EPA-650/4-74-029
May 1974
Environmental Monitoring Series
•:•:•:•:•??
.'V
I
5
o
LU
a
S^V:S$
iiiiii^
-------
EPA-650/4-74-029
COLLABORATIVE STUDY
OF METHOD FOR THE DETERMINATION
OF PARTICULATE MATTER EMISSIONS
FROM STATIONARY SOURCES
(Portland Cement Plants)
by
Henry F. Hamil and David E. Camann
Southwest Research Institute
8500 Culebra Road
San Antonio, Texas 78284
Contract No. 68-02-0626
ROAP No. 26AAG
Program Element No. LHA327
EPA Project Officer: M. Rodney Midgett
Quality Assurance and Environmental Monitoring Laboratory
National Environmental Research Center
Research Triangle Park, North Carolina 27711
Prepared for
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
May 1974
-------
This report has been reviewed by the Environmental Protection Agency
and approved for publication. Approval does no£ signify that the
contents necessarily reflect the views and policies of the Agency,
nor does mention of trade names or commercial products constitute
endorsement or recommendation for use.
11
-------
SUMMARY AND CONCLUSIONS
This report presents and analyzes the results of a collaborative test of Method 5-Determination
of Particulate Emissions from Stationary Sources. Method 5 was promulgated by the Environmental
Protection Agency in the Federal Register on December 23, 197 1. This method specifies that the
particulate concentration is determined from the particulate matter withdrawn isokinetically from the
source and determined gravimetrically after removal of uncombined water.
A Portland cement plant utilizing the wet feed process was the collaborative test site. The test-
ing was conducted with minimal supervisory interference to simulate actual "real world" usage of
Method 5 as much as possible. The four participating collaborative laboratory teams made a total of
15 simultaneous sampling runs on the stack over a two-week period. Because the cement plant fre-
quently experienced upset conditions during the testing period, the true average particulate emission
concentration sampled sometimes varied by a factor of two from one run to the next.
Various problems associated with the use of Method 5 become apparent upon preliminary analy-
sis of the data submitted by the collaborative laboratories. One collaborator's entire set of test data
had to be rejected because, due to an internal failure to communicate, the collection and weighing of
the particulate samples were done by another method that was inconsistent with Method 5. Numer-
ous minor calculation errors are contained in the remaining three collaborators' reported results.
After the calculation errors are corrected, it is evident that the sampling was often deficient. One
collaborator had difficulty achieving an isokinetic sampling rate, presumably because of inaccurate
calibration of his dry gas meter. Another collaborator frequently sampled a slightly insufficient gas
volume. The preceding problems are more attributable to operator error generally caused by inexper-
ience with Method 5 than to method inadequacy. Another usage problem with more serious conse-
quences for Method 5 is the occasional determination of extremely high particulate concentrations.
This high value phenomenon is believed caused by accidental scraping of particulate matter adhering
to the stack wall into the sampling probe tip during its insertion or removal through a sampling port.
In the dirty (and difficult) sampling environment in which Method 5 must be performed, occasional
accidental scraping with its concomitant high particulate determinations is likely to plague Method 5.
The valid corrected Method 5 determinations were given a thorough statistical analysis, including
data adjustment as appropriate, to estimate the precision of Method 5 in cement plant applications
and to investigate as many of the sources of precision variation as possible. The following conclusions
emerge from the statistical analysis:
A. Precision
The estimated precision of a Method 5 particulate concentration determination from a cement
plant stack is expressed in terms of the between-laboratory standard deviation and the within-
laboratory standard deviation. Since both of these standard deviations are proportional to their
respective group means, it is appropriate to utilize the weighted coefficient of variation technique to
estimate both standard deviations as a percentage of the "true" Method 5 concentration determina-
tion value, 8. Precision estimates for Method 5 in cement plant applications are derived for two rele-
vant situations: Method 5 as actually used by the collaborators, and Method 5 without the high
value phenomenon.
in
-------
(1) Method 5. The following precision estimates are obtained from those particulate determina-
tions obtained on the cement plant collaborative test that are valid and acceptable by
Method 5:
(a) Between-laboratory. The between-laboratory standard deviation of Method 5 is
estimated to be 58.4% of 5. Because there are a maximum of three valid collaborator
determinations per run, this estimate has only 2 degrees of freedom.
(b) Within-laboratory. The estimated within-laboratory standard deviation of Method 5
is 28.4% of 5 and also has 2 degrees of freedom associated with it.
(2) Method 5 Exclusive of the High Value Phenomenon. If the high value phenomenon could
be eliminated from Method 5 through suitable modification to the method, much better
precision would exist. The following estimates pertain to this hypothetical situation:
(a) Between-laboratory. The estimated between-laboratory standard deviation is 20.1%
of 5, with 2 associated degrees of freedom.
(b) Within-laboratory. The within-laboratory standard deviation is estimated with 20
degrees of freedom as 9.8% of 8.
Now the between-laboratory estimates contain only 2 degrees of freedom. So does the Method 5
within-laboratory estimate. The population of laboratories could differ substantially from the three
acceptable collaborators in this collaborative test. Thus, the true between-laboratory standard devia-
tions and the true Method 5 within-laboratory standard deviation may differ substantially from their
estimated values, perhaps by more than a factor of 2, in either direction. However, the hypothetical
version within-laboratory standard deviation estimate, based on 20 degrees of freedom, is reasonably
precise.
B. Sources of Precision Variation
The between-laboratory variance estimated for Method 5 from the cement plant collaborative
test can be apportioned into its components to indicate those areas in which precision improvements
would be profitable. Over 88 percent of the estimated between-laboratory variance is due to the high
value phenomenon effect. Of the usual variation sources, laboratory bias factors account for 9%,
while the within-laboratory factors cause less than 3%. If the cement plant precision of Method 5 is
to be substantially improved, means must be developed to minimize the effects of probe contamina-
tion through accidental scraping so that the high value phenomenon can be controlled. Further stan-
dardization of Method 5 by specifying more procedural details should reduce the laboratory bias
effect.
Method 5 now contains two sampling restrictions: limitation of the percent of isokinetic sampl-
ing I to the range 90 < 110, and a minimum dry gas sampling volume of 60 scf. These restrictions
were intended to prevent deterioration of Method 5's precision. The isokinetic restriction 90< 110
is essential to maintain the between-laboratory and the within-laboratory precision estimated herein
for cement plant applications. However, rigid adherence to the 60 scf minimum sampling volume is
unnecessary; acceptance of sampling volumes slightly below 60 scf has no detrimental effect on the
estimated Method 5 precision.
IV
-------
Based on the preceding conclusions and on the collaborators' experiences and suggestions,
several recommendations are presented for consideration in revising Method 5. The intent of these
recommendations is to minimize the high value problem and to further standardize the usage of
Method 5, so that its precision will be improved.
-------
TABLE OF CONTENTS
Page
LIST OF ILLUSTRATIONS viii
LIST OF TABLES - . viii
I. INTRODUCTION . . . . 1
II. COLLABORATIVE TESTING OF METHOD 5 . 2
A. Collaborative Test Site . . 2
B. Collaborators . . . .... 2
C. Philosophy of Collaborative Testing .... 3
III. STATISTICAL DESIGN AND ANALYSIS .... 6
A. Statistical Terminology . 6
B. The Experimental Design . 7
C. The Collaborative Test Data 9
D. The Precision of Method 5 13
E. Sources of Variability in a Method 5 Test Result . . .... 16
IV. RECOMMENDATIONS . . . 19
APPENDIX A—Method 5. Determination of Participate Emissions From Stationary Sources . 21
APPENDIX B-Statistical Methods 27
B.1 Preliminary Data Analysis . . .... .... 29
B.2 The High Particulate Concentration Determination Phenomenon ... 29
B.3 Examination of the Final Port Effect 31
B.4 Grouping the Samples into Blocks . 31
B.5 Data Adjustment for True Concentration Variation Within Block 34
B.6 Relationship of the Standard Deviation to the Mean 36
B.7 Weighted Coefficient of Variation Estimation 40
B.8 Precision Standard Deviation Estimation 43
REFERENCES 45
viz
-------
LIST OF ILLUSTRATIONS
Figure
1 Portland Cement Plant Test Site . . ...... • • 2
2 Platform Configuration at Sampling Level . . ....... 3
3 Typical Velocity Profiles-Lone Star Portland Cement Plant ...... 4
B-l Between-Laboratory Run Plot ....... • • . . 37
B-2 Between-Laboratory Run Plot Excluding High Values . . . .38
LIST OF TABLES
Table Pa9e
1 Design of the Cement Plant Collaborative Test of Method 5 ... 8
2 The Collaborators' Original Reported Data for the Cement Plant Collaborative
Test of Method 5 ... . . ..... - . .10
3 The Corrected Data for the Cement Plant Collaborative Test of Method 5 10
4 The Corrected Data for Between-Laboratory Analysis of the Published Version of
Method5 ................ 13
5 The Corrected Data for Between-Laboratory Analysis of Method 5 Improved to
Eliminate the High Value Phenomenon . ... . . . . . 14
6 The Adjusted Data for Within-Laboratory Analysis of Method 5 . . . 15
7 Method 5 Precision Estimates for the Published Version and for a Hypothetical
Version Modified to Eliminate the High Value Phenomenon . . 15
8 The Effects of Eliminating the High Value Phenomenon and Relaxing the
Isokinetic Sampling and Minimum Sampling Volume Restrictions . . . . 17
9 Sources of Between-Laboratory Variation in a Method 5 Cement Plant Test Result 1 8
B-l Examination of High Particulate Concentrations from All Three Collaborative Tests
of Method 5 . . . ....... . . 30
B-2 Youden Rank Test for Significance of Final Port Effect . . . . . 31
B-3 Kruskal-Wallis H Test for Differences in the True Particulate Concentration from
Run to Run ....... . . 32
vm
-------
LIST OF TABLES (Cont'd)
Table Page
B-4 Run Participate Concentration Ranks from Cement Plant Operating
Conditions ... ...... . . 33
B-5 Blocking the Runs by Rank Score .... . . 34
B-6 Block Coefficient of Variation Ratios . . . 34
B-7 Youden Rank Test for Significance of Laboratory Effect . . 35
B-8 The Adjusted Method 5 Data and Its Run Summary ... . . . 36
B-9 Proportionality Evidence from Comparative Regressions 39
B-10 Adequacy of Alternative Transformations to Achieve Equality of Variance ... 40
B-l 1 Sample Size Dependency of Coefficient of Variation Weights . 42
IX
-------
I. INTRODUCTION
This report describes the work performed and the results obtained on Southwest Research Institute
Project 01-3462-003, Contract No. 68-02-0626, which includes collaborative testing of Method 5 for par-
ticulate emissions as given in "Standard of Performance for New Stationary Sources."(1 **
This report describes the collaborative testing of Method 5 in a Portland cement plant and gives
the statistical analysis of the data from the collaborative test and the conclusions and recommendations
based on the analysis of data.
*Superscript numbers in parentheses refer to the References at the end of this report.
-------
II. COLLABORATIVE TESTING OF METHOD 5
A. Collaborative Test Site
Arrangements were made for a collaborative test of Method 5 at the Lone Star Industries Portland
cement plant in Houston, Texas.
The plant site was visited in February le>72 to evaluate suitability for collaborative testing. The
plant utilizes the wet feed process and operates three kilns. The Hue gas from each kiln passes through a
separate electrostatic precipitator. These Hue gases are then combined and led into a 300-tt-high stack.
The clinker cooler is used to preheat the air to the main kilns, so that the clinker cooler outlet does not
constitute an emission source. Plant capacity is 1700 tons/day, dry teed basis.
There are four sample ports spaced e>U deg apart, and located 150 ft above grade. The inside stack
diameter at the sample ports is 13 ft. The stack is of double-walled construction, with the outer support-
ing wall of reinforced concrete. The inner stack, separated from the outer stack by an annular space ol
one foot, is constructed of refractory brick. The sample ports are installed in the inner stack and are
accessable through openings in the outer stack wall. The sample ports are located ^0 teet (6.(> diameters)
above the inlet duct to the stack and 150 feet ( 1 1.5 diameters) below the stack outlet. Twenty
traverse points were used in the tests, ten per diameter. Permanent monorails were installed by Lone
Star Industries at each sample port. Access to the sample ports is provided by a 360-deg platform on
the stack. A second 360-deg platform on the stack at 125 ft above grade provided space lor the con-
trol consoles. The stack is equipped with a bolometer at the 125-ft level, which provides in-stack opacity
readings. Arrangements were made to obtain both daily raw material feed rates and m-stack plume
opacities during the test period.
An overall view of the stack and sampling platforms is shown in Figure 1. A schematic of the sampl-
ing platform and sampling port is given in Figure 2. Typical velocity profiles are shown in Figure 3.
B. Collaborators
The collaborators for the Lone Star Industries
Portland cement plant test were Mr. diaries Rodrigue/
and Mr. Nollie Swynnerton of Southwest Research
Institute, San Antonio Laboratory, San Antonio, Texas;
Mr. Mike Taylor and Mr. Ron Hawkins of Southwest
Research Institute, Houston Laboratory, Houston,
Texas; Mr. Quirino Wong, Mr. Randy Creighton, and
Mr. Vito Pacheco, Department of Public Health, City
of Houston, Houston, Texas; and Mr. Royce Alford,
Mr. Ken Drummond, and Mr Lynn Cochran of South-
western Laboratories. Austin, Texas.*
Collaborative tests were conducted under the gen-
eral supervision of Dr. Henry Hamil of Southwest
Research Institute. Dr. Hamil had the overall respon-
sibility lor assuring that the collaborators were com-
Fignre 1. Portland Cement Plant Test Site petent to perform the test, that the test was conducted
in accordance with the collaborative test plan, and that all collaborators adhered to Method 5 as written
in the federal Register, December 23, 11>7I.
* DmiUjjliout (he remainder o! this report, the collaborative laboratories are referenced by randomly assigned eode numbers as Lab 1()|
Lab 1(12, Lab 1(13, and Lab 104. These code numbers do not correspond to (lie above ordered listing of collaborators.
-------
Figure 2. Platform Configuration at Sampling Level
C. Philosophy of Collaborative Testing
The concept of collaborative testing followed in the tests discussed in this report involves conduct-
ing the test in such a manner as to simulate "real world" testing as closely as possible. "Real world"
testing implies that the results obtained during the test by each collaborator would be the same results
obtainable if he were sampling alone, without outside supervision, and without any additional informa-
tion from outside sources, i.e., test supervisor or other collaborators.
