COLLABORATIVE STUDY
of
REFERENCE METHOD FOR THE DETERMINATION
OF SUSPENDED PARTICULATES IN THE ATMOSPHERE
(HIGH VOLUME METHOD)
Herbert C. McKee
Ralph E. Childers
Oscar Saenz, Jr.
Contract CPA 70-40
SwRI Project 21-2811
Prepared for
Office of Measurement Standardization
Division of Chemistry and Physics
Air Pollution Control Office
Environmental Protection Agency
June 1971
SOUTHWEST RESEARCH INSTITUTE
SAN ANTONIO HOUSTON
-------
COLLABORATIVE STUDY
of
REFERENCE METHOD FOR THE DETERMINATION
OF SUSPENDED PARTICULATES IN THE ATMOSPHERE
(HIGH VOLUME METHOD)
Herbert C. McKee
Ralph E. Childers
Oscar Saenz, Jr.
Contract CPA 70-40
SwRI Project 21-2811
^ Prepared for
x. Office of Measurement Standardization
Division of Chemistry and Physics
Air Pollution Control Office
V" Environmental Protection Agency
\ June 1971
si
Approved:
Herbert C. McKee
Assistant Director
Department of Chemistry
and Chemical Engineering
-------
SUMMARY AND CONCLUSIONS
This report presents information obtained in the evaluation and collaborative testing of a method to
measure the mass concentration of suspended participate matter in the atmosphere. Minor variations of this
method have been used extensively by the National Air Sampling Network and by state and local air
pollution control agencies for approximately 15 years.
This method was recommended as a tentative standard method by the Intersociety Committee, a
cooperative group consisting of representatives of eight scientific and engineering societies.* It was pub-
lished as Tentative Method 11101-01-70T in Health Laboratory Science, October 1970, pp 279-286. It was
then tested as a part of this program, by means of a collaborative test involving 12 laboratories. A statistical
analysis of the data obtained provided the following results:
• The relative standard deviation (coefficient of variation) for single analyst variation (repeat-
ability of the method) is 3.0 percent.
• The relative standard deviation for multilaboratory variation (reproducibility of the method) is
3.7 percent.
• The minimum detectable amount of particulate matter is 3 mg (95 percent confidence level).
This is equivalent to l-2jug/m3 for a 24-hr sample. Values this low will rarely, if ever, be
observed in the atmosphere, and thus lack of sensitivity does not limit the use of this method
for ambient air quality measurement.
These results show that the method can give very good precision when followed rigorously. At the
same time, it is rugged, and variations in procedure and technique can occur with only a minor effect on the
results.
Based on these results, this method was adopted as a standard method for the measurement of
suspended particulate matter in the atmosphere by the Standardization Advisory Committee of the Air
Pollution Control Office, Environmental Protection Agency. It was published in the Federal Register,
April 30, 1971, and is reproduced as Appendix A of this report.
* Air Pollution Control Association
American Chemical Society
American Conference of Governmental Industrial Hygienists
American Industrial Hygiene Association
American Public Health Association
American Society for Testing and Materials
American Society of Mechanical Engineers
Association of Official Analytical Chemists
The Intersociety Committee receives partial financial support through APCO Contract 68-02-0004.
in
-------
ACKNOWLEDGMENT
The authors wish to express appreciation to the Project Officer, Mr. Thomas W. Stanley, and staff
members of the Office of Measurement Standardization, APCO, for assistance in planning the collaborative
study and in site preparation and preliminary sampling. Through this work, adequate space and facilities
were made available for the sampling, which was performed simultaneously by twelve participating labora-
tories, and for the calibration and other supplementary work which was required.
The assistance and cooperation of the participating laboratories is also acknowledged with sincere
appreciation for the voluntary efforts of the staff members who represented each organization. The
representatives and organizations participating in the collaborative test program were as follows:
Name
Organization
Harold K. Beatty
J. H. Blacker
Department of Environmental Control
Chicago, Illinois
Esso Research and Engineering Co.
Linden, New Jersey
Walter W. Cooney
Robert C. Crabtree
Edward J. Hanks, Jr.
W. Kenfield
State of Maryland
Division of Air Quality Control
Baltimore, Maryland
Jefferson County, Kentucky
Air Pollution Control District
Louisville, Kentucky
Air Pollution Control Office
Research Triangle Park, North Carolina
Montgomery County, Ohio
Air Pollution Laboratory
Dayton, Ohio
Rudy Marek, Jr.
Southwest Research Institute
Houston, Texas
M. R. Midgett
Air Pollution Control Office
Cincinnati, Ohio
Frank G. Norris
City of Steubenville
Air Pollution Department
Steubenville, Ohio
IV
-------
Name
Organization
James M. Peters
I. A. Schwabbauer
Van A. Wheeler
University of Texas at Austin
Austin, Texas
University of Iowa
Iowa City, Iowa
Air Pollution Control Office
Cincinnati, Ohio
-------
TABLE OF CONTENTS
Page
I. INTRODUCTION 1
II. EVALUATION OF THE METHOD 2
A. Calibration of Flow Rate 2
B. Selection of Sampling Locations 2
C. Effect of Possible Volatilization 3
III. COLLABORATIVE TESTING OF THE METHOD 4
A. Selection of Collaborators 4
B. Planning the Test Series 5
C. Site Evaluation 6
D. Conducting the Test Series 6
E. Summary of Statistical Analysis 7
F. Calibration Errors 9
APPENDICES
A. Reference Method for the Determination of Suspended Particulates in the Atmosphere
(High Volume Method) A-l
B. Statistical Methods B-l
vu
-------
I. INTRODUCTION
Of all the various methods available to measure
atmospheric contaminants, the so-called High Volume
Method (frequently called "Hi Vol" Method) has
probably been used more extensively than any other.
One reason for this is the widespread occurrence of
dust and particulate matter in the atmosphere, with
measurable quantities occurring as natural back-
ground even in remote areas. Another reason is that
the equipment required is relatively inexpensive,
whereas the measurement of other contaminants may
require much more expensive and elaborate instru-
ments. First developed by the predecessor organiza-
tion of the present EPA Air Pollution Control Office
in the mid-1950's, this method has been used on a
large scale by the National Air Sampling Network for
approximately 15 years. Many state and local air
pollution control agencies have also used this method,
or variations of it, for monitoring networks to supple-
ment the Federal effort.
The High Volume Method can be used by any
laboratory possessing normal equipment and skills if a
minimum of special purpose equipment is added. A
sampler capable of pulling air through a filter medium
is used to collect the dust and particulate matter,
which is then measured on a weight basis. Chemical
analysis of the collected deposit is also possible to
measure various constituents such as metals, nitrate,
sulfate, etc. (Analytical procedures for such analyses
are not included in this standard method at the
present time.)
Since no standardized procedure has been avail-
able to guide different laboratories in the use of this
method, each laboratory has had to develop and
evaluate different techniques for weighing filters, cali-
brating flow rates, and other details in conducting
tests. As an example, only 40 percent of the collabo-
rators in this test were routinely using a procedure
essentially similar to the method tested. Less than
half had access to a positive displacement meter to
use as a primary standard for calibration. While
80 percent had some sort of orifice calibration unit,
only 10 percent indicated that it was ever calibrated
using a primary standard of any kind. The calibration
of a sampler was indicated to be a standard practice;
however, only 20 percent calibrated routinely on a
monthly basis and 40 percent on a yearly basis. The
remaining 40 percent did not calibrate on any pre-
determined schedule.
In order to obtain comparable data so that
interlaboratory comparisons would be feasible, the
Air Pollution Control Office has been working for
some time to develop standard methods which could
be used by all persons making air quality measure-
ments. A number of scientific and engineering societies
have also been active in the development of standard
methods, including several of those now participating
in the Intersociety Committee, whose members are
listed in the Summary and Conclusions.
Following the development of a tentative stan-
dard method by the Intersociety Committee, the final
step in the standardization process is to conduct a
collaborative test, or interlaboratory comparison, of
the proposed standard method. This procedure, also
called "round-robin testing," has been used to evaluate
many different methods of measurement in such
diverse fields as water chemistry, metallurgy, paint and
surface coatings, food and related products, and many
others. A test of this nature by a representative group
of laboratories is the only way that the statistical limits
of error inherent in any method can be determined
with sufficient confidence.
This report presents the results of a col-
laborative test of the High Volume Method con-
ducted by Southwest Research Institute and the Air
Pollution Control Office, together with the statistical
analysis of the data obtained. In planning for the col-
laborative test, it was necessary to evaluate several
aspects of the recommended method with respect to
flow rate calibration and other details. The informa-
tion obtained in this evaluation is also presented as
background information relating to the collaborative
test program and as information helpful in under-
standing the capabilities and limitations of this stan-
dard method.
