United States
 Environmental Protection
 Agency
Atmospheric Research and Exposure
Assessment Laboratory
Research Triangle Park, NC 27711
 Research and Development
EPA/600/S3-90/008 May 1990
 Project  Summary
 Precision  and Accuracy
 Assessments  for  State and
 Local  Air Monitoring Networks:
 1988
 Luther Smith and Jack Wu
  Precision and  accuracy  data
obtained from  states  and local
agencies during  1988 are analyzed.
Average  biases,  pooled standard
deviations, and 95% probability limits
of percent differences are presented
for both accuracy and precision data.
The results  of a site-by-site linear
regression are reported for  the ac-
curacy data.  Reporting organizations,
states   and  regions   which
demonstrate consistent  precision
arid accuracy data as  the result of
effectively  admin-istered  quality
assurance programs are  identified.
The effectiveness  of the  quality
assurance programs on  a national
level is gauged by use of percentiles
for the percent differences. The PARS
arid NPAP data  sets for 1988  are
analyzed and  compared. This
information is in-tended as a guide
for identifying problem areas within
the  quality  assurance  programs
which  may need added attention and
for  allowing  more  knowledgeable
decisions concerning  attainment
status regarding ambient air quality
standards.
   This  Project  Summary  was
developed by EPA's Atmospheric
Research  and Exposure Assessment
Laboratory, Research Triangle Park,
NC,  to announce  key findings of the
research project  that is fully
documented  in a separate report of
the  same title (see Project Report
ordering information at back).
Introduction

  Revisions to Appendix A, 40 CFR Part
58 promulgated March 19, 1986, required
site-specific precision and accuracy data
to be submitted  as  actual test results
beginning m January 1987.  This report
analyzes these data for the year January.
1988 to December, 1988. The availability
of individual site data and the opportunity
to assess  the performance  of specific
instruments was  cited as a way to
improve the usefulness of the data quality
estimates  associated  with  the
NAMS/SLAMS monitoring network.  The
regulations did not, however, specify how
this  would  be accomplished  except that
EPA would now  be responsible  for
calculating the pooled precision and
accuracy  probability  limits formerly
calculated and reported by the reporting
organizations.  The  objectives  of this
report  are to analyze and interpret
individual Precision  and Accuracy
Reporting System  (PARS) site data as
they pertain to:

  1.   Identifying  extreme measure-
      ment errors.

  2.   Evaluating the  effectiveness of
      SLAMS  quality  assurance
      programs.

  3.   Validating models  used  to
      describe precision and accuracy
      data.

-------
    4.    Improving  decisions concerning
         attainment  of  air  quality
         standards  as they  relate  to
         specific instruments.

    The  goal  is to  provide  an overall
 assessment of various quality assurance
 programs at the reporting  organization,
 state,  and  regional  levels.  Unless
 otherwise noted,  region  as used  in this
 report refers to EPA regions.
    The  National  Performance  Audit
 Program (NPAP) also collects  accuracy
 data, and this data set was analyzed  as
 well.  At  those  sites  which   had
 measurements from both  networks,  the
 NPAP and  PARS accuracy data  were
 compared.
    Manual  S02 and  N02  data  were  not
 included in  this report because so  little
 data are available.

