ML19331E464
| ML19331E464 | |
| Person / Time | |
|---|---|
| Site: | Framatome ANP Richland |
| Issue date: | 08/14/1980 |
| From: | NRC OFFICE OF NUCLEAR REGULATORY RESEARCH (RES) |
| To: | |
| Shared Package | |
| ML19331E458 | List: |
| References | |
| REF-PROJ-M-3 NUDOCS 8009100238 | |
| Download: ML19331E464 (29) | |
Text
V RISK ANALYSIS OF POSTULATED PLUTONIUM RELEASES FROM THE EXXON NUCLEAR HIXED OXIDE FUEL PLANT AS A RESULT OF HIGH WINDS-AND EARTHQUAKES INTRODUCTION The Nuclear Regulatory Commission has sponsored a program to estimate the potential hazard to the general population as a result of the impact of high winds and earthquakes on the Exxon Nuclear Mixed Oxide Fuel Plant at Richland, Washington. This paper outlines the procedures used in combining the results of various increments of analysis obtained in this study to produce a measure of risk. The risk measure presented in this paper is the probability per year that a high wind or earthquake will result in doses above specific levels (complementary cumulative distributions). The two organs, lungs and bone, were chosen for the dose exceedance probability calculations since these organs are significant and generally dominate the 50-year committed dose equivalents from inhalation.
The doses were calculated for the population within an 80 km (50-mile) radius of the plant and for the nearest residence located within 150 meters NE of the plant. Three tornado wind speeds, 150 mph,190 mph, and 250 mph, and one earthquake event, 1.0g were evaluated for the analysis. Two earthquake events were reduced to one event since no significant plant damaf? was assessed for earthquakes of magnitude less than 1.0g peak ground acceleration.
TORNADO WIND SPEEDS-The estimated probabilities for the postulated tornado wind speeds were obtained from T. T. Fujita (Ref.1).
The frequency, F-scale, and associated 800o10o 2 32
l 4,
wind speeds of historical tornadoes (1950-1975) were also obtained from Ref.1 and are listed in Table 1 below.
4 TABLE 1 TORNADO FREQUEN';Y, F-SCALE, AND ASSOCIATED k'IND SPEEDS (1950-1978)
Number of Reference Point To rnadoes F-Scale Wind-Speed Range-mph Wind Speed-mph 8
0 40 - 72 59 5
1 73 - 112 92 i
9 2
113 - 157 131 0
3 158 - 206 177 0
4 207 - 260 277 0
5 261 - 318 276 To obtain confidence bounds on the probabilities of postulated wind speeds, an error factor of 10 was used throughout the analysis. Assuming that the postulated tornado wind speeds occur in accordance with a Poisson process, the error factor of 10 will, to an order of magnitude accuracy, provide conservative 90% confidence bounds for wind speed occurrence probabilities within the wind speed range of the observed data with one or more points.*
Estimates of complementary cumulative tornado wind speed probabilities and associated confidence bounds are provided in Table 2.
EARTHOUAKES Two earthquake events were considered in this analy'ses.
Each earthquake event consisted of a discrete range of peak ground acceleration levels These 90% confidence bounds will in 90% of the cases cover the true wind speed probability if the assuced model and distributions are correct.
+*e 9
e
-m y-
-w.3
--n+
w.
+e3 g
--r-=
9
,---y-e-
- 7
-=m
+-
. and a point value used to characterize the range.
No significant damage was assessed for any earthquake less than 1.0g peak ground acceleration.
Pmbabilities vs. peak acceleration with estimated standard deviations (a) are provided in Ref. 2 and reproduced here as Figure 1.
Table 2 presents peak ground acceleration levels vs. cc probabilities and associated uncertainty bounds for the significant earthquake events.
For the accompanying risk analyses, the bounds on the probability estimates were modified to a factor of 10 for an earthquake of greater than 1.0g.