-------
II. EVALUATION OF THE METHOD
Since the High Volume Method has been used
by many laboratories, it is not surprising that many
different variations of this method have occurred,
especially in flow rate calibration and other details of
measurement. To aid in resolving some of these dif-
ferences, an evaluation of the method was performed
during the development of plans for collaborative
testing This section of the report presents the data
obtained in some of these tests and discusses the
importance of different procedures in planning a col-
laborative test and in evaluating the capabilities and
limitations of the method itself.
A. Calibration of Flow Rate
In calibrating the flow measuring system of the
sampler, it is necessary to vary the flow rate over the
operating range of the sampler in order to obtain
accurate calibration. In the past, most workers have
done this by either of two methods: (l)use of a
series of perforated plates in place of the filter used
for sample collection, to vary resistance to flow (as
presently shown in the method); or (2) use of a vari-
able transformer to change the voltage applied to the
blower motor, thus varying the speed of the blower
and changing the flow rate. However, recent
workO)* indicated that varying the speed of the
blower led to possible errors, especially at low flow
rates, because this changed the flow pattern through
the sampler, which in turn caused changes in the flow
pattern past the variable-orifice meter located on the
discharge side of the blower. Therefore, both pro-
cedures were evaluated to see if errors would occur
with the method of calibration specified in the pro-
posed method.
Two types of high volume samplers were
checked. The first was equipped with a rotameter for
flow measurement, and the second had an orifice on
the discharge side of the blower for continuous
recording of flow rates. With both types of sampler,
flow rate was varied by using resistance plates, as
specified in the method, and by using a variable trans-
former to vary the speed of the blower. No differ-
ences were observed with the sampler equipped with
a rotameter for flow measurement. Identical readings
were obtained throughout the range of flow rate (30
to 65 cfm), regardless of whether the flow was varied
by using the resistance plates or by controlling blower
speed. With the other sampler, however, this was not
the case. Identical readings were obtained at the
higher flow rates, but at lower flow rates a difference
was observed. In all cases, the indicated flow was
lower when the voltage was varied than when the
resistance plates were used. The differences were not
great, ranging from 2 cfm at the low rate of 30 cfm to
no difference at 60 cfm. However, this confirmed the
fact that some error may occur with this method of
calibration, as was indicated by the University of
Cincinnati results, and, therefore, the Intersociety
Committee deleted the optional calibration procedure
based on voltage variation. The use of resistance
plates was specified as the standard method of vary-
ing flow during calibration and is included in the
method as now published. This method was used dur-
ing the collaborative test of the method.
B. Selection of Sampling Locations
A standard method can be developed to specify
details of calibration, flow measurement, analysis of
samples, etc., in considerable detail, and the statistical
accuracy of these various procedures can be estab-
lished. In the measurement of total particulate mat-
ter, however, a serious problem exists which is not
subject to statistical evaluation, except in rather
general terms. Since particulate matter is a ubiquitous
constituent of the atmosphere, and, since the amount
varies widely from place to place, the selection of
sampling times and locations is a matter of para-
mount importance. This cannot be standardized in a
specified method, but some general guidelines can be
given to indicate how these factors will influence the
results obtained.
Measurements of total particulate matter by the
High Volume method are usually made to determine
*Superscript numbers in parentheses refer to the List of References.
-------
overall community-wide patterns which exist. How-
ever, if the results obtained in any one location are
influenced to a major degree by a significant source
of dust located nearby, then the results will be typical
of dust levels over an extremely small area rather than
the portion of the community which a particular
monitoring station should represent.
This means, then, that if a monitoring station is
intended to represent typical levels over a large area,
it should be placed in a location free of local inter-
ferences. The most obvious interferences to avoid are
unpaved streets and parking lots, a major dust-
emitting industrial plant that would constitute a
single dominant source, nearby construction activi-
ties, and other obvious sources which affect only
limited geographical areas.
No specific figures can be given for the amount
of interference which such sources can cause since
this may vary over many orders of magnitude,
depending on the nature of the source and on the
distance from the source to the sampling location.
However, samples showing several hundred or even a
few thousand micrograms per cubic meter (jug/m3)
have been collected directly downwind from obvious
sources of this nature(2,3); which indicates that
values can be obtained which are several times the
community-wide levels that usually exist.
Another important variable in some circum-
stances is the height of the sampler above the ground.
Dust raised by automobile traffic and other extra-
neous sources frequently contains a large proportion
of relatively large particles which settle back to the
earth fairly rapidly, and therefore would not be col-
lected by a sampler located some distance above the
earth's surface. In one study'2), for example, samples
were collected 3 ft and 30 ft above ground level next
to a paved street with a light deposit of dust on the
surface of the pavement from nearby construction
activity; values obtained at the upper level averaged
about 50 percent of those obtained at the lower level.
In another study(^), 24-hr samples were collected in
four cities at levels of 3 ft and 30 ft; in this case,
average values for the four locations indicated that
samples collected 30 ft above ground level showed
from 60 percent to 90 percent of the dust loadings
measured 3 ft above ground level.
If samples are obtained for less than a 24-hr
period, the time of day is also an important variable.
This is illustrated by samples collected at 3-hr inter-
vals which show very low levels at night, occasionally
in the range of rural background levels such as 20 to
40 jUg/m3. During daylight hours, however, values are
usually much higher due to vehicular traffic and other
human activities, frequently reaching 3-hr levels as
much as four to five times the nighttime values! v
This shows the desirability of collecting 24-hr samples
to evaluate overall conditions in the community.
As stated previously, no specific standard pro-
cedures can be given to compensate for these prob-
lems. However, a considerable degree of judgment
and experience must be utilized in the selection of
sampling locations and in the evaluation of data to
avoid reaching erroneous conclusions because of near-
by sources of dust which may exert an undue influ-
ence on sampling results. Since the objective of a
monitoring network usually is to determine com-
munity-wide levels, sampling locations should be
chosen to avoid excessive influence from a single
dominant source immediately adjacent to the
sampler.
To conduct a collaborative test, results were
needed which would be representative of particulate
levels over a considerable area and which would not
be unduly influenced by a single source. As discussed
later, a location was selected which appeared to meet
this requirement.
C. Effect of Possible Volatilization
As outlined in the method, filters should be
equilibrated in the laboratory for 24 hr prior to
weighing, or reweighing after sampling, to determine
the weight of material collected. During the pre-
liminary sampling prior to collaborative testing, filters
were kept and reweighed after succeeding intervals of
time to obtain some information on possible changes.
-------
TABLE I. SUMMARY OF PRELIMINARY
DATA FOR SAMPLING OF PARTICU-
LATES AND THE EFFECT OF
EQUILIBRATION TIME
Parnculate Concentration, ng/m3
Sampler
Number*
1
•>
3
4
5
First Day
Equilibration
Time
1-Day
95
100
85
96
92
4 5 -Day
89
88
76
83
82
Second Day
Equilibration
Time
1-Day
80
73
82
83
80
4 5-Day
75
69
75
76
76
Third Day
Equilibration
Time
2-Day
82
73
83
77
82
5 5-Day
76
69
79
73
78
Table I tabulates the results of these tests. After an
additional 3.5 days of equilibration, the indicated
concentration of particulate matter had decreased by
5 percent or more. It is presumed that this was due to
evaporation of volatile "tarry" organic materials or
loss of water from the filters, with the tarry materials
being the more probable explanation. No completely
satisfactory method exists to compensate for these
changes, but this factor emphasizes the importance of
observing the 24-hr time limit specified in the method
if the results are to be comparable from one series of
tests to another. If the presence of volatile impurities
is suspected of causing any significant variation in
sampling results, separate investigation of this factor
would be advisable.
While the exact cause has not been determined,
greater amounts of tarry materials would be expected
if the particulate sample contained a substantial
amount of coal smoke, smoke from an operation such
as a hot mix asphalt plant or coke oven, particulates
from vehicle exhaust, or other materials that would
likely be high in organic content. Much smaller losses
would be expected in areas where the primary con-
stituents of atmospheric particulates would be natural
soil particles, dust from sandstorms, or other mate-
rials of mineral origin.