 Data Analysis
    A goal of this report is to assess the
 overall effectiveness  of quality assurance
 programs administered by the states and
 local  agencies.  The  use  of  95%
 probability  limits  for  this  purpose  rests
 upon  the  assumptions  which  were
 required for their calculation.
   The data has a nested structure (i.e ,
 sites within reporting organizations within
 states  within  regions) and  replicated  by
 quarters; for  the  precision data,  the
 biweekly measurements provide repeated
 sampling. The analysis of the 1988 data
 proceeded  by combining subgroups
 within  the hierarchical  structure and
 across quarters (and, for accuracy  data,
 levels) beginning with sites as the basic
 subgroup for accuracy data and  biweekly
 samples for the precision data.
   To test for homogeneity of dispersion,
 a test  known as Lev1:med was used; it
 has generally been found to perform well
 with  regard   to  size and  power   in
 comparison to  other methods.   Briefly,
 Lev1:med works as  follows: (1) within
 each subgroup form the quantities Z(J = |
med
                 where x,j is the  ith
observation  (in  the  case here,  percent
difference) in the jth subgroup and Xj.med
is  the corresponding  median value; then
(2) test  for homogeneity  by doing  a
standard  one-way ANOVA F-test on the;
Zij's. Strictly speaking, Lev1:med is not a
test  of  equality  of  variances but  of
dispersion or  spread in general.   An
advantage to using Lev1:med is that ones
is allowed to postpone the assumption of
normality until  the probability limits  are
actually computed. The significance level
used in this report was 5%.
    One  problem  with  Lev1:med in the
 above form is that it is too conservative
 for small (n < 8), odd sample sizes.  This
 results  from  values of  zero for  Zjj
 distorting the F test statistic.  To adjust
 for this  problem,  Zy   was calculated as
 above except when Xjj = Xj,med; in  this
 instance, Xj.med  was  replaced by the
 median of the subgroup with Xji removed.
 While this  does not completely cure the
 problem, it does provide  a measure of
 relief while still  retaining  all the original
 data.


 Precision Results
   The precision data analysis was begun
 by aggregating sites  to  the reporting
 organization  level  by  quarter  and  then
 across  quarters.   This  was  repeated
 through the state and regional levels.
   For  those reporting  organizations,
 states,  and regions which  consistently
 executed the quality assurance program
 for a given  quarter,  the results were quite
 good  in  1988; the  vast  majority  of
 average  percent  differences were  less
 than,  and a great many were well below,
 10% in magnitude.
   For  those  reporting  organizations,
 states,  and regions which  consistently
 executed the quality assurance program
 throughout  the year, the performance of
 the program was also good.
   Table 1  provides the percentages of
 the reporting organizations, states,  and
 regions which consistently  executed the
 PARS quality assurance  program.   It
 indicates that as the level of aggregation
 increased  either  geographically  or
 temporally the  percentage of cases
 where homogeneity occurred  declined;
 that  is, as  one combined  either larger
 geographic areas or blocks  of time,  the
 difficulty  of consistently  executing   the
 quality assurance program increased.

 Accuracy Results
   The accuracy data analysis was done
 in an analogous manner to  that for  the
 precision data  The nature of the  data
 necessitated one difference.  Generally,
 the accuracy data was very sparse within
 quarters  for the  PARS data (often  only
 one site  in  a reporting organization); for
 the NPAP  data,  usually only one value
 was available for the year.  The accuracy
 measurements  are made  at  (1  to  4)
different levels  Therefore,  accuracy data
was  aggregated  to  the reporting
organization, state,  and regional scales
across  quarters;  the analyses were
performed by audit  level separately  and
also across  audit levels.
    For those  reporting  organizations,
 states, and  regions which consistently
 executed the  PARS quality  assurance
 program, the results were very good. The
 overwhelming  majority  had average
 percent differences less  than  10%  in
 magnitude (with many well below).   This
 was true for individual  levels, and  when
 aggregating data across levels.  Table 2
 gives  the percentages  of the reporting
 organizations, states, and  regions which
 consistently executed the PARS program.
    As  with  the PARS  data, where the
 NPAP  program  was  consistently
 executed,  the  percent  differences  were
 not generally large.  However,  the
 percentages   of  groupings  which
 maintained  consistency  varied between
 PARS and NPAP (Tables 2 and 3). But a
 better way  to  compare the  NPAP  and
 PARS programs is to use collocated data;
 this is discussed later.
    The accuracy data was also analyzed
 with  a  regression approach.  Linear
 regression was attempted as a means of
 assessing the overall performance of the
 accuracy program  by  examining  the
 network on  a  site-by-site  basis.
 Unfortunately,  the regressions  at  the
 individual sites suffered from a  lack of
 data;  often only  3 data  points were
 available. The  limited  amount  of  data
 reduced the power of the regressions and
 prevented adequate checking  of  the
 basic  regression  assumptions (i.e.,
 normally distributed error terms  with
 homogeneous variances).  Therefore, the
 results of the regression analysis should
 not be viewed as firm conclusions, but as
 general indicators of where future efforts
 might best be directed.
   The measured value was regressed on
 the audit (target) value on a site-  by-site
 basis,  and the resulting regression  line
 was compared  to the "ideal"  line  which
 has a  slope of one and passes through
 the origin.
   The comparison  to the ideal line was
 done by making the joint hypothesis  test
 that the  intercept  estimated  by  the
 regression  was  zero  and  that  the
 estimated slope was 1.  It may be useful
 to consider  the interpretation of these
 estimated parameters. While theoretically
 the intercept is the value that would be
 measured if no pollutant  were present,
 the estimated intercept is  best  viewed
 here as an indicator of the general  bias of
 the measurement process over the range
 of values established by the audit levels.
 (This  is  because  audit  levels   are
 necessarily set at positive values,  and
thus no  data  are  available  about  the
measurement process at a pollutant level
of zero.)