These modified factor of -10 bounds, in general, include more than 24 variations from the best estimate probability and are conservative
(>90%) if the er estimates and rounding uncertainties introduced in the gener% ion of the cc curves for the risk analyses. The exact confidence represented by the bounds is not critical to the subsequent risk analysis.
TABLE 2 A.
COMPLEMENTARY CUMULATIVE (cc) PROBABILITIES OF
-TORNADO WIND SPEEDS AND ASSOCIATED CONFIDENCE LIMITS Tornado Wind cc Probability Conservative 90% Confidence Speed per year Bounds on the Probability 150 mph 3.0E-7 (3.0E-8,3.0E-6) 190 mph 6.0 E-8 (6.0E-9,6.0E-7) 250 mph 3.0E-9 (3.0E-10,3.0E-8)
TABLE 3 B.
EARTHQUAKE PROBABILITY AND ASSOCIATED UNCERTAINTY Peak Ground Probability Approximate 90% Bounds Acceleration per year on the Probability i
1.0g 1.0E-5 (1.0 E-6, 1.0E-4)
L l
1 t-
-4 CC CURVES FOR CONSEQUENCES FROM ACCIDENTAL RELEASES The 50-year committed dose equivalents from inhalation following a
~
natural phenomena event of tornadoes or earthquakes were calculated by Watson and McPherson (Ref. 2) and presented in Table 3 below. Table 3 provides the dose to the nearest residence and to the population within an 80 km (50-mile) radius of the plant from tornadoes and earthquakes.
The table provides calculations of doses using more likely estimates and conservative estimates for the source releases and dispersion (meteomlogical) which were treated as random variables.- The most likely estimates were computed using the median (50%) values for source releases and dispersions and were assigned a probability of.95.
(The median value was used as the appmximate midpoint of the probability interval from 0 to 0.95.)
The conservative estimates were calculated using 95% values and were assigned a probability of 0.05.
The probabilities of possibl'e sources and the probabilities of possible dispersions were thus discretized into two intervals, O to 0.95 represented by the median value and.95 to 1.0 i
represented by the 95th percentile.
This breakdown of probabilities is i
gross, and care should be taken in interpreting any subsequent risk results to no more than an order of magnitede type of precision.
)
Figures 2 and 3 give the step function cc curves of doses to lungs and bones for the population within an 80 km (50-mile) radius of the plant i
due to damage from tornadoes. These complementary cumulative distributions give the probability per year that tornado-induced damage will result in doses greater than various values shown in the figures.
Figures 4 and 5
. provide the corresponding cc distributions of nearest resident doses for high winds.
Figures 6 through 9 contain the corresponding step function cc distributions for earthquakes.
These cc step functions and associated approximate confidence bounds have.a similar interpretation as those presented for tornadoes.
i
10,000
/,
/
./
l
/
/
./
/
/
l/
/
/
/i i
/
/
/
i j
f
,/
l/
1
,/
Y r
s
)'
,/
u I
/
/
/
/
f 1,000 -
z 1
i
/
i r
I i
J f
i*
i
/
/
/.
8 0
l i
1
/
/
s e
e 1
/
/,
/
i
[
/
/ l /
e i
?
It
/ /
1
[
E l
/
/ /-
h
/ / I I,
O I
/
9
/
/
i
/
/:
/
I C
/
l t
W 100 '
C.
z.
9
. I I h
8 i//
4 t
///
p
! l
+
///
i
///
I l
/// i I
i
//
l i
I l
' f l
.l I
j; l 10 11
,f l
il !
!.I l
i f,
5 ii
. i}
i i
,' l 1
0 5
10 15 20 25 30 l1 PEAK HORIZONTAL ACCELERATION (%g) l:_
r.
1 -
i ?
FIGURE I i1 !
