. Thus, doctor's records, hospital
records and public health records are quite inadequate for our purposes.
Since we are primarily concerned with finding the economic costs
associated with inadequate drinking water, it is reasonable to consider
only those incidents of illness which resulted in absenteeism.
26
-------
Although this simplified the task of measuring community health
to some extent, it clearly was not without problems. Direct measures
of adult absenteeism are difficult to gather and are of questionable
value even where available. As we shall demonstrate, indirect measures
appear to be superior. First, however, let us deal with the problem of
direct measures.
Any community of 5, 000 or more is large enough to have a reason-
ably diverse set of business enterprises. There may be, for example, one
or two medium-sized factories, a few small manufacturing concerns, an
array of retail establishments, a small number of construction firms,
some government officials, and perhaps some mining or agriculture-related
businesses.
Records of absenteeism are not kept uniformly. Some companies
have detailed records, other are very incomplete. The variation occurs
along at least two dimensions: size of business and style of management.
Normally large concerns keep better records than small businesses as a
matter of necessity. Furthermore, just as a matter of personal style,
some managers simply keep better books than do others. This variation
is unknown, immeasurable, and not necessarily random. Thus, even
given some acceptable sampling of employees in a community, it is
questionable whether the measure of absenteeism would even reflect the
number of days for which work was missed. Even if the actual number of
days missed were accurately estimated, there are a number of intervening
variables which would make the interpretation of these data of questionable
value. Occupation-related illness, variations in paid sick leave, insurance
policies, commuters who do not live in the cities where they work, etc. , all
pose problems which would be extremely cumbersome to overcome.
Given these problems, we chose to select a different measure of
absenteeism which, although not perfect, has fewer difficulties:^lernentary
school absenteeism. This measure has a number of characteristics which
make it a reasonably good measure of water-related absenteeism.
It is such a large proportion of the poulation of elementary school age
children, that the measure can be considered to be the population reflecting
the socio-cultural and economic variations of the communities within
which they reside.
27
-------
Unfortunately, the purity of population is often somewhat diminished
by the presence of students bussed in from smaller rural communities.
Obviously the water which these "outsiders" drink is different from that of
the local children. Although complete compensation for the problem cannot
be achieved given the resources available to us, there are three factors
which tend to relieve the problem. First, the selection of elementary
schools (rather than middle or high schools) reduces the problem con-
siderably. School centralization is far less frequent for elementary schools
than for higher grade levels. During the course of this contract we were
pleased (for scientific reasons) to find that large numbers of small
communities will have their own very small elementary schools rather than
selecting to send their children to larger cities a few miles away. Secondly,
although when they are at home, children from outside a school district are
exposed to drinking water of possibly different quality than was measured
by the CWSS study, when they are at school, they will drink the same water
as other students.
Finally, in one state (Vermont), the records of children who are bussed
in from outside the school district are kept separate from those within the
school district. Thus, in this state, the records available to us reflected
primarily only those children who actually lived in the school district.
Since the CWSS study recorded all public water systems in each town visited,
this means that for Vermont, we have a high confidence that the student
population is nearly entirely drawn from within the water system boundaries.- ''
A second advantage of elementary school data over its adult counterpart
is that occupational-related absentees simply are not there. While variations
in school health programs may produce some variations in absenteeism, the
problem cannot compare with the variations that surround different occupa-
tions.
A third advantage is that there is no equivalent to paid sick-leave
days. The adage that "sick days taken expands to meet sick days allotted"
is not applicable.
A fourth advantage of elementary school data (over higher grades)
is that unexcused absence is not as prevalent. Certainly it exists, but
the overwhelming number of educators consulted expressed the opinion
that xinexcused absenteeism is more severe in the high schools than the
elementary schools.
37) Since we were able to analyze data at the state level, it will be
seen later on in this report that we used Vermont as a check for
those areas where the data were aggregated only by school district.
.28
-------
Finally, although there are a number of socio-economic factors
underlying much of the absenteeism including both excused and unexcused
absences these are systematic and analyzable. The economic state of the
community, housing characteristics, etc. , can all be engaged to account
for some of the variation unrelated to water.
Probably the most widely raised question about the use of school
absenteeism data has been the fact that we do not know what illnesses
caused children to miss school. The child with the broken arm is recorded
exactly the same as is the child with gastroenteritis. The former obviously
is not water related - the latter may well be. How, our critics will argue,
can we possibly make statements about the health effects of poor drinking
water when we have no idea what illnesses the children had? The problem
is resolvable given one assumption: that controlling for all other systematic
influences on health, non-water related illnesses occur randomly. That is, once
we account for all systematic effects from economic well being, air pollution,
urbanization, etc.8 the remaining illness is distributed randomly except
for water-related illness. If the assumption is true, we have a classic
problem of data with random noise. Its effect (Rummel 1965) will serve
only to reduce the "real" correlation. In summary to our critics on this
point, we can say that if they accept the assumption of randomness, then
results cannot possibly be inflated and probably are reduced from what we
would expect if we were able to control for that randomness. *°'
Data Collection Procedures
Elementary school attendance data was collected for school year
1969-70 for all communities having at least one elementary school and
having been studied in the 1969 CWSS study.
Since both the 1968-69 and 1969-70 school year's overlap the calendar
year 1969, we had the option of choosing either for our data base. Although
the 1968-69 school year covers more months of 1969 than does the 1969-70
school year, we chose the latter because of the wide-spread Hong Kong flu
epidemic from November 1968 through March of 1969. We believe that
this could have introduced additional uncontrolled variance. At its peak,
the epidemic was afflicting 10 million people per week. ''
In the collection .of our data, we found that central records are
normally available in the state capitols or county seats. We were courteously
38) Although the correlations will be low, the regression coefficients will
still be unbiased.
39) Wilder, Charles S. , "Acute Conditions Incidence and Associated
Disability. " U. S. Department of Health, Education and Welfare, Public
Health Service, 1972.
29
-------
given access to these records by state and county officials. In nearly
every state at least a small number of individual schools had to be
contacted because the state did not have their communities recorded
separately. Where central records were available only the smallest
of the communities had to be contacted separately. In five states
(Kansas, Missouri, Louisiana and California) attendance records were
not kept in a central location and all schools had to be contacted
individually.
Water Quality Data
We used two direct measures of drinking water
quality and 18 indirect measures covering various aspects of water
treatment facilities and their direct operators. These data, shown in
Table 3 below were taken from the Community Water Supply Study
(CWSS) performed in 1969 by the Public Health Service (PHS). 40)
Unfortunately, the CWSS did not take adequate precautions to
insure against the possibility of stochastic variation in drinking
water quality. The major portion of the data, in fact, was collected
at one point in time. This would be adequate if drinking -water quality
was relatively stable across time. However, it is a widely held
belief that for many reasons, the quality of U.S. drinking water can
vary considerably across time. Since we have only a one time
measure of water quality for any particular tap, it is necessary to
assume that water quality is stable across time if the measure is
to be acceptable as a linear function of an average acres s an
entire year. In most instances, this is a dangerous assumption.
Therefore, the results which will be discussed in this report must
be evaluated with cognizance for the fact that the water quality data
were not collected in a rigorous time series.
40) McCabe, op. cit.
30
-------
Table 3 '
Total Coliform
Total Plate Count
Range of least and most monthly samples
Is the lab certified ?
Frequency of sampling of finished water
Common walls between finished and lesser quality water?
Are there inter-connections to other systems, of known acceptable quality?
Of unknown quality without protection?
Detectable chlorine residual in distinct parts of the distribution system?
Chlorination process interrupted XX times in last 12 months
Highest level of formal education
Length of time on this job
Total years water purification experience
Is operator full time employee?
Salary range
Level of training in water treatment
Level of study in microbiology
Level of study in water chemistry
41) Heavy metals and toxics, etc. were not included in the study.
The reasons are that 1) the presence of heavy metals is so in-
frequent as to make statistical analysis unreliable; and 2) the
effects of heavy metals are not likely to show up in school
attendance data.
31
-------
Fortunately, the CWSS did provide us with a number of potential
surrogate measures which make intuitive sense, and which we also found
to bear some interesting relationships to school absenteeism. These
measures are the indicators of plant facilities, procedures and of the
operators' capabilities. Logically we expected that poor quality operators
and/or facilities would tend to produce poor quality water. Although we
have no data which specifically enable us to compare the surrogate with
the actual variable, ahe linkage seems to'be a reasonable one. Additionally,
the surrogates (salary of operator,education of operator) are stable across
time and have some significant theoretical linkages to drinking water quality
which will be discussed below.
Air Quality Data
The air quality data was obtained from EPA's National Aerometric
Data Bank: Yearly Frequency Distribution. Although SO_ and particulates
are both important measures, only the latter was sufficiently available to
permit its usage.
Unfortunately, air quality monitoring stations are not necessarily
located in the same communities which were studied in the CWSS study.
We, therefore, had to attempt to estimate the quality of the air in the
CWSS communitiesi by using data for the quality of air in nearby communities,
The estimations were made by geographically locating the set of CWSS
communities being examined in this stxidy and the most proximate air
monitoring stations. Our first rule was that we assume that the air quality
is equal to the air quality of the closest monitoring station within 15 miles.
If within 15 miles, two or more stations were approximately equidistant from
the CWSS community, the one which came closest to being upwind, given
the prevailing winds, was selected.
42) We cannot compare water quality with the surrogates because we do not
trust the direct measure of water quality. Clearly, if we had direct
measures of water quality which we believed, the surrogates would not
be necessary.