-------
Table 1.  Percentage of Cases with Homogeneous Dispersion for
         PARS Precision Data

         Key: N  = Total number possible
             S  = Number meeting homogeneity criterion
             %  = Percentage to nearest whole percent
          A.  Reporting organizations across sites by quarter


           Pollutant     CO     NO2    O3   Pb  PM10
  SO,
                                                               TSP
N
S
%
274
167
61
155
118
76
293
196
67
17
16
94
95
77
81
287
195
68
297
250
84
          B.  Reporting organizations across sites and quarters


          Pollutant     CO    N02    03    Pb   PM10
           C.  States across reporting organizations by quarter


           Pollutant     CO    NO2   O3    Pb   PM10
SO,
 SO,
                                                             TSP
N 80
S 34
% 43
48
32
67
93
62
63
16
12
75
47
37
79
87
49
56
103
76
74
       TSP
N
S
%
78
35
45
47
28
60
71
43
61
20
16
80
68
51
75
66
28
42
76
40
53
         D.  States across reporting organizations and quarters


          Pollutant     CO    NO2    03    Pb   PM10     S02   TSP
N
S
%
23
4
17
13
4
31
21
10
48
8
5
63
20
9
45
18
6
33
22
4
18
    The slope indicates how the accuracy
 measurement depends on audit level.  A
 slope  of  zero would indicate  that  a
 sampler reports  numbers  essentially
 independently of  what the true pollutant
 levels are;  a slope  between zero and  1
 indicates  that  the  sampler  does  not
 increase its reported value fast  enough as
 pollutant level increases, while a slope
 value  greater  than  1  indicates  that
 reported values  increase too  rapidly;  a
 slope of 1 says that as the pollutant  level
 changes,  the machine  responds  with
 exactly the same  change in  its reported
 value.
    The joint hypothesis  tests  were all
 conducted  at   the   5%   level;  the
 hypothesis  was  rejected considerably
 more often than would be expected by
 chance.  (Note:  the joint hypothesis  may
 be  rejeccted because either  or both
 estimated  parameters  may  be  too far
 from its  "ideal  " value.) However,  in
 judging the effectivenesses of the  quality
 assurance program, the estimates  for the
 intercept and slope are  more relevant
 since they indicate the degree  to which
 the  parameters  depart from their ideal
 values.
   In general there did  not appear to be a
 large overall  bias (i.e.,  intercept estimate)
 in the accuracy measurements for  CO,
 N02, 03, or S02  in the PARS network.
 Similarly, the NPAP data did  not show a
 large  bias  for  CO  or Pb     Of more
 concern  were the  intercept estimates for
 Pb in the PARS results. For this pollutant
 intercepts were estimated which were
 quite large  in magnitude, both positive
 and negative, in several cases.
   The  slope estimates  were  generally
 within 10% to 20% of their ideal value of
 1.
   In summary,  then, the  regression
 results indicated that for  CO,  N02,  03,
 and  SO2,  accuracy   audits  generally
 conformed to the desired results. For Pb,
 there may have been  some  bias  in the
 accuracy audit results  at certain sites in
 the  PARS network.  (Note:  Accuracy
 audits for TSP  and PM-10  were only
 done at  one level, and regression was
 therefore inappropriate in these cases.)