The function of the test supervisor in such a testing scheme is primarily to see that the method is
adhered to as written and that no individual innovations are incorporated into the method by any col-
laborator. During the test program, the test supervisor observed the collaborators during sampling and
sample recovery. If random experimental errors were observed, such as mismeasurement of volume of
impinger solution, improper rinsing of probe, etc., no interference was made by the test supervisor. Since
such random errors will occur in the everyday use of this method in the field, unduly restrictive super-
vision of the collaborative test would bias the method with respect to the performance test results which
will be obtained when the method is put into general usage. However, if gross deviations were observed
of such magnitude as to make it clear that the collaborator was not following the method as written, these
would be pointed out to the collaborator and corrected by the test supervisor.
While most of the instructions in the Federal Register are quite explicit, some areas are subject to in-
terpretation. Where this was the case, the individual collaborators were allowed to exercise their professional
judgment as to the interpretation of the instructions.
-------
Figure 3. Typical Velocity Profiles
Lone Star Portland Cement Plant
-------
The overall basis for this so-called "real-world" concept of collaborative testing is to evaluate the
subject method in such a manner as to reflect the reliability and precision of the method that would
be expected in performance testing in the field.
-------
I. STATISTICAL DESIGN AND ANALYSIS
A. Statistical Terminology
To facilitate the understanding of this report and the utilization of its findings, this section
explains the statistical terms used in this report. The procedures for obtaining estimates of the per-
tinent values are developed and justified in the subsequent sections.
We say that an estimator, 8, is unbiased for a parameter Q if the expected value of 0 is 6, or
expressed in notational form, E(9) = 6. Letx1,x2, .,xn be a sample of n replicate method deter-
minations. Then we define:
1 "
(1) x =~ ^, X{ as the sample mean, an unbiased estimate of the true mean, 5, of the determinations.
ni=\
This term gives an estimate of the center of the distribution of the Xj's.
1 "
(2) s2 = ^ (xt - x)2 as the sample variance, an unbiased estimate of the true variance,
«-!,-=!
a2 This term gives a measure of the dispersion in a distribution.
(3) s - A/52 as the sample standard deviation, an alternative measure of dispersion, which esti-
mates a, the true standard deviation.
The sample standard deviation, s, however, is not unbiased for a,(2) so a correction factor needs
to be applied. The correction factor for a sample of size n from a normal population is &„ , and the
product of an and s is unbiased for a. That is, E(ans) = o. As n increases, the value of an decreases
to eventually approach 1.0, going for example from «3 = 1.1284, «4 = 1.0854 to a10 = 1.0281.
We define
a
»"
as the true coefficient of variation for the distribution of_the method determinations. To estimate
this parameter, we use a sample coefficient of variation, j3, defined by
OLnS
P = ^r
x
where j3 is the ratio of the unbiased estimates of a and 5 , respectively. The coefficient of variation
measures the percentage scatter in the observations about the mean and thus is a readily under-
standable way to express the precision of the observations.
The collaborative test plan for this test called for 16 runs. On each run, the collaborative teams
were expected to collect simultaneous samples from the stack in accordance with Method 5. Since
the actual particulate emission concentration in the stack fluctuates, one can generally expect dif-
ferent true concentrations for each run. To permit a complete statistical analysis, the individual
runs are grouped into blocks, where each block has approximately the same true emission con-
centration level.
-------
We can apply the statistical terms of the preceding paragraph both to the collaborators' values
during a given run, and to each collaborator's values in a given block. In this report, statistical
results from the first situation are referred to as run results. Those from the second situation are
referred to as collaborator block results. For example, a run mean is the average of each collaborator's
concentration level for the run as obtained by Method 5. A collaborator block coefficient of variation
is the ratio of the unbiased standard deviation estimate to the sample mean for all of a collaborator's
runs grouped in the block.
The variability associated with a Method 5 concentration determination is estimated in terms of
the within-laboratory and the between-laboratory precision components. In addition, a laboratory
bias component can be estimated. The following definitions of these terms are given with respect
to a true stack concentration, /u:
• Within-laboratory—The within-laboratory standard deviation, a, measures the dispersion in
replicate single determinations made using Method 5 by one laboratory team (same field
operators, laboratory analyst, and equipment) sampling the same true concentration, p..
The value of a is estimated from within each collaborator block combination.
• Between-laboratory—The between-laboratory standard deviation, a§ , measures the total
variability in a concentration determination due to simultaneous Method 5 determinations
by different laboratories sampling the same true stack concentration, p.. The between
laboratory variance, a\, may be expressed as
°l= °L + °2
and consists of a within-laboratory variance plus a laboratory bias variance, a\ The
between-laboratory standard deviation is estimated using the run results.
• Laboratory bias—The laboratory bias standard deviation, a/, = ^Ja]) - a2 , is that portion of
the total variability that can be ascribed to differences in the field operators, analysts and
instrumentation, and due to different manners of performance of procedural details left
unspecified in the method. This term measures that part of the total variability in a deter-
mination which results from the use of the method by different laboratories, as well as
from modifications in usage by a single laboratory over a period of time. The laboratory
bias standard deviation is estimated from the within- and between-laboratory estimates
previously obtained.
B. The Experimental Design
The collaborative test of Method 5 at the Lone Star Industries' Cement Plant in Houston, Texas
was conducted on the kiln stack under ambient operating conditions. No attempt was made to block
the concentration of the particulate matter emitted from the stack into distinct levels. Of a planned
16 sampling runs, 15 were actually conducted, at a rate of one or two runs per day. A run involved 2
hours of simultaneous sampling by each of the four collaborators. Approximately one-fourth of a run's
sample, which consisted of 30 min of sampling, was collected at each of the four ports by each col-
laborator. At each port, the collaborator made a radius traverse of the circular stack, sampling for
six minutes each from the five traverse points according to Method 1. Table 1 shows the layout of
the test. The starting port for each collaborator, and the direction of port rotation during the run,
were randomized each day. When two runs were made on the same day, sampling on the second
run was rotated in the opposite direction of the first run, avoiding entanglement of the umbilical
cords connecting the sampling equipment to the control consoles.
7
-------
Table 1. Design of the Cement Plant Collaborative Test of Method 5
Day
4/2/73
4/3/73
4/4/73
4/5/73
4/6/73
4/9/73
4/10/73
4/11/73
4/12/73
4/13/73
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Rotation
Direction*
Ccl.
Ccl.
Ccl.
Cl.
Ccl.
Ccl.
Cl.
Ccl.
Ccl.
Cl.
Ccl.
Cl.
Cl.
Ccl.
Cl.
Final Port Sampled
Lab 101
C
B
B
C
B
B
D
A
A
B
D
A
A
A
C
Lab 102
A
D
A
B
A
A
C
D
D
A
A
B
B
B
B
Lab 103
D
C
C
D
C
C
A
B
B
C
B
C
C
C
D
Lab 104
B
A
D
A
D
D
B
C
C
D
C
D
D
D
A
*C1. -The laboratory teams sampled from the four ports clockwise around the
stack.
Ccl. -Sampling was conducted counterclockwise around the stack.
Because the runs were conducted under ambient operating conditions, with no intention of hold-
ing the particulate concentration constant from run to run, one would expect some drift in the true
average particulate concentration in the stack from run to run. In fact, the run particulate concentra-
tions varied from around 8 X 10~7 Ib/scf to 60 X 10"7 lb/scf.* Some of this run variation may be
attributed to the curtailment from capacity three kiln operation on the first week of the collaborative
test to a two kiln operation on the second week. However, the in-plant bolometer indicated that the
preponderance of the run variation resulted from sporadic upset conditions during the testing periods.
Even if the runs were grouped into blocks of approximately equivalent concentration levels, some
variation in the true concentration could still be expected for the runs in a block. Hence, in order
to estimate the within-laboratory precision variability of Method 5, it is likely that the particulate
concentration data will have to be adjusted for true variation from one run to another.
Particulate matter will precipitate out of any particulate-air or particulate-acetone mixture.
Therefore, one cannot prepare a known particulate concentration in air for conducting a Method 5
accuracy test; nor can one prepare a known particulate solution in acetone for use in determining
*EPA policy is to express all measurements in Agency documents in metric units. When implementing this policy will result in undue
cost or difficulty in clarity, NERC/RTP is providing conversion factors for the particular nonmetric units used in the document For
this report, the factors are:
10" lb/scf =1.6018 X 103
1 scf = 0.028317m3
1 in. = 0.0254m
-------
the precision of just the analytical phase of Method 5. Consequently, this collaborative test report
cannot assess the accuracy of Method 5, nor can it separate Method 5 precision variability into its
sampling phase and analysis phase components.
C. The Collaborative Test Data
The essential Method 5 data that were originally reported by the four participating laboratories
in this collaborative test are displayed in Table 2. Shown in Table 2 are c, the Method 5 particulate
matter concentration, /, the percent of isokinetic sampling, and Vm std, the volume of gas sampled
corrected to standard conditions (70°F, 29.92 in. Hg), reported by each collaborator for each run.
I is tabulated because the acceptable isokinetic sampling range for Method 5 is 90% < / < 110%
(cf. Appendix A). Vm std is presented because a minimum sampling volume of 60 scf corrected to
standard conditions on a dry basis is specified in the Federal Register^"1 for Portland cement plants.
Note that many of the runs fail to meet one or both of these criteria. There are three missing
data points. Lab 101 incurred a leak causing loss of vacuum on Run 14. Lab 102 had to abort
on Run 5 because a power failure caused water to surge into the filter holder; this necessitated
replacing the filter. Some of the equipment used by Lab 103 was malfunctioning during Run 1.
There are wide discrepancies in the collaborators' particulate concentration values for a single
run, with one collaborator's concentration frequently differing from another's by an order of
magnitude or more. Lab 103 was especially prone to discrepant reported concentrations. Discus-
sion with Lab 103 personnel, including an on-site review of their analytical procedures, revealed
that Lab 103 was performing Texas' Rule 9 rather than EPA 's Method 5 to determine particulate
matter concentration. The procedure used by Lab 103 involves extraction of condensibles from the
impinger catch. The extracts are added to the acetone probe and glassware wash before evaporation.
Therefore, the combined probe catch and extracted condensibles are obtained and reported. Since
these methods are incompatible, it is impossible to convert Rule 9 concentrations into their Method 5
equivalent. Consequently, Lab 103 had to be excluded from the entire data analysis.
According to the customary procedure, the calculations of the outlying sample concentrations
reported by the three remaining collaborators were rechecked. Numerous sizable calculation errors
were discovered that cast doubt on the validity of all the reported data in Table 2. Therefore, all
the calculations involved in computing Vm s(d, c, and / were done over from the raw data for every
run of the three remaining collaborators. The corrected data are presented in Table 3.
In the process of recalculation, numerous Method 5 errors were encountered. Section B.I of
Appendix B describes the calculation errors in detail. The most disturbing mistakes were made by
using the wrong sets of data and using one wrong formula, in addition to excessive rounding on
intermediate steps of calculations. Table 3 shows that the reported Table 2 values were frequently
in error by as much as 10 percent due to improper calculation. Thus, calculation is a major source
of error in the use of Method 5.
There are two ways to remedy the calculation errors that plague the usage of Method 5 and
the associated Methods 2 and 3 that it requires. One approach is to design a standard data form on
which all of the raw data required in the calculations of Methods 2,3, and 5 could be recorded
for easy keypunching. The keypunched data could then be input to a computer program to perform
-------
Table 2. The Collaborators' Original Reported Data for the Cement Plant
Collaborative Test of Method 5
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Lab 101
c
15
13
11
150
11
8.8
18
29
150
140
29
33
230
Missing
27
7
111.3
114.3
113.0
117.7
106.3
108.7
96.3
104.0
102.7
106.0
107.3
108.0
109.7
-
103.7
*Xtd
63.2
63.7
63.5
58.6
53.4
58.2
53.4
58.6
58.4
58.9
58.7
58.7
58.5
-
53.4
Lab 102
c
12.9
13.1
10.3
18.8
Missing
8.6
15.3
15.9
6.8
12.3
20.7
38.9
75.2
19.8
26.9
7
108
103
109
106
-
107
87
92
101
98
106
103
103
99
102
*Wd
80.9
80.2
83.9
76.2
-
74.8
69.4
77.1
79.5
77.0
74.5
77.7
75.4
67.9
72.1
Lab 103
c
Missing
43.5
71.2
83.3
130.3
79.0
202.9
270.8
128.6
39.3
54.3
71.0
80.4
42.2
98.9
I
—
103
90.9
97.6
97.7
97.6
91.8
91.1
95.0
97.6
102
101
98.9
102
101
•Xtd
_
69.2
84.0
63.0
61.7
62.8
59.5
62.1
66.7
67.2
65.2
65.0
64.0
60.3
62.1
Lab 104
c
14.2
11.8
11.2
9.2
10.6
8.2
9.4
14.4
9.0
9.4
17.4
23.6
57.8
10.5
18.7
'
112.0
120.1
97.5
132.4
112.4
108.3
99.7
107.2
108.3
109.4
105.7
107.7
96.3
105.9
110.4
>Wd
66.4
77.3
70.6
86.7
67.3
68.3
78.1
77.3
73.5
70.5
65.1
66.0
61.2
59.4
64.6
Note:
c— Paniculate matter concentration from Method 5, 10"' Ib/scf.
/-Percent of isokinetic sampling.
Vm t(j-Volume of gas sample at standard conditions, cu. ft.
Table 3. The Corrected Data for the Cement Plant Collaborative
Test of Method 5
Run
IvUIl
l
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Lab 101
r
16.0
14.5
12.4
137. Ot
10.8
10.5
17.0
29.2
148. 8f
137.4f
28.7
33.4
214.2t
Missing
27.0
/
96.8
100.5
99.5
108.2
101.4
99.1
89.2*
90.1
103.6
93.5
98.3
97.7
99.9
-
99.5
Vmstd
63.7
64.0
64.4
62.4
58.2*
59.5*
57.0:):
59.2*
58.4*
61.5
60.6
60.7
61.7
-
56.4*
Lab 102
c
12.9
13.1
10.3
18.8
Missing
8.6
15.3
15.9
6.8
12.3
20.7
38.9
75.3
19.8
26.9
/
108.8
102.7
108.8
106.1
-
107.0
87.0*
92.6
100.6
98.0
101.6
103.4
102.8
99.4
102.2
F^std
80.9
80.2
83.9
76.2
-
74.8
69.4
77.1
79.6
76.6
74.5
77.8
75.3
68.0
72.1
Lab 104
c
13.1
11.8
10.9
11.7
10.7
7.8
9.4
15.5
9.0
6.3
15.7
21.2
52.2
9.5
18.9
/
107.8
115.2*
101.2
111.1*
113.8*
114.0*
100.6
108.4
109.5
110.9*
115.8*
117.3*
107.4
115.2*
111.6*
K"Jstd
72.4
77.2
72.1
67.9
66.5
71.1
73.2
71.6
73.5
70.4
72.1
73.3
67.9
65.6
64.1
Note:
c-Particulate matter concentration from Method 5, 10~7 Ib/scf.