.32
-------
In many instances no air quality stations existed within 15 miles.
Indeed, in some areas no measures were to be found within a 50 mile
radius. This posed a dilemma because,on one hand, we wanted to retain
the option of using all the data -even if questionable - but we also wanted
to have the option of rejecting those air quality data associated with stations
more than fifteen miles from the city in question. Therefore, we created
a dummy variable whose value is 1. 0 for cities with air data gathered
within 15 miles and -0. 0 for those whose data were gathered at a distance
greater than 15 miles. This gave us the option of treating the air quality
data as "missing" if it -were gathered at a distance of over 15 miles.
Socio-Economic Data
The socio-economic characteristics of the communities were obtained
from the 1970 census. We used the tapes containing the first count data.
From the list of variables available on the first count, we selected the
following subset:
1 Total population
2 Percent rural
3 Percent SMSA
4 Percent urban portion of SMSA
5 Percent rural less than 2500
6 Percent in urbanized areas
7 Percent less than 18 years old
8 Percent 18 to 62 years old
9 Percent 62 and older
10 Percent white
11 Percent black
12 Percent of families with a husband and wife
13 Percent other male head of household
14 Percent other female head of household
15 Average home value
16 Average rent
11 Percent units less than 1.0 persons/room
18 Percent units 1.01 - 1. 50 persons/room
19 Percent units greater than 1. 50 persons/room
20 Percent units with complete kitchen
21 Percent units with complete plumbing
22 Percent without complete plumbing
23 Percent units with shared toilet
24 Percent persons with complete plumbing
25 Percent persons without complete plximbing
26 Percent families with complete plumbing
33
-------
We believe that this subset of community characteristics are the
most likely to have some influence on health. The 26 variables are not
all mutually independent. In fact, they fall into six identifiable categories
as follows:
Variables Category
1-6 Urban/Rural
7-11 Age/Ethnic
12-14 Family Characteristics
15-16 Wealth
17-19 Density (persons/room)
20-26 Home Plumbing
Of these six categories, we expected only the last five to enter
directly into regressions. Given the results of the CWSS study, we
expected that the urban/rural distinction would present particular problems
which would require a dichotomization of the sample rather than simply
an additional term in the regression equations. The necessity to dichoto-
mize the sample will be discussed with the other hypotheses in the following
section.
It should be noted that for each of the remaining five sets, it is
only reasonable to use one of the possible variables at a time because they
are either statistically or sometimes linearly related to each other. For
example, since variables 12 to 14 must eqxial 100%, any two of necessity
define the third. Furthermore, the correlations between any two are
expected to be so high as to prohibit the use of more than one because of
the problem of multi-collinearity. We chose to include each of the variables
in the data set for pragmatic purposes. There is no research or theory which
would give us good reason to choose one version of the same variable over
another. Rather than arbitrarily exclude the data, we chose to include each
version in our initial analysis. We employed a modified stepwise regression
by limiting our final variable set to one variable per category. We felt that
this relatively mild limitation (one variable per category) made the inter-
pretation more manageable as well as alleviating somewhat a troublesome
degree of freedom problem.
Climate
We included the climatic measures of average hours of sunshine, the
ratio of temperature/humidity, and inches of precipitation to attempt to
estimate the effect of climate on health. These climatic variables were
34
-------
included because it is widely recognized that weather does influence
health. The fact that sub-clinical illness is the greatest during winter
months is virtually unquestioned. Additionally, it is normally believed
that the effect is most severe in the colder and more humid areas. It
should be recognized, however, that in this study we had no time series
absentee data to pick up seasonal illness variations and we had only nine
geographical regions thus giving us only nine data points for our climate
variables. Because of this, our expectations were not particularly high that
we would find that sunshine or the temperature/humidity index would
produce interesting results.
Hypotheses
If health is to be statistically linked to water quality in a natural
experimental setting, it is necessary to us to recognize that, in un-
controlled situations, people become ill for a wide variety of reasons.
Some of these may have to do with specific illness-producing substances
such as air pollution. Others may have to do with inadequate diets or
poor health care. Still others may result from living in overcrowded
conditions or sub-standard housing. To the extent that these factors
systematically influence health, we attempted to account for them in this
study. At the most general level, then, we are hypothesizing that,
accounting for all other macro-systemic influences on health, the
occurrence of sub-clinical illnesses will be the highest in those communities
with the worst water quality. More specifically, we are arguing that these
factors account for absenteeism roughly according to Figure 1 (next page).
We are arguing that the elements in the boxes act directly on
elementary school absenteeism. The elements in the diamonds act on
absenteeism through the elements in the boxes. All of these are encompassed
in the broader circle which includes those factors which have complex
and wide ranging effects on all of the factors and/or their relationships
with absenteeism. Because these factors act in concert, we cannot simply
pull out water quality and analyze it by itself. Although the other variables
a-re "just" control variables, they are vital to the successful analysis of
the absentee data. Let us consider the control variables first.
Health care is one of the most obvious links to absenteeism. Clearly we
would expect that improved nutrition, housing, medical attention, would
directly influence health. Assuming that health is linked to absenteeism,
we can presume that there is a close linkage between health care and
absenteeism.
Similarly, cultural factors can be argued to link directly Ic absenteeism.
The ethics of social or cultural groups in all probability influence the
motivation to attend school. This may, on one hand, result in simple
unexcused absence. On the other hand, some cultural/economic groups
35
-------
Figure 1
Socio-Economic
Status
Social
Groups
Cultural
Norms
Elementary
School
Absenteeism
Air
Pollution
Drinking
Water
Quality
Regional Differences and
Urban/Rural Differences
36
-------
may provide students with encouragement to attend school even when
they are ill.
Both health care and cultural factors, we argue, are related to
socio-economic status. Normally we expect most of the factors associated
with health care to be positively related to socio-economic status (SES).
For example, it is common knowledge that the children of low income
parents suffer greater physical and mental problems from poor nutrition
than do children of more affluent parents. We might also expect that
there is some SES level above which there is no general improvement in
health care. We also expect that the protestant ethic of the upper and
middle classes would have an effect on those cultural norms partially
influencing school attendance.
In addition to the economic determinants of cultural norms, it is
possible that other cultural descriptors are salient. For example, many
people argue that those communities with high percentages of blacks
have different behaviors toward education above and beyond the fact that
they are normally in lower economic groups. As a point of interest, it
should be noted that there was no empirical confirmation of this position.
Air pollution and climate are two of the other factors presumed to
directly cause health problems and, therefore, school absenteeism. The
exact details of the relationship between air pollution and health are
somewhat inconsistent across the numerous studies.
Nonetheless, there seems little question that air pollution is related to
health in some manner. The relationship between climate and absenteeism
is two-fold. On the one hand, as we discussed above, climatic variation
influences health. Additionally, adverse weather can close schools,
temporarily, obviously increasing absenteeism.
All of these relationships mentioned above are themselves
partially and complexly influenced by regional variations and by urban/
rural differences. As a simple example, adverse weather often has a
more serious effect on the closing of rural schools than on urban schools
because of the relatively large miles of highway per person requiring plowing.
Popular wisdom argues that the protestant ethic in rural America
when coupled with the allegedly healthier life style should reduce
elementary school absenteeism. On the other hand, the frequently lower
incomes of rural communities, the more limited access to medical
facilities and other factors conceivably could counteract this effect. All
37
-------
of this is further complicated by the fact that we know for a fact that
water treatment facilities in rural areas are inferior to those of
urban facilities. '
All of these factors basically lead to the fact that these broad
complex factors (urban/rural and regional differences) must be
handled carefully in the analysis of the data.
Having discussed the control variables, let us now move to the
independent variable of interest - drinking water quality. We discussed
in the literature review the arguments and analysis supporting the
hypothesis that poor quality drinking water can cause illness. That this
should logically be the case is not the least bit surprising. What is
somewhat surprising is the fact that the water provided through public
systems may on a widespread basis, be of low enough quality to produce
illness . Given the results of the CWSS and other studies, we have good
reason to believe that people do in fact become ill from low quality
drinking water. As we see in figure 1« variations in drinking water
quality are hypothesized to be influenced by variations in either the quality
of the operators or the quality of the facilities or both.
44)
It is widely accepted 'that the reductions in quality resulting from
human error tend to be sporadic rather than steady state. Workers ranging
from assembly line operators to air traffic controllers tend to err by
making occasional mistakes. They tend not to perform in a constant
sub-standard manner. The poorer the worker the more frequent and more
severe are the mistakes. Thus, if we were to directly measure the quality
of the water produced by poor operators, we would have to have a time
series recording the frequency and magnitude of the deviations from the
average. Because we do not have data of that quality, we must use measures
of operator qualifications as surrogates for water quality.
The CWSS reports that water quality varies regionally and by level of
43) McCabe, Op. cit.
44) See for example: 1) Niece, E. H. , "Statistical Quality Control and
Value Analysis, " in Carl Heyel (cd. ) The Foreman's Handbook,
McGraw Hill, New York, 19&7; 2) Levinson, Harry, Executive
Stress, Harper and Row, 1970; and 3) Barnes, Ralph, Motion and
Time Study: Design and Measurement of Work, Wiley, New York,
1964.
