Sample Essay on Water Fluoridation Legislation

Water Fluoridation Legislation


The author analyzes water legislation and measurement of the fluoride level in water. The study was carried out in South Africa. The dosing tolerances established by the water fluoridation legislation seem to have been violated at some instances by water providers.  This is the reason why the author used inter-laboratory data to compare the level of fluoride in water that is manageable. The following observations were made by the researcher. First, to measure the natural sample of water was more difficult as compared to synthetic samples. Therefore, water that is extracted from natural sources had more fluoride level than water sourced out from artificial sources. Second, errors in measuring natural water have not been minimized by the advancement in technology over the last 25 years. Thus, this article is concerned with the issues of measurement of fluoride level in water, as well as the regulations and strict measures that would ensure the reduction of fluoride level in natural water.

The Author’s Intended Audience

            The author audiences are the water providers and analytical chemists who are yet to comply with South African Regulations. Thus, water distributed to the population contains fluoride that exceeds the 0.2 mg/l mark. This poses danger to the lives of people that in turn may result in sanctions by the regulatory body. At the same time, the operators are liable to the public through provision of clean and healthy water. If this is overlooked, then, the population will suffer from teeth decay, bone weakness and other forms of diseases connected with high intake of fluoride.

Relevance to the chosen legislation. The article is relevant to Water Fluoridation Legislation which denotes that water providers should ensure that the fluoride level is 0.2mg/l or less. There are errors in measuring the concentration of fluoride. The tools used have not shown any improvement in reducing the errors. Therefore, volume of water distributed to the community fails to meet the set standards and this poses health risk to the society. Data recorded by South African bureau of standard (SABS), Division of water technology (DWT) and Medical University of South Africa (MEDUNSA) were extremely helpful in understanding the processes through which water is measured.  However, the institutions expressed some alterations or errors when it came to the measuring of fluoride concentration. At the same time, the implementation process fails to mitigate the risks associated with the measuring processes. Mostly, other chemicals interfere with the measurement of fluoride. This means that before measuring samples, an analysis should be conducted to ensure that the sample contains only fluoride. By doing this, the process will be free of errors.


University of Southern Africa (MEDUNSA)

Comment on author’s use of theories, models and references. The author utilizes colorimetric methods and ion chromatography (IC) for the measurement of fluoride in water samples. These methods are widely used in the fluoridation process all over the world. However, they have loopholes in terms of chemical interferences that affect fluoride concentration. This means that new and destructive methods need to be put into place to oversee the measuring of water. Information is derived from credible sources that deal with the regulation of water. They include the Data recorded by South African bureau of standard (SABS) and Division of water technology (DWT).

Implications and recommendations of the article. The author recommends that sophisticated monitoring and dosing equipment tools be applied. At the same time, a set of practical guidelines should be established by the South African water industry in order to reduce errors and keep fluoride level adequate. The implication of the article is that the regulation body needs to re-evaluate its stand and regulations that control the water industry.


Haarhoff, J. The accuracy of fluoride measurement in water and its implications for water fluoridation: rapid communication. Water SA. 2003, 29(2):219-224. Retrieved from:

The accuracy of fluoride measurement in water and its

implications for water fluoridation

Johannes Haarhoff

Rand Afrikaans University, PO Box 524, Auckland Park 2006, South Africa


The accurate measurement of the fluoride concentration in water is an essential prerequisite to stay within the allowable dosing

tolerances required by the South African water fluoridation legislation.  In the absence of reliable error estimates for fluoride

measurement in natural water samples, a study was conducted utilising data from interlaboratory comparison studies conducted

by the CSIR, the SABS and MEDUNSA.  This study shows that:

  • natural samples are more difficult to measure than synthetic samples;
  • technology advances over the last 25 years did not reduce the measurement error significantly;
  • all analytical methods suffer to some extent when natural samples are analysed; and
  • the measurement error will have to be appreciably reduced if the legal requirements of the pending water fluoridation are

to be met.