 National Results
   If the assumption is made  that  the
 percent differences were taken  from  the
 same normal population for  a  given
 pollutant (and level),  then  annual
probability limits could be calculated on a
national basis, as shown in Tables  4 and
5.  However,  the earlier results displayed

-------
   Table 1. (cont'd)  Percentage of Cases with Homogeneous
                   Dispersion for PARS Precision Data
            Key: N =  Total number possible
                S =  Number meeting homogeneity criterion
                % =  Percentage to nearest whole percent
         E.   Regions across states by quarter


           Pollutant     CO    NO2    O<,    Pb   PMiO   SO2    TSP
N
S
%
40
4
10
28
8
29
39
10
26
22
11
50
35
20
57
36
12
33
39
25
64
         F.  Regions across states and quarters


           Pollutant    CO    NO2   O3    Pb   PM10   SO2   TSP
N 10
S 1
0/0 10
7
0
0
10
1
10
8
3
38
9
1
11
9
2
22
10
3
30
 in Tables  1,  2,  and  3 indicate that
 variance (or dispersion) is  not  uniform
 across  the  country.  In addition,
 examination of data  plots  led  to  the
 conclusion  that  the  assumption  of
 normality is probably incorrect.
   An alternative  method of examining
 the data which makes no  assumptions
 about  the  underlying  statistical
 distribution(s) is provided in Tables 6 and
 7. (The  n'h  percentile is that value such
 that n% of the data  were  less  than  or
 equal  to it.   Note  that the 5th and 95^
 percentiles  bracket the middle 90%  of
 the data.)
   While the normality  assumption may
 not be  appropriate, the distributions  of
the percent differences for both precision
and accuracy data generally  appeared to
be unimodal and symmetric,  and thus  the
means and medians  are quite  similar
(Tables 6 and 7).
   For the precision  data,  the  ranges
established by the 95% probability limits
(Table 4) and the 5th and 95th percentiles
(Table  6)  are  very  similar  for each
pollutant. However, this is not  quite the
case for the accuracy data (Tables 5 and
7). The accuracy data  show these
exceptions  the percentile spread is wider
 han the probability limits for  N02  and
narrower for 03 at level 4 in the PARS
data and is narrower for lead and TSP in
 he  NPAP  data.  These  observations
(except 03 at level 4 in PARS and Pb and
TSP in NPAP for the accuracy  data)
indicate that the  violations  of  the
assumptions necessary  for calculating
 he probability limits leads to an under-
estimate of the amount of data in the tails
(i. e., farther reaches) of the distributions.
   Table 6 shows that the middle 90% of
 he  precision  percent  differences
occurred roughly within the range of (-
10%,  10%) for CO, N02, 03,  and SO2
  For  Pb the range  was roughly (-20%,
  20%), and the spreads for PM-10 and
  TSP were broader than for the gases but
  less than for lead.
    Table 7  indicates that the percent
  differences for the accuracy data tended
  to be larger at  the lowest audit  level.
  (Note the larger widths  between the 5'h
  and  95th percentiles at  these lowest
  levels.)  The widest spread  of the middle
  90% of the accuracy percent differences
  is the range from -12%  to  23% (for the
  NPAP PM-10 data).
    Thus,  on  a  national   scale,  the
  precision and  accuracy  quality assur-
  ance  programs seem to  be  operating
  well, in general.