/-Percent of isokinetic sampling.
1/mstd~Vo'ume of gas sample at standard conditions, scf.
*Outside acceptable isokinetic sampling range : 90 < / < 1 1 0.
•fHigh concentration; probable probe tip particulate scraping from stack wall.
*Less than minimum sampling volume of 60 cu. ft.
10
-------
all the calculations and print the necessary results. This would eliminate calculation errors. However,
there are difficulties to implementing this approach. Most available equipment does not measure in
the desired metric units. Voluminous data are obtained on every sample. Finally the number of
traverse points is not fixed but will vary from one site to another. An alternative approach is to
clarify the calculation procedures of Methods 2 and 5. The formulas and data sources for all variables
calculated in these methods should be stated explicitly. Furthermore, a note of caution should be
added warning against intermediate rounding of calculations. The analyst should retain at least one
more significant digit on all intermediate calculation steps than he intends to report in his results.
A perusal of the Table 3 participate matter concentration data obtained by Method 5 reveals four
values, all of which appear to be approximately one order of magnitude too high. These high con-
centration samples were all obtained by Lab 101, on Runs 4, 9, 10, and 13. Because no violation
of the Method 5 procedures could be found, there are no physical grounds for rejecting these high
concentrations. They must be treated as valid Method 5 determinations. But because these four
high determinations are certain to cause most of the Method 5 precision variability reported for
this test, it is essential to seek an explanation for their occurrence.
In all three of the Method 5 collaborative tests that have been conducted, high particulate
matter concentrations have occurred. An analysis of the 12 high concentrations that occurred
in ihe three tests is presented in Section B.2 of Appendix B. In 11 of the 12 cases, the high con-
centration is due to an excessive amount of particulate matter, ranging from 310 mg to 860 mg,
present in the acetone washings of all sample-exposed surfaces prior to the filter. The probable
explanation is that the probe tip was scraped against the inner stack wall during the insertion or
removal of the sampling train from one of the sampling ports on these runs; this scraping would
contaminate the probe tip with the particulate adhering to the wall. Subsequent sampling during the
run might suck the contaminating particulate material into the probe body or through it to the fil-
ter. Apparently, the principal cause of the high particulate concentration phenomenon that has
afflicted all three Method 5 collaborative tests is accidental scraping of the probe tip against the
stack wall while the probe is being inserted into or removed from the stack.
Table 3 lists 43 Method 5 particulate concentration determinations. Four of these determina-
tions, those followed by a dagger(f), are the high values. All four of these high values were obtained
by Lab 101—on Runs 4, 9, 10, and 13. Since these appear to be valid Method 5 determinations,
they should be included with the other determinations in evaluating the precision of Method 5.
However, many auxiliary aspects of Method 5 and this collaborative test are also investigated in
the statistical analysis. Since the magnitude of the high value effect usually obliterates the finer
effects of these other aspects, the four high determinations are generally eliminated from the
data for these side analyses.
Method 5 stipulates that the isokinetically acceptable sampling range is 90 < / < 110. For 1 1
of the 43 samples shown in Table 3, / lies outside the acceptable range. But a single laboratory.
Lab 104, obtained 9 of these 1 1 isokinetically unacceptable samples. Lab 104 reports that its
isokinetic sampling problem probably arose from using an old wet test meter, subsequently found
to be inaccurate, to calibrate its dry gas meter. Using this inaccurate primary standard meter may
have given inaccurate calibration constants both for the nomograph setting and for the dry gas meter
correction factor. Hence, the frequent failure to maintain isokinetic sampling on this collaborative
test does not suggest any general difficulty in adhering to the restriction 90 < / < 110. Now, of
course, the 11 isokinetically unacceptable concentration samples have to be excluded from the
analysis of Method 5 precision. But because there are so few determinations per run with Lab 103
11
-------
excluded, the isokinetic restriction is sometimes relaxed to 85 < 115 or 80 < 120 in the
side analyses in order to provide sufficient data for a substantive treatment.
The corrected dry gas volume sampled is below the Federal Register limitation of Vm std> 60 scf
on six of the 43 samples, all of which were obtained by Lab 101. Enforcement of this minimum
sampling volume condition, in addition to the isokinetic acceptance condition, with Lab 103
excluded, would have left too little data on which to base a meaningful analysis of the Method 5
precision. The minimum sample volume condition is never violated by more than 10 percent.
Consequently, this volume condition is considered the least vital of the data exclusion criteria, and
it is ignored in nearly all of the statistical analysis.
In taking its 2-hour sample by Method 5 on a run of this collaborative test, each collaborator
was supposed to sample isokinetically for 30 min from each of the four ports. Normally, one would
not expect a port effect influencing the data under these circumstances. Yet due to the probable
probe tip contamination discussed above, an analysis of the final port effect was made. This
analysis is conducted in Section B.3 of Appendix B. No final port effect is detected in the Method 5
samples by Youden's Rank Test. It can be concluded that the final port from which a collaborator
sampled had no systematic effect on his Method 5 concentration determination.
However, the true paniculate concentration sampled did vary greatly from one run to another
during the collaborative test. For example, the average of the acceptable collaborators' Method 5
particulate concentration determinations sampled, with high values eliminated, is 7.9 X 10~7 Ib/scf
on Run 9, 11.2 X 10~7 Ib/scf on Run 3, 24.7 X 10'7 Ib/scf on Run 11, and 63.8 X 10'7 Ib/scf
on Run 13. In Section B.4 of Appendix B, the Kruskal-Wallis H Test is applied to the acceptable
determinations. The H Test shows that the variation in the true particulate concentration from
one run to another during the collaborative test is indeed significant.
The effect of the tremendous variation in the true particulate concentration from run to run
must be removed from the collaborative test data prior to estimating the within-laboratory precision.
Otherwise, the calculated within-laboratory precision would be greatly inflated. Several statistical
techniques can be employed to remove this true concentration effect. In Section B.4 of Appendix B,
the runs having roughly the same true particulate concentration are grouped into blocks. The runs
are ranked in ascending order of predicted true particulate concentration according to their in-stack
bolometer opacity charts and the daily feedrate. Then the runs in the same rank score range are
assigned to the same block. Runs 9, 3, 1, 10, 5, 6, 2, and 7 with rank scores ranging from 11.5 to
27.0 are placed in Block 1. Runs 4, 11, 14, 8, 12, and 15 having rank scores from 32.0 to 50.0
are assigned to Block 2. Run 13 with a rank score of 57.5 is given a separate block, Block 3. Table B-5
in Appendix B illustrates the blocking assignment of each run based on its rank score.
Blocking the runs does remove much of the true concentration effect, but not all of it. The
analysis of Section B.5 of Appendix B demonstrates that there is considerable variation in the true
particulate matter concentration of the runs grouped in the same block. This variation is substantial
enough to overwhelm the sensitive statistical procedure used to separate the between-laboratory
variance into its within-laboratory and laboratory-bias components. Calculation by the usual pro-
cedure yields no laboratory bias in the collaborative test data. On the other hand, simple inspec-
tion of the test data and statistical analysis by laboratory using Youden's Rank Test both reveal a
large laboratory bias effect. To reconcile the calculation procedure results with the evident nature
of the test data, the variation in true concentration among the runs in the same block must be
separated from the method's actual precision variation and eliminated. In the absence of any
alternative indicator of the true particulate concentration, the collaborators' Method 5 run mean is
12
-------
used as the true concentration indicator. The true concentration effect is removed from the blocks
by adjusting every collaborator's determination on a run by the amount needed to give the run the
same mean as its block. The resulting adjusted data is shown in Table B-8 of Appendix B. While
this data adjustment technique probably over-adjusts the within-laboratory precision estimates, it
does permit the calculation of a laboratory-bias precision component of reasonable magnitude in
comparison with the corrected concentration data of Table 3.
D. The Precision of Method 5
The precision of Method 5 at cement plants is examined in two situations. The first situation
relates to Method 5 as published in Appendix A: the high particulate concentration phenomenon
occurs occasionally, the stated isokinetic sampling restriction (90 < 110) is enforced, and
approximately 60 scf of dry gas is sampled (Vm std > 56 scf). The resultant acceptable data from
Table 3 under these restrictions are displayed in Table 4, with a run summary included. The four
runs with high values stand out sharply from the other runs in Table 4. These four runs have
standard deviations ranging from 81.4 X 10~7 Ib/scf to 88.5 X 10~7 Ib/scf, while the maximum
for the other runs is only 7.8 X 10~7 Ib/scf.
Table 4. The Corrected Data for Between-Laboratory Analysis of
the Published Version of Method 5
Block
1
2
3
Run
9
3
1
10
5
6
2
7
4
11
14
8
12
15
13
Acceptable Method 5 Particu-
late Concentration Determina-
tions, 10~7 Ib/scf
Lab 101
148.8
12.4
16.0
137.4
10.8
10.5
14.5
R
137.0
28.7
M
29.2
33.4
27.0
214.2
Lab 102
6.8
10 3
12.9
12.3
M
8.6
13.1
R
18.8
20.7
19.8
15.9
38.9
26.9
75.3
Lab 104
9.0
10.9
13.1
R
R
R
R
9.4
R
R
R
15.5
R
R
52.2
Run Summary
Mean,
X.j
54.87
11.20
14.00
74.85
10.80
9.55
13.80
9.40
77.90
24.70
19.80
20.20
36.15
26.97
113.90
Standard
Deviation,
si
81.36
1.08
1.73
88.46
1.34
0.99
83.58
5.66
7.80
3.89
0.07
87.63
Coef. of
Variation,
k
1.6732
0.1090
0.1398
1.4812
0.1763
0.0899
1.3447
0.2870
0.4355
0.1348
0.0033
0.8681
Weight,
lit •
W,
1.366
1.366
1.366
0.738
0.738
0.738
0.738
0.738
1.366
0.738
0.738
1.366
Note:
M-Missing determination.
R-Rejected determination: outside acceptable isokinetic sampling range, 90 < I < 110.
It is felt that Method 5 can be improved to eliminate some, and perhaps most, of the high par-
ticulate concentration phenomenon. To assess the impact of this improvement on Method 5's
precision, a second situation is also considered. The 90 < / < 110 isokinetic sampling restriction
is retained, and a loose Vm td ^56 scf interpretation of the minimum sampling volume limitation
is again taken. But now the four high particulate concentrations are excluded from the valid
collaborative test data. Table 5 presents the collaborative test data for this second situation and its
run summary Here the standard deviations and the coefficients of variation for the runs are much
more uniform.
13
-------
Table 5. Ttie Corrected Data for Between-Laboratory Analysis of
Method 5 Improved to Eliminate the High Value Phenomenon
Block
1
2
3
Run
9
3
1
10
5
6
2
7
4
11
14
8
12
15
13
Acceptable Method 5 Particu-
late ConcsntiS-tion D6tcrrniri3~
tions with High Values
Excluded, 10'' Ib/scf
Lab 101
E
12.4
16.0
E
10.8
10.5
14.5
R
E
28.7
M
29.2
33.4
27.0
E
Lab 102
6.8
10.3
12.9
12.3
M
8.6
13.1
R
18.8
20.8
19.8
15.9
38.9
26.9
75.3
Lab 104
9.0
10.9
13.1
R
R
R
R
9.4
R
R
R
15.5
R
R
62.2
Run Summary
Mean,
X i
1
7.90
11.20
14.00
12.30
10.80
9.55
13.80
9.40
18.80
24.70
19.80
20.20
36.15
26.95
63.75
Standard
Deviation,
si
1.56
1.08
1.73
1.34
0.99
5.66
7.80
3.89
0.07
16.33
Coef. of
Variation,
Pi
0.2468
0.2090
0.1398
0.1763
0.0899
0.2870
0.4355
0.1348
0.0033
0.3211
Weight
wi
J
0.797
1.474
1.474
0.797
0.797
0.797
1.474
0.797
0.797
0.797
Note:
E-Ex eluded determination: high value concentration probably caused by scraping probe tip
against stack wall.
M-Missing determination.
R-Rejected determination: outside acceptable isokinetic sampling range, 90 < I < 110.
As discussed in the previous section and in Section B.5 of Appendix B, data adjustment is
necessary to eliminate the true concentration effect and thus to permit a valid within-laboratory
precision analysis. Because the effect of the high value phenomenon obscures all finer effects, data
adjustment is applied to the Table 5 data, in which the high values are excluded, to yield the
appropriate data for the within-laboratory analysis. Table 6 presents the adjusted data and gives
a statistical summary for each collaborator block that is the basis of the within-laboratory analysis.
A weighted coefficient of variation approach is used to obtain the precision estimates for Method
at cement plants. This procedure is only valid when the collaborative test data exhibit a proportional
relationship between the standard deviation and the mean. There are two essential proportional
relationships: that between laboratories over the test runs and that within each laboratory on the
collaborator's blocked samples adjusted to replication conditions. The Table 4 and Table 5 run sum-
maries provide the between-laboratory mean and standard deviation data for the two situations under
consideration. The corresponding within-laboratory mean and standard deviation data for each col-
laborator block, which is appropriate for both situations, are shown in Table 6. Graphical and more
sophisticated statistical techniques are employed in Section B.6 of Appendix B to examine these
relationships of standard deviation to mean. The examination yields enough favorable evidence to
warrant the conclusion that the proportionality of standard deviation to mean is characteristic of
Method 5 concentration determinations at cement plants, both between laboratories and within a
single laboratory.
The precision estimates obtained for Method 5 at cement plants are summarized in Table 7
for both situations. These estimates are derived by the weighted coefficient of variation
approach in Section B.8 of Appendix.B.
14
-------
Table 6. The Adjusted Data for Within-Laboratory Analysis of Method 5
Block
1
2
3
Note:
Collab-
orator
Lab 101
Lab 102
Lab 104
Lab 101
Lab 102
Lab 104
Lab 101
Lab 1 02
Lab 1 04
Acceptable Adjusted Method 5 Particulatc
Concentration Determinations with High
Values Excluded, 10~7 Ib/scf
Runs in the Block
1
E
10.0
12.2
9
E
24.4
R
15
E
75.3
52.2
9
P.3
10.2
10.8
10
28.4
20.4
R
3
13.2
10.0
10.2
1 1
M
24.4
R
4
E
11.1
R
12
33.4
20.1
19.7
5
11.1
M
R
13
21.7
27.2
R
1
6
12.1
10.2
R
14
24.5
24.4
R
7
11.8
10.4
R
8
R
R
11.1
Collaborator Block Summary
Mean,
12.10
10.32
11.07
27.00
23.48
19.70
75.30
52.20
Standard
Deviation,
si. k
0.76
0.41
0.84
5.07
2.73
Coef. of
Variation,
Pik
0.0672
0.0420
0.0821
0.2040
0.1222
Weight,
Wik
1.000
1.231
0.769
0.769
1.231
E-Excluded determination: high concentration probably caused by scraping probe tip against stack wall.