III. COLLABORATIVE TESTING OF
THE METHOD
An important step in the standardization of any
method of measurement is the collaborative testing of
a proposed method to determine, on a statistical
basis, the limits of error which can be expected when
the method is used by a typical group of investiga-
tors. The collaborative, or interlaboratory, test of a
method is an indispensable part of the development
and standardization of an analytical procedure to
insure that (l)the procedure is clear and complete,
and (2) the procedure does give results whose pre-
cision and accuracy are in accord with those claimed
for the method/^) Among other organizations, the
American Association of Analytical Chemists
(AOAC) and the American Society for Testing and
Materials (ASTM) have been active in the field of
collaborative testing and have published guidelines of
the proper procedure for conducting collaborative
tests and evaluating the data obtained/^) Publica-
tions of both of these organizations were used
extensively in planning and conducting the collabora-
tive test of this method to measure atmospheric
particulates.
After the preliminary evaluation of the pro-
posed method had been completed and various ques-
tions regarding procedure had been clarified, a
detailed collaborative test was undertaken to obtain
the necessary data to make a statistical evaluation of
the method. This section of the report describes the
test plan that was developed, presents the data
obtained, and provides a statistical analysis of the
data, together with conclusions based on the results
and statistical analysis.
A. Selection of Collaborators
Since the preliminary evaluation indicated a
possible source of error due to loss of weight after
sample collection, this factor was also considered in
conducting the collaborative test. The results are
discussed subsequently in outlining various con-
siderations which may affect the precision of the
standard method.
If a collaborative test is to achieve the desired
objectives, it is necessary that the participants in the
collaborative test be representative of the large group
that will ultimately make use of the method being
tested. Since air pollution measurements are of
interest to many different groups, it was desirable to
include in the group of collaborators a variety of
-------
governmental agencies, universities, industrial labora-
tories, and others. The final selection included three
participants from federal laboratories, five from state
and local air pollution control agencies, one from
industry, two from universities, and one from a
research organization. A complete list of the partici-
pants and their affiliation is given elsewhere in this
report.
Each laboratory that agreed to participate was
asked to select a staff member with previous experi-
ence in measuring particulates by the high volume
sampler method. This was done to avoid errors that
otherwise would result from lack of experience, and
thus provide a more realistic appraisal of the capa-
bility of the method being tested. Each laboratory
also was engaged in making routine measurements
with this method or some variation of the same
method, and possessed the necessary equipment for
calibration of samplers, conditioning and weighing of
filters, and other work necessary to follow the
method as outlined.
B. Planning the Test Series
Ideally, a collaborative test should be con-
ducted by a group of participants, each working in his
own laboratory in his usual manner. For some
methods to measure air pollution, this can be accom-
plished by sending known samples to various labora-
tories for analysis. In this case, however, no method
was available to send a "standard atmosphere" which
could be used for test purposes, and, therefore, it was
necessary to make measurements by sampling a real
atmosphere. In order to assure uniformity, the partic-
ipants were brought together to one location to
sample the same atmosphere simultaneously. A
laboratory building used by the Air Pollution Control
Office in Cincinnati, Ohio, was selected where a large
area on the roof of a two-story wing of the building
provided adequate space. A clear space approximately
SOX 100ft was available without obstructions. An
exhaust fan outlet on the roof was equipped with a
duct which discharged the exhaust air down below
roof level along the side of the building so that no
disruption of normal atmospheric turbulence would
occur. This location was also convenient in that
laboratory facilities were available in the same
building to provide a work area for calibration and
assembly of equipment.
The building chosen was located in a neighbor-
hood which contained both residential and industrial
property. Several industrial plants in the area pro-
duced visible emissions, although none were located
close enough to the site so that a dominant effect
from a single source would be expected. Cincinnati is
also in an area where coal is used extensively as fuel,
and thus some coal smoke would be included in the
atmospheric particulates present. The nearest freeway
with heavy traffic was several hundred yards distant.
By sampling on top of a two-story wing of the build-
ing, localized effects from vehicle traffic in the
immediate neighborhood would be minimized. Thus,
all preliminary evidence indicated that this site was
typical of many urban locations that might be used
for air pollution monitoring, and would be expected
to give results representative of particulate levels over
a wide area of that portion of the city. Subsequent
experience in the collaborative test appeared to con-
firm this expectation, and, therefore, the area of the
test site was considered to be well suited to the objec-
tives of the test.
In order to minimize any effects which might
result from working in unfamiliar surroundings, each
participant was instructed to calibrate the sampler,
condition and weigh all filters, and do all other pre-
paratory and followup work in his own laboratory.
Work in Cincinnati was then to be limited to recheck-
ing the flow rate and to actual sample collection.
Thus, insofar as possible, the final results should indi-
cate what each participant routinely accomplishes in
his own laboratory. In the original schedule, time was
allowed for rechecking sampler calibration in
Cincinnati prior to sampling, to be sure that shipping
of equipment had not adversely affected the calibra-
tion. However, it was found that many laboratories
did not have access to a positive displacement meter
to use as a primary calibration standard and other
laboratories did not use the standard type of orifice
meter for calibration. Therefore, the entire
-------
calibration procedure specified in the method was
repeated by all participants after arrival in Cincinnati,
including both the calibration of the orifice meter
with a primary standard and the calibration of the
sampler with the orifice meter.
C. Site Evaluation
In order to be sure that the site chosen for the
collaborative test was suitable, a preliminary evalua-
tion of this particular site was performed prior to the
actual test. The primary objective of this work was to
confirm the fact that all samplers used in the test
would be sampling the same atmosphere. For this
purpose, five samplers were placed in representative
locations on top of the building and were operated
for three sampling periods of approximately 24 hr
each. While some variation occurred among samplers,
as expected, there was no systematic pattern which
would indicate a consistent difference in atmospheric
dust levels. The data are shown in Table I and indi-
cate that location within different sections of the
roof area was essentially immaterial in affecting the
particulate levels measured. Therefore, when
12 samplers were used simultaneously during the
actual test, it could be assumed that all participants
were measuring the same test atmosphere.
Another factor considered in evaluation of the
site was the location of the 12 samplers, within the
clear space available, when the collaborative test was
conducted. It was decided to use two rows of
6 samplers each, maintaining a 12-ft spacing both
ways between samplers. With this arrangement, no
sampler was placed closer to the edge of the roof than
12ft, thus avoiding turbulence and other flow dis-
turbances that might exist at the edge or corner of
the building. In the preliminary tests, a 12-ft spacing
was used with 2 of the 5 samplers, and the results
indicated that no interference resulted. In fact, a
superficial examination of airflow patterns around a
sampler indicated that any disturbance of airflow was
dissipated rapidly, and interference with an adjacent
sampler would not be expected unless a spacing of
3 ft or less was used. Therefore, all available evidence
indicated that the site and the proposed arrangement
of samplers were adequate for the purposes of the
test.
D. Conducting the Test Series
Actual sampling at the Cincinnati site was
accomplished in a single week in October 1970,
from Monday morning to Friday afternoon. Weather
conditions during the week were variable, covering
the normal range of conditions expected at that sea-
son of the year. Light rain fell during portions of the
second and third day. The morning period on Mon-
day was used by all participants to unpack equip-
ment, calibrate orifice meters and samplers, and set
up shelters. The first sampling period was started in
the afternoon on Monday when all participants were
ready.
All samplers were started simultaneously for
the actual sampling period and each participant
observed and recorded the initial flow rate. At the
end of each sampling period (approximately 24 hr),
each participant determined the final flow rate and
then all samplers were stopped simultaneously. The
filters were removed, new filters were installed for the
next sample period, and the entire procedure was
repeated. While the samplers were running, everyone
stayed away from the test site except for occasional
checks of equipment, to minimize errors due to dust
that might be raised by walking on the roof.
At the conclusion of the final test period on
Friday, final flow rate figures were obtained, samplers
were stopped, the filters were removed from the
samplers, and shelters and other equipment were pack-
ed for shipment. Each participant was asked to return
the filters to his own laboratory, condition them as
specified in the method on Monday and Tuesday (for a
total of 24 hr), and then obtain a final weight. This
weight was used for calculations and, in addition, sub-
sequent weighings were also made to obtain additional
data on weight loss. The question of weight loss over a
period of time is discussed in connection with the
evaluation of the precision of the method.
One participant lost a sample when the sampler
motor brushes failed. However, this person replaced
-------
the brushes and recalibrated the sampler prior to the
beginning of the next period, thus limiting the loss to
a single sample. This gave a total of 47 samples from
all participants out of a possible 48 (12 participants,
4 samples each). Data from these 47 samples were
then used for the statistical analysis which was per-
formed.
E. Summary of Statistical Analysis
The fundamental purpose of the statistical
design was to determine the existence and amount of
both random and systematic laboratory errors. The
experiment was designed so that, in addition to inter-
laboratory comparisons, some implications could be
made regarding the day-to-day variations within an
individual laboratory. No estimate is possible, how-
ever, for replication errors, since a true replicate can-
not be accomplished. Since it was impossible to pre-
pare and submit samples with known reference
values, no statements regarding the accuracy of the
method can be made.