  Comparison of the 1988  Pars
  and NPAP Data
    Some sites had  data from  both the
  NPAP and  PARS programs. Using this
 collocated data, the two networks  were
 compared. The quantities examined  were
 the accuracy percent differences for CO,
 Pb flow rate, PM-10, and TSP. Generally
 there was only one NPAP observation at
 a site (or, for CO, a level).  There  were
 usually two or  more PARS  observations
 at a  site (or level),  but  in  some cases
 there was also only one PARS value.
   The  NPAP  observations were
 examined to see whether the NPAP value
 was above, below, or within  the range of
 the PARS values. This was done on a
 site by site (and, for CO, level  by  level)
 basis.
   In  cases where only  one value  was
 available from  each program,  binomial
 tests  indicated  that the NPAP values
 were  evenly split between the above and
 below categories for level  1 of CO and for
 PM-10; however, there were significantly
 more above occurrences for  levels 2 and
 3 of CO  and TSP.  The significance level
 used was 5%.
   Where more  than one  PARS value
 was available, the data were  examined to
 see whether (in each individual case) the
 NPAP value was so far removed from the
 PARS values as to  be considered to have
 come  from  a different  distribution.
 Dixon's r10 outlier test was used.
   The null hypothesis was that  for each
 individual case  the  PARS  and NPAP
 values  came from  a  single   normal
 distribution.  First,  the  ratio  of  the
 difference  between  the  NPAP value
 (when it was the highest or lowest value)
 and the closest  PARS value to the range
 of all  values was formed.  A p-value  was
then either calculated from a formula or

-------
Table 2.   Percentage of Cases with Homogeneous Dispersion for
         PARS Accuracy Data

         Key: N  = Total number possible
             S  = Number meeting homogeneity criterion
             %  = Percentage to nearest whole percent
        A.  Reporting organizations across sites and quarters by level


          Pollutant   CO    NO2    03    Pb    Pb-flow   PM10   SO2    TSP
N
S
%
150
90
60
83
48
58
166
105
63
29
24
83
5
3
60
54
38
70
157
94
60
84
52
62
        B.  Reporting organizations across sites, quarters, and levels


          Pollutant    CO   NO2    O3    Pb    Pb-flow   PM10   SO2   TSP
N
S
°/o
78
63
81
46
40
87
92
69
75
17
14
82
5
3
60
54
38
70
88
62
70
84
52
62
         C.   States across reporting organizations and quarters by level


           Pollutant   CO   NO2    O3     Pb    Pb-flow    PM10   SO2    TSP
N
S
%
61
46
75
37
33
89
61
54
89
22
12
55
1
1
100
20
15
75
55
47
85
20
15
75
         D.  States across reporting organizations, quarters, and levels