M— Missing determination.
R-Rejected determination: outside acceptable isokimetic sampling range, 90 < / < 1 10.
Table 7. Method 5 Precision Estimates for the Published Version and for a Hypothetical
Version Modified to Eliminate the High Value Phenomenon
Situation
Criterion
Application:
Site:
Data Characteristics:
High Values
Isokinetic Restriction
Min. Volume Limit
Data Sources:
Application:
Site:
Data Characteristics:
High Values
Isokinetic Restriction
Min. Volume Limit
Data Sources:
Value
Published Method 5
Portland Cement Plants
Included
90 < 110
56 scf
Tables 4 and 6
Method 5 Modified to
Eliminate High
Value Phenomenon
Portland Cement Plants
Excluded
90 < / < 1 1 0
56 scf
Tables 5 and 6
Precision
Measure
Between Labs
Within Labs
Lab Bias
Between Labs
Within Labs
Lab Bias
Coefficient
of Variation,
3
(36 = 0.584
/3 = 0.284
fa = 0.510
pbfe) = 0.201
P(e) = 0.098
k(e) = °'176
Standard
Deviation,
a
ob = (0.584)6
o= (0.284)5
OL = (0.510)6
Bb(e> = (0.201)5
S(ej = (0.098)5
°L(e) = (°-176)5
The precision estimates for Method 5 as written, the situation in which the high value phenomenon
occasionally occurs, are presented in the upper half of Table 7. The between-laboratory coefficient
of variation fo is estimated as the weighted average of the run coefficients of variation shown in Table 4.
The run coefficient of variation weights are assigned based on the number of collaborators making a
determination on the run (cf. Section B.7 of Appendix B). The weighted average is fa = 0.584, with
a maximum of 3 participating collaborators per run. Let 5 denote the true particulate concentration
15
-------
in a cement plant stack at which Method 5 particulate determinations are made. The estimated
between-laboratory standard deviation is ab = (0.584)8 with 2 degrees of freedom. With so few
degrees of freedom, the true between-laboratory values could vary markedly from these estimates.
The within-laboratory coefficient of variation is estimated via a weighted average of the collaborator-
block coefficients of variation in Table 6, with an adjustment to reintroduce the high value effect
(cf. Section B.8 of Appendix B). A within-laboratory coefficient of variation of 0 = 0.284 is
obtained. This gives an estimated within-laboratory standard deviation of 5 = (0.284)5, which has
no more than 2 degrees of freedom because the adjustment involves the ratio ($bl$b(e)~) of between-
laboratory precision estimates each of which have only 2 degrees of freedom. The concomitant
laboratory bias standard deviation estimate is OL= (0.510)6.
Method 5 precision estimates are also obtained for a second situation, which assumes the
development of a hypothetical version of Method 5 that could prevent the high value phenomenon
from occurring. These precision estimates, which are obtained from the Table 5 run summary and
the Table 6 collaborator block summary, are shown in the lower half of Table 7. If the high value
phenomenon could be eliminated from Method 5, the estimated between-laboratory standard
deviation would be abfej = (0.201)6, and the estimated within-laboratory standard deviation
a(ej = (0.098)5. The subscript e denotes elimination of the high value phenomenon. Preventing
the high value phenomenon from affecting Method 5 would tremendously improve the method's
precision in measuring the particulate matter concentration at cement plants.
One should exercise caution in the use of the precision estimates shown in Table 7. The between-
laboratory estimates are based on a maximum of 3 collaborators' determinations per run, and thus
they only possess 2 degrees of freedom. The published Method 5 within-laboratory estimates are
calculated using these between-laboratory estimates and thus also have just 2 degrees of freedom.
The entire population of laboratories performing Method 5 might differ substantially from the three
laboratories, Labs 101, 102, and 104, employed in this cement plant collaborative test. Hence it is
quite conceivable that the true between-laboratory, within-laboratory, and laboratory bias standard
deviations for Method 5 might vary from the estimates shown in Table 7 for cement plant appli-
cations by a factor of 2 in either direction. Since the hypothetical version's within-laboratory esti-
mates have 20 degrees of freedom, these estimates are more precise. It should also be noted that
real world collaborative testing places some extraordinary burdens on the collaborative teams that
result from sampling in cramped quarters on a fairly rigid schedule for long hours every work day for
two consecutive weeks. These burdens, which would not be experienced in compliance testing
of the method, would tend to cause additional operator errors and to thereby inflate the precision
estimates above the true precision of the method.
E. Sources of Variability in a Method 5 Test Result
The preceding section has disclosed that the lack of precision appears to be a major weakness
in the published version of Method 5, at least for its cement plant applications. The between-lab-
oratory standard deviation associated with a particulate concentration determination at a cement
plant is more than half as large as the true value of the determination itself. The other two Method 5
collaborative tests(4'5) reveal that imprecision is also a deficiency of Method 5 in fossil fuel-fired
steam generators and in incinerators. This section will analyze many of the sources of variability
in the estimated cement plant between-laboratory and within-laboratory standard deviations.
As Table 3 shows, the cement plant collaborative test yielded many questionable or unaccept-
able Method 5 samples (non-isokinetic sampling,deficient volume sampling, and/or high particulate
concentration determinations). Hence, the cement plant test offers a rare opportunity to evaluate
these sources under field conditions as contributors to the between-laboratory and within-laboratory
16
-------
precision variability in a Method 5 determination. One approach to ascertaining whether a given
factor is a source of variation is to compare the between- and within-laboratory standard deviations
calculated by a weighted coefficient of variation analysis, when the factor is allowed against the
comparable standard deviations, for the published Method 5, in which the factor is not allowed.
This approach is applied to the high value phenomenon, the isokinetic sampling restriction, and
the minimum sampling volume limitation. The results are reported in Table 8.
Table 8. The Effects of Eliminating the High Value Phenomenon and Relaxing the
Isokinetic Sampling and Minimum Sampling Volume Restmctiom
Data Base
Use of
High
Values
Isokinetic
Sampling
Restriction
Minimum
Sampling
Volume Limit
Between- Lab Standard Deviation
Estimated
Value
°b
Percent Change
from Current
Restriction
Within-Lab Standard Deviation
Estimated
Value
a
Percent Change
from Current
Restriction
High Value Phenomenon
Included
Excluded
90 < /< 110
90 < /« 110
56 scf
56scf
(0.584)6
(0.201)6
-66%
(0.284)6
(0.098)6
-66%
Isokinetic Sampling Restriction — With High Values Included
Included
Included
Included
90 «/< 110
85 < 115
80 « 120
56 scf
56 scf
56 scf
(0.584)6
(0.596)6
(0.575)6
+ 2%
-2%
(0.284)6
(0.360)6
(0.317)6
+ 27%
+ 8%
Isokinetic Sampling Restriction- High Values Eliminated
Excluded
Excluded
Excluded
90 < 110
85 < 115
80 < 120
56 scf
56 scf
56 scf
(0.201)6
(0.244)6
(0.275)6
+ 21%
+ 37%
(0.098)5
(0.147)5
(0.151)6
+50%
+ 55%
Minimum Sampling Volume Restriction
Excluded
Excluded
80 «/< 120
80 < 120
60 scf
56 scf
(0.270)6
(0.275)6
+ 1%
(0.158)5
(0.151)6
- 4%
The high value phenomenon is examined by comparing the between-laboratory standard
deviation oj, (e) = (0.201)5 and the within-laboratory standard deviation O(e) = (0.098)6 estimated
for the second situation discussed in the previous section (high values excluded, 90 < 1 10,
Vm > 56 scf) against the corresponding estimated values, Of, = (0.584)5 and a = (0.284)6, for
the published Method 5 situation (high values included, 90 < / < 110, Vm std > 56 scf). As
Table 8 shows, this comparison reveals that an estimated 66 percent reduction in the precision
standard deviations would occur if the high value phenomenon could be prevented.
The effect of relaxing the isokinetic sampling restriction is shown in Table 8 both when the
high value phenomenon occurs and when it is eliminated. When the high value samples are excluded
from the data, both the between-laboratory and the within-laboratory precision estimates undergo
pronounced deterioration as the isokinetic restriction is relaxed from 90<110to85<115
and 80 < 120. For example, the between-laboratory standard deviation estimate increases
from (0.201)5 for90< 110 to (0.244)6 for 85 < 115 (+21 percent), to (0.275)5 for
80 < / ^ 120 (+ 37 percent). Inclusion of the high sample values partially obscures this deter-
ioration. There is still a 27 percent increase in the within-laboratory standard deviation estimate
in relaxing from 90 < 110 (a =0.2846) to 85< 115 (a =0.3605). But the high values
obliterate the effect of isokinetic sampling relaxation on the between-laboratory precision. The
17
-------
high value exclusion analysis demonstrates the necessity of maintaining the Method 5 isokinetic
sampling restriction at 90 < / < 110: any relaxation of this restriction will exacerbate the imprecision
of Method 5.
To avoid confounding the effect of the minimum sampling volume restriction with the high value
phenomenon effect, the minimum sampling volume restriction is only examined with the high values
excluded. Table 8 shows that decreasing the minimum sampling volume limit from 60 scf to 56 scf
when 80 < / < 120 has little effect on either the between-laboratory or the within-laboratory standard
deviation estimates. Apparently, samples falling slightly below the minimum sampling volume limit
could be accepted without damaging the precision of Method 5.
Table 8 has shown that the high value phenomenon is the primary source of the between-
laboratory variability in a determination obtained at a cement plant, according to the published
version of Method 5. The portion of the between-laboratory variance caused by the high value
phenomenon is the reduction in the between-laboratory variance excluding the high values compared
against the between-laboratory variance calculated including them. These variance calculations are
shown in Table 9. Over 88 percent of the between-laboratory variance in a concentration deter-
mination obtained by the published Method 5 at a cement plant is caused by the high value pheno-
menon. Only 11.9 percent of the between-laboratory variance results from all other sources. //
Method 5 is to be substantially improved, the cause of the high value phenomenon must be isolated
and eliminated. Based on the evidence of Section B-2 of Appendix B, we believe scraping the
probe tip against the stack wall during insertion or removal of the sampling train is the cause of
the high value phenomenon.
Table 9. Sources of Between-Laboratory Variation in a
Method 5 Cement Plant Determination
Source of Between-Laboratory Variance
Total Between-Laboratory, a^
• High Value Phenomenon
• All Other Sources, crLe.
— Within Laboratory, a}e\
— Laboratory Bias, °L(e)
Estimated Variance
Component
(0. 3407)8 2
(0.3002)5 2
(0.0405)6 2
(0.0096)6 2
(0.0309)52
Percentage of
Total Variance, %
100.0
88.1
11.9
2.8
9.1
Table 9 also divides the between-laboratory variance from sources other than the high value
phenomenon into its within-laboratory and laboratory bias components. This breakdown allots 9.1
percent of the between-laboratory variance to laboratory bias sources and 2.8 percent to within-
laboratory sources. These percentages are 76 percent and 24 percent, respectively, of the other
source variance. Since the within-laboratory analysis is based on a data adjustment technique that
is probably too severe, it is probable that the actual percentage due to within-laboratory sources is
between 30 and 35 percent, instead of the calculated 24 percent. However, even at 65 to 70 percent,
the laboratory bias sources are still the major component of the other source variance. To improve
Method 5 beyond eliminating the high value phenomenon, the main emphasis should be placed on
further standardization of the method whenever possible so that each laboratory's performance of
Method 5 is more similar.
18
-------
IV. RECOMMENDATIONS
A number of problems relating to the precision and the usage of Method 5 in cement plant
applications have been discussed in the Summary and Conclusions section. The following recom-
mendations are made to alleviate these problems:
(1) The cement plant collaborative test appears to indicate that the precision of Method 5,
both between laboratories and within a single laboratory, is poor. In addition, all of the
laboratory teams participating in the collaborative test experienced difficulties using
Method 5. In view of these facts, the method should be rewritten. The revised version of
the method should incorporate the following suggestions or others that have the equi-
valent effect:
• Develop a procedure to prevent, or to at least inhibit, the occurrence of the high
value phenomenon. This phenomenon is probably caused by accidentally scraping
the probe tip against the particulate-laden stack wall while inserting or removing
the probe through a sampling port. Thus a partial solution is to examine the probe
tip for extraneous particulate matter each time it is removed from a port. However,
if the scraping hypothesis is true, particulate could also be scraped into the tip while
inserting the probe into a port, and sucked into the probe body or filter during the
sampling at that port. To prevent or inhibit this, some redesign of the probe tip
might be necessary.
• Standardize the method to reduce laboratory bias by specifying in greater detail how
procedures are to be performed. To minimize the possibility of sample loss or
contamination, both ends of the probe assembly prior to the filter should be capped
as soon as the probe is extracted from the last sampling port. Specific instructions
should be given for cleaning the probe assembly at the sample recovery area. Also,
the equation for calculating the stack gas pressure Ps should be specified.
• Retain the isokinetic sampling restriction 90 < / < 110. However, allow some latitude
in the minimum dry gas sampling volume limitation. For instance, it might only be
required that the average of the three test result samples' volumes exceed 60 scf, with
each sample volume at least 50 scf. This latitude would ease somewhat the difficulty
in obtaining acceptable Method 5 samples.
(2) After Method 5 is revised, it should be intensively field tested for reliability and ruggedness
by an initiating laboratory, as Youden'6-* recommends. The revised Method 5 should then
be subjected to a thorough collaborative test, at one or more sites, to determine its precision
and sources of variability.
(3) Thought should be given to developing standard data reporting forms suitable for keypunching
and an accompanying computer program for the revised version of Method 5. The raw
measurements from Methods 2, 3 and 5 would be recorded in specified units on the
reporting forms, keypunched, and input to the computer program which would calculate
the particulate matter concentration determination, the percent of isokinetic sampling, and
the other necessary results. This procedure would eliminate the serious Method 5 calculation
error problem, but it would be difficult to implement. Complicating factors include the lack
of metric measuring equipment, the voluminous amount and variety of data collected on
every sample, and the dependence of the number of traverse points, and hence, data points
needed on the size and configuration of the stack cross section.
19
-------
APPENDIX A
METHOD 5. DETERMINATION OF PARTICULATE EMISSIONS
FROM STATIONARY SOURCES
Federal Register, Vol. 36, No. 247
December 23, 1971.
21
-------
RULES AND REGULATIONS
Air PoiUitlaa .EBgiionettttg AtaBuaL, t>anl«U
son, I, A. , OS DHEW, pas, ffaWojHrt
Center Jos Atr Poll-atlas Control. Onclonatt.
OJjlfl, PBS Publication No, 909~AP<-40, J8*T,
Iweortdn^ Koward, et »t,, AJt Pflil-utiao
Soute« T-estlftg 5,t»oya1, Air Poltutjoft Coo-
Method* for
D«s».