In consideration of the logistics involved, a very
comprehensive test is not feasible;however, a modest
test design was possible. The approach was to have all
collaborators sample from the same real atmosphere
at the same site at the same time. There were
12 collaborators, each from a different laboratory,
who worked independently but simultaneously at the
same site collecting data for 4 consecutive days. Thus,
each day constituted a separate sample or material, of
unknown concentration, but common to all
participants. It should be made clear that there was
no intent to measure day-to-day variations for the
site. Neither was it desirable that all days be the same,
a condition not under experimental control. A single
result was collected by each analyst for each of the
4 days, yielding the data shown in Table II. Samples
were collected over a 24-hr interval; therefore, no
replication was possible for a particular analyst-
sampler combination. Although the data do not cover
a wide range, they are certainly satisfactory. Notice
that there are 2 days with means near 120 micro-
grams per cubic meter (jug/m3) and 2 days with
means significantly lower and near 80 jug/m3, a
fortuitous but agreeable circumstance.
TABLE II. COLLABORATIVE TESTING
DATA-HIGH-VOLUME METHOD
FOR PARTICULATES
Paniculate Concentration, Mg/m3
Laboratory Number
222
311
320
341
345
509
572t{**
575ttt
578t
600
7871
799
Day 1
138
25
28
26
27
28
28
08
26
25
25
131
Day 2
*
80
72
75
78
74
82
73
77
72
76
76
Day 3
87
82
81
83
87
86
84
72
83
80
83
86
Day 4
114
113
112
114
124
121
112
93
1 11
110
117
120
*Missmg data due lo equipment maltunction
T Recalculated results because of errors m units ot measurement
I Recalculated results because ot arithmetic errors
**Recalculated results because ot ncorrect data entry tor time of
sampling
ttRecalculated results because ot errors in calibration curves
The statistical analysis, presented in more detail
in Appendix B, is summarized in the following
paragraphs.
The standard deviation for each day fluctuated
systematically with the respective means, indicating
that it was advisable that the data be transformed to a
different scale. The appropriate tests were applied,
resulting in the use of the simple logarithmic trans-
formation. All subsequent analyses were made on
transformed data.
The missing sample for one laboratory required
either that all data for that laboratory be rejected or
that an appropriate substitution be made for the
missing value. The latter approach was followed, as
outlined in Appendix B.
Following standardized procedures for the
analysis of data of this type, Figure 1 was prepared.
This illustration is based on transformed data and
shows a separate line for each laboratory, plotting the
observed values against reference values equal to the
mean of all laboratories for each day. This
makes it possible to visualize the grouping of the
data and to identify the variation between
laboratories. One obvious advantage of this
presentation is that any outliers can be easily
identified; in this case, it is obvious that the
results of Laboratory 575 are significantly dif-
ferent. Other outlier tests also substantiated this
-------
2.20
2.15
1.80
1.85 1.90 1.95 2.00 2.05 2.10 2.15
Reference Value
FIGURE 1. GRAPHIC DATA SUMMARY FOR ALL
LABORATORIES, LOGARITHMIC DATA
TRANSFORMATION
conclusion, and the final statistical analysis was
made omitting Laboratory 575.
An analysis of variance of the transformed data
for 11 laboratories was made, and the model and
more detailed data are given in Appendix B. The ulti-
mate purpose for making the analysis of variance is to
derive components of variance which are shown in
Table III. The mean square from the deviation from
linear term is the component V(d) which contains the
unknown replication error plus the irreducible experi-
mental error of the method. Components V(u) and
V(b) express the variability of the means and the
TABLE HI. COMPONENTS OF VARIANCE
Component
V(u)
V(b)
V(d)
Value*
0.000087
0
0.000166
Percent
of Total
34
66
Significance
>99%
<50%
Relative Deviation,
Percent
Estimate
-i i
0
30
Confidencet
Interval
06 to 4 3
0 to 1 4
2 3 to 4 4
*Based on transformed data
f95 percent confidence interval
slopes, respectively, from one laboratory to another.
These components were calculated according to
recommended practice.^) A negative value for V(b)
was obtained and therefore replaced with a value of
zero, since negative variances are meaningless. V(u) is
highly significant while V(b), of course, is not signifi-
cant. The confidence intervals (95 percent level of
significance) for each component are shown in
Table III. Since the component V(b) is not significant
at the 95 percent level of significance, the lower con-
fidence limit is naturally zero.
The quantities V(d), V(u), and V(b) are the
basic elements from which the precision of the results
can be estimated for any set of conditions.'^) The
reproducibility and the repeatability(8) may thus be
determined.
Two-thirds of the variance is accounted for by
the component V(d). The relative standard deviation
for a single material analyzed repeatedly in the same
laboratory by the same analyst is 3.0 percent, the
repeatability of the method. Any two such values
should be considered suspect (95 percent confidence
level) if they differ by more than 4.3 percent. Con-
versely, two materials analyzed in the same labora-
tory may not be considered significantly different if
the results differ by less than 4.3 percent.
One-third of the variance is accounted for by
the component V(u). The relative standard deviation
for a single material analyzed in different laboratories
is 3.7 percent, the reproducibility of the method.
Any two such values should be considered suspect
(95 percent confidence level) if they differ by more
than 5.3 percent. Conversely, two materials, one
analyzed in each of two laboratories, may not be con-
sidered to be significantly different if the results
differ by less than 5.3 percent.
Some additional information was obtained dur-
ing the collaborative test relating to errors from
individual steps within the method. Each collaborator
was asked to submit the results of a series of weights
of the filters that were supplied. A senes of three
weights at 24-hr intervals was requested for each of
-------
ten filters provided. These data were obtained by the
collaborator prior to coming to the test site. Follow-
ing the formal collaborative test, each analyst
returned to his own laboratory and made a series of
four weights, at 24-hr intervals, of each of the ten
filters, not all of which were used in the test. The
weight after the first 24-hr equilibration was used, as
specified by the method, to calculate the total sus-
pended particulate concentrations which were
reported.
The supplementary weight data for the succeed-
ing 3 days are useful for two purposes: first, to pro-
vide an estimate of a limit of detectability of the
method; and, second, to give an indication of the
stability of the suspended particulate matter.
These data can be divided into three subsets:
• Unexposed filters prior to test period
• Unexposed filters after test period
• Exposed filters used in the test.
These data indicate the combined random effects of
equilibration, weighing, and decomposition or volatil-
ization of particulate matter. The standard deviation
of the weights for each of the cases is 0.7, 0.9, and
1.7 mg, respectively. The number of sets of replicates
for each case is 100, 56, and 43, respectively.
Bartlett's test(ll) for homogeneity of variances leads
to the acceptance of the null hypothesis that there is
no difference between the variances in the weights of
the unexposed filters before and after the test period.
The variance in the weights of the used filters was
shown to be significantly greater than that of the
unexposed filters, indicating that there are significant
changes in the weights of exposed filters over the
96-hr period.
The data, which are too voluminous to report
here, show weight losses up to 10 mg, or about 5 per-
cent, for an individual used filter over the 4-day
period of equilibration. This loss, though less than
that indicated in the preliminary work, is significant
in terms of the precision data reported above. This
illustrates the importance of careful control and
uniformity of equilibration in order to achieve satis-
factory reproducibility. The presence of volatile
materials can cause unpredictable errors.
In the absence of significant amounts of volatile
material, an estimate of the lower detection limit can
be made. The standard deviation of the difference in
the mean weights of the unexposed filters before and
after the test period could be directly determined for
a set of 56 observations. The result obtained was
1.5 mg, which is almost exactly the same value
obtained by the less direct method of doubling the
standard deviation obtained for all unexposed filter
weights. Based upon this result, the 95 percent con-
fidence limit for these differences is 3.0 mg. There-
fore, a difference in weight of 3 mg or less cannot be
considered to be significantly different from zero.
The minimum detectable amount of particulate
matter is thus 3 mg, and, for a 24-hr sample at
1.5 cubic meters per minute (53 cfm), this would
correspond to a minimum detectable concentration
of 1.4 ;Ug/m3. A concentration this low will rarely, if
ever, be observed in the atmosphere.