           Pollutant   CO   NO2     03    Pb    Pb-flow   PM10    SO2   TSP
N 22
S 15
% 68
14
10
71
21
5
24
11
4
36
2
2
100
20
15
75
19
11
58
20
15
75
 linearly interpolated from a table. (Cases
 where all  PARS values were  the  same
 were excluded since in such a case, the
 test  automatically  would  declare  the
 NPAP value  an  outlier, no matter how
 close it  was  to the  PARS value.  There
 were very few such cases.)
    Table 8 shows the number of times a
 significant result was obtained from  this
 procedure at the 5%  level. This  table
 indicates  that generally  NPAP percent
 differences  are in agreement with  PARS
 percent  differences  for  accuracy
 measurements on a case by case basis.
    These results for the individual sites or
 levels were expanded to a network basis
 by using  Fisher's method of combining
 tests. Combining the  individual  test
 results allows a comparison of the overall
 NPAP and  PARS networks  based on
 collocated  data.  Fisher's  method has
 good  statistical  power  and may detect
 differences  which were  hidden  because
 the individual  tests were based on so few
 data  points.  Basically,  Fisher's method
 transforms   the  p-values from  the
 individual tests and adds them to obtain a
 chi square distributed test statistic.
    The results  from  combining the
 individual tests were  that only for PM-10
 were the NPAP values different from the
 PARS  values.  The difference   was
 significant at the 5%  level, but not at the
 1% level.
   The fact that CO values were available
 from  three  levels permitted a different
 method of comparing the two networks.
 At  each site, the percent difference
 values from both  networks together were
 linearly regressed on  level,  and the
 regression  diagnostic  Cook's D   was
 calculated  for each point. Cook's  D
 basically measures how strongly a single
 data   point   affects the  estimated
 regression parameters.
   Of the 197  CO regressions performed,
 only 13 NPAP data points had Cook's D
 statistics  exceeding  the  50th  percentile
 value of  the appropriate  F distribution.
 Thus,  based  on  this  examination of
 regression diagnostics, it does not appear
 that the  PARS and NPAP  CO  percent
 differences are substantially different.
   In summary, then,  there do not seem
 to  be  large  differences  between  the
 NPAP  and PARS  data  sets. One  pos-
 sible exception to this is the case of PM-
 10.  Overall,  NPAP  PM-10 percent  dif-
 ferences  appeared to  be  higher  than
 PARS values at the 5% significance level;
 however,  under  the more stringent
criterion of a 1% significance levei, there
is not a significant difference between the
two networks.

-------
   Table 2.  (cont'd)  Percentage of Cases with Homogeneous
                    Dispersion for PARS Accuracy Data
            Key: N  = Total number possible
                S  = Number meeting homogeneity criterion
                %  = Percentage to nearest whole percent
         E.  Regions across states and quarters by level


           Pollutant    CO    NO2    O3   Pb    Pb-flow   PM10   SO2    TSP
N
S
%
30
12
40
18
16
89
29
14
48
14
6
43
3
3
100
9
4
44
30
23
77
8
3
38
         F.  Regions across states, quarters, and levels


           Pollutant    CO    NO2    03   Pb    Pb-flow   PM10   SO2    TSP
N
S
%
10
1
10
6
5
83
10
4
40
7
4
57
3
3
100
9
4
44
9
2
22
8
3
38
 Summary

   For both precision and accuracy data
 in  those  cases  where  the quality
 assurance program (PARS or NPAP) was
 consistently executed across  different
 strata  of the network (e.g., geographic
 region, quarter, or  audit  level),  the
 performance was generally good.  Only
 rarely  did  weighted  average  percent
 differences exceed  10%  in magnitude
 when data were  combined across  strata,
 and  often the average  level  was  well
 below 10% in size.
   Regression  applied on a site-by-site
 basis to the accuracy data indicated that
 for CO (PARS  and  NPAP), N02, 03, and
 S02, accuracy audits generally were not
 biased to any large  degree  over the
 range of pollutant levels  established by
 the audit levels and generally instruments
 responded well as pollutant levels varied.
 For Pb, there may have been some  bias
 in  the  accuracy  audit results at certain
 PARS  sites. However, the nature of the
data limited the utility of the regressions,
and these results should be viewed only
as rough indicators  of the  state of the
accuracy quality assurance program.
   On a national basis, the middle 90%of
the PARS  precision percent  differences
occurred roughly within the range of (-
10%, 10%) for CO, N02, 03, and SO2.
For  Pb  the  range was  roughly  (-
20%,20%), and the spreads for  PM-10
and TSP were broader than for the gases
but less than  for lead.  Nationally,  the
percent differences  for  the PARS  and
NPAP accuracy data tended to be larger
at the lowest  audit  level.  The widest
spread  of  the  middle  90%  of  the
accuracy percent differences is the range
from -12% to  23%  (for  PM-10 in NPAP
data). Thus,  on  a   national  scale,  the
precision and accuracy quality assurance
programs  seem to  be operating  well, in
general.
   Based on analyses of collocated data,
there do  not  seem  to  be  large  dif-
ferences  between the NPAP  and PARS
data sets  One possible exception to this
is the case of PM-10.
   Taken  together,   the  results  above
indicate  that  the  quality  assurance
programs  for  accuracy  and  precision
generally  seem to   be  operating  well,
though  there  may   be  pockets which
could be improved.