>f Ve3oaltj.
iflst Gonleit of
Co., lx»
Caltf,.
Wp-450.,
METHOD 5 — DETERMINATION or PARTICVLATE
EMISSIONS FROM STATIONARY SOURCES
1. Principle and applicability.
1.1 Principle. Paruculat* matter is with-
drawn isokinetically from the source and its
weight Is determined gravimetrlcally afi*r re-
moval of uncomblned water.
1.2 Applicability. This method is applica-
ble for the determination of paniculate emis-
sions from stationary sources only when
specified by the test procedures for determin-
ing compliance with New Source Perform-
ance Standards.
2. Apparatus.
2.1 Sampling train. The design specifica-
tions of the paniculate sampling train used
by EPA (Figure 5-1 1 are described in APTD-
0581. Commercial models of this train are
available.
2.1.1 No:z!e — Stainless steel (316) with
sharp, tapered leading edge.
21.2 Probe — Pyrex * glass with a heating
sv5-em capable of maintaining a minimum
r.-s tcinper.itvire of 250' F. nt the exit end
dxiring sampling to prevent condensation
from occurrmfr. When length limitations
isrriMter than about 8 ft.) are encountered at
temperatures less than COO* F.. Incoloy 825 '.
cr equivalent, may be used. Probes for s-un-
plms gas streams at temperatures in excess
of 600° F. mu»t have been approved by the
Administrator.
2.13 Pilot tube — Type S. or equivalent,
attached to probe to monitor stack gas
wloclty.
' Trade name.
2.14 Filter Holdpr — Pyrex ' glatfl with
heating system capable of maintaining mini-
mum temperature of 225' F.
2.1.5 Implngers / Condenser— Four Impln-
gers connected In series with glasa ball Joint
nt tings. The first, third, and fourth Impln-
gers are of the Greenburg-Smlth design.
modified by replacing the tip with a '/2-lnch
ID glass tube extending to one-half Inch
from the bottom of the flaik. The second Im-
plnger Is of the Grccntaurg-Smlth design
with the standard tip. A condenser may be
used in place of the Implngers provided that
the moisture content of the t»tack gas can
still be determined.
2.1.6 Metering system — Vacuum gauge.
leak-free pump, thermometers capable of
measuring temperature to within 5" P., dry
gas meter with 27, accuracy, and related
equipment, or equivalent, as required to
maintain an Isokinetlc sampling rate and to
determine sample volume.
2.1.7 Barometer — To measure atmospheric
pressure to ±0.1 Inches Hg.
2.2 Sample recovery.
r wat« -rapor
la tlw «&,
r =Vo3«*n* of water -u&por _„„„,—
(standard cocdjttioos i. foxjcn^te voltusetr^ft proportioQ
of v»ter vapor ITJ tht ga»
22.1 Probe brush—At least aa long as
probe.
2.2.2 Olase wash bottles—Two.
2.2.3 Close sample storage containers.
2.2.4 Graduated cylinder—250 ml.
2.3 Analysis.
2.3.1 Glass weighing dishes.
2.3.2 Dcelocator.
2.3.3 Analytical balance—To measure to
±0.1 mg.
2.3.4 Trip balance—300 g. capacity, to
measure to ±0.05 g.
3. Reagents.
3.1 Sampling.
3.1.1 Filters—Glass fiber, MSA 1106 BH>,
or equivalent, numbered for Identification
and prewelghed.
3.1.2 Silica gel—Indicating type, 6-16
mesh, dried at 175- C. (350' F.) for 2 hours.
3.1.3 Water.
3 1.4 Crushed Ice.
3.2 Sample recovery.
3 2.1 Acetone—Reagent grade.
3.3 Analysis.
3.3.1 Water.
IMPINGER TRAIN OPTIONAL. MAY BE REPLACED
BY AN EQUIVALENT CONDENSER
PROBE
REVERSE-TYPE
PITOT TUBE
HEATED AREA FJLTER HOLDER / THERMOMETER CHECK
,VACUUM
LINE
IMPINGERS ICE BATH
BY-PASS.VALVE
THERMOMETERS
DRY TEST METER
Figure 5-1. Paniculate-sampling train.
3.3.2 Deslccant—Drierite.' indicating.
4 Procedure.
4.1 Sampling
4.1.1 After selecting the sampling site and
the minimum number of sampling points,
determine the stack pressure, temperature,
moisture, and range of velocity head.
4.1.2 Preparation of collection train.
Weigh to the nearest gram approximately 200
g. of silica gel. Label a filter of proper diam-
eter. desiccate5 for at least 24 hours and
woigh to the nearest 0.5 mg. in a room where
the relative humidity Is less than SOS-. Place
100 ml. of water In each of the first two
Impingcrs. leave the third Lmplnger empty.
and pliice approximately 200 g. of prewelghed
silica gel In the fourth impinger. Set up the
train without the probe as In Figure 5-1.
Leak chock the samplini: train at tlie sam-
pling site by plugging up the inlet to the ni-
ter holder and pulling a 15 In. Hg vacuum. A
leakage rate not In excess of 002 c.fJn. at a
vacuum of 15 In. Hg Is acceptable. Attach
the probe and adjust the heater to provide a
gas temperature of about 2f'05 F. at the probe
outlet. Turn on the Qlier heating system.
Place crushed Ice around the Implugers. Add
1 Trade name.
= Dry using Drleritc ' at 70' F.±
.±10" F.
more Ice during the run to keep the temper-
ature of the gases leaving the last Impinger
as low as possible and preferably at 70" F,
or less. Temperatures above 70' F. may result
In damage to the dry gas meter from either
moisture condensation or excessive heat.
4.1.3 Partlculate train operation. For each
run. record the data required on the example
sheet shown In Figure 5-2. Take readings at
each sampling point, at least every 5 minutes,
and when significant changes In stack con-
ditions necessitate additional adjustments
In flow rate. To begin sampling, position the
nozzle at the first traverse point with the
tip pointing directly Into the gas stream.
Immediately start the pump and adjust the
flow to Isokinetlc conditions. Sample for at
least 5 minutes at each traverse point; sam-
pling time must be the same for each point.
Maintain Isoklnettc sampling throughout the
sampling period. Nomographs are available
which aid In the rapid adjustment of the
sampling rate without other computations.
APTD-0576 details the proccdxirc for using
these nomographs. Turn off the pump at the
conclusion of each run and record the hnal
readings. Remove the probe and norL'le from
the stack and handle In accordance with the
sample recovery process described In section
4.3.
FEDERAL REGISTER, VOL. 36, NO. 247—THURSDAY, DECEMBER 73, 1971
23
-------
RULES AND REGULATIONS
21889
nju.7
wrr
KM HO.
unrria
fcCTIUH
C FACT W
TBAVWS row
Kuwtt
TOTAL
cano
1MO,
_
^
AVEU3E
STATIC
STACf
rrji.'f
AMTNTTfH
ASOJWC HC
*««ATiJ* ,_ ._
KTUK,*
i^imimtfnwa
i/:>f WATIC o» STACI c»w MCTKX
vnocm
friiSUK
office
to. HjO
i
i
VO.UW
tv»J. tr
™
41 DtTGXS WUI
IWUT
n. iKl.'r
i
1
*-,-
ourui
(T.^l. 'f
*-»
*^
*iri.i.
»^/tu^~.
tu^lf *01
or ui
ILIINC
CO»C>£KC*0»
LATT W»0t«
•F
4.2 Sample recovery. Exercise care in mov-
ing the collection train 'from the test site to
the sample recovery area to minimize the
loss of collected sample or the gain of
ertraneous paniculate matter. Set aside a
portion of the acetone used in the sample
recovery as a blank for analysis. Measure the
volume of water from the first three im-
pingers, then discard. Place the samples in
containers as follows:
Container No. J. Remove the filter from
its holder, place in this container, and seal.
Container No. 2. Place loose participate
matter and acetone washings from all
".mple-exposed surfaces prior to the filter
hi 1his container and seal. Use a razor blade,
brush, or rubber policeman to lose adhering
particles.
Container No. 3. Transfer the silica gel
Jiom the fourth impinger to the original con-
tainer and seal. Use a rubber policeman as
an aid in removing silica gel from the
Implnger.
4.3 Analysis. Record the data required on
the example sheet shown in Figure 5-3.
Handle each sample container as follows:
Container No. I. Transfer the filter and
any loose partlcxilate matter from the sample
container to a tared glass weighing dish,
desiccate, and dry to a constant weight. Re-
port results to the nearest 0.5 mg.
Container No. 2. Transfer the acetone
washings to a tared beaker and evaporate to
dryness at ambient temperature and pres-
sure. Desiccste and dry to a constant weight.
Report results to the nearest 0.5 mg.
Container No. 3. Weigh the spent silica gel
and report to the nearest gram,
5. Calibration.
Use methods and equipment which have
been approved by the Administrator to
calibrate the orifice meter, pilot tube, dry
gas meter, and probe heater. Recalibrate
after each test series.
6. Calculations.
6.1 Average dry gas meter temperature
and average orifice pressure drop. See data
sheet (Figure 5-2).
6.2 Dry gas volume. Correct the sample
volume measured by the dry gas meter to
standard conditions (70° F., 29.92 inches Hg)
by using Equation 5-1.
y _v
"" °
P ,
Pb"+
AH\
13.6)._
* /
AH
. equation 5-1
where:
Vtu,[d= Volume of gas sample through the
dry gas meter (standard condi-
tions), cu. It.
Vm = Volume of gas sample through the
dry gas meter (meter condi-
tions) , cu. ft.
T,td=Absolute temperature at standard
conditions, 530° R.
T^~ Average dry gas meter temperature.
'R.
Pb.,— Barometric pressure at the orifice
meter. Inches Hg.
AH = Average pressure drop across the
orifice meter. Inches H2O.
13.6— Specific gravity of mercury.
Pild-- Absolute pressure at standard con-
ditions, 29.92 inches Hg.
6.3 Volume of water vapor.
(»
0474C"-,ft>,
ml. / •
equation .7-2
where:
Vw(ld—Volume of water vapor in the gas
sample (standard conditions),
cu. ft.
Vi,= Total volume of liquid collected in
impingers and silica gel (see Fig-
ure *-3), ml.
ptt o= Density of water. 1 g./ml.
Mn/i = Molecular weight of water, 18 lb./
Ib.-mole.
R= Ideal gas constant, 21.83 inches
Hg—cu. ft./lb.-mole-°R-
T.,,, = Absolute temperature at standard
conditions, 530* R.
p>ta = Absolute pressure at standard con-
ditions, 29.92 inches Hg.
6.4 Moisture content.
equation S -3
whoie:
= Proportion l.y volume of water vapor in tin- ir:is
strt-am, dinn'Tisionloss.
^ *,L.i=Volume of water in the pas sample (stan-Kird
conditions,), cu. ft.
^
-------
21890
RULES AND REGULATIONS
PLANT.
OATE_
RUN NO.
CONTAINER
NUMBER
1
2
TOTAL
WEIGHT OF PARTICULATE COLLECTED.
mg
FINAL WEIGHT
J^XL
TARE WEIGHT
:XL
WIGHT GAIN
FINAL
INITIAL
LIQUID COLLECTED
TOTAL VOLUME COLLECTED
VOLUME OF LIQUID
WATER COLLECTED
IMPINGER
VOLUME,
ml
SILICA GEL
WEIGHT.
9
g* ml
CONVERT WEIGHT OF WATER TO VOLUME BY DIVIDING TOTAL WEIGHT
INCREASE BY DENSITY OF WATER. (1 g. ml):
INf,REff.-9 = VOLUME WATER, ml
(1 g/ml)
Figure5-3. Analytical dala.
6.6.3 Concentration in Ib./cu. ft.
•.= Col'.conlrjtiun of p.irlKul.it.'IM.III.'! i:i '.
0=Mf'fb.
tfV.l', \ .
.
ViB—Toial volume of ll'inl'l coll«xt«l In Impln^rs
and irtJkJ* fl (fc^ Up. fr-3), ml.
*H,O~L>Gnft!ly Of W*lfJ, 1 £ /(111.
It-Mra) rafl consuuil, 21.83 Inclica HK-CU. ft /l!i.
rnolir • 11.
MH^i«=MoUjculjj wlt'M of watrr. 18 ]b /Ib.-nuile.
V»—Volume of cw^rnplo through tlic dry gus mHT
(ni''lT t»ll'lit1OIL«), CU. ft.
T«">Absolute avtru'o Ory gas ratter t*mpir.itiire
(MI> Knu-ci 2). "It.
Pb,r= B.irf.in1 tile pn..Tiire dr'.p a'.T&ss the orlflr« (s««
Hi' 5-i), Inchis Ili«.
T. = Altv,ltit« avfrri;-e stJfk gas f mfK-rature (see
hip. i-2).°U.
« = Tot-il s-tnipllnc tlrnp, mln.
V.^SIaik I'tS v.lo.ny rakul.itrd by Mi'lt.ijil 2.
Eqii:tl<.n 2-2. [1 fti-c.
P. = Atj^jjule slryk ct»3 prt.'surf*, Inrhcs Hp.
AB = Cro^-J si-ctl'jiial iirt-a of nozzle, s^. fl.
6.8 Acceptable results. The following
range sets the limit on acceptable Isoklnetic
sampling results:
If 90^
-------
APPENDIX B
STATISTICAL METHODS
27
-------
APPENDIX B. STATISTICAL METHODS
This appendix is composed of various independent sections, each of which contains a statistical
analysis pertinent to a particular question or problem encountered in the analysis of the Method 5
collaborative test data. Reference to these sections has been made at various junctures in the Statis-
tical Design and Analysis part of the main report.
B.1 Preliminary Data Analysis
An initial scan of the original participate concentrations reported by the four collaborators and
shown in Table 2 of the main report disclosed consistently high values from Lab 103. No calculation
errors could be located to account for this discrepancy. As discussed in Section III.C of the main re-
port, it was discovered during a subsequent visit that Lab 103 had included extracted condensibles
with its probe participate catch to obtain its total collected particulate matter. Consequently, Lab 103
had to be excluded from the entire data analysis.
An outlier analysis was performed on the original particulate data reported by the three remaining
collaborators. When any of the three collaborators' values for a run differed by more than 10 percent
from either of the other two, the Method 5 particulate matter calculation of the outlying value was
rechecked. During this process, some calculation errors and frequent deficient round-off procedures
were discovered. Since the calculation deficiencies were so prevalent, all the calculations pertinent to
Method 5 were recomputed from the raw data for each collaborator on every run. These pertinent
calculations include the corrected gas volume Vm ,, the particulate matter concentration c, and the
percent of isokinetic sampling/. The recalculated values for c, I, and Vm std are displayed in Table 3
in the main report.