EARTHQUAKE RISK AT THE EXXON FACILITY
!{
l
!i
~
TABLE 3 FIFTY-YEAR COMMITTED DOSE EQUIVALENTS FROM INHALATION FOLLOWING NATURAL PHENOMENA EVENTS (CLASS Y MATER Population Dose Dose at Nearest Residence -
(person-rem)
(rem)
Case (a)
Case Case Case Case Case Case' Case Event Organ I (0.90)
II(SE-2)
III(SE-2)
IV (3E-3)
I (0.90)
II (SE-2)
III(SE-2)
IV (3E-3)
Tornado wind Speeds 150 mph Lung
.l.7E3 1.7E4 1.7E3 1.7E4 3.4 E-2 3.4E-1 3.4E-2 3.4E-1 Bone 2.5E3 2.5E4 2.5E3 2.5E4 5.0E-2 5.0E-1 5.0E-2 5.0E-1 190 mph Lung 3.lE4 2.8E5 5.lE4 4.2E6 5.0E-1 5.0E0 7.8E-1 7.8E0 Bone 4.5E4 4.lES 7.4E4 6.1E6 7.3E-i 7.3E0 f.lE0
- 1. l E' 250 mph Lung 1.9E4 1.3E5 2.6E4 1.2E6 3.7E0 3.7El 4.lE0
- 4. l E' Bone 2.8E4 1.8E5 3.7E4 1.7E6 5.4E0 5.4El 5.9E0 5.9E' Earthquake Accelera tion 1.0g Lung 1.6E4 1.lES 2.2E4 3.6E5 2.4E0 2.8El 2.7E0 3.lEl Bone 2.3E4 1.5ES 3.2E4 5.2E5 3.5E0
- 4. l El 3.9E0 4.5El Case I:
Most Likely Release (95%) and Most Likely Dispersion (95%)
Case II:
Most Likely Release (95%) and Conservative Dispersion (5%)
Case III: Conservative Release (5%) and Most Likely Dispersion (95%)
Case IV:
Conservative Release (5%) and Conservative Dispersion (5%)
(a) The probabilities in parenthesis are tha conditional probabilities given the event, where the conditional probabilities are calculated using best estimate or conservative value for source releases and dispersion.
TORNR00 RISK RNRLYSIS FOR EXXON.POPULRTION. LUNG UNSM00THED CCDF 10-3 1 7
'5 A
3 10-e 3 a
g 7
a Aa 5
6 3
X 10-7 1 j
p A
5 w
U3 C
C t.L 104 1
i l
l C
7 i
I:
m em P
l 5
l p
e a
3 w
J A
A w
$c 10-8 1 03 7
C lu g
5 m
O O
10-10 1 l
-l 1
3 5
5 10-11
10s j:02 3
3 2
5 104 103
' ice f07 50 YR DOSE COMMITMENT (PERSON-REM)
4 TORNRDO RISK ANALYSIS FOR EXXON.POPULRTION. BONE UNSM00THED CCDF 10-8 i g
i i
7 I
i A
3 10-4 i l
7 Aj A A l
5 i
e 3
X 10-7 1 i
el 1:
I I
l h
3 l
e se
=
1
-w a
A A
03 3
0 O
t.i_
10-e i O
mi l
l I I 7
9 mm 3
l 8 1 5
i p
e m!
1 m
g A
A m
CDc 10-8 1 l
ID 7
O l
s OC a.
8 3
O e
l 10-18 1 j
i l l 7
I i i 3
m m
10-11
- 08 10s 1'04
' ' 1*08 3
5 7 3
~ ' l'O s
[07 50 YR DOSE COMMITMENT (PERSON-REM) i
,-7
-r-.-,n,
4 TORNADO RISK RNRLYSIS FOR EXXON.NR.LUNO UNSM00THED CCDF 10-8 1 7
5 10-8 1 l
IA e
l
~
- I
{
o 10-7 i e
c Ep a
g 5
E A a 3
X A
10-8 1 E
7 W
W W
e stb A
o 3
Q.
A L1.