-------
Table 3.   Percentage of Cases with Homogeneous Dispersion for
          NPAP Data
          Key: N =  Total number possible
              S =  Number meeting homogeneity criterion
              % =  Percentage to nearest whole percent
      A.   Reporting organizations across sites and levels

       Pollutant     CO     Pb    Pb-flow    PM10   TSP
N
S
%
59
53
90
0
0
0
0
0
0
0
0
0
0
0
0
     B.   States across reporting organizations by level

     Pollutant      CO    Pb    Pb-flow    PM10   TSP
N
S
%
44
24
55
22
3
14
0
0
0
5
2
40
19
15
79
   C.   States across reporting organizations and levels

    Pollutant     CO    Pb    Pb-flow    PM10   TSP
                  20     11       0         5      19
                  16     11       0         2      15
                  80    100      0        40      79
    D.  Regions across states by level

     Pollutant      CO    Pb    Pb-flow   PM10    TSP
N
S
%
27
22
81
27
12
44
1
0
0
9
8
89
9
5
56
    E.   Regions across states and levels

     Pollutant     CO     Pb    Pb-flow    PMIO   TSP
N
S
%
10
6
60
10
7
70
1
0
0
9
8
89
9
5
56

-------
Table 4. National Probability Limits for PARS Precision Data
Pollutant
CO
NO2
03
Pb
PM-10
S02
TSP
Table 5-A.
Pollutant
CO



NO2



03



Pb

Pb-fllow

PM-10
S02



TSP
N
14143
6294
16980
1108
4634
18087
11839
Mean (°A
-0.03
-0.67
-0.84
-0.15
0.84
-1.09
0.09
x Standard Lower 95% Upper 95%
Deviation prob. limit prob |jmit







3.42
5.41
4.14
13.28
8.73
4.28
7.76
-6.74
-11.27
-8.95
-26.19
16.26
-9.47
-15.13
6.68
9.93
7.28
25.88
17.94
7.29
15.31
National Probability Limits for PARS Accuracy Data
Level
1
2
3
4
1
2
3
4
1
2
3
4
1
2
1
2
2
1
2
3
4
2
N Mean (°/
1006
1050
895
7
473
458
420
15
1364
1349
1161
114
1000
856
3
107
1226
1235
1211
1003
101
2814
0.03
0.38
0.24
1.97
0.40
0.01
-0.47
0.58
-0.82
-0.58
-0.75
-2.64
0.11
-0.99
-2.09
0.57
-0.05
-0.14
-0.10
-0.44
-0.62
0.56
,Q > Standard
Deviation
5.51
2.79
2.97
0.14
4.83
3.66
4.34
1.66
5.57
3.64
4.58
8.75
5.07
4.59
1.41
3.25
3.05
5.38
4.91
5.13
4.27
322
Lower 95%
prob. limit
-10.82
-5.09
-5.57
1.69
-9.07
-7.16
-8.99
-2.68
-11.74
-7.71
-9.73
-19.78
-9.83
-9.97
-4.85
-5.80
-6.04
-10.69
-9.73
-10.50
-8.99
-5.75
Upper 95%
prob. limit
10.77
5.84
6.05
2.24
9.87
7.18
8.04
3.83
10.09
6.55
8.24
14.50
10.05
8.00
0.67
6.95
5.94
10.41
9.53
9.62
7.75
6.87