F. Calibration Errors
It was possible to investigate the random error
associated with the calibration of the orifice units
with the primary standard positive displacement
meter. The experimental design was such that a
participant was randomly assigned one of six identical
orifice calibration units and one of five essentially
identical positive displacement meters. The partici-
pant then calibrated the orifice unit according to the
tentative method. These independent calibration
curves have been superimposed in Figure 2 which is a
log-log plot of the flow rate through the orifice versus
the pressure differential across the orifice. With the
exception of one laboratory (787) and one atypical
point for another laboratory (311), linear results were
obtained. Another laboratory (572) switched primary
standard meters in the middle of the calibration pro-
cedure; however, there were no noticeable irregu-
larities in its results. A least squares regression
-------
15
12
10
S 9
I 8
! 7
<0
o
f 6
oT
£
? 5
a
o
1.5
Least Squares Line
95% Confidence Limits
0.6 0.7 0.8 0.9 1.0 1.2
1.5
2.0 2.3
Flow Rate Through Positive Displacement
Meter, m^/min.
FIGURE 2. ORIFICE METER CALIBRATION DATA
analysis was made (omitting Laboratory 787 and the
extraneous point of Laboratory 311) and the line of
best fit is shown in Figure 2. The magnitude of the
derived constants for the regression line is of second-
ary importance. The important consideration is the
standard error of estimate which is 2.1 percent. This
error combines the errors from variation among
orifice units, variation among positive displacement
meters, and experimental measurement in the calibra-
tion procedure. The 95 percent confidence interval
for the error associated with calibration is ±4.1 per-
cent. All the orifice units used in the test were new
and in good condition.
The manufacture^12) furnishes an "Average
Calibration Curve" with each orifice. The orifice units
are quoted as being accurate to ±1.0 percent. These
calibration data are almost identical with the line of
best fit in Figure 2. The temperature and pressure
conditions are approximately the same for each.
It should not be concluded that the calibration
of the orifice should be omitted on the basis of these
estimates; rather, that the main purpose of periodic
orifice calibration is to provide assurance that the
orifice has not been damaged or otherwise changed in
accuracy. A difference of more than 4 percent from
the manufacturer's curve probably means that the
orifice has been damaged.
LIST OF REFERENCES
1. Lynam, D. R., Pierce, J. O., and Cholak, J.,
"Calibration of the High-Volume Air Sampler,"
American Industrial Hygiene Association
Journal, No. 30, pp 83-88 (January-February
1969).
2. Unpublished data, Southwest Research Insti-
tute, Houston, Texas.
3. Unpublished data, Texas Air Control Board,
Austin, Texas.
4. Unpublished data, Air Pollution Control Pro-
gram, City of Houston Health Department,
Houston, Texas.
5. Youden, W. J., "The Collaborative Test,"
Journal of the AOAC, Vol 46, No. 1, pp 55-62
(1963).
6. Handbook of the AOAC, Second Edition,
October 1, 1966.
7. ASTM Manual for Conducting an Interlabora-
tory Study of a Test Method, ASTM STP
No. 335, Am. Soc. Testing & Mats. (1963).
8. Recommended Practice for Developing Preci-
sion Data on ASTM Methods for Analysis and
Testing of Industrial Chemicals, ASTM Designa-
tion: El80-67. 1968 Book of ASTM Standards,
Part 30.
9. ASTM STP No. 335, op. cit., p 37.
10. Ibid, pp 37-40.
11. Dixon, Wilfred J., and Massey, Frank!., Jr.,
Introduction to Statistical Analysis, Chap-
ter 10, pp 179-180, McGraw-Hill Book Com-
pany, Inc., New York (1957).
12. General Metals Works, Cleves, Ohio.
10
-------
RULES AND REGULATIONS
8191
APPENDIX A
REFERENCE METHOD FOR THE DETERMINATION OF
SUSPENDED PARTICULATES IN THE ATMOSPHERE
(HIGH VOLUME METHOD)
Reproduced from Appendix B, "National Primary and Secondary
Ambient Air Standards," Federal Register, Vol 36, No. 84, Part II,
Friday, April 30, 1971.
APPENDIX B—REFERENCE METHOD FOR THE
DETERMINATION OF SUSPENDED PAHTICULATES
IN THE ATMOSPHERE (HIGH VOLUME
METHOD)
1. Principle and Applicability.
1.1 Air Is drawn into a covered housing
and through a niter by means of a high-flow-
rate blower at a flow rate (1.13 to 1.70 m.5/
min.; 40 to 60 ft.'/min.) that allows sus-
pended particles having diameters of less
than 100 ion. (Stokes equivalent diameter)
to pass to the filter surface. (I) Particles
within the size range of 100 to O.ljim. diame-
ter are ordinarily collected on glass fiber fil-
ters. The mass concentration of suspended
particulates in the ambient air (/tg./m.s) is
computed by measuring the mass of collected
particulates and the volume of air sampled.
1 2 This method Is applicable to measure-
ment of the mass concentration of suspended
particulates in ambient air. The size of the.
sample collected is usually adequate for
other analyses.
2. Range and Sensitivity,
2 1 When the sampler is operated at an
average flow rate of 1.70 m.Vmin. (60 ft.V
mm.) for 24 hours, an adequate sample will
be obtained even in an atmosphere having
concentrations of suspended particulates as
low as 1 /ig./mA If particulate levels are
unusually high, a satisfactory sample may be
obtained in 6 to 8 hours or less. For deter-
mination of average concentrations ol sus-
pended particulates in ambient air, a stand-
ard sampling period of 24 hours is
recommended.
2.3 Weights are determined to the near-
est milligram, airflow rates are determined to
the nearest 0.03 m.Vmin. (1.0 ft.'/mln.),
times are determined to the nearest 2
minutes, and mass concentrations are re-
ported to the nearest mlcrogram per cubic
meter. •
3.~ Interferences.
3.1 Partdculate matter that Is oily, such
as photochemloal «nnng or wood smoke, may
block the filter »nd cause a rapid drop in
airflow at a nonunlform rate. Dense fog or
high humidity can cause the fitter to become
too wet and severely reduce the airflow
through the filter.
3.2 Glass-fiber filters are comparatively
Insensitive to changes in relative humidity,
but collected particulates can be hygro-
scopic. (2)
4. Precision, Accuracy, and Stability.
4.1 Based upon collaborative testing, the
relative standard deviation (coefficient of
variation) for single analyst variation (re-
peatability of the method) is 3.0 percent.
The corresponding value for multilaboratory
variation (reproduclbillty of the method) is
3.7 percent. (3)
4.2 The accuracy with which the sampler
measures the true average concentration
depends upon the constancy of the airflow
rate through the sampler. The airflow rate Is
affected by the concentration and the nature
of the dust in the atmosphere. Under these
conditions the error in the measured aver-
age concentration may be in excess of ±50
percent of the true average concentration, de-
pending on the amount of reduction of air-
flow rate and on the variation of the mass
concentration of dust with time during the
24-hour sampling period. (4)
5. Apparatus.
5.1 Sampling.
5.1.1 Sampler. The sampler consists of
three units: (1) the faceplate and gasket,
(2) the filter adapter assembly, and (3) the
motor unit. Figure Bl shows an exploded
view of these parts, their relationship to each
other, and how they are assembled. The
sampler must be capable of passing environ-
mental air through a 406.5 cm.2 (63 In.8)
portion of a clean 20.3 by 25.4 cm. (8- by
10-in.) glass-fiber filter at a rate of at least
1.70 m.Vmin. (60 ft.Vmin.). The motor must
be capable of continuous operation for 24-
hour periods with input voltages ranging
from 110 to 120 volts, 50-60 cycles alternat-
ing current and must have third-wire safety
ground. The housing lor the motor unit
may be of any convenient construction so
long -as the unit remains airtight and leak-
free. The life of the sampler motor can be
extended by lowering the voltage by about
10 percent with a small "buck or boost"
transformer between the sampler and power
outlet.
5.1 2 Sampler Shelter. It is important
that the sampler be properly installed In a
suitable shelter. The shelter is subjected to
extremes of temperature, humidity, and all
types of air pollutants. For these reasons
the materials of the shelter must be chosen
carefully. Properly painted exterior plywood
or heavy gauge aluminum serve well. The
sampler must be mounted vertically in the
shelter so that the glass-fiber filter is paral-
lel with the ground. The shelter nrnst be
provided with a roof so that the filter is pro-
tected from precipitation and debris. The
internal arrangement and configuration of
a suitable shelter with a gable roof are shown
in Figure B2. The clearance area between the
main housing and the roof at Its closest
point should be 580 5± 193.5 cm.! (90±30
In =). The main housing should be rectangu-
lar, with dimensions of about 29 by 36 cm.
(ll'/2 by 14 in.).
513 Rotameter. Marked in arbitrary
units, frequently 0 to 70, and capable of
being calibrated. Other devices of at least
comparable accuracy may be used.