Keywords: Water fluoridation, fluoride measurement, accuracy, measurement error, natural samples, synthetic



South Africa is poised for the mandatory fluoridation of drinking

water up to a general target concentration of 0.7 mg F/l. This often

emotional issue had been debated during years of preparation of

draft legislation, until the approval of the final regulations by

Parliament and their promulgation in September 2001. The atten-

tion of the water industry has therefore moved to the many smaller

but important technical issues of practical implementation.  One of

the obvious keystones of water fluoridation is the ability to measure

the fluoride concentration in water – firstly for determining how

much supplementation is required, and secondly for checking that

the target is reached within the allowable tolerances.  This paper

thus directs its attention to the measurement of fluoride in water; a

vital, difficult aspect of water fluoridation which has seemingly

been overlooked.

Dosing accuracy legally required in SA

The SA Regulations (Government Gazette, 2001) are explicit in

their requirements for dosing accuracy, thereby closely following

similar UK regulations.  The first stipulation is:

  • Instantaneously measured fluoride concentration must be within

0.2 mg/l of the target.

The fluoridation plant operator therefore has to measure the fluo-

ride concentration in the raw water; measure the water flow rate;

calculate the fluoride dosing rate to make up the shortfall towards

the target concentration; and  check the procedure by measuring the

fluoride concentration in the fluoridated water.  For the purposes of

this paper, assume that there are no calculation errors, that the water

flow rate and the fluoride solution strength in the day tank are

exactly known, and that the fluoride dosing pump can be set with

absolute accuracy.  The only error to consider is therefore the error

due to the measurement of the fluoride concentration before and

after dosing.

It is obvious that any measurement error will be compounded

during the required feedback loop. This is illustrated by the

following example, where a constant under-measurement error of

30% is assumed for illustration:

  • The raw water fluoride concentration is actually 0.20 mg/l, but

measured as 0.14 mg/l.  The shortfall is therefore actually 0.50

mg/l below the target level of 0.70 mg/l, but perceived by the

operator to be 0.56 mg/l.

  • The dosing pump is set to add 0.56 mg/l, resulting in an actual

concentration of 0.76 mg/l or 0.06 mg/l above the target.  The

operator measures this as 0.53 mg/l and still perceives a short-

fall of 0.17 mg/l.

  • The operator, correctly, increases the dosing rate by 0.17 mg/l

to get an actual concentration of 0.93 mg/l, which is still

perceived as 0.65 mg/l or 0.05 mg/l short of the target.

  • This cycle continues until the concentration stabilises at the

perceived target concentration.  By this time, the actual concen-

tration is about 1.00 mg/l, or about a 40% overdose.

As an apparent independent check on the dosing accuracy, the SA

Regulations further stipulate that:

  • Average dosage over a week must be within 0.1 mg/l of the


The fluoridation operator should check this requirement by a

careful site inventory of  on-site fluoridation chemicals and the

volume of water fluoridated since the previous check. (It should

!+27 11 489  2148; fax: +27 11 489  2148; e-mail:

Received 3 March 2003.

ISSN 0378-4738 = Water SA Vol. 29 No. 2 April 2003220 Available on website

certainly not be checked by mathematical averaging of the spot

measurements taken during the week, as this would not detect an

error such as demonstrated above!). An inventory check will

provide an external check on how much fluoride was dosed.  To

then obtain the average concentration in the fluoridated water, the

fluoride  dosed has to be added to the concentration originally

present, which can be obtained in no other way than taking the

volume-weighted average of the raw water concentrations meas-

ured during the week.  It is clear that an inventory check is still

dependent on the accuracy of measuring the fluoride concentration

in the raw water.  Using the same example as above where the raw

water concentration is undermeasured by 0.60 mg/l, the target will

therefore also be missed by 0.60 mg/l without knowing.

This line of reasoning underscores the vital importance of

measuring the fluoride concentration accurately.  For the proper

estimation of risk, it is essential to get a quantitative grip on the

practical limits of the error of fluoride measurement.  Despite

numerous enquiries by the author, no comprehensive or convinc-

ing information could be readily obtained.  As the author is neither

an analytical chemist nor a statistician, it is hoped that this cursory

analysis will stimulate further, more comprehensive consideration

of this question by others.