-------
   Table 5-8.   National Probability Limits for NPAP Data


    Pollutant      Level     N
Mean      Standard    Lower 95%    Upper 95%
 (%)      Deviation     prob. limit     prob. limit
CO


Pb


Pb-flow
PM-10
TSP
1
2
3
1
2
3
2
2
2
232
229
231
113
115
114
23
133
511
-0.62
1.85
2.18
-250
3.96
-252
4.46
1.18
2.29
5.65
3.81
3.54
10.68
10.98
10.10
7.09
10.12
9.45
-11.69
-5.63
-4.75
-23.43
-17.56
-22.31
-944
-18.65
-16.22
10.45
9.33
9.11
18.43
25.49
17.27
18.36
21.02
20.81
Table 6. National Percent Difference Percentiles for PARS
        Precision Data

                                               5th        g5th
 Pollutant        N        Mean      Median   Percentile   Percentile
CO
NO2
03
Pb
PM-10
S02
TSP
14143
6294
16980
1108
4634
18087
11839
0
-1
-1
0
1
-1
0
0
0
0
0
0
-1
0
-7
-13
-9
-22
-13
-10
-13
8
10
7
23
18
8
13

-------
Table 7-A.    National Percent Difference Percentiles for PARS
            Accuracy Data
Pollutant
CO



NO2



03



Pb

Pb-flow

PM-10
S02



TSP
Level
1
2
3
4
1
2
3
4
1
2
3
4
1
2
1
2
2
1
2
3
4
2
N
1006
1050
895
7
473
458
420
15
1364
1349
1161
114
1000
856
3
107
1226
1235
1211
1003
101
2814
Mean
0
0
0
2
0
0
0
1
-1
-1
-1
-3
0
-1
-2
1
0
0
0
0
-1
1
Median
0
0
0
2
0
0
0
-1
0
-1
0
-2
0
0
-3
0
0
0
0
-1
0
0
Percentile
-10
-5
-5
0
-14
-10
-10
-4
-11
-8
-8
-9
-10
-8
-4
-5
-6
-11
-9
-10
-8
-5
Percentile
12
7
6
3
15
10
8
7
9
6
6
2
11
6
1
6
6
11
9
9
9
6
                                    10

-------
Table 7-S   National Percent Difference Percentiles for NPAP Data
Pollutant
CO


Pb


Pb-flow
PM-10
TSP
Level
1
2
3
1
2
3
2
2
2
N
232
229
231
113
115
114
23
133
511
Mean
-1
2
2
-3
4
-3
4
1
2
Median
0
2
2
-1
5
-2
2
0
2
Percentile
-8
-3
-2
-18
-13
-12
-4
1 1
-10
Percentile
8
7
7
7
16
7
24
23
15
 Table 8.  Number of Outlier Tests Significant at the
         5% Level
Pollutant
CO
level 1
level 2
level 3
Pb-flow
PM-10
JSP
Number of
tests

137
136
118
4
59
217
No.
significant

1
9
5
1
7
16
°/

1
7
4
25
12
7
                                                          11

-------
   Luther Smith and Jack  Wu are with  NSI-ES Technology Services Corporation,
        Research Triangle Park, NC 27709
   Jack C. Suggs is the EPA Project Officer (see below).
   The complete report, entitled  "Precision and Accuracy Assessments for State
        and Local Air Monitoring Networks," (Order No. PB 90-183 401/AS; Cost:
        $23.00 subject to change) will be available only from:
            National Technical Information Service
            5285 Port Royal Road
            Springfield, VA22161
            Telephone: 703-487-4650
   The EPA Project Officer can be contacted at:
            Atmospheric Research and Exposure Assessment Laboratory
            U.S.  Environmental Protection Agency
            Research Triangle Park, NC 27711
United States                   Center for Environmental Research
Environmental Protection         Information
Agency                         Cincinnati OH 45268
Official Business
Penalty for Private Use $300

EPA/600/S3-90/008

-------