38
-------
of urbanization. Urban regions tend to have both higher levels of water
quality, better trained operators and more sophisticated facilities. Since
we suspect that school absenteeism, water quality and the control variables
all are influenced by the urban/rural and regional variation, it is apparent
from the outset that a generalized model applying to all cases may encounter
difficulties. Because of this, we have to recognize the possibility that while
it may be possible to develop a general model of the effect of poor quality
drinking water, it is likely that we will also need alternative models to deal
•with the problems of multicollinearity and systematic bias in both the
dependent and independent variables.
In addition to multiple models, we will also use dummy variables
which themselves include some of the correlated interactions we are trying
to simplify. In this study, we added dummy variables representing each
of the SMSA's being studied. For example, we had one variable each for
Vermont, New York, Cincinnati, etc. If a community were in the region, it
received a 1.0, otherwise it was scored a 0.0. This approach enables us
to try to include unaccounted systematic regional differences in elementary
school attendance. It is important to note that this method is only marginally
effective in controlling for any regional bias in the independent variable set.
The primary hypothesis is that one or more versions of the following
equation should predict absenteeism.
A..= C.-M W.+ 4. ,. E.+ A.- F.-f 4 I. + j D.+ ((,. S.+ / _. P.-f j^.CL +e.
ij J lj i 2j i ^3j i Mj i \5j i %j i A 7j i ^8j r i
where
Subscript indicates a water supply within a subset area j
A = elementary school absenteeism
C = constant
W = water quality measured either by plate count of coliforms
or by quality of facilities
E =; one variable from the age/ethnic category
F = one variable from the family characteristics category
I = one variable from the wealth category
D = one variable from the density category
S = one variable from the home sanitation category
P = level of air pollution
Cl = climate
e = the error term
39
-------
Frequently, hypotheses such as these are considered proven or
disproven simply on the basis of the statistical confidence of the partial
correlation or of the regression coefficient. In this analysis, it should
be emphasized that we have some definite expectations about the direction
of the relationships. These linkages are most important for the pollution
variables and particularly for water quality. We would expect both water
quality and air quality to be negatively related to absenteeism. Except for
some narrow ranges at very high levels of purity, it is not reasonable
that higher quality water or air should result in increased illness. '
For the other variables, we would expect younger populations to be
healthier; non-white population to be less healthy; wealth should be nega-
tively related to absenteeism; density should be positively correlated ; and
home sanitation should be negatively linked with health. Finally, given
the pressures of time and, frequently, income on single parent families,we
would expect absenteeism to be higher in single parent families than in
the more traditional husband-wife family.
The socio-economic measures would be used in the analysis in an
additive manne'r, but only one variable per group was permitted in the
final regression. These were selected using standard stepwise regression.
Analytical Methodology
The primary statistical tool which we employed in the study is step-
wise multiple regression. This is a technique which fits well with the
general assumptions that health is a result of the simultaneous effects of
multiple factors. The methodology provides us with a set of coefficients
which indicate the expected magnitude change in the dependent variable as
a result of a given change in one of the independent variables.
45) At extremely high levels of purity, small increases in bacteria levels
may serve as boosters to keep anti-body levels high enough to ward off
more serious infectious material. This fact is particularly important
for those persons who are a part of a stable population.
40
-------
In the current study, we are primarily concerned with the health
effects of water quality and have taken considerable effort to eliminate any
serious problems of multi-collinearity. Since we are less concerned with
the individual (as opposed to combined) effects of the control variables, we
placed less effort on maintaining statistical independence within this group
of variables. We did, of course, take the effort to insure against linear
independence of the control variables.
Missing Data
In nearly any statistical analysis using surveys or multiple data files,
missing data becomes a problem. There are three logical approaches which
can be used to solve the problem only two of which are worth serious con-
sideration. The one logically feasible but impractical solution is to ignore
the problem and run the analyses as though the data were present. This
alternative inserts zeroes in the missing cells and thereby gives the
missing cells arbitrarily incorrect values.
The second alternative is to simply omit any case from the analysis
for which there are one or more missing data cells. This alternative
has two negative effects •which under some circumstances are not
particularly difficult to accept. The first is that the sample size is reduced
thus reducing the degrees of freedom available for analysis. The second
is that if missing cells are systematically distributed, the actual
characteristics of the sample are altered by eliminating a non-random
subset of cases
The third alternative is to estimate the missing cells given informa-
tion already in the data set. For example, a data set in which age, income,
race, education, and sex were independent variables, a missing "income"
cell could be statistically predicted using the remaining four variables. The
method is used most frequently in studies having a rather numerous random
distribution of missing cells. For example, if each of five variables in a
data set had ten percent randomly distributed missing data there would be
. 5N empty cells. This logically could eliminate as many as one half of the
cases.
The difficulty of estimating missing cells is that to do it reasonably
successfully creates a conflict of the assumptions of the various techniques.
Multiple regressions, as we pointed out above, assumes the independent
variables to be independent of each other. The missing data routines, on
the other hand, assume that the more highly intercorrelated the independent
variables are the more accurate the estimates are. Thus, the closer one
moves toward the ideal assumption of regression, the less accurate the
41
-------
missing data estimates become. There is, admittedly, a gray area in which
some researchers believe that the technique is preferable to eliminating large
numbers of cases .
In the current analysis, there are two significant types of missing data.
The first of these are the census variables which, if any are missing, are
generally all missing because we were unable to find the community on the
master enumeration district list. Normally this occurs in very small un-
incorporated communities which have their own elementary school. '
The second primary source of missing data is in the water quality/
facility inventory variables. In some instances, the water quality measures
were not present. For our sample this was relatively infrequent. More
frequently, some of the questions concerning characteristics of the operators
or failures went unanswered. These missing cells tended to be scattered
although there was a slight tendency for some respondents to refuse to
answer several questions.
Given the above configuration of missing data and given that the water
quality/facility inventory variables were of primary significance (and,
therefore, should not be estimated) we chose to eliminate those cases
which had any missing data on the particular variables being analyzed at a
specific time. Clearly this would change the sample size as a function of
the particular variables being employed in the analysis. For example,
consider the following hypothetical data set:
Var. 1 Var. 2 Var. 3 Var. 4
Case 1 1 1 0 25
Case 2 2 5 0 19
Case 3 313
Case 42 3
Case 5 3 1 0 10
Case 6 18 60
Case 7 1 6 12 8
Case 83 32
Case 9 115
Case 10 12 84
46) If there were no elementary school, there would be no dependent viiriablc
and the case would not even be included in the data set.
42
-------
If variable 1 were the dependent variable and variables 2, 3, and
4 were independent variables, cases 1, 2, 5, 6, 7, 10 would be examined
because they have complete data on all three variables. If, on the other
hand, variables 2 and 3 were the independent variables, cases 3 and 9 would
be added to the above list because there is data for them for variables 2
and 3 but not 4.
In most circumstances, missing data tended to occur in the smallest
communities and those communities with at least one piece of missing
data tended to have several missing pieces. The deletion of cases with
missing data, therefore, biased the sample in the direction of larger
communities relative to the CWSS sample. There is , of course, no
assurance that the CWSS sample was itself randomly selected from any
larger population.
Case Selection and Aggregation
The CWSS, as mentioned before, covered 969 public water systems
across the United States. We were unable to individually examine some of these
for three basic reasons:
1) Two or more public systems supplied one community.
2) A public system served a community which did not have an
elementary school.
3) The system was in California and the school's records were
not easily accessible.
If two or more water systems supplied a single community, a weighted
average water quality figure was computed using the number of people served
as the weighting figure. For example, consider the following typical hypo-
thetical situation of three systems supplying one medium sized community.
Population Served Total Plate Count
System 1
System 2
System 3
Community means =
2000
4000
10000
16000
32 x 10<
16 x 10
3000
4000
1000
8000
2000
P x C
6 x 10J
16 x 10
10 x 10
32 x 10
43
-------
A large number of the systems examined in the CWSS provided water
to very small communities, (under 500 population) or to special populations
such as hospitals, trailer parks, camping grounds, boarding schools,
nursing homes, etc. Of these, all special systems and systems serving
those smallest communities with no elementary schools were dropped from
the study.
The State of California was one of the states which does not keep
centralized absenteeism records. Therefore, individual schools were
contacted to collect the absenteeism data for the school year 1969-70.
Because absenteeism in California means an unexcused absence, we
did not believe that extensive collection efforts were warranted. However,
every school district serviced by one of the water systems measured by
the CWSS was contacted. If the data for a given school were not easily
accessible, we did not request the school's officials to extend themselves
to provide it.
44
-------
The Test
A General Model
Despite all of the potential pitfalls of a general model, we found the
results encouraging. The bivariate correlations showed the
following variables to be most strongly linked to absenteeism.
Variable Correlation
Precipitation . 40
Vermont Dummy -.36
Temperature . 34
New York Dummy . 30
Urbanization . 26
Frequency of Testing
of Finished Water -. 24
Louisiana Dummy . 20
Operator Salary . 19
Particular attention should be paid to the incorrect signs for the
relationship between temperature, operator salary and absenteeism. We
shall find shortly, that these problems disappear when we control for
urbanization and regional variation. It is also worth identifying those
variables which did not appear in the list of significant bivariate re-
lationships.