10-9 i l
i i
i i
c3 7
I i m i f
5
! E td !S l
3 H
e i
_J 10_gg 1
=g s
s l
s x
t l'
l 5
g
.'5 E
3 C
5 E-10-11 i I
I 0
i
- l l
7 o
e i
i e
i 6
5 i
i I
3 10-12 1 I
1 l
3 1
i W f I
7 i
i i
i i
i e
i l
10-13 3 2
~
101 l'02 3
i
~'
1'00
~
1:0-1 l'0-1 t
SO YR DOSE COMMITMENT (PERSON-REM)
I
TORNADO RISK ANALYSIS FOR EXXON.NR.80NE UNSM00THE0 CC0F 10-8 1 7
s A
10-8 1 a
5 J
l 10-7 1 e
i i
c 7
=
.=
i i
i t
As e
in L
X A
10-8 1 m i I
W 5
CO e s a
c 3
C a
La 10-s g
C 7
i m
i s
l ELS 3
l--
w d 10-10 1
1 00
?
i i
T l
m G3 3
C M
M C 10-11 1 w
s i
3 10-12 1 i
ua i 7
i i
l 3
10-13 {.g-t 3
l'0-1 100 1'01 102 l
50-YR DOSE COMMITMENT (PERSON-REM)
EARTHQURKE RISK ANALYSIS EXXON POPULATION LUNG UNSM00THED CC0F 10-8 1 7
s 3
10-4 1 7
s 3
X 10-8 1 A
l I
7 3
l A
s LtJ CO 3
o O
Lt.
10-e t
=
l l
l c3 7
i I
H m
s p
H 3
a w
_J m
In Cr 10-7 1
=
I l
7 C
I m
l l
cr l
L 3
10-8 1 I
h I
3
=
10-s :,
l'08 1*04 1'08 l'Os 50 YR DOSE COMMITMENT (PERSON-REM)
EARTHQURKE RISK RNALYSIS EXXON BONE POPULATION UNSM00THED CCOF 10-8 1 7
5 3
10-4 1 7
s 3
X 10-8 1 0
^
I 7
'A l
A 5
LL1 3
,m oo
- u. 10-s a
C I
I y
I m
i l-l 3
_J s
CDcr 10-7 1 CD 7
o m
of 0
3 10-e 3 1
7 s
3 5
10-s {:0 i
3 l'04 108 10s 50 YR DOSE COMMITMENT (PERSON-REM)
EARTHQURKE RISK ANALYSIS EXXON NR LUNG UNSM00THED CCDF 10-3 1 7
s 3
10-4 1 A
7 s
3 x 10-s i ei 7
A i
j W
us 3
o o
11.
10-s 3 ee C
7 I
a; s
p H
l2 3
g J
w 00cc 10-7 1
?
G"I 7
O m
l 0::
O-
)
e 10-e t I
s 3
10-s :,
1'01 f02 l'08 50 YR DOSE COMMITMENT (PERSON-REM) 1
ERRTHQURKE RISK RNRLYSIS EXXON NR BONE UNSM00THED CCDF 10-8 1 1
7 5
3 10-4
^
7a s
3 X 10-8 1
?^
[
A i
LU u) 3 oo 11.
10-e e=
l o
I I
i 1
e i
l--
I 3
y J
e-.
G3 a-10-7 1 7
a l
l C
a E
i L
3 0
10-e 3 I
I l
s 5l 3
10-e :,
I f08 1'01 1~01 50 YR DOSE COMMITMENT (PERSON-REM)
. For all figures, the confidence bounds on the smallest dose point included in the cc swanation were used as the confidence bounds for the cc distribution.
This approximation assumes the cc probability is dominated by the probability of the smallest dose point.
If the assumption is not true (e.g., at smallest dose values of the cc curve) then the confidence bounds may be somewhat conservative.
The confidence bounds used are those gifen earlier and summarized in Table 2.
Because of the approximations used in obtaining them, the confidence bounds should be interpreted as only indicating the order of magnitude precision associated with the cc Cu rves.