Data sources and exclusion criteria

The Division of Water Technology (DWT) of the CSIR ran an

interlaboratory comparison programme from the middle of the

1970s to 1998.  A wide variety of sample types and analytical

methods were covered for water, sewage, sediments and sludges.

As part of this programme, samples for fluoride analysis were

distributed on eight occasions, comprising both synthetic samples

(made up of distilled water with reagent-grade chemicals) and

natural samples (surface water, effluent and groundwater).  In total,

21 samples were sent out between 1977 and 1998, with an average

of 35 laboratories participating in each survey.  A total of 725 data

points were collected.  The DWT followed a consistent approach

of removing all the outliers with a statistical procedure before

reporting.  For the study reported here, all the original data points,

outliers included, were retrieved from old files and recaptured


The South African Bureau of Standards (SABS) Test House

runs an ongoing WATERCHECK programme which covers more

than 70 analytical laboratories in Southern Africa.  Fluoride is part

of a group of determinants which is measured every third month on

three samples (one synthetic and two natural samples).  The latest

annual data set for 2002 was retrieved, during which a total of 631

fluoride values were returned from 55 laboratories.  No outliers are

removed in the reporting process adopted by the SABS, but a

socalled robust statistical evaluation is used to reduce the effect of

outliers.  For this study, all the original values, outliers included,

were recaptured electronically.

A small once-off  survey was conducted by the Medical

University of Southern Africa (MEDUNSA) during 2002.  Five

samples of tap water, spiked to varying degrees with fluoride

compounds, were sent to 10 commercial laboratories in the Gauteng

area for fluoride analysis, with a clear note that it formed part of an

interlaboratory comparison.  This data set, in its entirety, was

included in this study.

In an operational environment, the operator has to react to each

measurement as it is taken.  Outliers, on the other hand, can only be

detected and rejected after some time, when a retrospective statis-

tical analysis becomes possible. For this reason, all data points were

critically considered for this analysis.  Six complete sample sets

were excluded on the basis of:

  • Two samples were spiked to more than 20 mg/l, which is way

outside the range where water fluoridation plants will ever


  • One laboratory returned values for a set of four samples where

every single value was deemed to be an outlier by the DWT

procedure, which indicated a gross, systematic error which

warranted exclusion from the data set.

A number of individual data points were removed next, as they

were obviously extreme outliers:

  • Individual measurements which were less than 20% of the

average concentration were excluded. This eliminated 12 of the

data points.

  • Individual measurements which were more than five times the











0.1 0.3 0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9

measured F / true F

fraction of data points

synthetic samples

natural samples

Figure 1

Histogram of data

points for synthetic

and natural samples


ISSN 0378-4738 = Water SA Vol. 29 No. 2 April 2003 221

Available on website

average concentration were excluded. This eliminated 31 of the

data points.

This left a total of 1 278 data points, including 36 samples over

25 years, covering a range of average concentrations between

0.07 mg/l and 4.37 mg/l.  Eighty per cent of the samples had

average concentrations of between 0.2 mg/l and 2.1 mg/l.

In synthetic samples, the “true” fluoride concentration is

known from the way the samples are made up.  In some cases, they

were reported, but they were not used for this study.  For the

purposes of this study, the “true” fluoride concentration of each

sample was taken as the arithmetic mean of all the measurements

reported for that sample, after removal of the outliers as described

above.  All the data points were then finally normalised by dividing

each value by the “true” value for that sample.

Data analysis

Natural vs. synthetic samples

As with most analytical procedures, care has to be taken to

eliminate interferences by background chemical compounds.  Fluo-

ride, to name but one example, will form complexes with alu-

minium which will affect both aluminium and fluoride measure-

ments if no precautions such as preliminary distillation or digestion

are taken.  During the DWT programme, the consistency of fluoride

measurements generally suffered from a high percentage of outliers

attributed to not adequately pretreating the samples.