Variable Correlation
Total Plate Count . 09
Home Value . 04
Living Density . 02
Shared Bathroom . 02
Presence of common
walls with inferior
water . 01
Presence of Chlorine
Resident .03
Air Quality (particu-
lates) .07
-------
In the general model, none of these variables ever showed significance
despite controls for other variations. '
After running the bivariate correlations, we examined the data using
stepwise regressions. Because of the complex intercorrelations in the
data, we ran the regression several times using different variables. The
overall findings from the regressions of all cases is that controlling for
urbanization and regional variation,' the only consistently significant
variable is operator salary. The frequency of testing finished water is
consistently related in the proper direction although the relationships are
weak. The regression coefficients are in the correct directions in all
cases. Operator salary is generally significant at the . 10 level. In some
instances, however, its significance drops to . 20 . As we mentioned, the
frequency of testing is a weaker variable and is, at best, significant at the
. ZO level. Interestingly, the product of operator quality and frequency of
testing is significant at the . 05 level in the following regression equation.
Multiple R = . 33 Significance of the Regression = . 001 R-Squared = . 10
Variable Partial Coefficient Std. Error T-Stat. Significance
Correlation
Constant 5.7 .57 10.0 .001
Percent in
Urbanized
Area .09 .8 .7 1.2 .25
New York
Dummy .20 1.5 .6 2.7 .01
Louisiana
Dummy .26 3.4 1.0 3.5 .001
Fqx Op48 -.15 -.08 .04 -2.1 .05
47) All of thein were significant in at least one regional analysis.
48) An index comprised of frequency of testing finished water times
operator salary.
46
-------
We would argue frequency of testing times the salary of the operalor(s)
is a powerful index of the operation of a water treatment facility. Whereas, a
weakness in one or the other variable may not necessarily mean poor
operation, infrequent testing coupled with an underpaid operator almost
certainly signifies substandard operation.
Another interesting point is that although temperature and precipitation
showed significant correlations in the bivariate analysis, in no instance
are they even close to significant in the regressions. Clearly, whatever
variance climate was able to explain at the bivariate level was better
accounted for by the regional dummy variables in the regressions.
The general model clearly includes a great deal of uncontrolled or
semi-controlled variance. The correlations are not particularly strong
and a multitude of sins are buried in the regional dummy variables. None-
theless it is very encouraging to note that variables linked to operation of
water treatment facilities are consistently stronger than any of the other
competing specific variables such as home value (a surrogate for income),
density, cultural factors, etc.
Regional Analyses
In an attempt to gain additional insight into the relationships between
drinking water and health, we subdivided the sample into more homogeneous
sets. In' choosing these subsets it was nec.essary for us to balance the
homogeneity of the subset against the number of cases in the subset. On the
one hand, we wanted to have as great homogeneity as possible while also
retaining enough cases to permit meaningful analyses. In doing this we
finally decided upon four subsets two of which are quite homogeneous, the
other two of which are less so by reason of necessity.
The homogenous subsets are Vermont and the greater New York area
excluding New York City. The former includes all of the analyzable CWSS
communities in Vermont. The data cover nearly every county in Vermont
a-nd systematically exclude only the rural-farm population. We had 82
communities with complete data.
The New York data cover Westchester, Nassau, Suffolk and Rockland
counties. The communities studied are generally suburban and smaller
urban centers. New York City was excluded because it has an extremely
serious unexcused absence problem. After examining the New York City
data rather closely, we concluded that uncontrollable variations in unexcused
absence made the absentee data hardly meaningful as an indicator of
illness. Excluding the New York City data we had 54 cases with complete
data.
47
-------
For the rest of the areas covered by CWSS, we did not have sufficient
cases to permit truly regional subsetting. Therefore, we left the areas
together and subset them only by level of urbanization. Urban and sub-
urban communities •were placed in one category. Rural communities as
classified by the Bureau of the Census, were placed in the other subset.
We had 62 cases with complete data in the urban/suburban subset and 47
cases in the rural subset.
Summarizing, we had four subsets which we subjected to further
analysis. They are listed below.
Name
(1) Vermont
(2) New York
Cases with
Complete Data
82
54
(3) Six Area Rural
47
(4) Six Area Urban
62
Comments
Includes data from most Vermont
counties. Generally small
communitie s
Four counties surrounding New York.
Includes small communities and
suburbs with a few medium sized
cities.
Rural communities surrounding
Cincinnati, Ohio, Charleston, W. Va. ,
Charleston, S. C. , Kansas City, Mo. ,
New Orleans, La. , Pueblo, Colorado
Urban and suburban areas including and
surrounding six areas in (3)
The results of the final regressions for the four areas are shown below.
The interpretation and discussions will follow the presentation of the
equations.
Vermont
N = 82
Significance of regression = . 97
Multiple R = . 508 R-Squared = . 259
Absenteeism = 5. 04 - .568 -.33Ic +16.48D
48
-------
Standard Error of
Significance Coefficient
S ' = Salary of operators
(index) .99 .14
50) = Interconnections with
Ic
poor quality systems .85 .21
D = Percentage of living units
with more than 1. 01
persons/room .90 8. 8
New York51)
N = 54 Significance of Regression - . 99
Multiple R = . 614 R-Squared = . 377
Absenteeism = 7. 02 - 11.4 Env + 34. 25D0 - . 000033H - . 53R
u
Standard Error of
Significance Coefficient
Env = Environment
Index .98 4.4
D = Percentage of
living units "with
greater than 1. 5
persons/room 095 16.2
H = Home value .99 .00001
R = Percent rural .85 .34
u
49) 1 = 1999, 2 =2000-4999, 3 = 5000-7499, 4 = 7500-9999, 6 = 10000 or
50) 1 = Y, 2 =N
51) Computed by dividing index of operator salary by average particulates.
The result is that a high value indicates high environmental quality.
49
-------
Six Area Grouping - Rural
In the above mentioned six areas, the CWSS studied water systems
in 57 rural communities under 2500 persons which also had an elementary
school. Of these, 10 had missing data on at least one of the variables
which were finally entered in the regression. The regression results are
summarized below.
N = 47 Significance of Regression = . 99
Multiple R = .501 R-Squared = . 251
Absenteeism = 13.9 - 1.04F - 1. OS
p52) -. Frequency of testing
of finished water
S ' = Salary of operator (Index)
Significance
.99
.90
Standard Error
of Coefficient
. 32
.54
In this subset we tried to employ the index of frequency of testing
times salary of the operator. The product was significant at the .99
level but the total explanatory power of the regression was smaller than
when we employed the additive model. Because the additive model is
easier to extrapolate and because statistical confidence for the additive
model was quite satisfactory we remained with that solution.
52) 1 = each month, 2 ~ each year, 3 =2 years
4=3 years, 5 = infrequent, 6 = never
53) 1 = < 1999, 2 = 2000-4999, 3 = 5000-7499
4 = 7500-999, 5 = 10000 or
50
-------
Six Area Grouping - Urban /Suburban
N = 62 Significance of Regression = . 99
Multiple R = . 79 R-Squared = . 62
Absenteeism = 3. 81 + . 0006C + . 69R + 54. 6T + 1. 4K
Significance
C = Total plate count . 99
R = Presence of Chlorine residual . 95
T = Percentage of living units sharing
bathroom facilities . 99
K = Kansas City Area Dummy . 90
In examining the results there are two facts which are initially
apparent. The first is that the strengths of the relationships are con-
siderably stronger for the subsets than for the general model. Multiple
R's range from a low of . 501 in the six area-rural to . 79 in the six
area-urban/suburban, as compared to multiple R's in the . 30-. 35 range
for the general model. The significance of the water-related variables
is also appreciably higher.
The second major point to observe is the fact that in all regressions
except one, operator salary is a significant predictor of absenteeism. In
Vermont and the six area-rural areas the relationship is additive with one
other measure of plant operation. In New York, operator salary interacts
multiplicatively with air quality. It would appear that where there is
appreciable air pollution, it lowers resistance to disease such that contaminants
in the water will more easily produce illness.
Although operator salary is a strong predictor of elementary school
absenteeism in the national sample and three of the subsets, it is not linked
to absenteeism in the six area-urban/suburban grouping. Here we find
chlorine residual and total plate count as the primary factors. Because it
is only in this isolated sub sample that these variables appear, we have to
be somewhat skeptical of the confidence we can place in them. Caution is
particularly important when we recall that the direct measure of water quality
•was taken only at one point of time. Thus, the reliability of the independent
51
-------
variable in this regression equation is in question. Keeping these
important caveats in mind, the findings do make sense if we were to
hypothesize that for some reason, the contaminants in the water in
this subset arose from systematic non-human related problems. For
example, if there were some serious problems with the distribution
systems, we could argue that the one time measure of total plate count
would, in fact, identify the presence of generally contaminated water.
Similarly, any facility problems such as these would be more likely to
produce a steady level of contaminants than would operator malfunctions.
If for some reason, the six area urban/suburban systematically has
different problems than all other areas, the results have meaning. Again,
however, these results must be taken with caution.
Across the total sets of results, we find that urbanization and geo-
graphical location appear to play predominant roles. These, in fact,
prompted subsetting in order to control for their associated complex
variances. Aside from these factors there is the unmistakable presence of
something we might call housing quality. One or more of the variables
measuring home value, occupants/room, and the frequency of shared
bathrooms occur in three of the four subsets. The importance of these
variables is mostprominent in the Vermont and New York subsets. As
the models covered broader geographic ranges (i. e. , six area groupings
and the general model) the effect of housing variables tended to wash out.
This is not particularly surprising given the geographic differences in
property and housing values and types. ^
Notably, none of the climatic variables enter into any of the equations.