FEDERAL REGISTER, VOL. 36, NO. 84—FRIDAY, APRIL 30, 1971
A-l
-------
8192
RULES AND REGULATIONS
6.1.4 Orifice Calibration Unit. Consisting
of a metal tube 7.6'cm. (3 in.) ID and 15.9
cm. (6% In.) long with a static pressure tap
5.1 cm. (2 in.) from one end. See Figure
B3. The tube end nearest the pressure tap is
flanged to about 10.8 cm. (4% in.) OD with
a male thread of the same size as the Inlet
end of the high-volume air sampler. A single
metal plate 9.2 cm. (3% in.) in diameter and
0.24 cm. (%2 in.) thick with a central orifice
2.9 cm. (1% in.) in diameter is held in place
at the air Inlet end with a female threaded
ring. The other end of the tube is flanged to
hold a loose female threaded coupling, which
screws onto the inlet of the sampler. An 18-
hole metal plate, an Integral part of the unit,
is positioned between the orifice and sampler
to simulate the resistance of a clean glass-
fiber filter. An orifice calibration unit is
shown in Figure B3.
5.1.5 Differential Manometer. Capable of
measuring to at least 40 cm. (16 in.) of
water.
5.1.6 Positive Displacement Meter. Cali-
brated in cubic meters or cubic feet, to be
used as a primary standard.
5.1.7 Barometer. Capable of measuring at-
mospheric pressure to the nearest mm.
5.2 Analysis.
5.2.1 Filter Conditioning Environment.
Balance room or desiccator maintained at
15° to 35°C. and less than 50 percent relative
humidity.
5.2.2 Analytical Balance. Equipped with
a weighing chamber designed to handle un-'
folded 20.3 by 25.4 cm. (8- by 10-in.) filters
and having a sensitivity of 0.1 rog.
5.2.3 Light Source. Frequently a table of
the type used to view X-ray films.
5.2.4 Numbering Device. Capable of print-
ing identification numbers on the filters.
6. Reagents.
6.1 Filter Media. Glass-fiber filters having
a collection efficiency of at least 99 percent
for particles of 0.3 um. diameter, as measured
by the DOP test, are suitable for the quanti-
tative measurement of concentrations of sus-
pended particulates, (5) although some other
medium, such as paper, may be desirable for
some analyses. If a more detailed analysis is
contemplated, care must be exercised to use
filters that contain low background concen-
trations of the pollutant being investigated.
Careful quality control is required to deter-
mine background values of these pollutants.
7. Procedure.
11 Sampling.
7.1 1 Filter Preparation. Expose each filter
to the light source and inspect for pinholes,
particles, or other imperfections. Filters with
visible imperfections should not be used. A
small brush is useful for removing particles.
Equilibrate the filters in the filter condition-
ing environment for 24 hours. Weigh the
niters to the nearest milligram; record tare
weight and filter identification number. Do
not bend or fold the filter before collection
of the sample.
712 Sample Collection. Open the shelter,
loosen the wing nuts, and remove the face-
plate from the filter holder. Install a num-
bered, preweighed, glass-fiber filter in posi-
tion (rough side up), replace the faceplate
without disturbing the filter, and fasten
securely. Undertightening will allow air leak-
age, overtlghtening will damage the sponge-
rubber faceplate gasket. A very light applica-
tion of talcum powder may be used on the
sponge-rubber faceplate gasket to prevent
the filter from sticking. During inclement
weather the sampler may be removed to a
protected area for filter change. Close the
roof of the shelter, run the sampler for about
5 minutes, connect the rotameter to the
nipple on the back of the sampler, and read
the rotameter ball with rotameter in a verti-
cal position. Estimate to the nearest whole
number. If the ball is fluctuating rapidly,
tip the rotameter and slowly straighten it
until the ball gives a constant reading. Dis-
connect the rotameter from the nipple; re-
cord the initial rotameter reading and the
starting time and date on the filter folder.
(The rotameter should never be connected
to the sampler except when the flow is being
measured.) Sample for 24 hours from mid-
night to midnight and take a final rotameter
reading. Record the final rotameter reading
and ending time and date on the filter folder.
Remove the faceplate as described above and
carefully remove the filter from the holder,
touching only thevouter edges. Fold the filter
lengthwise so that only surfaces with col-
lected particulates are In contact, and place
In a manlla folder. Record on the folder the
filter number, location, and any other factors,
such as meteorological conditions or razing
of nearby buildings, that might affect the
results. If the sample is defective, void it at
this time. In order to obtain a valid sample,
the high-volume sampler must be operated
with the same rotameter and tubing that
were used during its calibration.
7.2 Analysis. Equilibrate the exposed fil-
ters for 24 hours in the filter conditioning
environment, then reweigh. After they are
weighed, the filters may be saved for detailed
chemical analysis.
7.3 Maintenance.
7.3.1 Sampler Motor. Replace brushes
before they are worn to the point where
motor damage can occur.
7.3.2 Faceplate Gasket. Replace when the
margins of samples are no longer sharp. The
gasket may be sealed to the faceplate with
rubber cement or double-sided adhesive tape.
7.3 3 Rotameter. Clean as required, using
alcohol.
8. Calibration.
8.1 Purpose. Since only a small portion
of the total air sampled passes through the
rotameter during measurement, the rotam-
eter must be calibrated against actual air-
flow with the orifice calibration unit. Before
the orifice calibration unit can be used to
calibrate the rotameter, the orifice calibra-
tion unit Itself must be calibrated against
the positive displacement primary standard.
8 1.1 Orifice Calibration Unit. Attach the
orifice calibration unit to the intake end
of the positive displacement primary stand-
ard and attach a high-volume motor blower
unit to the exhaust end of the primary
standard. Connect one end of a differential
manometer to the differential pressure tap
of the orifice calibration unit and leave the
other end open to the atmosphere. Operate
the high-volume motor blower unit so that
a series of different, but constant, airflows
(usually six) are obtained for definite time
periods. Record the reading on the differen-
tial manometer at each airflow. The different
constant airflows are obtained by placing a
series of loadplates, one at a time, between
the calibration unit and the primary stand-
ard. Placing the orifice before the inlet re-
duces the pressure at the inlet of the primary
standard below atmospheric; therefore, a
correction must be made for the increase in
volume caused by this decreased inlet pres-
sure. Attach one end of a second differential
manameter to an inlet pressure tap of the
primary standard and leave the other open
to the atmosphere. During each of the con-
stant airflow measurements made above,
measure the true inlet pressure of the
primary standard with this second differen-
tial manometer. Measure atmospheric pres-
sure and temperature. Correct the measured
air volume to true air volume as directed in
9.1.1, then obtain true airflow rate, Q, as
directed in 9.1.3. Plot the differential manom-
eter readings of the orifice unit versus Q.
8.1.2 High-Volume Sampler. Assemble a
high-volume sampler with a clean filter in
place and run for at least 5 minutes. Attach
a rotameter, read the ball, adjust so that the
ball reads 65, and seal the adjusting mech-
anism so that it cannot be changed easily.
Shut off motor, remove the filter, and attach
the orifice calibration unlit in its place. Op-
erate the high-volume sampler at a series of
different, but constant, airflows (usually six) .
Record the reading of the differential ma-
nometer on the orifice calibration unit, and
record the readings of the rotameter at each
flow. Measure atmospheric pressure and tem-
perature. Convert the differential manometer
reading to m.'/min., Q, then plot rotameter
reading versus Q.
8.1.3 Correction for Differences in Pressure
or Temperature. See Addendum B.
9. Calculations.
9.1 Calibration of Orifice.
9.1.1 True Air Volume. Calculate the air
volume measured by the positive displace-
ment primary standard.
(P.-Pm)
Va= - (VM)
P.
V»=True air volume at atmospheric pres-
— sure, m.3
P. = Barometric pressure, mm. Hg.
Pn = Pressure drop ait inlet of primary
standard, mm. Hg.
VM=Volume measured by primary stand-
ard, m.s
9.1.2 Conversion Factors.
Inches Hg.X25.4=mm. Hg.
Inohee water X 73 .48 X 10-" = inches Hg.
Cubic feet air x 0.0284 = cubic meters air.
9'.1.3 True Airflow Rate.
V.
Q=—
T
Q=Plow rate, m.Vmin.
T=Time of flow, min.
9.2 Sample Volume.
9.2.1 Volume Conversion. Convert the ini-
tial and final rotameter readings to true
airflow rate, Q, using calibration curve of
81.2.
9.2.2 Calculate volume of air sampled
QiQi
V=- - XT
2
V — Air volume sampled, m.3
Qi = Initial airflow rate, m.Vmin.
Qt = Final airflow rate, m.Vmin.