Even a cursory comparison of Table 3 against Table 2 reveals that calculation is a major source of
error in Method 5. Hence, an examination of where mistakes and excessive intermediate round-off
commonly occurred in the calculations is appropriate. Lab 101 consistently misused the stack gas
temperature (Ts)^vg and velocity head (v/Ap)avg from the preliminary traverse for setting the nomo-
graph rather than from the actual run sampling in computing the stack gas velocity (Fj.)avg and the
percent of isokinetic sampling/. Lab 102 calculated the stack gas pressure/^ incorrectly, correcting
the barometric pressure by the tabulated orifice meter average instead of using the tabulated static
pressure average; since no formula is presented in the "Standards of Performance for New Stationary
Sources" for computing Ps, this or some similar Ps calculation error could frequently occur. Lab 104
was inconsistent in the application of its dry gas meter volume correction factor. All three laboratories
were somewhat remiss in rounding intermediate calculations, but Lab 101 often rounded the ratio
Mn/Vm d of particulate mass to gas sampled to one significant digit; never did it retain more than two
significant digits in this critical factor of the particulate concentration calculation equation 5-5 (cf.
Appendix A). In computing Vm^A, Lab 101 always truncated the dry gas meter volume-to-tempera-
ture ratio (Vm /Tm ) to two digits before inserting this factor in the Vm std calculation equation. In
computing the velocity head (^/Ap),dVg, Labs 101 and 104 only summed the two most significant digits
from the square roots of the Ap readings; this frequently introduced errors of 1- to 2-percent in the
velocity head calculation, which were carried directly in the (Fx)avg and / calculations. Lab 101 only
carried its calculation of the nozzle areaAn to one significant digit, thereby introducing a 14 percent
positive error into its calculation of/.
B.2 The High Particulate Concentration Determination Phenomenon
Table 3 contains four extremely high particulate concentration determinations, all of which were
obtained by Lab 101, on Runs 4, 9, 10, and 13. Since no violation of Method 5 procedures could be
ascertained, these must be considered valid Method 5 determinations. However, because they are bound
to cause the majority of the Method 5 precision variability reported for this collaborative test, it is
important to find out why they occurred.
29
-------
This high participate concentration phenomenon has occurred in all three of the Method 5
collaborative tests that have been conducted. In addition to the four high values in this cement plant
test, six high values were obtained by the three included laboratories in the Allen King Power Plant
collaborative test(7), and two high values occurred in the Holmes Road incinerator collaborative test(8)
An examination of these 12 high determinations revealed that, in 11 of the 12 cases, the reason for
the high particulate concentration was an extraordinary amount of particulate matter present in
the acetone washings. The pertinent data are shown in Table B-l. The only high sample that had a
normal particulate content in its acetone wash was Run 3 by Lab 104 in the incinerator plant test.
Table B-l. Examination of High Particulate Concentrations from
All Three Collaborative Tests of Method 5
Site
Cement
Plant
Power
Plant
Incinerator
Lab
101
101
101
101
102
102
103
104
104
104
101
104
Run
4
9
10
13
7
9
7
1
3
11
8
3
Particulate Matter
Cone.
10-' Ib/scf
137.0
148.8
137.4
214.2
335.7
405.3
313.9
334.4
375.1
351.2
469.0
374.6
Mass, mg.
Total
387.4
394.4
383.3
609.2
1065.7
1270.5
1036.8
878.3
1349.9
1188.4
1288.0
1050.0
Acetone
Wash
374.1
381.1
371.0
510.0
581.3
788.7
490.6
542.9
1012.2
611.2
675.0
30.5
Normal Acetone Wash
Range for Lab
12 to 50
12 to 50
12 to 50
12 to 50
200 to 450
200 to 450
80 to 420
80 to 270
80 to 270
80 to 270
30 to 225
10 to 100
Median
for Lab
35
35
35
35
250
250
180
150
150
150
60
40
Excessive
Acetone Wash
340
345
335
475
330
540
310
390
860
460
615
-10
On all the other 11 high determination samples, which we will term the high acetone wash samples, the
laboratory found much more particulate matter in these acetone washings than in the acetone washings
of any of the normal concentration range samples it obtained in the test. The excessive amount of
particulate matter in the acetone washings of these samples over the laboratory median is listed in the
last column of Table B-l. For the 11 high acetone wash samples, the excessive particulate matter in
the acetone wash ranges from 310 mg to 860 mg, with a median of 390 mg.
In our opinion, the cause of the excessive amount of particulate matter in the acetone wash of
these 11 samples is the scraping of particulate matter from the stack wall into the probe tip dur-
ing the insertion or removal of the sampling train from a sampling port. This scraping hypothesis
explains all the observed features of the high value phenomenon. If the probe tip did scrape up partic-
ulate matter as the sampling train was being inserted or removed, this excessive particulate matter would
be washed from the probe during sample recovery and would consequently appear in the acetone
washings. Scraping of the probe tipe against the stack wall would only occur occasionally; therefore,
our hypothesis does account for both the discrete nature (either obviously present or entirely absent)
and the infrequent occurrence of the high value phenomenon. Since the likelihood of contaminating
the probe tip by scraping would vary with the dirtiness of the stack inner wall and with the carefulness
of the sampling team, the scraping hypothesis does explain the observed variation in high value fre-
quency from one site to another and from one laboratory team to another. Finally, brushing the probe
tip against the stack wall would scrape a considerable amount of particulate matter into the tip; the
scraped amount could well equal the observed excessive amount which ranged from 310 mg to 860 mg.
Thus, the scraping hypothesis is the probable explanation for the high particulate concentration
determinations that inflate the precision of Method 5.
30
-------
B.3 Examination of the Final Port Effect
Each collaborative team was supposed to sample isokinetically for 30 min from each of the four
ports in taking a Method 5 sample; therefore, one would not expect a port effect with Method 5.
However, it is believed that on some samples, the probe tip was accidently scraped against the inner
stack wall during insertion or removal of the probe, thereby contaminating the tip with excessive
particulate matter. Because there is a possibility of a port effect in this collaborative test, the question
was subjected to a statistical test.
Youden's Rank Testt9) is used to determine whether there was any significant detectable final
port effect. The test for final port effect is presented in Table B-2. Each port is assigned a rank on
each run based on the particulate matter concentration determined for the sample having that port as
its final sampling port. The final ports corresponding to all rejected samples (Lab 103 determinations,
missing values, or unacceptable isokinetic sampling at 80 < / < 120) are assigned median ranks of 2.50.
The other final ports are then assigned the ranks
Table B-2. Youden Rank Test for
Significance of Final Port Effect
Run
i
->
3
4
5
6
7
8
9
10
11
12
13
14
15
Final Port Rank Scores:
Null Hypothesis:
Alternative Hypothesis:
Approximate 5 Percent
Two-Tailed Limits:
Conclusion:
Ranked Data by Final Port Sampled
Port A
1.25
1.25
1.25
1.25
2.50
2.50
2.50
3.75
3.75
2.50
2.50
2.50
3.75
2.50
1.25
35.00
Port B
2.50
3.75
3.75
2.50
3.75
3.75
1.25
2.50
2.50
3.75
2.50
3.75
2.50
3.75
2.50
45.00
Port C
3.75
2.50
2.50
3.75
2.50
2.50
2.50
1.25
2.50
2.50
1.25
2.50
2.50
2.50
3.75
38.75
Port D
2.50
2.50
2.50
2.50
1.25
1.25
3.75
2.50
1.25
1.25
3.75
1.25
1.25
1.25
2.50
31.25
The true particulate concentration
sampled is equivalent, regardless of
the final port sampled.
The particulate concentration sam-
pled differs systematically, depend-
ing on the final port sampled.
(28.1,46.9)
Because there are no final port
scores outside the 5 percent limits,
the alternative hypothesis is rejected
in favor of the null hypothesis.
There is no significant final port
effect.
1.25, 2.50, and 3.75, according to the correspond-
ing concentration determinations. The port ranks
are summed over all 15 runs to give the final port
rank score.
The null hypothesis to be tested is that there
is no final port effect, i.e., that the true particulate
concentration sampled is equivalent, regardless of
the final port used. This hypothesis should be re-
jected in favor of the alternative that there is a sys-
tematic final port effect, at the 0.05 significance
level, if any of the final port rank scores fall out-
side the two-tailed limits of 28.1 and 46.9. Since
all the final port rank scores (35.00, 45.00, 38.75,
and 31.25) are between 28.1 and 46.9, the null
hypothesis should not be rejected. Thus, any sig-
nificant final port effect is not detectable.
B.4 Grouping the Samples into Blocks
Even with the high values eliminated. Table 3
shows tremendous variation in the Method 5 par-
ticulate concentration determinations from one
run to another. It is reasonable to assume that
Method 5 concentration determinations for vari-
ous runs reflect the ordering of the true particu-
late concentrations in the stack while those runs
were taken. With this assumption, the
Kruskal-Wallis//Test(10) can be applied to the
acceptable Method 5 concentration determina-
tions for the 15 runs to determine whether the
true particulate concentration did indeed vary
from run to run. This H Test is summarized in
Table B-3. Acceptable Method 5 concentration
determinations for this test are those sampled
isokinetically (90 < /< 110) and not high values.
31
-------
Table B-3. Kmskal- Wallis H Test for Differences in the True
Paniculate Concentration from Run to Run
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Acceptable Method 5
Particulate Cone. Determinations,*
10"7 Ib/scf
12.9, 13.0, 16.0
13.1,14.5
10.3, 10.9, 12.4
18.8
10.8
8.6, 10.5
9.4
15.5,15.9, 29.2
6.8,9.0
12.3
20.7,28.7
33.4,38.9
52.3,75.3
19.8
26.9,27.0
Null Hypothesis:
Alternative Hypothesis:
H Test Statistic'
Approximate 5 Percent
and 2.5 Percent Limits:
Conclusion :
Ranks of
Determinations
11, 12.5,17
12.5,14
5,8, 10
18
7
2,6
4
15,16,24
1, 3
9
20,23
25,26
27,28
19
21,22
Number
Accept. Determ.,
3
2
3
1
1
2
1
3
2
1
2
2
2
1
2
TV =28
Rank
Sum,
40.5
26.5
23
18
7
8
4
55
4
9
43
51
55
19
43
The true particulate matter concentration of every run is the
same.
The true particulate matter concentration differs from one run
to another.
12 15 /*A 12
fj ^ — > l 1 \ •} / A/" _i_ 1 \ ( H £L -\ Z C\ Q/ T Q\ "") ^ xj.95 (14)= 23.7, the null hypothesis is
rejected at the 0.05 significance level. The true particulate
matter concentration does differ from one run to another.
*Acceptable under the isokinetic restriction 90 < I < 110; high values eliminated.
Each acceptable determination is assigned a rank in ascending concentration order. Rank sum Rt is
computed for run / as the sum of its nt ranks. With a total of TV = 28 ranks assigned, the H Test sta-
tistic is
H =
12
N(N+ 1) £*
-3(/V+ 1) = 25.54
It tests the null hypothesis of identical true particulate matter concentrations on every run; a high
H statistic rejects this null hypothesis. Since H= 25.54 exceeds the 5 percent limit x?9 5 (14) =
23.7, the null hypothesis is rejected at the 0.05 significance level. The true particulate matter concen-
tration did differ from one run to another. Grouping the runs into blocks having approximately the
same true particulate concentration is the first-order statistical technique for coping with this true
variation problem.
32
-------
The approximate ordering of the true particulate matter concentrations for the 15 runs can be
estimated from the appropriate cement plant operating conditions. Data on two operating variables,
the daily feedrate and the stack opacity, are available. The daily feedrate is calculated from the tons of
dry feed materials input to the cement plant each day and from the number of kilns (two or three) in
use that day. Since the daily feedrate is a 24-hour average, it provides little direct information about
the true particulate concentration in the stack during the 2 hours in which a run was made. The opacity
of the stack plume being tested was measured by an in-plant bolometer on a strip chart recorder. Both
the peak opacity recorded during the sample's collection and the time-weighted average opacity over
the sampling period are useful measures of the true particulate concentration. The daily feedrates, the
time-weighted average opacity, and peak opacity for each run are listed in Table B-4. Each run is
assigned a feed rank Rf, an average opacity rank Ra, and a peak opacity rank Rp based on the values of
the corresponding operating condition variables. A rank score is computed for each run from the
Table B-4. Run Particulate Concentration Ranks from Cement Plant Operating Conditions
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Date
4-2
4-3
4^
4-5
4-5
4-6
4-9
4-9
4-10
4-10
4-11
4-11
4-12
4-12
4-13
Daily Feed Rate
Tons Dry Feed
per Kiln
481.7
511.3
533.5
573.1
573.1
517.0
604.9
604.9
548.1
548.1
603.6
603.6
609.4
609.4
602.4
Rank,
Rf
1
2
4
7.5
7.5
3
12.5
12.5
5.5
5.5
10.5
10.5
14.5
14.5
9
Opacity from Transmissometer
Time Wtd. Average
Average, %
9.5
10.5
10.0
14.0
10.7
9.0
10.5
13.0
9.5
13.0
9.5
14.0
15.5
14.5
13.3
Rank,.Ra
3
6.5
5
12.5
8
1
6.5
9.5
3
9.5
3
12.5
15
14
11
Peak Opacity
Peak, %
26
29
12
20
16
42
16
66
12
16
39
69
79
21
88
Rank,^p
8
9
1.5
6
4
11
4
12
1.5
4
10
13
14
7
15
Rank
Score,
Rf+Ra + 2Rp
20.0
26.5
12.0
32.0
23.5
26.0
27.0
46.0
11.5
23.0
33.5
49.0
57.5
42.5
50.0
formula, rank score = Rf + Ra + 2Rp. The peak opacity rank Rp receives a double weight because it
is believed to be the best particulate concentration indicator. The ranks and rank scores for each run
are also presented in Table B-4. The ordering of the runs that is produced by the rank score is shown
in Table B-5. Our assumption is that the rank score provides an objective criterion for ordering the runs
according to their true particulate concentration. Since the correlation between the rank score in the
first column of Table B-5 and the collaborators' Method 5 mean in the last column of Table B-5 is
r = 0.8144, an extiemery significant correlation, this assumption seems justified. Thus, Runs 9 and 3,
with rank scores of 11.5 and 12.0, respectively, had the lowest average true particulate concentration,
while Run 13 with a 57.5 rank score had the highest true concentration.
The runs are grouped into blocks by their rank scores. Listing the rank scores in ascending order
in Table B-5 shows two useful breaks, between 27.0 and 32.0 and between 50.0 and 57.5. These
break points divide the runs into the three groups that minimize the within-group mean square varia-
tion of the rank score-ordered Method 5 run means. Thus these break points are used as the block
boundaries. Therefore, as TableB-5 shows, Block 1 consists of Runs 9, 3, 1, 10, 5, 6, 2, and 7. Runs
4 11; 143 8, 12, and 15 comprise Block 2. Run 13 is in a block by itself, Block 3.