It should be noted that weather variables play a strange role in studies which
are not time series oriented. Although from one month to the next, weather
variation is known to influence absenteeism, this variation can be seriously
obscured by high level aggregation. This is particularly important when the
communiti*?*? being studied are geographically proximate to each other.
For example, we would not expect the weather in Burlington, Vt. to be
much different from Rutland Vt. We know, however, that absentee rates
do differ between these two communities. Weather may theoretically
explain a good part of the stable base of absenteeism but it cannot reason-
ably be expected to explain the differences between two geographically proxi-
mate communities.
Even in the six area groupings, we have only 3 or 4 appreciably different
climates. Here, one results show that the influence of climate is subsumed
in the geographical dummy variables. These latter variables apparently
incorporate more than climate and are, therefore, the stronger predictors.
54) For example, it is difficult to assume that differences in home
values between Mt. Kisco, New York arid Abita Springs, Louisiana
reflect differences in standard of living.
52
-------
In concluding the interpretation, we would point out that in both the
general model and in three of the subsets, operator salary shows a strong
and consistent relationship to elementary school absenteeism. There are
some differences from region to region but they basically support each
other. The one significant deviation occurred in the six area urban/
suburban grouping in which operator salary did not enter at all but was
replaced by chlorine residual and total plate count. Given the data
collection difficulties, one would have to intrepret this one deviation with
considerable caution. If the results of the six area urban/suburban
grouping are meaningful, they must be taken to indicate some non-human
problem which
wate r.
As support for the major findings concerning operator salary, we
should recall that the CWSS reported that 84 percent of the principle
operators were earning less than $7500 per year and that 37 percent
were earning less than $2000. With such poor salary structures for
operators, it is not surprising to find this as a significant element in the
production of poor quality water.
The fact is quite significant. While the regression equations broadly
predict an improvement for all regions studied of . 2 to .5 percent in attend-
ance for each $1000 increase in operator salary, we should not assume that
this means that simply by paying the operators more money, we will improve
performance by that amount. While there may be some improvement in perform-
ance with increased salary, the problem is obviously much deeper than that.
Given 1969 general salaries in the areas we were examining, the
majority of operators were in the lower 50% income range. This simply
makes recruitment of individuals with college training in-chemistry,
microbiology aid related subjects quite difficult. In fact, the CWSS showed
that over 60% of the operators never had even a short course in water treat-
ment. To a larger extent, the public's image of municipal water treatment
operators is often much closer to a manual laborer than to a professional
with a significant role in community health. According to some people with
whom we discussed the problem, the primary role of the operator is to keep
the machinery working properly. While this is an important function
performed in a water treatment facility, it hardly seems reasonable that
that it be the primary role of the principle operator.
Somewhat illustrative of the type of problem encountered is the
following anecdote reported by Harris and Breckner : 56
55) The averages fluctuated from area to area between $6000 and $8000
per year per worker,
56) Harris, Robert and Edward Breckner, Consumer Reports, June 1974,
p. 436-443.
53
-------
The operator in charge of one New England
community water supply, for example, heard
somewhere that activiated carbon will remove
foul tastes and odors. So, when the water
smells bad, he fills a cloth bag with activated
carbon powder, ties a rope to it, climbs into a
rowboat, and tows the bag back and forth across
the town reservoir. This is about as effective
as waving a wand over the reservoir.
Obviously to simply increase salaries would not be enough. There
would have to be a program to upgrade the operator quality through training
programs and, in the long range, improved hiring.
Another important water-related variable is the frequency of testing
of finished water This is significant at the . 99 level in the rural areas
not including New York and Vermont. In the general model it consistently
appears with the proper sign although the statistical significance is erratic
across different regressions. This finding is also consistent with a major
deficiency uncovered by the CWSS. In their report, they discussed the fact that
90 percent of the systems studied failed to meet the bacteriological sur-
veillance criteria.
Finally, in Vermont there is a weak relation ship between absenteeism
and systems with interconnections with water systems of unknown quality
without protection. This problem is relatively rare showing up in only six
percent of the total systems examined. Interestingly, thirty-five of these
were located in Vermont.
Effects
The effects of degraded drinking water can be expressed in a number
of ways. The regression equations themselves are the most abstract form
of this effect. They essentially provide an estimate of the percentage change
in absenteeism» per unit change in the independent water-related variable(s).
In the following section we will translate these into more concrete forms
moving progressively from the soundest conclusions to the most tenuous.
We will initially calculate the magnitude reduction of elementary school
absentees in those counties studied resulting from incremental improvements
in drinking water quality. 'The above findings will be expanded to the; dollar
57)Because of the tenuousness of the six area urban findings, we will not
extrapolate the costs for this group.
54
-------
value of lost wages and medical bills saved due to reduced absenteeism/
illness. Finally, making some very bold assumptions, we will extra-
polate the total dollar sayings to the entire country.
Elementary School Absenteeism - Vermont
Because the CWSS covered nearly every county in Vermont, we
believe that the findings apply to all except the rural-farm population of
the state. The study applies to a population of 409,000 and an elementary
school enrollment of 66, 600. The regression coefficient of the variable
"salary of operators" is . 56, and the coefficient for "interconnections
with unprotected systems" is . 33. Because the salary variable was
coded in a variable interval manner, we can strictly say only that an
increase of one point in the salary index will decrease absenteeism by
. 56 percentage of the total student days. Assuming a 180 day school
year, there are 12, 000 student days per year and each unit increase in
the salary index would result in an estimated 67, 100 fewer absences
from elementary schools. The standard error of the regression co-
efficient is . 14. Thus, a reduced absenteeism of from 50, 000 to
84, 000 days is within the plus or minus one standard deviation unit. A
range of 33,600 to 101, 000 is the range defined by plus or minus two
standard units.
Since the salary index intervals are roughly $2500 increments, we
introduce only a small error by translating these into actual salary values.
Doing this we would estimate that, state wide, every $1000 increase in
the average salary of the principle operator would reduce total elementary
absenteeism by 27, 000 days per year.
The interconnections with unprotected systems have a regression
coefficient of . 33 indicating that for each ten percent of the total systems
which eliminate the problems, absenteeism rates will be reduced by . 033
percent of the total school days. This translates to an increase of 4000
student days per year. The standard error in the regression coefficient
is .21. The comparable range of reduced absenteeism for one standard
deviation is 1500 to 5500 per ten per cent increase in eliminating linkages
with unprotected systems. The range for the plus or minus two standard
units is 0 to 9000.
If we are to draw any inferences about the linkages between drinking
water quality and adult absenteeism, we must be able to make some
55
-------
assumptions about the relationship between adult absenteeism and
elementary school absenteeism. Our first assumption is that the portion
of elementary school absenteeism explained by drinking water variation
is a subset of acute illness. That is, we are not interested in other types
of absenteeism such as chronic illness, broken bones, or other types of
absence which could not even logically be related to drinking water.
Obviously, any discussion concerning increases in operator salaries
implies that it is the lower salaries which are most in need of improve-
ment.
The second assumption is that across a large population, the
occurrence of acute illness in adults is a responsibly stable proportion of its
occurrence in children. This does not imply a one-to-one relationship nor does
it imply that the constant of proportionality is the same for any specific
adult-child pair.
A.
Finally, we assume that this constant is measured by the ratio
(_>.
where A. is the average, across the population being studied, of annual l
adult days lost from work with infective and parasitic diseases and
digestive system conditions. C. is the counterpart for children's days
lost from school. We have eliminated respiratory conditions from the
ratio, not because it is impossible to communicate them via drinking water,
but because the others are those most commonly associated with drinking
water.
The National Health Survey of 1969 found that there were 100. 9 days of
work lost per 100 employed due to the above described acute conditions.
They found that there were 163. 8 days lost from school per 100 students
due to the above conditions. A check across an SMSA vs non-SMSA
population showed that despite different rates of bed disability with acute
illnesses, A./C. ratios differ by a maximum of .006. Checks across
i i
different age groups show that in the 15-14 vs 45-64 age brackets, the
ratios are identical. Given the other error in the data, we believe this
difference to be insignificant. Given that, we will use the ratio " ~ . 616
* 163.8
as the constant of proportionality. This means that we expect that the
percentage of estimated water-related absenteeism in the adult population is
. 6l6 times that of children in elementary school.
Additionally, the survey found that, excluding hospital care, dental
bills, and special expenses, persons •with no chronic illnesses spent an
average of $69 per year on doctors and medicine. The non-respiratory and
non-injury acute illnesses account for 30 percent of the total disability
56
-------
CQ\
days. Thus, we estimate that the combined doctor and medicine expenses
per person per year for the acute illnesses which we are considering is
roughly $20. 7.
The labor force in Vermont to which this study is applicable numbers
153, 094 with a total income of 873 million dollars per year. Each one
point increase in average operator salary index should reduce absenteeism
by . 34 percent of the total worked days. Annually for the total state, this
would mean a savings of $2, 980, 000 in wages or a savings of roughly $20
per working person per year for each average increase of one point on the
operator salary index. Using the rough conversion factor of $2500 per
index point, the savings are $1, 190, 000 or $8. 00 per worker for each
$1000 average increase in operator salary.
In the area of interconnections with unprotected systems, the savings
is considerably less. For each ten percent average improvement in the
interconnections, we can expect a reduction in absenteeism of . 02 percent.
Statewide this amounts to a savings of $170, 000 per year in wages or slightly
more than $1 per -worker.
Total medical costs in the illness with which we are concerned are
$9.6 million dollars.