Figures 19 through 17 present the step function cc curves obtained by applying isotonic regressions to the probability mass functions (probability versus dose) used to construct the basic cc curves in Figures 2 through 9.
The isotonic curves in Figures 10 through 17 are thus smoothed versions of the basic step function cc curves in Figures 2 through 9.
Isotonic regression is a nonparametric method of smoothing the basic step function cc curves whicie does not require assuming specific distribu-tion forms for the cc curve. -(Other approaches are called parametric approaches and involve, for example, assuming that a Weibull distribution fits the points and then finding the parameters of the best Weibull.)
Since the isotonic regression does not require as many assumptions as the parametric approaches, it is more suited to situations where there are relatively few points calculated for the cc curve--as was the case in this analysis. _The isotonic regression approach, however, does have the disadvantage that it still produces step functions and not smooth,
. continuous curves.
The isotonic regression method is explained in greater detail in the appendix to this report.
RISK TABLES 4
Table 4 tabulates the risk, defined as probability times consequence for
~
the various events analyzed in Table 3.
The risk tables indicate the contribution to the total risk fmm the various events considered. The total risk is the sum of the various contributions. The error factors on the risk contributions are mughly the error factors on the pn>bability for the event, assuming the uncertainties on the probability estimates dominate (or at most, are comparable to the cons quence uncertainties.*
- The error factors are the upper confidence level divided by the best estimate divided by the lower confidence bound.
TORNADO RISK RNAL YSIS FOR EXXON. POPULATION. LUNG SMOOTHED CCDF 10-8 1 1
7 s
3 10-e 1
^
T<
5 0
3 X 10-7 t
7m L
A I
,s
- )
m 03 3
' r i
C O
I Lt 10-8 3 o
T1 L
1 l
l l
[
--g l
l I
I
- i a
i H
e 3
w w
s,
~
G3 Cr 10-9 lM CD 7
I l-C Q::
- J a
.,I O
10-10 1 7
c l
l l
3 3
I:)
10-11 k' 2
' 1n3 104 1*05
' l'O s i
3 6
2 3
5 i 8
ro7 50 YR DOSE COMMITMENT (PERSON-REM)
TORNRDO RISK RNRLYSIS FOR EXXON.NR. BONE SMOOTHED CCDF 10-8 1 1
7 5
10-e 3 y_
7 5
s
, )
J 10-7 1 f___.
a 7
5
)
dL 3
g IL A
10-s 3 g
9 7
se 1d s
'.I w
0 -a O
3
- )
+
1 l
~~
LL 10-s 3 g
, __g_,
u O
7 kJ 5
a -o H
+i1 w
, )
l d 10-10 1
e i
e CC
-n CD 3
C E
]
l I
0-10-11 1 i
I o N
I 10-18 1 I
i l
I d '
i i
g 3
10-18 3
3
,g 10-1 100 101 1'02 50 YR DOSE COMMITMENT (PERSON-REM)
\\
i EARTHQURKi AfSK RNALYSIS EXXON POPULATION LUND SMOOTHED CCOF 10-a s
3 10-4 1 7
1 s
3 i
i I
X 10-8 1 A
5
_m_
l.LJ 03
)
O O
I 12. 10-8 1 S
o I
l l
l l
li I
l I
n l
l p
H l
3 w
T J
w CD
'l cc 10-7 1 E
7 O
ry 3
E I
CL I
3 a,
10-e i 1
1 7
s 3
C 1
10-e l
kOs 3
5 1'04 l'08 f0s 50 YR DOSE COMMITMENT (PERSON-REM)
ERRTHQURKE RISK ANALYSIS EXXON BONE POPULRTION SMOOTHED CCDF 10-s 3 7
5 3
10-4
+
7 s
3 i
X 10-s 3
~
l A
n s
IM v3 3
o C
in 10-e 3 9
o i
7 5
- s y
l-3 L
w
_J fD cc 10-7 O
i O
7 O
Q::
! L 3
ir l
10-e 3 I
I l
3 10-e 3,
5 2
l'Os 104 1'03 l'Os 50 YR DOSE COMMITMENT CPERSON-REM)
TORNRDO RISK RNALYSIS FOR EXXON.NR.LUNO SMOOTHED CC0F 10-8 1 7
5
_m_
10-s 2 7
5
)
3 10-7 1 g
i t
a n
u s
3 X
WL
^
10-s i g
g w
Y y)
% -a h)
C wL tt.