In natural samples, as opposed to synthetic samples, there are

many more unknowns that could affect the fluoride measurements.

The hypothesis then has to be that accuracy of fluoride measure-

ments in natural samples has to be poorer than in synthetic samples.

The data set was therefore split into a ”natural” group (747 data

points from 21 samples) and a “synthetic” group (525 data points

from 15 samples).  Figure 1 shows the distribution of the data in

histogram form.  Two points emerge from Fig. 1:

  • The synthetic samples are more narrowly distributed, as hy-

pothesised above.  The remainder of this analysis will thus be

carried through by keeping natural and synthetic data separate.

  • There is a definite asymmetry in the distribution, especially

evident for the natural samples.  This is due to the fact that the

distribution is constrained on the left by a zero concentration

limit.  The normal distribution is therefore not adequate for

describing fluoridation accuracy; the log-normal distribution

would be a more suitable candidate should someone wish to

pursue the modelling of the measurement error.

“Old” vs. “new” technology

The DWT data set spans the period 1977 to 1998, therefore

including data collected by what many would consider to be “old”

technology.  The SABS and MEDUNSA data sets were collected

during 2002 and are a reflection of the current state of technology

used in South African laboratories.  An obvious question is whether

the much-touted advances in instrument technology led to a signifi-

cant, if any reduction in measurement error.

The data set was next split by both sample type (natural and

synthetic) and data set (DWT, SABS and MEDUNSA), yielding

the following number of samples per subset:

  • DWT synthetic 314 data points from 11 samples
  • DWT natural 297 data points from 8 samples
  • SABS synthetic 209 data points from 4 samples
  • SABS natural 402 data points from 8 samples
  • MEDUNSA natural 50 data points from 5 samples

These data were used to construct cumulative distribution lines,

which are shown in Fig. 2.  The following points emerge from

Fig. 2:

  • The difference between natural and synthetic samples observed

earlier, remains strongly evident for the individual data sets.

  • There is very little, if any, difference between the “old” data set

of the DWT and the “new” data set of the SABS.  This is evident

from a comparison of both the synthetic and natural sample

lines.  Advances in instrument technology have made no

appreciable difference to the empirically observed measure-

ment accuracy.












0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

measured F / true F

fraction less than

DWT synthetic

DWT natural

SABS synthetic

SABS natural

MEDUNSA natural

Figure 2

Cumulative distribution

plot of fluoride

measurement error,

showing difference

between data sets and

sample types

ISSN 0378-4738 = Water SA Vol. 29 No. 2 April 2003222 Available on website

  • The MEDUNSA data set does not clearly follow either the

natural or the synthetic data.  This is understood if the nature of

the MEDUNSA samples is explained.  One sample was a tap

water sample with an average concentration of 0.07 mg/l.  The

other four samples of the set were made up of the same tap

water, progressively spiked to concentrations of 0.19 mg/l,

0.45 mg/l, 0.87 mg/l and 1.18 mg/l respectively.  The samples

themselves could therefore be considered to be progressively

more “synthetic” than natural.

Comparison of analytical methods

Fluoride in water can be measured in a number of ways.  There are

colorimetric methods (of which the SPADNS method is the most

commonly used), the ion-selective electrode (ISE) method, and ion

chromatography (IC), all comprehensively described in Standard

Methods (1985).  For routine measurement on site, the practical

choice is between SPADNS (only an option when spot analysis is

opted for rather than continuous monitoring) or ISE (suitable for

both spot and continuous measurement).  During the course of the

DWT study, it seemed that the SPADNS method was more suscep-

tible to these interferences and it was suggested to SPADNS users

that they should consider switching to the ISE technique.

The SABS and MEDUNSA data sets do not have information

on the methods used by the participating laboratories, but the DWT

data set included such information between 1977 and 1990, which

includes 35% of all the data considered in this study.  The method-

specific data points break down to:

  • Natural samples measured colorimetrically 47 data points
  • Synthetic samples measured colorimetrically 76 data points
  • Natural samples measured by ISE 95 data points
  • Synthetic samples measured by ISE 148 data points
  • Natural samples measured by IC 33 data points
  • Synthetic samples measured by IC 44 data points

The cumulative distribution plots for these six groups are shown in

Fig. 3.  The following points emerge from Fig. 3:

  • For synthetic samples, there is little difference amongst the

three methods.  All follow the same narrow distribution seen in

the earlier analyses.