Since each $1000 average salary increase reduces absenteeism by
. 136 per cent of the total worker days, this accounts for * "0* ^o = . 055
. 25
of the illnesses in the acute non-respiratory, non-injury category. This
amounts to a savings in medical costs of $545, 000 per year per $1000 average
salary increase. Similar calculations show a savings of $77, 000 per year per
ten percent improvement in interconnections with unprotected systems.
The total savings (salaries plus medical costs) are 1.7 million dollars
per year per $1000 average increase in operator salaries and $247,000 per
year per ten percent improvement in the interconnections.
"New York
The New York study examined non-farm communities and suburbs in
Nassau, Rockland, Suffolk and Westchcster counties. We also examined
New York City,but for reasons explained above, had to leave it out of the
final analyses. The applicable population in the above-mentioned com-
58)National Center for Health Statistics, "Personal Health Expenses -
Per Capita Annual Expenses, United States: July-December 1972,"
U. S. De partment of Health, Education and Welfare, Public Health Service,
Series 10, Number 27, February 1966.
57
-------
munities number 3. 67 million persons and 645, 000 elementary school
students. The final regression employed an index of environmental quality
formed by dividing the index of operator salary by the annual average
particulates in the air. The range of the index is from . 12 to . 01. The
regression coefficient for the index is 11.4 with a standard deviation of
4.4. Thus, for each . 01 change in the index, absenteeism will drop by . 11
percent of the total population.
With the 118 million student days under consideration, this means
that for each improvement of . 01 in the environmental index, absenteeism in
these four counties will drop by 133,000. The one and two standard
deviation unit ranges respectively, are 81,000 to 185,000 and 31,000 to
235, 000 student days per . 01 improvement. Interpreting the salary index
of $2500 per increment, an average improvement of $1000 in the salaries
of the operators will,^9) holding air pollution constant, reduce absenteeism
by 71, 500.
The New York population which is affected by the study, totals 3. 7
million persons with a working force of 1. 5 million and a total income
of $15. 6 billion. Using the environmental index comprised of the index
of operator salary divided by average particulates, the adjusted regression
coefficient for the working population is 7.25. Thus, for every .01 increase
in the index, worker absenteeism will fall by . 07 percent of the total
population. For the population under consideration this means a savings
in lost wages of $10, 900, 000 per average index improvement of one per-
cent. This averages $7. 5 per worker in the four county areas surrounding
New York City.
Assuming average and unvarying particulate counts, this translates
to a total savings of $5. 95 million dollars per year per $1000 average
increase in operator salary.
Medical expenses for the illnesses under consideration in the four
county area total 94 million dollars per year. Each $1000 average increase
in operator salaries under the above air pollution conditions will produce
a . 038 percent reduction in absenteeism, due to our subject of acute
2. 5
diseases. This amounts to a savings of 1.42 million dollars per year. Thus,
the total savings per $1000 increase in operator salaries is 7. 37 million
dollars per year.
59) This is the empirically observed average.
58
-------
Six Area Group-Rural
In the other six metropolitan areas examined in the study, we divided
our sample into urban and rural sectors. The applicable rural population
affected is 480, 000 with 82, 000 elementary students. Both frequency of
testing of finished water and operator salaries are strong predictors.
The regression coefficient of testing of finished water is 1. 04. The
variable is scaled into intervals roughly equal to one year. The regression
coefficient for operator salary is 1. 0.
There are 14. 8 million student days covered by the sample. For each
average increase in the frequency of testing finished water we can expect
an increase of 160, 000 student days attendance in these areas. The one
standard deviation range would be 113, 000 to 208, 000 and the two standard
deviation range would be 65, 000 to 255, 000.
The expected reduction in absenteeism resulting from increased
operator salary is 148, 000 student days per one point average increase in
the salary index. This translates roughly to an increase of 63, 000 student
days attendance per $1000/year increase in the average salary. The one
and two standard deviation ranges are 31, 000 to 95, 000 and 0 to 127, 000,
respectively.
The population of this six area group in the study is 479, 000 with a
labor force of 170, 000 and a total income of $1". 2 billion dollars. The
primary predictor variable is the frequency of testing finished water with
an adjusted regression coefficient for the adult population of . 64. Thus,
for each one year average reduction in the frequency of testing,absenteeism
will be reduced by . 64 percent of the population. This amounts to $7. 2
million or approximately $45 per worker.
The second significant variable is the salary of the operator which
has anadjusted regression coefficient of .616 . This would lead us to
predict a savings of 2. 9 million dollars annually for each $1000 average
increase in operator salaries. This is a savings of roughly $18 per
worker per year.
The related medical expenses for this group is 9. 9 million dollars
per year.
The medical savings per average one year improvement in testing
frequency is 2. 6 million dollars per year. A $1020 average salary increase
will produce a savings of $970, 000.
59
-------
The total savings for improving the testing procedures are 9.8
million dollars per one year increased frequency. The total savings
resulting from a $1000 operator salary increase is 3.9 million dollars.
National Extrapolation
The above calculations have been based on a limited sample of 8.4
million persons, concentrated within eight relatively small geographical
areas. The original CWSS study of nine areas made no claims that the
sample was representative of a larger universe. The sample was reason-
ably diverse geographically although only three of the nine were west of
the Mississippi River.
It included a diversity of economic and cultural characteristics and
covered a wide spectrum of community sizes. Despite this, they did not
choose a scientific sample from which we can generalize to the total
population with any known degree of statistical confidence. Thus, it is
only prudent for us to advise the reader that the following national estimates
are based only on the unsupported assumption that the sample has drinking
water vs. health characteristics which are similar to those of the entire
nation.
We will compute the rational extrapolations from those of our models :
(l)the general model, (2) the Vermont model, and (3) the six area rural
model. The six area-urban/suburban model is too,tenuous to permit
such extrapolations. National extrapolations from the New York model are
seriously hampered by the strong interactions with air pollution.
The Vermont findings of the effect of interconnections with unprotected
systems is difficult to generalize given that the CWSS found that across
these total samples only six percent of the cases have this problem. We
will, therefore, not extrapolate this portion of the Vermont results.
The national estimates are straight line extrapolations using the
total number of enrolled elementary students as the basis for elementary
school absenteeism, total wages paid as the basis for total lost wages
saved and total population for special medical costs.
Extrapolation from the General Mode]
Recalling that in the general model, we found that the idea of operator
salary times the frequency of testing was the strongest water related
variable. The regression coefficient was . 08, meaning that for every 1. 0
reduction in the index value,absenteeism would be reduced by . 08 percent
+ . 08 at the . 05 confidence level. With 5. 4 billion student days nationally,
60
-------
an improvement of 1. 0 in the index would reduce elementary school
absences by 4. 3 million student days.
Expanding this to the adult population in 1969 there, were 57. 9 million
persons in the private sector of the labor force excluding rural farm workers.
These persons were employed for approximately 15 billion days and earned
nearly 390 billion dollars. Each 1. 0 reduction in the index of salary times
frequency of testing would reduce absenteeism due to illness by 7.4 million
days +_ 7. 1 million days resulting in a savings of 194 million dollars + 186
million dollars from wages alone. The total of 7. 4 million days of worker
illness plus 4. 3 million days of school illnesses means a reduction of
11. 7 million days in sub-clinical illness apparently resulting from drinking
water.
The National Health Survey reports indicate that illnesses in this
general category cost an average of $5.46 per illness, including medicine
and physicians. We do not know the average length of illness in this category.
Assuming, however, that it is close to 1. 0 days per illness, and accepting
the very gross assumption of the national extrapolation, let us estimate a
cost of $5. 00 medical costs per day of sub-clinical illness. Given one
estimate of 11. 7 +_ 11. 2 million days illness, the medical savings are
$58. 5 million dollars +_ 56 million dollars. From the general model,
therefore, we would estimate savings of 252 million dollars per year +_ 242
million dollars for each 1. 0 improvement in the index.
It should be remembered that the composition of the index makes
substantive interpretation somewhat difficult. Since better (more highly
paid) operators could be expected to engage in better testing procedures,
we cannot expect that realistically we can hold testing constant while
improving operator quality (salary). Similarly we probably do not expect
a general improvement in practice until municipalities decide to improve
what are clearly inadequate salaries. Since the relationship in the general
model is greatly strengthened by the multiplicative term, the reader must
realize that a 1. 0 increase implies a probable improvement in both
elements.
Vermont Extrapolation
.Although we did not exclude urban areas from the Vermont studies,
those which were included were so small relative to the other large
metropolitan areas in the country, that it appeared more reasonable to
extrapolate only to the rural non-farm portions of the United States.
We are extrapolating to a working population of 12.4 million persons
earning 70. 9 billion dollars.
61
-------
The Vermont analyses applied to 409, 000 persons and concluded that
an average increase of $1000 per year in operator's salaries would reduce
elementary school absenteeism by 27, 000 days per year and save 1. 7
million dollars per year in -wages and medical expenses.
To the extent that these findings are applicable to all rural non-farm
communities there would be a nationwide reduction of 3. 1 million elementary
school absentee days per year per $1000 average increase in the salary
of the operators. This is based on a nationwide rural non-farm elementary
school population of 7. 7 million students.
The reduction in lost wages would be 138 million dollars per year
and the savings in medical expenses would be 50 million dollars . Thus,
using the Vermont model the total savings to the rural non-farm popula-
tion per average $1000 increase in operator salary would amount to 188
million dollars per year.