10-s i g
o 7
-TG p
l ll l
E3.-o H
-f 3 0j w
d 10-10 1
ai CD i
I CC
(
l
--- n l
LD s
Og J.
0-10-11 1 i
e u
l 7
i s
3 1 0-12 3 e3, i
t t
I 3
10-18 1
=
=
s 10-2 1*0-1 1'00 1'01 l'02 50 YR DOSE COMMITMENT (PERSON-REM)
TORNP00 RISK RNALYSIS FOR EXXON.POPULRTION. BONE SMOOTHED CCDF 10-8 1 7
5 10-8 1
-J_
1 L
l 5
O 3
=
X 10-7 1 i'
Y
^
- W s
pg 3
u a
03 O
ti.
10-e 3 I
%u
-n c
'E l I
l 1:"
5 g
p I
3 g
J
~
=
Oc 10-8
- 1 m
O r
a::
u Q.
3 i
- )
l 10-10 t 1
I I
I i
I l
3 l
P i
l' 10-11 2:
3
'7 3
5 2
08 l'Os 104
' 1*05 los 107 50 YR DOSE COMMITMENT (PERSON-REM) 1
1 ERRTHQURKE RISK RNRLYSIS EXXON NR LUND SMOOTHED CCOF 10-3 1 j
7 s
3 10-4 1 7
s i
X 10-s 3 l
L l
l i
I l
A tu Cr) 3 o
O tr. 10-8 1 O
7 5
O p
H 3
w r
J w
coc2-10-7 i 53 l
in 7
C P1 I
e E
Q-
-n, 3
10-e i 7
5 3
-!]
10-s :
~
f08 1*01 f02 i
50 YR DOSE COMMITMENT (PERSON-REM)
ERRTHQURKE RISK RNRLYSIS EXXON NR BONE SMOOTHED CCDF 10-8 1 7
s 3
10-4 1 1
7 I
s 3
>< 10-8 1 A
5 IM CO 3
oo I.i 10-8 1 O
7 M
s y
W 3
n s
a3a-10-7 i
'?
E3 7
O F1 r
ct::
3 0
10-e i 7
s 3
y 10-0 2,
f08 1'01 1'02 50 YR DOSE COMMITMENT (PERSON-REM)
TABLE 4 RISK TO NEAREST RESIDENT AND NEARBY POPULATION FROM POSTULATED DAMAGE DUE TO NATURAL PilEN0MENA Population Dose Dose at Nearest Residence (person-rem)
(remi Case *I Case Case Case Case Case Case Case I
~ Event-Organ I(0.90)
II (SE-2)
III(SE-2)
IV (3E-3)
I(0.90)
II (SE-2)
III(SE-2)
IV(3E-3)
Tornado wind Spceds 150 eph Lung 4.6E-4 2.6E-4 2.6 E-5 1.5E-5 9.2E-9 5.lE-9 5.lE-10 3.lE-10 Bone 6.8E-4 3.8E-4 3.8E-5 2.3E-5 1.4 E-8 7.5E-9 7.5E-10 4.5E-10 190 mph Lung 1.7E-3 8.4 E-4 1.5 E-4 7.6E-4 2.7E-8 1.5E-8 2.3E-9 1.4E-9 Bone 2.4E-3 1.2E-3 2.2E-4 1.lE-3 3.9E-8 2.2E-8 3.3E-9 2.0E-9 250 mph Lung
- 5. l E-5 2.0E-5 3.9E-6 1.1E-5 1.0E-8 5.6E-9 6.2E-10 3.7E-10 Bone 7.6E-5 2.7E-5 5.6 E-6 1.5E-5 1.5E-8 8.lE-9 8.9E-10 5.3E-10 Earthquake Acceleration 1.0g Lung 1.4 E-1 6.0E-2 1.0E-2 1.0E-2 2.2E-5 1.4E-5 1.4E-6 9.3E-7 Bone 2.lE-1 8.0E-2 2.0E-2 2.0E-2 3.2E-5 2.lE-5 2.0E-6 1.5E-6 Case I:
Most Likely Release (95%) and Most Likely Dispersian (95%)
Case II:
Mast Likely Release (95%) and Conservative Dispersion (5%)
Case III:
Conservative Release (5%) and Most Likely Dispersion (95%)
Case IV:
Conservative Release (5%) and Conservative Dispersion (5%)
(a) The probabilities in parenthesis are the conditional probabilities given the event, where the conditional probabilities are calculated using best estimate or conservative value for source releases and dispersion.