  • For natural samples, all the methods show significantly poorer

accuracy than for synthetic samples.  The effect of interfering

compounds cannot be eliminated by simply switching the

analytical method.

  • The colorimetric method returns higher concentration values

than the ISE and IC methods.

  • The ISE and IC methods return comparable results for natural













0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

measured F / true F

fraction less than

colorimetric – syntheti c

colorimet ric – natural

ISE – synthetic

ISE – natural

IC – synthetic

IC – natural












0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

measured F / true F

fraction less than

synthetic samples

natural samples

Figure 3

Cumulative distribution plot

of fluoride measurement

error, showing difference

between analytical methods

and sample types

Figure 4

Cumulative distribution plot

of fluoride measurement

error, showing difference

between natural and

synthetic samples

ISSN 0378-4738 = Water SA Vol. 29 No. 2 April 2003 223

Available on website

Likely fluoride measurement accuracy

Analytical laboratories commonly use the “standard deviation

approach” as a scoring criterium during interlaboratory compari-

son studies.  Both the DWT and the SABS programmes consider

results within 1 standard deviation from the mean as “good”,

between 1 and 2 deviations away as “average” and further than

2 standard deviations as “poor”.  For normal distributions (nor-

mally implicitly or explicitly assumed), this translates into a

criterium that 68% of all measurements would be “good” and a

further 27% of all measurements would be “average”.  A more

conservative, arbitrary assumption of only 50% “good” values is

adopted for the purposes of this paper – an assumption of the author.

Figure 4 shows the cumulative distribution plot of the entire

pooled data set, split between natural and synthetic samples.  The

25th and 75th percentile lines from Fig. 4 provide error estimates of

the measurement error that could reasonably be expected from

“good” laboratories.  These values are shown in Table 1.

Questions pertaining to pending water


Does the performance of analytical laboratories give a

reliable indication of measurement performance at an

operational water fluoridation plant?

This is a crucial question which cannot yet be answered.  Intui-

tively, there are reasons why analytical laboratories should do

better than operational plants:

  • Recognised analytical laboratories that care to participate in

nation-wide comparisons invariably have trained chemists

working under professional supervision and clean, controlled

laboratory conditions.

  • It is only human that each participating laboratory will put its

best possible foot forward at the time of the comparative

analyses.  It is difficult if not impossible to sustain the same

diligence over the many years of continuous operation of a

water fluoridation plant.

Conversely, it can be argued that operational plants should do


  • The interlaboratory comparisons normally request a suite of

analyses of which fluoride forms only a small part.  A control

laboratory dedicated to a single compound may be better

focused and equipped to produce better results.

  • Many of the participating laboratories do not specialise in

water analyses, but probably in some other areas, for example

air, food, coal, soils, etc..

Were the samples of this study representative of

general water treatment conditions?

Those samples with unrealistically high concentrations were ex-

cluded from the analysis.  The data from synthetic samples were

analysed separately from the data from natural samples.  A small

number of borehole and sewage effluent samples remained in the

data set, but in practice these sources could also be encountered.  In

general, the synthetic samples were spiked to F concentrations

above 1 mg/l, while most natural samples had concentrations less

than 1 mg/l.  The very low concentrations of say less than 0.3

mg/l were poorly represented in the data set and it seemed from the

few available data points as if the variability of their results was


In other words, the data sets were reasonably representative

of actual water sources, but further studies have to focus specifi-

cally on the measurement of concentrations at or below say 0.5

mg/l, as this will be the critical region for water fluoridation plant


Does the uncertainty around fluoride measurement

present an obstacle to water fluoridation?