Six Area Grouping Rural Extrapolation
In the six area rural grouping we found that operator salary was a
significant variable with a regression coefficient of nearly twice that of
Vermont.
Applying those findings to the total U.S. rural non-farm population,
we -would predict a total reduction of 5. 5 million absentee days from
elementary schools per $1000 average increase in operator salary. We
would also predict a savings of 246 million dollars per year in lost wages
and another 80 million dollars per year in medical expenses.
In the six area groupings, we found that the frequency of testing
finished water was also quite significant. In the directly affected 480, 000
population, we predicted a reduction of 160,000 days elementary school
absenteeism and a savings of 7. 2 million dollars in wages and another 2. 6
million in medical expenses for each 1 year average increase in the freqency
of testing of finished water
Extending these'findings nationwide to all rural non-farm communities,
we would expect a reduction of 15 million days elementary school absenteeism,
savings of 599 million dollars in lost wages and 230 million dollars in
medical expenses.
If we were to assume a testing improvement of frequency/month for each,
rather than frequency/year we would predict a reduction of 1. 3 million days
elementary school absenteeism, savings of 49 million dollars in lost wages and
19 million dollars in medical expenses.
62
-------
Recalling the fact that improvement in salary and improvement in
testing are likely to partially co-occur, it is not reasonable to compute
a joint savings of salary improvement and testing frequency.
Again we would urge the reader to be cautious in interpreting
these national extrapolations. The extrapolations are best estimates
given existing information but we have only limited basis for making
any statements of statistical confidence.
63
-------
Conclusions
In concluding this report, there are several points we would like to
make. The first is that the existence of a relationship between drinking
water and sub-clinical absenteeism in elementary schools is strongly
supported by the study. The results are highly statistically significant
consistent with our hypotheses, and stable across different sub-samples.
Even in the weakest cases, partial correlations were above . 33, thus
accounting for more than ten percent of the unexplained variance.
The problem of operator salary is one which can easily be miscon-
strued. Therefore, it is worth reemphasizing a point made earlier.
Although operator's salaries are strong predictors of absenteeism,
we would caution against the conclusion that pay raises will immediately
improve community health. The astoundingly low salaries have resulted
in the recruitment of generally poorly educated, undertrained operators.
While in time, higher salaries will make possible the selection of higher
caliber operators, in the short term, the salaries would have to be
supplemented by vigorous operator improvement programs if any signifi-
cant result is to be expected.
Therefore, we computed benefits on the basis of increments in
improvements in the various measures of water quality. The various
increments predict reductions of 3% to 10% of the digestive and non-
respiratory illnesses. In the specific areas studied, this translated to
reductions in elementary school absences on the order of 300, 000 to
600, 000 absences per year per incremental improvements. Converting
this to worker absenteeism using a constant of proportionality of .62,
this becomes a savings in the order of 23 million dollars per year per
increment."0' For the nationwide sample, we would cautiously estimate
that elementary school absences would be reduced on the order of
magnitude of 11-12 million absences per year per incremental improve-
ment. The wages and medical expenses saved were in the order
of 250 million dollars per year per incremental improvement.
60) These ranges.are very broad and should be interpreted only as "ball
park" estimates. The more precise estimates which are presented
in the body of the report, require breakdown by categories of drinking
water problem and geographic area.
64
-------
In performing the study, we made every attempt to be conservative
in our estimates. The medical cost figures, for example, assume that
there are no respiratory diseases resulting from drinking water. If they
were transmitted in a manner similar to gastro-intestinal illness, the
medical cost figures would be more than twice what we have estimated.
At all other points in the research program, when we were faced with
two alternatives, one of which would tend to inflate the results and the
other of which would tend to deflate them, we systematically chose the
more conservative.
We believe that the results indicate that if improvements are made
in the quality of drinking water, substantial savings will be realized. As
the CWSS report suggested and our study has confirmed, the problems are
most severe in the rural non-farm communities. ' There the economic
savings alone would be most, dramatic both because the effects are re-
latively large and the costs of resolving the problem are relatively small.
In most cases, upgrading the personnel and procedures appears to be all
that is needed.
61) Rural farm populations are excluded from the conclusions because they
were not included in either study.
135
-------
SECTION VI
BIBLIOGRAPHY
Alter, Milton. "The Distribution of Multiple Sclerosis and Environmental
Sanitation", Israel Journal of Medical Sciences, 1972, 99-105.
Barrett, Larry B. and Waddell, Thomas E. "Cost of Air Pollution Damage:
A Status Report, " Environmental Protection Agency, Research Triangle
Park, North Carolina, February 1973.
Berg, G. (ed. ). Transmission of Viruses by the Water Route, New York:
Interscience Publishers (1967).
Borneff, J. Public Health Aspects of Viruses in Water, N. A. Clarke (ed. )
Cincinnati, Ohio, 1 (1970), 1.
Chapman, Robert S. et al. "Chronic Respiratory Disease Symptom Fre-
quency in Late Adolescence, In Parents of School Children and in Elderly
Persons, " Presented at the American Medical Association's Air Pollution
Medical Research Conference, Chicago, Illinois, October 1972.
Chelsky, Morris. "The Propagating Rate: A Method for Interpreting In-
fectious Disease Incidence". American Journal of Public Health,
September 1969, 1661.
Chin, T. D. Y. , W. H. Mosley, S. Robinson and C. R. Gravelle, "Detec-
tion of Enteric Viruses in Sewage and Water, Relative Sensitivity of the
Method", Transmission of Viruses by the Water Route, G. Berg (ed. ), New
York: Interscience Publishers (1967).
Clarke, N. A. "Viruses in Water, " Journal of the American Water Works
Association, 61:10 (1969), 491-494.
Cohen, A. , et al. "Asthma and Air Pollution from a Coal-Fueled Power
Plant, " Asthma and Air Pollution, American Journal of Public Health,
September 1972, 1181-1.188.
Craun, Gunther and Leland McCabe. "Waterborne Disease Outbreaks,
1961-1970", Presented at the Annual Meeting of the American Waterworks
Association, Colorado, June 1971.
-------
Dauer, C. C. "1951 Summary, Food and Waterborne-Disease Outbreaks".
Public Health Reports, 1952, 1089.
Dohan, F. C. , G. S. Everts, and R. Smith. "Variations in Air Pollution
and the Incidence of Respiratory Disease". Journal of the Air Pollution
Control Association, 1962. 418-436.
Eliassen, R. and Cummings, R. H. "Analysis of Waterborne Outbreaks,
1938-45", Journal of the American Waterworks Association, 1948, 509.
Enviro Control, Inc. "National Assessment of Trends in Water Quality"
Final Report submitted to Council on Environmental Quality, June 1972.
Environmental Protection Agency, "National Water Quality Inventory Re-
port", Office of Water Planning and Standards, November 1973.
Fairchild, Glen A. , Judy Roan, and James McCarroll. "Atmospheric
Pollutants and the Pathogenesis of Viral Respiratory Infection", Arch.
Environmental Health. September 1972, 174-182.
French, J. G. , et al. "The Effect of Sulfur Dioxide and Suspended Sul-
fates on Acute Respiratory Disease in Children and Adults in Four Areas
of the United States, " Presented at the AMA's Air Pollution Medical Re-
search Conference, Chicago, October 1972.
Glass, A. C. and K. H. Jenkins. "Statistical Summary of Municipal Water
Facilities in the United States", USPHS Pub. No. 1039. January 1958.
Gorman, A. E. , and Wolman, A. "Waterborne Outbreaks in the United
States and Canada, and Their Significance". Journal of the American Water'
works Association, 31:225, 1939.
Harris, Robert H. and Edward M. Brecher, and the Editors of Consumer
Reports. "Is the Water Safe to Drink?" Consumer Reports, June 1974,
436-443.
Harvard University, "The Economics of Water Supply and Quality". Proj-
ect No. 16110 DTF, Environmental Protection Agency, February 1971.
Hill, William F. , Elmer W. Akin and William H. Benton. "Detection of
Viruses in Water: A Review of Methods and Application", Proceedings of
the 13th Water Quality Conference, Illinois 1971.
67
-------
Kramer, H. P. "Research Need for Drinking Water", Journal of the
American Waterworks Association. 57:1 (1965), 3-9.
Krishnaswami, S. K. "Health Aspects of Water Quality", American
Journal of Public Health. Vol. 61:11. November 1971. 2259-2268.
Lackner, Jack. "Safe Drinking Water Act of 1973: Estimated Benefits
and Costs", Environmental Protection Agency, Office of Planning and
Evaluation, January 1973.
Lave. L. B. and E. P. Seskin. "Air Pollution and Human Health",
Science 169 (3947): 723-733. August 21, 1970.
"An Analysis of the Association between U. S. Mortality and Air
Pollution", Carnegie-Mellon University, September 1970.
Liu, Oscar C, M. D. Private Communication to Dennis Tihansky, En-
vironmental Protection Agency, August 29, 1972.
Lyon, W. A. "Water and Health - Are We Concerned Enough?" Journal
of Sanitary Engineering Division. October 1970.
McDermott, J. H. "Virus and Water Quality: Occurrence and Control -
Conference Summary", Environmental Protection Agency, 1971.
McKee, J. E. and H. W. Wolf. "Water Quality Criteria". California
State Water Quality Control Board, Pub. No. 3-A, 1963.
McLean, D. M. et al. "Virus Dispersal by Water" Journal of the
American Waterworks Association. 1966, 920.