33
-------
Table B-5. Blocking the Runs by Rank Score
B.5 Data Adjustment for True Concen-
tration Variation Within Block
It is demonstrated in Section B.6 that the
proportionality of standard deviation to mean is
characteristic of Method 5 data. The between-
laboratory proportionality ob = fe5 is derived
from analysis of the Method 5 run results, while
the within-laboratory proportionality a = |3§
emerges from a study of collaborator block
results. By definition of the laboratory bias_
standard deviation CT/, = j3/,5 where j3/, =<
Thus the coefficient of variation is a legitimate
device for expressing the Method 5 variation.
In the Method 6 collaborative test report/11 -
it was shown that the within-to-between coef-
ficient of variation ratio j3/fo should usually lie
between 0.3 and 0.95. Values of j3/j3j > 1.0
result in a zero laboratory bias standard deviation estimate, OL = o, implying that all laboratories obtain
equivalent Method 5 concentration determinations.
Rank
Score
11.5
12.0
20.0
23.0
23.5
26.0
26.5
27.0
32.0
33.5
42.5
46.0
49.0
50.0
57.5
Run
9
3
1
10
5
6
2
7
4
11
14
8
12
15
13
Block
1
1
1
1
1
1
1
1
2
2
2
2
2
2
3
Collaborators' Method 5 Mean
with High Values Eliminated
80
-------
Table B-7. Youden Rank Test for Significance
of Laboratory Effect
Run
9
3
1
10
5
6
1
1
4
11
14
S
12
15
13
Lab Rank Scores
Null Hypothesis
Alternative Hypothesis
Approximate 5 Percent
Two-Tailed Limns
Conclusion:
R,in>"-f1 T-infat
Lab 101
3
3
3
3
3
3
3
3
3
1
3
3
34
Lab 102
1
1
1
1
1
1
1
1
2
3
1
2
16
Lab 104
2
2
2
2
2
2
2
2
1
2
2
1
22
All laboratories perform Method 5
in an equivalent manner
Method 5 determinations differ
systematically from laboratory to
laboratory
(18 1,299)
Since the Lab 101 and Lab 102 rank
scores both lie outside the two-
tailed limits, one must reje t the
nulj hypothesis in favnr of it alter-
native Melhod 5 possesses a bora-
tory effect which produces sy tema-
tic differences in ihe Me lod 5
determinations of different bora-
tones.
*Only data acceptable under the isokinetic restriction 90 < I
< 110 are included.
Because the true particulate concentration did
differ substantially, even on the runs in the same
block, the within-laboratory standard deviation estimates
obtained from a straightforward coefficient of variation
analysis of the acceptable Table 3 data will be grossly
inflated. Thus to obtain more realistic within-laboratory
estimates, the data need to be adjusted for the true con-
centration before making these estimates. However, the
true particulate matter concentration existing in the stack
during a run is unknown. No valid particulate concen-
tration monitor took simultaneous readings during the
runs. Even the approximating total rank score developed
in Section B.4 and presented in Table B-5 is too crude
for this purpose— it is essentially an ordering rather than
a measuring variable. The only valid indicator of the
true particulate matter concentration during a run is the
collaborators' Method 5 run mean. With the high value
samples eliminated, these run means are listed in
Table B-5 for both the 80 < / < 1 20 and the 90 < /
isokinetic acceptance ranges. We will assume that the
differences between these run means, for the runs in a
block, are equal to the differences in the true concen-
trations for these runs. A workable data adjustment
technique for obtaining realistic, albeit somewhat con-
servative, within-laboratory estimates results from this
assumption. Each collaborator's determination on a run
is adjusted by the amount necessary to give the run the
same mean as its block. So, every run in a block of
adjusted data has the same run mean. The data adjust-
ment formula used is
1 10
where .v,y is the corrected concentration determination
reported in Table 3 for collaborator / on run /, .Y ; is the
run/ mean, and x__ is the block mean. This adjustment
formula is applied to the Block 1 and Block 1 determina-
tions in the acceptable 90 < / < 110 isokinetic range,
with the high values eliminated. The resulting adjusted
data are presented along with its run summary in Table B-8. The corresponding collaborator
block summary is shown in Table 6 of the main report. In Table B.8, the symbol E follows an
eliminated high value, M denotes a missing determination, and R identifies an isokinetically rejected
sample. The jS/jSj ratios of Blocks 1 and 2 are reduced from 1.40 and 1.10 for the corrected data to
0.44 and 0.76 for the adjusted data. So the p/f}/, ratios for the adjusted data are in the usual range.
Although the adjustment technique is probably too severe, the adjusted data provide a more suitable
basis on which to perform the within-laboratory precision analysis.
35
-------
Table B-8. The Adjusted Method 5 Data and Its Run Summary
Run
9
3
1
10
5
6
2
7
4
11
14
8
12
15
13
Acceptable Method 5 Particu-
late Concentration Determina-
tions with High Values
Excluded, 10"' Ib/scf
Lab 101
E
12.3
13.2
E
11.1
12.1
11.8
R
E
28.4
M
33.4
21.7
24.5
E
Lab 102
10.0
10.2
10.0
11.1
M
10.2
10.4
R
24.4
20.4
24.4
20.1
27.2
24.4
75.3
Lab 104
12.2
10.8
10.2
R
R
R
R
11.1
R
R
R
19.7
R
R
52.2
Run Summary
MG3T1 ,
X.j
11.10
11.10
11.13
11.10
11.10
11.15
11.10
11.10
24.40
24.40
24.40
24.40
24.45
24.45
63.75
Standard
Deviation,
si
1.56
1.08
1.79
1.34
0.99
5.66
7.80
3.89
0.07
16.33
Coef. of
Variation,
&i
0.1756
0.1100
0.1817
0.1510
0.1118
0.2906
0.3606
0.1994
0.0036
0.3211
Note:
E-Excluded determination: high concentration probably caused by scraping
probe tip against stack wall.
M— Missing determination.
R-Rejected determination: outside acceptable isokinetic sampling range,
90< 110.
B.6 Relationship of the Standard Deviation to the Mean
The coefficient of variation approach is an ideal procedure for estimating the precision of a test
method from collaborative test data, both because it yields percentage precision estimates which the
non-statistician can quickly comprehend, and because estimates of the basic between-laboratory and
within-laboratory standard deviations derive naturally from it. Application of the coefficient of varia-
tion procedure requires justification of its fundamental assumption, namely, that the precision
standard deviations of the method's determinations are actually proportional to their mean value. The
precision standard deviations for which the proportional relationship needs to be verified are the
between-laboratory standard deviation for run results and the within-laboratory standard deviation for
collaborator block results. This section presents the necessary verification information for Method 5
from the corrected cement plant collaborative test data.
The direct evidence for proportionality comes from plotting and regressing the pertinent stan-
dard deviation versus mean data. The values of the between-laboratory standard deviations and means
for each run are presented in the run summaries of Tables 4 and 5 in the main report. Table 4 includes
the four high values, along with all the 90 < 110 isokinetically acceptable particulate concentra-
tion determinations. Table 5 excludes these values, but retains the 90 < 110 isokinetic sampling
limitation. The run standard deviation and mean results from Table 4 are plotted in Figure B-l.
These data do not exhibit a very proportional relationship because the four high value runs, having
standard deviations between 80 and 90 X 10~7 Ib/scf, are so distinct from the other runs. Removing
the high concentration determinations from these four runs yields the corresponding Table 5 results
which are plotted in Figure B-2. A straight line passing through the origin provides a good fit of the
Figure B-2 points; this suggests a valid proportional relationship. A qualitative assessment of the
36
-------
U)
-0
10 r-4- I -"-i
90 100
y, Run Mean, 10~7 Ib/scf
Figure B-l. Between-Laboratory Run Plot
-------
OJ
00
80 •
Run Standard Deviation
10-7'Tb/scf !
10 --r
90 100 110
ry, Run Mean, 10"7 Ib/scf
120
Figure B-2. Between-Laboratory Run Plot Excluding High Values
-------
proportionality of run standard deviation to mean is obtained by comparative regression of the run
standard deviation against various functions of the mean. The functions of the mean x, that are
regressed throughjhe origin against the standard deviation s, as s = af(x), are x'1, x1' 4, x l' 2 ,
"I, ,x-x ' ,x 3/2,.^2, and \ogx A summary of the regression results obtained is presented in
table B-9. For the Table 4 run results (high values included, and a 90 < 110 isokinetic sampling
restriction), the regression of .Y 5/4 against 5 gives the best correlation, rB = 0.9338, but the proportional
Table B-9. Proportionality Evidence from Comparative Regress
•ions
Type of
Std. Dev.,s,
vs. Mean, x,
Results
Repressed
Run
Collaborator
Block
(Adjusted Data)
Data Base
Source
4
5
6
Use of
High
Values
Included
Excluded
Excluded
Excluded
Excluded
Isokinetic
Sampling
Restriction
90 < 110
90 < 110
80 < 120
90 < 110
80 < 120
Regression Summary*
Best Single Variable
Regression through
Origin
Best
Variable
-5/4
~3/2
X
P
X
Correlation
CoeL,rB
0.9338
0.9241
0.9594
0.9874
0.9759
Mean Regression
through Origin
Equation
Coef., a
0.8969
0.2000
0.2623
0.1351
0.1446
Correlation
Coef.,r-
0.9283
0.9001
0.9594
0.9346
0.9759
*The functions of the mean x regressed against the standard deviation s as s = affx) are x~ ' , x"4, x"2.
x3'4 , x, xs'" , x3'2 , x2 , and logic.
relationship s - 0.8969 x , with a correlation coefficient r- = 0.9283, is almost as good. For the
Table 5 run results, with high values excluded and the 90 < / < 110 isokinetic sampling limitation,
.x~3/2 gives a considerably better correlation to s (rB = 0.9241) than the mean x itself (r- =0.9001).
But when the isokinetic sampling limitation on the Table 5 run results with high values excluded is
relaxed to 80 < 120 allowing the utilization of all concentration determinations, then the mean
x yields the best regression s = 0.2623 x with r- = 0.9594. When enough concentration data are
available, comparative regression also supports the idea that the run standard deviation is proportional
to the run mean.
The within-laboratory standard deviations and means for each collaborator block of the adjusted
data of Table B-8 are shown in Table 6 in the main report. The four high values are excluded, and
all samples collected outside the isokinetic limitation 90 < 110 are rejected from the Table B-8
data from which the Table 6 results are calculated. The collaborator block standard deviation is
regressed against various functions of the collaborator block mean, according to the procedure de-
scribed in the previous paragraph. As the Table B-9 summary shows, the mean square _v~2 produces a
considerably better regression through the origin (rB = 0.9874) than the mean regression (r- = 0.9346).
But again, when the isokinetic restriction is relaxed to 80 < / < 1 20 to permit enough data to support
a firm conclusion, the mean regression s - 0.1446 x , with r- - 0.9759 as its correlation coefficient,
proves best. It thus appears that the collaborator block standard deviation is also proportional to its
mean.
Indirect evidence regarding the proper relationship of standard deviation to mean is obtained
from Bartlett's Test for equality of variance/12) This test is used to evaluate the linear (no transforma-
tion of x), the logarithmic (log! Ox), and the square root (>/x) transformations as procedures for
39
-------
achieving equality of variance for both the run and collaborator block comparisons of the appropriate
data. The Bartlett's Test results are shown in Table B-10. On the run evaluations of both the Table 4
and the Table 5 data, the logarithmic transformation yields better equality of between-laboratory
Table B-10. Adequacy of Alternative Transformations to Achieve Equality of Variance
Type
of
Comparison
Run
Collaborator
Block
(Adjusted Date)
Data Base
Source
Table
4
5
6
Use of
High
Values
Included
Excluded
Excluded
Isokinetic
Sampling
Restriction
90«/«110
90 <110
90 «/«110
Test for Equality of Variance
Transformation
Linear: ;f =x
Log:jF =log10x
Sqr. root: y = \Jx
Linear : y = x
~Log:y = log,0x
Sqr. root: y = \fx
Linear : y = x
Log:y =log10x
Sqr. root: y = ^fx
Bartlett's
Test
Result
52.8
32.8
40.9
20.9
10.7
14.2
25.4
9.8
17.0
Significance
Pr{x2(H) >52.8
Pr
Pr
Pr
Pr
Pr
Pr
Pr
Pr
X2(1D»32.8
fx2Ul)>40.9
(
X2(9)»20.9}
X2(9)»10.7}
X2(9)» 14. 2\
(X2(4)»25.4)
[x2(4)»9.8}
|X2(4)> 17.0}
< 0.0001
= 0.0006
< 0.0001
= 0.02
= 0.30
= 0.13
< 0.0001
= 0.04
= 0.002
variance than either the linear or the square root transformations. Because the high values are ex-
cluded from the Table 5 data, there is a 30-percent chance that an actual underlying equality of run
variance process would have produced data with as much between-laboratory scatter as that exhibited
by the logarithmically transformed Table 5 data. The collaborator block transformation evaluation
based on the adjusted data summarized in Table 6 shows that the logarithmic transformation is also
superior in producing equality of within-laboratory variance. In the first Method 7 collaborative test
report( 13), it was demonstrated that when the logarithmic transformation of a set of data yields equal-
ity of variance (so that the data probably have a lognormal underlying distribution), then the standard
deviation is proportional to the mean. Thus, the results of Table B-10 imply that both the run and the
collaborator block standard deviations are directly proportional to their respective means.
The preceding evidence enables us to conclude that the proportionality of standard deviation to
mean is characteristic of Method 5 concentration determinations at cement plants, both between labor-
atories (ai, = j3fc6) and within a single laboratory (a = |36). The proportional between-laboratory relation-
ship follows from the run results discussed above, while the within-laboratory relationship is inferred
from the collaborator block results. Hence, the assumption of proportionality of standard deviation to
mean, which is the fundamental prerequisite to use of the coefficient of variation procedure, is justified.
B.7 Weighted Coefficient of Variation Estimation
The usual procedure for estimating the between-laboratory coefficient of variation fo for a test
method from a collaborative test is to average the run coefficient of variation estimates 0, from each
of the K runs with a sample size of more than n,- = 1 collaborator's determination:
Kj-l '
Although Sj is a biased estimate of oy, the correction factor an(J) applies the proper adjustment so that
is unbiased for oy and
40
-------
is unbiased for 0 when the run / determinations have a normal distribution.
Now intuitively one feels that the run estimates j3/ based on a large sample size nf, i.e., on simul-
taneous determinations by a large number of laboratory teams, are better than those based on fewer
determinations. With large nf, the estimate j3/ should have smaller variance. Indeed, this is true.