If nothing is done to improve measurement accuracy, or to demon-

strate with a more comprehensive study that the problem is smaller

than it now seems, then it would.  In the example at the beginning

of the paper, it was shown that undermeasurement of 30% would

make it extremely difficult to stay within the allowed dosing

tolerances.  But the author has no doubt that the South African

water industry, internationally known for decades for its drive and

innovation, has adequate expertise and resources to successfully

resolve this problem, provided its leadership shows the necessary


Is fluoride measurement error the only significant

source of error?

The only systematic study of total fluoride dosing error known to

the author has recently been published (Lalumandier et al., 2001).

Operators at 1 280 fluoridation plants in 12 US states responded,

amongst others, to a question of how close they could maintain the

fluoride dosing level to the optimal level.  This effectively means

that operators had to assess themselves without any external

controls, which would probably lead to optimistic estimates.  The

results, briefly summarised, were:

  • A total of 25.9% responded that they could maintain the

concentration within 0.1 mg/l of the optimal level; 49.3%

could maintain it between 0.1 and 0.2 mg/l from the optimal

level and 19.5% could stay between 0.2 to 0.3 mg/l of the

optimal level.

  • For the large plants (above a capacity of about 4 Ml/d) the

performance was significantly better, with 33.5% of the large

plants maintaining the dosage within 0.1 mg/l of the optimal,

as opposed to 21.3% for the small plants.

  • The two main causes for the dosing error were reported to be

problems with feeding equipment (18.4%) and variations in

raw water flow (12.8%).

  • Poor training of operators was blamed for the dosing error in

8.7% of the responses.

These survey results are interpreted as follows:

  • Most US treatment plants would not be able to stay within the

tolerances allowed by the SA Regulations.


Fluoride measurement limits based on

50% of all measurements being acceptable

Lower limit Upper limit

Synthetic samples -14% +11%

Natural samples -39% +17%

ISSN 0378-4738 = Water SA Vol. 29 No. 2 April 2003224 Available on website

  • Measurement error (more difficult to detect) is not amongst the

top two reasons for dosing error, indicating that the allowable

measurement error should be set at a significantly lower level

than the allowable total dosing error.

  • Proper training of operators provides one way to improve

dosing performance.


This paper presents more questions than answers.  It is primarily

intended to alert those water chemists and operators grappling with

the pending implementation of water fluoridation in South Africa

to the uncertainties surrounding the measurement of fluoride in

water.  It was shown that very good accuracy for fluoride measure-

ment is a prerequisite for staying within the narrow dosing toler-

ances allowed by the SA Regulations.

Water providers are currently considering their technological

options for expensive, sophisticated dosing and monitoring equip-

ment for the pending fluoridation.  Now is the time to undertake an

urgent, systematic study of the fluoridation measurement options,

pertaining to our typical source waters with their own unique

potential for analytical interference.  In order to get water fluorida-

tion right, the SA water industry needs a set of practical, locally

relevant guidelines for fluoride measurement, which can only be

provided through the collaboration of the water chemists and

process controllers in water treatment practice.


The CSIR (Dr. Carlsson and Ms. Elizabeth Truter of Watertek), the

SABS (Mr. Chris Fouche and Ms. Marie McKay of the

WATERCHECK programme) and MEDUNSA (Prof. JB du Plessis)

generously shared the data reported on in this paper.  Prof. Paul

Coetzee of the Department of Chemistry at the Rand Afrikaans

University provided better perspective on the analytical methods

and their susceptability to interferences.


DEPARTMENT OF HEALTH (2000) Regulation  No. 873 Regulations

under the Health Act, 1977 (Act No. 63 of 1977) Regulations on

Fluoridating Water Supplies. Government Gazette, 8 September 2000,

Vol. 423, No. 21533, Pretoria.


(2001)  US drinking water: Fluoridation knowledge level of water

plant operators.  J. Public Health Dent.  61 (2).

STANDARD METHODS (1985)  Standard Methods for the Examination

of Water and Wastewater (16th edn.)  Published jointly by the Ameri-

can Public Health Association, the American Water Works Associa-

tion and the Water Pollution Control Federation, Washington DC