T= Sampling time, min.
9.3 Calculate mass concentration of sus-
pended particulates
SP =
i-Wi) xlO«
S.P.^Mass concentration of suspended
particulates, /tg/m.s
Wi = Initial weight of filter, g.
Wt = Final weight of filter, g.
V = Air volume sampled, m.3
10G — Conversion of g. to fig.
10. References.
(1) Robson, C. D., and Foster, K. E.,
"Evaluation of Air Particulate Sam-
pling Equipment", Am. Ind. Hyg.
Assoc. J. 24, 404 (1962).
(2) Tierney, G. P., and Conner, W. D.,
"Hygroscopic Effects on Weight Deter-
minations of Particulates Collected on
Glass-Fiber Filters", Am. Ind. Hyg.
Assoc. J. 28, 363 (1967).
(3) Unpublished data based on a collabora-
tive test involving 12 participants,
conducted under the direction of the
Methods Standardization Services Sec-
tion of the National Air Pollution Con-
trol Administration, October, 1970.
(4) 'Harrison, W. K., Nader, J. S., and Fug-
man, P. S., "Constant Flow Regulators
for High-Volume Air Sampler", Am.
Ind. Hyg. Assoc. J. 21, 114-120 (1960).
FEDERAL REGISTER, VOL. 36, NO. 84—FRIDAY, APRIL 30, 1971
A-2
-------
RULES AND REGULATIONS
8193
(5) Pate, J. B., and Tabor, E. C., "Analytical
Aspects of the Use of Glass-Fiber Fil-
ters for the Collection and Analysis of
Atmospheric Partlculate Matter", Am.
Ind. Hyg. Assoc. J. 23. 144-150 (1962).
ADDENDA
A. Alternative Equipment.
A modification of the high-volume sampler
Incorporating a method for recording the
actual airflow over the entire sampling pe-
riod has been described, and Is acceptable
for measuring the concentration of sus-
pended partlculates (Hendefson, J. S., Eighth
Conference on Methods In Air Pollution and
Industrial Hygiene Studies, 1967, Oakland,
Calif.). This modification consists of an ex-
haust orifice meter assembly connected
through a transducer to a system for con-
tinuously recording airflow on a circular
chart. The volume of air sampled is cal-
culated by the following equation:
V = QXT.
Q = Average sampling rate, m.Ymin.
T = Sampling time, minutes.
The average sampling rate, Q, is determined
from the recorder chart by estimation If the
flow rate does not vary more than 0.11 m.V
min. (4 ft.Vmin ) during the sampling pe-
riod. If the flow rate does vary more than
0.11 m.3 (4 ft.Vmin.) during the sampling
period, read the flow rate from the chart
at 2-hour Intervals and take the average.
B. Pressure and Temperature Corrections.
If the pressure or temperature during
high-volume sampler calibration is substan-
tially different from the pressure or tempera-
ture during orifice calibration, a correction
of the flow rate, Q, may be required. If the
pressures differ by no more than 15 percent
and the temperatures differ by no more than
100 percent (°C), the error In the uu-
corrected flow-rate will be no more than 15
percent. If necessary, obtain the corrected
flow rate as directed below. This correction
applies only to orifice meters having a con-
stant orifice coefficient. The coefficient for
the calibrating orifice described in 5.1.4 has
been shown experimentally to be constant
over the normal operating range of the high-
volume sampler (0.6 to 2.2 m.Vmln.; 20 to 78
ft.Vmin.). Calculate corrected flow rate:
Q2=Corrected flow rate, m.'/min.
Q!=FIOW rate during high-volume sampler
calibration (Section 8.1.2), m.Vmin.
T!=Absolute temperature during orlnce
unit calibration (Section 8.1.1), 'K
or °R. ^
P!=Barometric pressure during orifice unit
calibration (Section 8.1.1), mm. Hg.
Ta=Absolute temperature during high-
volume sampler calibration (Section
8.1.2), °Kor 'R.
Ps = Barometric pressure during high-vol-
ume sampler calibration (Section
8.12), mm. Hg.
ADAPTER
MOUNTING MOTOR
PLATE CASKET
Figure B1 Exploded view of typical high-volume air samplei parts,
No. 84—Ft. II-
FEDERAL REGISTER, VOL. 36, NO. 84—FRIDAY, AFKIL 30, 1971
A-3
-------
8194
RULES AND REGULATIONS
Figure B2. Assembled sampler and shelter.
ORIFICE
RESISTANCE PLATES
Figure B3. Orifice calibration unit.
FEDERAL REGISTER, VOL. 36, NO. 84—FRIDAY, APRIL 30, 1971
A-4
-------
APPENDIX B
STATISTICAL METHODS
1. Data Transformation
The total suspended particulate measurement
has often been assumed to be log-normally distri-
buted. This is generally advisable whenever the stan-
dard deviation tends to fluctuate systematically with
its respective mean. The simple logarithmic trans-
formation is widely applicable but not necessarily the
only one to use.
Bartlett's testO)* for homogeneity of variance
was applied to both transformed and raw data. The
null hypothesis that the variances for each day are
homogeneous was rejected (95 percent level of signifi-
cance) for the full set of raw data and accepted for
the full set of transformed data. Therefore, in per-
forming the analysis of variance and all subsequent
analyses, the logarithms (base 10) of the total
suspended particulate measurements were used, since
they can be assumed to be normally distributed. The
transformed data are shown in Table B-I.
2. Missing Data
Missing data in the analysis of variance always
pose a problem, but, fortunately, there was only one
data gap and a simple approach was possible. The
TABLE B-I. SUMMARY OF DATA FOR ALL
LABORATORIES, WITH LOGARITHMIC
TRANSFORMATION
Laboratory
Code Number
222
311
320
341
345
509
572
575
578
600
787
799
Mean
Maximum
Minimum
Standard Deviation
Day 1
2 1399
20969
2 1072
2 1004
2 1038
2 1072
2 1072
20334
2 1004
20969
20969
2 1173
21006
21399
20334
00243
Day 2
9138*
9031
8573
8751
8921
8692
9138
8633
8865
8573
8808
1 8808
1 8828
1 9138
1 8573
00199
Day3
1 9395
1 9138
19085
1 9191
1 9395
1 9345
1 9243
8573
9191
9031
9191
9345
1 9177
1 9395
1 8573
00224
Day 4
20569
20531
20492
20569
20934
20828
20492
1 9685
20453
20414
20682
20792
20537
20934
1 9685
00314
Mean
20125
1 9917
1 9806
1 9879
20072
1 9984
1 9986
1 9306
1 9878
19747
1 9912
20029
1 9887
20125
1 9306
00212
Standard
Deviation
0 1053
00979
0 1171
0 1078
0 1074
01151
00950
00854
01016
01130
01072
0 1133
0 1171
00854
*Data value substituted for missing data
method(2) consists of inserting an estimate of the
missing observation, so chosen to minimize the
residual variance. The inserted value makes no contri-
bution to the residual sum of squares, and it is un-
likely to have any serious effect on the conclusions.
Application of the method yielded an estimate equiv-
alent to approximately 82 /ug/m3, but it should be
emphasized that this value does not represent an
observation. It allows an approximation of the results
obtained by a less powerful nonorthogonal analysis of
the incomplete data.
3. Linear Model Analysis
This experiment is an illustrative example of
the nonavailability of known reference values. There
is no knowledge of the true values for each of the
materials (days). The alternate approach^ >4) is to use
the mean of all laboratories for each day to replace
the unknown reference values. These reference values
may be seen in Table B-I. The assumption is made
that systematic differences exist between sets of mea-
surements made by the same observer at different
times, and that these systematic differences are linear
functions of the magnitude of the measurements.
Hence, the scheme is called "the linear model." The
linear model leads to a simple design, but requires a
special method of statistical analysis, geared to the
practical objectives of a collaborative test.
For each day, we may plot the measured values
versus the reference value substitutes. The measured
values should be a linear function of the reference
values (all data normally distributed), and the points
corresponding to each line may be represented by
three parameters: a mean; a slope; and a quantity
related to the deviation from linearity, the standard
error of estimate. These parameters are determined
by a least squares regression analysis. These results for
the transformed data are shown in Table B-II which
*Superscnpt numbers in parentheses refer to the List of References at the end of this Appendix.