Var(fy) = Var
\ •*/
/ S'
~°(-nf\ Var f~^~} since a.n ,., is a constant on run /.
\J ) I Y . / (//
r &2
~ <*«(/) — (1 + 2j3ft) I for normally distributed run /
7 determinations (cf. Cramer'14)N
The variance of j3y varies with n/ as [<*„/• •>]/»/• Table B-l 1 enumerates this dependency. There is a
drastic reduction in the proportional variance of (3/ as «/ increases.
In this situation, the standard statistical technique for estimating the mean from individual values
is to weight each value. Applied to the between-laboratory coefficient of variation, this technique
assigns a weight Wj to each run coefficient of variation estimates /?/:
The proper weight to obtain the unbiased between-laboratory coefficient of variation estimate with
minimum variance is to set the weight inversely proportional to the variance of j3;- and then standardize
the weights so that the average of the Wj is 1 .OS15 ^ This is essentially the same as the procedure
recommended for weighting observations in linear least-squares regression/ ' Denote the initial
weight for run / as u. Then
1
Ui = v^
The average of the ufs is
u = — >
Kfr
41
-------
Table B-l 1. Sample Size Dependency of
Coefficient of Variation Weights
Sample
Size,
n
2
3
4
5
6
7
8
9
10
12
16
20
24
Unbiasing
Correction
Factor
<*n
1.2533
1.1284
1.0854
1.0638
1.0509
1.0424
1.0362
1.0317
1.0281
1.0230
1.0168
1.0132
1.0109
Proportional Variance
of Coefficient of
Variation Estimate,
a>
0.7854
0.4244
0.2945
0.2263
0.1841
0.1552
0.1342
0.1183
0.1057
0.0872
0.0646
0.0513
0.0426
Raw Coefficient
of Variation
Weight,
"K
1.273
2.356
3.395
4.418
5.433
6.442
7.451
8.456
9.461
11.466
15.475
19.482
23.485
Then Wj is given by
Uj
u i I
If
K [
w. -
«20, lei (i
2
_fe2d+2fe2
"/
+ 2/£)J
1 f "'
) ^^-
yJ i= 1 "(0
Thus the proper weighted between-laboratory coefficient of variation estimate is
K
7=1
njSj
K
The same adjustment for differing sample size is also appropriate for the within-laboratory coef-
ficient of variation estimate j! Let us denote the collaborator block index by / and the number of
collaborator blocks with more than one determination by L. Then the weighted within-laboratory
coefficient of variation estimate is
42
-------
/=!
/=!
/=!
B.8 Precision Standard Deviation Estimation
In this section, the measures of the precision of Method 5 at cement plants are estimated by the
weighted coefficient of variation procedure. Two sets of precision measures are calculated, those from
all the acceptable collaborative test data shown in Table 4 of the main report, and those from these
data with the high values excluded (Table 5 of the main report). The weighted coefficient of variation
calculation formulas to be used were derived in the first Method 5 test report(4) and in Section B.7.
The first set of precision measures is applicable to the use of Method 5 as published in the Federal
Register (cf. Appendix A) at the Portland cement plants. The appropriate source of concentration
determination data is Table 4, in which the published isokinetic sampling restriction (90 < 110) is
imposed; the minimum sampling volume limitation (Vm std > 60 scf) is ignored; and the four high
value samples are included. The between-laboratory coefficient of variation ft is estimated from the
run point estimates for the 12 runs in Table 4 with more than n,- = 1 collaborator's determination. The
coefficient of variation point estimate ft for run; is weighted by the factor Wf (cf. Table 4) to account
for the varying number of collaborators' determinations per run:
12
7=1
1 2
/=!
njSj
12
a x- V
u« (/) / Zj
!= 1
/«,' \
\a"(o/J
= 0.58368
At a true particulate concentration 5, the between-laboratory standard deviation estimate for Method 5
from the cement plant collaborative test is
ab = ft6 = (0.58368)5
The usual procedure for estimating the within-laboratory coefficient of variation (3 is invalidated by
the magnitude of the true concentration effect (cf. Sections B.4 and B.5). Adjusted data are needed
to eliminate the true concentration effect, but unfortunately the existence of the high values , in the
Table 4 data, renders appropriate data adjustment impossible. Our only recourse is to use the within-
laboratory and laboratory-bias coefficient of variation estimates, fte) and ft, (e) respectively, from the
Table 5 and Table 6 data, in which the high values have been eliminated, as a basis for the corresponding
Table 4 estimates. Where the subscript e denotes high value elimination, fte) and ft, (e) are calculated
in the second set of precision measures below. From the nature of high value phenomenon discussed
in Section B.2, it is apparent that the high values make a substantial contribution to both the within-
laboratory and between-laboratory variability components. Lacking any quantitative information, the
safest approach is to apportion the between-laboratory coefficient of variation estimate ft between
jSandfo in the same proportion as ft,(e) =0.20123 is divided into ft e) =0.09788 and (3L(e) =0.17582.
Thus, the within-laboratory coefficient of variation |3 for the Table 4 data is estimated by proportion:
fO.09788)
0.20123J
(0.58368) = 0.28390
43
-------
The within-laboratory standard deviation estimate is
a = 06 = (0.28390)5
The laboratory bias coefficient of variation $L is similarly estimated:
(0.58368) = 0.50999
'
Thus OL = j3/, 5 = (0.50999)5 is the estimated laboratory-bias standard deviation. The between-labora-
tory estimates, based on a maximum of n/ = 3 collaborators per run, have 06 = 2 degrees of freedom.
Since the within-laboratory estimates are calculated from the fo and fe(e), each of which possess only
2 degrees of freedom, the within-laboratory estimates also contain just 2 degrees of freedom.
A second set of calculable precision measures would apply to Method 5, at Portland cement plants,
if the high value phenomenon could be removed from the method. The proper data are contained in
Table 5 along with their run summary; the collaborator block summary of the associated adjusted data
is presented in Table 6. Here the isokinetic sampling restriction (90 < 1 10) is still imposed and
the minimum sampling volume limitation (Vm std > 60 scf) is again ignored, but now the four high
value samples are excluded from the acceptable concentration data. The following precision estimates
are then obtained:
1 10
fo (e) = 77 Z ^7=0.201 23
1U/=1
ab(e) -(0.20123)5
5
5 i=l
& = 0-09788
d(e} = (0.09788)5
$L(e) = V(0.20123)2 -(0.09788)2 =0.17582
aL(e) = (0.17582)6
These between-laboratory estimates again have 2 degrees of freedom. But the within-laboratory values
are estimated from the five collaborator blocks and possess a cumulative 20 degrees of freedom. Note
that the between-laboratory and within-laboratory standard deviation estimates for Method 5 at cement
plants would be reduced to about one-third of their present levels if the high concentration phenomenon
could be eliminated.
44
-------
REFERENCES
1. Environmental Protection Agency, "Standards of Performance for New Stationary Sources,"
Federal Register, Vol. 36, No. 247, December 23, 1971, pp 24876-24890.
2. Dixon, W. J. and Massey, F J., Jr., Introduction to Statistical Analysis, 3rd Ed., McGraw-Hill,
New York, 1969,p 136.
3. Environmental Protection Agency, op. cit., p 24880.
4. Hamil, H. F., and Thomas, R. E., "Collaborative Study of Method for the Determination of
Particulate Matter Emissions from Stationary Sources (Fossil Fuel-Fired Steam Generators),"
Southwest Research Institute report for Environmental Protection Agency, June 30, 1974.
5. Hamil, H. F., and Thomas, R. E., "Collaborative Study of Method for the Determination of
Particulate Matter Emissions from Stationary Sources (Incinerators)," Southwest Research
Institute report for Environmental Protection Agency, July 1, 1974.
6. Youden, W. J., "The Collaborative Test," Journal of the AOAC, Vol. 46, No. 1, 1963, pp 55-62.
7. Hamil, H. F., and Thomas, R. E., op. cit., (Fossil Fuel-Fired Steam Generators).
8. Hamil, H. F., and Thomas, R. E., op. cit., (Incinerators).
9. Youden, W. J., op. cit., pp 59-62.
10. Brownlee, K. A., Statistical Theory and Methodology in Science and Engineering, 2nd Ed., Wiley,
New York, 1965, pp 256-258.
11. Hamil, H. F., and Camann, D. E., "Collaborative Study of Method for the Determination of Sulfur
Dioxide Emissions from Stationary Sources (Fossil Fuel Fired Steam Generators,)" Southwest
Research Institute report for Environmental Protection Agency, December 10, 1973, pp 35, 36.
12. Brownlee, K. A., op. cit., pp 290-295.
13. Hamil, H. F., and Camann, D. E., "Collaborative Study of Method for the Determination of
Nitrogen Oxide Emissions from Stationary Sources (Fossil Fuel Fired Steam Generators),"
Southwest Research Institute report for Environmental Protection Agency, October 5, 1973,
pp B-14to B-16.
14. Cramer, H., Mathematical Methods of Statistics, Princeton University Press, Princeton, 1946, p 358.
15. Brownlee, K. A., op. cit., pp 95-97.
16. Draper, N. R., and Smith, H., Applied Regression Analysis, Wiley, New York, 1966, pp 77-81.
45
-------
7. AUTHOR(S) " •
Henry F. Hamil and David E. Camann
. REPORT NO.
EPA-650/4-74-029
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
2.
4. TITLE AND SUBTITLE
Collaborative Study of Method for the Determination of Particulate Matter
Emissions from Stationary Sources (Portland Cement Plants)
5. REPORT DATE
May 1974 (date of approval)
6. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Southwest Research Institute
8500 Culebra Road
San Antonio, Texas 78284
12. SPONSORING AGENCY NAME AND ADDRESS
Quality Assurance and Environmental Monitoring Laboratory
National Environmental Research Center
Research Triangle Park, North Carolina 27711
15. SUPPLEMENTARY NOTES
3. RECIPIENT'S ACCESSION-NO.
10. PROGRAM ELEMENT NO.
Task Order 3
11. CONTRACT/GRANT NO.
68-02-0626
13. TYPE OF REPORT AND PERIOD CO vEREC
Task Order
14. SPONSORING AGENCY CODE
16. ABSTRACT ~~ ""'"""*" ------
This report presents and analyzes the results of a collaborative test of EPA Method 5-Determination of Particulate
Emissions from Stationary Sources. The test was conducted by four participating laboratories at a Portland cement plant under
simulated "real world" Method 5 testing conditions. This report describes the collaborative test, examines problems
encountered in the use of Method 5, estimates the between-laboratory and within-laboratory precision of Method 5, ascribes
precision variability to its sources, and evaluates the necessity of sampling restrictions required for compliance testing. The
magnitude of the Method 5 precision estimates obtained is due primarily to occasional extremely high particulate determina-
tions. This high value phenomenon, present in all three Method 5 collaborative tests that have been conducted by Southwest
Research Institute is believed caused by accidental scraping of the probe tip against the stack wall. Hypothetical estimates
of Method 5 precision are made with the high values excluded from the data.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.IDENTIFIERS/OPEN ENDED TERMS
COSATI Field/Group
Air Pollution 1302
Flue Dust
Collaborative Testing
Methods Standardization
Particulates
13-B
07-B
18. DISTRIBUTION STATEMENT
Release Unlimited
19. SECURITY CLASS (This Report)
Unclassified
21. NO. OF PAGES
54
20. SECURITY CLASS (Thispage)
Unclassified
22. PRICE
EPA Form 2220-1 (9-73)
-------
INSTRUCTIONS
1. REPORT NUMBER
Insert the EPA report number as it appears on the cover of the publication.
2. LEAVE BLANK
3. RECIPIENTS ACCESSION NUMBER
Reserved for use by each report recipient.
4. TITLE AND SUBTITLE .,
Title should indicate clearly and briefly the subject coverage of the report, and be displayed prominently. Set subtitle, if used, in smaller
type or otherwise subordinate it to main title. When a report is prepared in more than one volume, repeat the primary title, add volume
number and include subtitle for the specific title.
5. REPORT DATE
Each report shall carry a date indicating at least month and year. Indicate the basis on which it was selected (e.g., date of issue, date of
approvd, date of preparation, etc.).
6. PERFORMING ORGANIZATION CODE
Leave blank.
7. AUTHOH(S)
Give name(s) in conventional order (John R. Doe, J. Robert Doe, etc.). List author's affiliation if it differs from the performing organi-
zation.
8. PERFORMING ORGANIZATION REPORT NUMBER
Insert if performing organization wishes to assign this number.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Give name, street, city, state, and ZIP code. List no more than two levels of an organizational hirearchy.
10. PROGRAM ELEMENT NUMBER
Use the program element number under which the report was prepared. Subordinate numbers may be included in parentheses
11. CONTRACT/GRANT NUMBER
Insert contract or grant number under which report was prepared.
12. SPONSORING AGENCY NAME AND ADDRESS
Include ZIP code.
13. TYPE OF REPORT AND PERIOD COVERED
Indicate interim final, etc., and if applicable, dates covered.
14. SPONSORING AGENCY CODE
Leave blank.
15. SUPPLEMENTARY NOTES
Enter information not included elsewhere but useful, such as: Prepared in cooperation with, Translation of, Presented at conference of,
To be published in, Supersedes, Supplements, etc.
16. ABSTRACT
Include a brief (200 words or less) factual summary of the most significant information contained in the report. If the report contains a
significanl bibliography or literature survey, mention it here.
17. KEY WORDS AMD DOCUMENT ANALYSIS
(a) DESCRIPTORS - Select from the Thesaurus of Engineering and Scientific Terms the proper authorized terms that identify the major
concept of the research and are sufficiently specific and precise to be used as index entries for cataloging.
(b) IDENTIFIERS AND OPEN-ENDED TERMS - Use identifiers for project names, code names, equipment designators, etc. Use open-
ended terms written in descriptor form for those subjects for which no descriptor exists.
(c) COSAT1 FIELD GROUP - Field and group assignments are to be taken from the 1965 COSATI Subject Category List. Since the ma-
jority of documents are multidisciplinary in nature, the Primary Field/Group assignment(s) will be specific discipline, area of human
endeavor, or type of physical object. The application(s) will be cross-referenced with secondary Field/Group assignments that will follow
the primary posting(s).
18. DISTRIBUTION STATEMENT
Denote releasability to the public or limitation for reasons other than security for example "Release Unlimited." Cite any availability to
the public, with address and price.
19. & 20. SECURITY CLASSIFICATION
DO NOT submit classified reports to the National Technical Information service.
21. NUMBER OF PAGES
Insert the total number of pages, including this one and unnumbered pages, but exclude distribution list, if any.
22. PRICE
Insert the price set by the National Technical Information Service or the Government Printing Office, if known.
EPA Form 2220-1 (9-73) (Reverse)
------- |