Melnick, J. L. "Detection of Virus Spread by the Water Route", Proceed-
ings of the 13th Water Quality Conference, Illinois, 1971.
Metcalf. T. G. "Biologic Parameters in Water Transmission of Viruses"
Proceedings of the 13th Water Quality Conference, Illionis, 1971.
Mosley, James W. "Transmission of Viral Diseases by the Water Route."
New York: J. Wiley and Sons, 1965.
68
-------
New York, State of. "Medical Fee Schedule", Workmen's Compensation
Board, July I960.
Neri, L. C. , J. S. Mandel, D. Hewitt. "Relation Between Mortality and
Water Hardness in Canada", The Lancet. April 29, 1972, 931-934.
Ongerth, Henry, et al. "Public Health Aspects of Organics in Water"
Journal of the American Water Works Association, July 1973, 495-498.
Petersen, Norman J. and Virginia D. Hines. "The Relation of Summer-
time Gastroentestinal Illness to the Sanitary Quality of the Water Supplies
in Six Rocky Mountain Communities", American Journal of Hygiene. I960,,
Poynter, S. F. B. "The Problem of Viruses in Water", Water Treatment
and Examination 1968, 187-204.
Purdue University, "The Utilization of Health Services: Indices and
Correlates", Health Services Research and Training Program, Indiana,
1972.
Ridker, Ronald G. Economic Costs of Air Pollution, New York: Frederick
A. Praeger, 1967.
Schafer, E. Public Health Aspects of Viruses in Water. N. A. Clarke, (ed. ),
Cincinnati, Ohio, 1 (1970), 14.
Schroeder, H. A. "Municipal Drinking Water and Cardiovascular Death
Rates", Journal of the American Dietetic Association 195: 2, 1966, 125-
129.
Shakman, Robert A. "Nutritional Influences on the Toxicity of Environmen-
tal Pollutants," ^ff^^^^^f^^^iJlf^l^i' February 1974, 105-113.
Shy. Carl M. and John F. Finklea. "Air Pollution Affects Community
Health", Environmental Science and Technology, Volume 7: 3, March 1973,
204-208.
Shy, C. M. , et al. "An Overview of CHESS". In Health Consequences of
Sulfur Oxides: A Report from CHESS. 1972.
Smith, L. S. et al. "Public Health Aspects of Water Pollution Control",
Water Pollution Control, 1969, 544-549.
69
-------
Social Security Administration, "Medical Care Costs and Prices: Back-
ground Book", Department of Health, Education and Welfare.
Stevenson, A. H. "Studies of Bathing Water Quality and Health".
American Journal of Public Health. 43:5, 1953, 529-538.
Thompson, Donovan, J. et al. "Health and the Urban Environment: Air
Pollution, Weather and the Common Cold, " American Journal of Public
Health 60:4. p. 731-739.
Tihansky, D. "Economic Damages from Residential Use of Mineralized
Water Supply", Water Resources Research, April 1974.
USPHS. Drinking Water Standards - 1962. P. H. S. Publication No. 956,
1963, 61.
Viraraghavan, T. "Water Quality and Human Health". Journal of the
American Water Works Association, October 1973, 647-650.
Weibel, S. R. et al. "Waterborne-Disease Outbreaks, 1946-60", Journal
of the American Water Works Association, August 1964, 947-958.
William, D. B. "Viruses in Water Supply - Are Present Treatment Prac-
tices Adequate?" Paper Presented at the Ontario Sect. Annual Conference,
Toronto, 1971.
70
-------
APPENDIX A
CALIFORNIA RESULTS
N = 32 Significance of Regression =.99
Absenteeism = 6. 47 - . 09 Sw + 2. 6E - 1.45F
Significance
Sw = Number of water samples
collected/month .95
E - Education of operators .90
F = Frequency of testing of water „ 95
The dependent variable in the California study supposedly was
unexcused absences from elementary schools. Because of this we had
originally thought that it would be a good test of the null hypothesis
(i. e. no relationship between water quality and training). Unfortunately,
we were reminded, lower income children and children of transit
parents (numerous in the San Bernardino-Riverside area) frequently
do not have excuses even when legitimately ill. Therefore, the measure
is quite "dirty. " By this we mean that we do not know with certainly
what we are measuring.
The results of the regression reflect this confusion. The two
operations variables measuring frequency of sample collection and
frequency of testing are statistically significantly linked in the
hypothesized direction to absenteeism as though it were truly measuring
illness. The education of the operators is strongly linked in the
opposite direction. We would conclude from the latter that for each
incremental improvement in education of the operators, truancy (or
illness) would increase by 2. 6 percent of the population. These findings
make sense only if we interpret the dependent variable as truancy. In
that event, truancy, which tends to be higher in urban areas than in rural
areas, is positively correlated with operator education which also tends
to be higher in urban areas than in rural areas. If we assume that the
dependent measure captures variance both of truancy and of illness, then
the results are somewhat reasonable. In that case, the two operations
variables are capturing the illness portion of the variance while education
of the operators is capturing the truancy portion.
71
-------
Admittedly, the argument is less than convincing, and, to some
extent, that supports the original purpose of the test of the California
data. That is, because of the dependent variable, it is the only area
in which we have to make such strained arguments defending the
linkage between water quality and absenteeism. In all other areas the
drinking water quality results were much, more sensible and straight-
forward. We think that this demonstrates that the methodology itself
does not necessarily produce positive results which are statistically
significant and consistent with one hypotheses,,
72
-------
APPENDIX B
The Profile of Medical Practice
Average Fee for Initial Office Visits
by Census Division and Specialty, 1970
Census Division General Practice
Total $ 8. 46*
New England 9. 19
Middle Atlantic 8. 66
East North Central 8. 18
West North Central 7. 27
South Atlantic 8. 56
East South Central 7. 13
West South Central 7. 21
Mountain 7. 78
Pacific 10. 67
Average 7.49
JU
"^Based on 852 observations
Notes
The less populated regions (with relatively low concentrations of
physicians) experienced lower levels of fees for certain office visits than
the more populated regions (•with relatively high concentrations of phy-
sicians).
Variations in fees presumably reflect differences in the orientation
and training of physicians in various specialties, as well as patient de-
mand for their services and other factors.
In this study fees have been compared for a very few general cate-
gories of service. Medical specialists differ widely in the types of service
they provide, and the demand for these services differ among geographic
regions. Specific comparison of fees are therefore difficult. At the very
least,, explanations for the marked variations require a great deal more
knowledge than is contained in the available data. Given these considera-
tions, assessment of variations and general trends in physicians' fees
should be approached with caution.
Source: Sixth Periodic Survey of Physicians. Nov. 1970. Center for
Health Services Research to Development, AMA.
73
-------
SELECTED WATER \ i. Report KO. 2.
RESOURCES ABSTRACTS
INPUT TRANSACTION FORM
4.
7.
9.
12.
IS.
The Health and Economic Effects of
Drinking Water
Author(s) David McCormick
Cristine Candela
Organization
Applied Science and Technology Divisa
Bendix Corporation c
First National Bldg.,201 S. Main St.";
_, -, , ,., ^,-Ann . Arbor. , , Michigan .48108 : s ,,. .
Sponsoring Organization ,
Supplementary Notes
J 3. Accession No.
j w
, 5. Report Date \
6. •'..-- •'.•;.
s- - -
j -
' S. Performing Organization
Report No.
10. Project No.
T.T T\ "7 o o *3 d D
WA /J KJbo
•°n 11. Contract/ Grant No.
Suite 68-01-1867
303 13. Type ol Report and
•:•• -••• • Period Covered
16. Abstract This study examines the relationship between sub-clinical common j
illness such as upset stomach (gastrointeritis) and other low level di- j
eases, and the quality of drinking water and water treatment plant char- !
acteristics. Such illnesses account for the greatest occurrence of health1
roblems in the country. For example, in 1969 the National Health Survey i
found that between 67 and 72 percent of disability days arising from ill-
ness fell into the sub-clinical category.
The basic hypothesis tested was that if one accounts for all
nacro-systems influences on health, the occurrence of sub-clinical ill-
lesses will be highest in those communities with the worst water quality.
The measures of drinking water quality and water treatment plant character-
istics were obtained from a Public Health Survey study entitled, "Community)
tfater Supply Study." Elementary school absenteeism was used as a surrogate
neasure for sub-clinical disease. The primary statistical tool employed j
stepwise multiple regression. j
A linear regression was performed of elementary school absentee-
ism onto total plate counts, measures of water treatment facilities, and a!
series of control variables. The number of drinking water-related absences
and their dollar value was then estimated. i
--I
i?a. Descriptors Drinking water, waterborne disease, epidemics, viruses, water
pollution control, sub-clinical illness, low level illness, school absen-
;eeism, community health, treatment plants, operators, socio-economic
status, urban/rural, coliforms, total plate count, purification
facilities.
17b. Identifiers
Immunity Water Supply Study, elementary school absenteeism, water puri-
Eication, operator
17c. CO WRR Field & Group
18. Availability
Abstractor Cristine
salary, adult population.
19. Security Cls:-s. 21 Nv. ni
(Rcpot't) Pago.
20. Securiiy Class. 22.' Price
(Page)
Send To :
WATER RESOURCES SCIENTIFIC INFORMATION CtNTUK
U.S. DEPARTMENT OF THL INTERIOR
WASHINGTON. DC 2O24O
Candela institution Bendix Corporation
WRSIC 102 ( REV. JUNE 1971)
------- |