B-l
-------
TABLE B-II. ESTIMATES OF THE PARAMETERS
OF THE STRAIGHT LINES CORRESPONDING
TO THE VARIOUS LABORATORIES, LOGA-
RITHMIC DATA TRANSFORMATION,
ALL LABORATORIES
Laboratory
Code Number
222
311
320
341
345
509
572
575
578
600
787
799
Mean
Maximum
Minimum
Mean
2.0125
1.9917
1 .9806
1.9879
2.0072
1.9984
1 .9986
1.9306
1.9878
1.9747
1.9912
2.0029
1 .9887
2.0125
1.9306
Slope
0.9929
0.9294
1.1150
1.0272
1.0122
1.0869
0.8998
0.7966
0.9678
1.0762
1.0193
1.0768
1.0000
1.1150
0.7966
Standard Error of
Estimate
0.0189
0.0110
0.0069
0.0047
0.0197
0.0191
0.0131
0.0218
0.0049
0.0049
0.0098
0.0102
0.0141*
0.0218
0.0047
•Pooled Estimate
was used to prepare Figure 1 found in Section III-E
of this report. This figure shows the data for Labora-
tory 575 to be notably different from the other
laboratories.
Control limits for the means and for the slopes
may be computed, based upon the pooled estimate of
standard error. Control charts, in standard deviation
units, for these parameters are shown in Figures B-l
and B-2. No control limits are possible for the
CONTROL CHART FOR MEANS IN SIGMA UNITS
(WITH LOG TRANSFORMATION)
STD. ERROR BASED ON DEVIATION FROM LINEARITY 0.007063
8S2 311 3£0 341 3^5 509 572 575 578 600 767 799
LAB MS. 575
-------
OF SL0PES fY> VS MEWJS (X)
CWITH LOG TRANSF3HMATI0N>
C1KHELATION C3EF.
SLOPE
INTERCLPT
0.50050
2.14690
-3.26951
- Laboratory 575
FIGURE B-4. COMPUTER PLOT OF SLOPES VERSUS
MEANS, ALL LABORATORIES
Although the individual data for each remaining
laboratory are the same as shown in Table B-I, there
are important differences in the summary statistics.
New reference values have been established, and the
standard deviations are more uniform.
TABLE B-JII. SUMMARY OF DATA, OMITTING
LABORATORY 575, WITH LOGARITHMIC
TRANSFORMATION
Statistic
Mean
Maximum
Minimum
Standard Deviation
Day 1
2 1067
2 1399
20969
00126
Day 2
1 8845
1 9138
1 8573
00199
Day 3
1 9232
1 9395
1 9031
00124
Day 4
20614
20934
20414
00171
Mean
1 9940
20125
1 9747
00113
Standard
0 1171
00950
Using the new reference values and the remain-
ing 11 laboratories, a final data analysis using the
linear model was made. The results of the regression
analysis are shown in Table B-IV, which is analogous
to Table B-II, but only the individual means are the
same. The slopes and the standard errors of estimate
are more uniform.
Control limits for the means and for the slopes
may again be computed. The control chart for means
is not shown because it is practically the same as
Figure B-l with the point for Laboratory 575
omitted. The standard error based on deviation from
TABLE B-IV. ESTIMATES OF THE PARAMETERS
OF THE STRAIGHT LINES CORRESPONDING
TO THE VARIOUS LABORATORIES,
OMITTING LABORATORY 575,
LOGARITHMIC DATA
TRANSFORMATION
Laboratory
Code Number
222
311
320
341
345
509
572
578
600
787
799
Mean
Maximum
Minimum
•Pooled Estimate
Mean
20125
99 1 7
9806
9879
0072
9984
9980
9878
9747
99 1 2
20029
1 9940
20125
1 9747
Slope
09729
09113
1 0948
1 0089
09955
1 0691
08817
0 9496
1 0566
1 0013
1 0582
1 0000
1 0948
08817
Standard Error ol
Estimate
00205
00124
00063
00028
00181
00170
00149
00064
00047
0 0085
00081
00129*
00205
00028
linearity is lower and is 0.006435, which, coupled
with a slightly higher average mean, results in shifting
each point in Figure B-l downward about 0.4 unit.
The control chart for slopes is also not shown because
it is practically the same as Figure B-2 with the point
for Laboratory 575 omitted. The standard error
based on deviation from linearity is lower and is
0.06954, and, of course, the range of slopes is sub-
stantially reduced. The net effect is to move each
point in Figure B-2 downward approximately
0.2 unit.
The control chart i'or standard errors is similar
in appearance to Figure B-3 with the point for
Laboratory 575 omitted, and, therefore, is not
shown. Comparing the standard errors of estimate in
Tables B-1I and B-1V, it can be seen that the pooled
estimate of standard errors is reduced, but that the
range is approximately the same after elimination of
Laboratory 575.
The plot of slopes versus means is shown in
Figure B-5 which is analogous to Figure B-4, but
indicates that the deletion of the single atypical point
due to Laboratory 575 results in a substantial
reduction in the correlation between the slopes and
the means. This is apparent both visually as well
as from the correlation coefficients shown in Fig-
ures B-4 and B-5.
B-3
-------
CWITH L0G TRANSF3RMATI3N)
C0RRELATI0N CHEF* 0.28731
SLOPE -I.71JH2
INTERCEPT 4. 41729
0.8817- *
1.975 1.981 1.987 1.994 2.000 2.OOF, J.U1J
FIGURE B-5. COMPUTER PLOT OF SLOPES VERSUS
MEANS, OMITTING LABORATORY 575
No further elimination of outliers is required,
and the following conclusions can now be drawn:
• There is no significant variation in the
means and slopes among the eleven
laboratory lines which cannot be
entirely explained by the scatter due to
the deviation from linearity.
• There is no correlation between the
slopes and the means of the individual
lines.
M,
= main effect of laboratory /
= main effect of material (day) /'
= interaction effect between labora-
tory / and material / and includes
unknown replication error
= 1, 2, 3 . . . p = number of
laboratories
1, 2, 3 .
materials
. . q = number of
A more meaningful analysis is obtained by parti-
tioning the interaction term as follows:
where the first term is the linear term in which b,
is the slope determined by the /th laboratory, b
represents the slope of the average total suspended
particulate (TSP) observed-versus-mean line (aver-
aged over all laboratories), Cj represents the true
TSP value for the y'th material, and c represents
the true mean TSP of all materials. The second
term, d,j, is the deviation from linear term. The
linear term indicates the difference in slope of the
line for a particular laboratory and the average
slope for all laboratories, and the nonlinear term
expresses the departures from linearity for this
individual line.
4. Analysis of Variance
The general model for the analysis of the
results, classified according to two criteria, labora-
tories and materials (simulated by different days),
where
= an individual measurement
A = overall average
Starting with the ordinary two-factor analysis
of variance, the deviation from linear component
of the interaction sum of squares can be com-
puted.^) The sum of squares for the linear
component of interaction is then obtained by
TABLE B-V. ANALYSIS OF VARIANCE OMITTING
LABORATORY 575, WITH LOGARITHMIC
TRANSFORMATION
Source of Variation
Laboratories
Days
Labs X Days
Linear
Concurrence
Nonconcurrence
Deviation trom Linear
Sum of
Squares
0005126
0 376777
0 004874
0001562
000012')
0001433
0003313
Degree's
ol hreeclom
10
30
11)
1
•I
20
Mean Squaie
I) 0005 1
0 12v5')
000016
000015
1)1)001.'
0000159
0000106
B-4
-------
difference. For the sake of completeness, the
linear component may be further partitioned into
a concurrence and a nonconcurrence term,v°)
although it is apparent that no appreciable correla-
tion exists between the means and the slopes of
the laboratory lines. This is done without further
explanation and is shown in Table E-V. The pur-
pose for making the analysis of variance is to
derive the components of variance. These com-
ponents are computed from Table B-V and are
shown in Table III in Section HI-E of this report.
LIST OF REFERENCES
1. Dixon, Wilfrid J. and Massey, Frank J., Jr.,
Introduction to Statistical Analysis, Chap-
ter 10, McGraw-Hill Book Company, Inc.,
New York, pp 179-180 (1957).
Bennett, Carl A. and Franklin, Norman L.,
Statistical Analysis in Chemistry and the
Chemical Industry, John Wiley and Sons,
Inc., New York, pp 379-385 (1954).
ASTM Manual for Conducting an Interlabora-
tory Study of a Test Method, ASTM STP
No. 335, Am. Soc. Testing & Mats. (1963).
Mandel, John and Lashof, T.W., "The Inter-
laboratory Evaluation of Testing Methods,"
ASTM Bulletin, No. 239, p 53 (TP133) (July
1959).
Recommended Practice for Dealing with Out-
lying Observations, ASTM Designation:
E178-68, Book of ASTM Standards, Part 30
(1968).
ASTM STP No. 335, op. cit., p 27.
Ibid, p 34.
Ibid, p 28.
B-5
------- |