References 1.
" Review of Severe Weather Meteorology at Exxon Nuclear Company, Inc.,
Richland Washington," T. T. Fujita, University of Chicago, March 31, 1977.
2.
" Seismic Risk Analysis for the Westinghouse Electric Facility, Cheswick, Pennsylvania," Tera Corporation, 2150 Shattuck Avenue, Berkeley, California, 947-4, October 21, 1977.
McPherson, R.
B., and Watson, E. C., " Environmental Consequences of Posf.ulated Plutonium Releases from the Exxon Nuclear M0FP, Richland, Washington, as a Result of Severe Natural Phenomena," Battelle Pacific Northwest Laboratory, Richland, Washington 99352, PNL-2984, February 1980.
4.
Barlow, R. E., et. al., Statistical inference Under Order Restrictions, The Theory and Application of Isotoroic Regression, John Wiley and Sons, London,1972.
O
g].
a-APPENDIX ISOTONIC REGRE5SION (Ref. 5)
~
Isotonic regression was used to dev'elop the risk curves in Figures 10 thrdugh17. The only basic assumption in an isotonic regression is that the probabilitp of dose to the population or,to the nearest residence is non, increasing,as the dose increases. The assumption'is that the probability decreases' (or is constant) as the consequence increases,
~
which is not an unreasonable assumption fortrisk analyses. We should
^
note that'we make the monotonic assumption on the probability.versus dose and not'on the cc curve (which decreases by its definition'). A general statement of our isotonic regression problem is as follows:
}
s We are given a sequence of doses (D,
...D ) where Dj $_ D,j, j
n j
3 i = 1,...n-1 and we give estimates of the probability P(Dj) that the population or nearest residence receives dose Dj. We are interested in minimizing the expression:
^
n iY P(Dj) - P(Dj) 2
^
T D.
~
j i
among all isotonic functions P on the sequence (D),.. D )*
N' n
call the function that minimizes this sum of sequences (P*) the isotonic regression of P.
The isotonic regressian is thus similar to a least squares type of analysis (a us0al regression analysis) where we impose the restri.: tion that P(D ) is non-increasing as Dj j
increases Qg.
The Pool-Adjacent 'fiolators Algoritlw was used to compute P*.
Plots of the isotonic regressions.of P* versus dose are presented in Figuris 10 through 1}. The probability mass functions were used to obtain the isotonic curves.
A The isotonic regression P* of P has the following desirable properties.
1.
The isotonic regression P* of P minimizes the weighted squared error 16ss, i.e.:
n-g n
g P(D ) - P*(D )
Dj i
-P(Dj)
P(D )
D j
9 i=
i=
~
for any isotonic function P.
3 2.
The isotonic regression P* of P minimizes the error in the risk,'
},
i.e.:
max n
n.
E D P*(Dj)
E D P(Dg) i=1 i
j g
i=1 max n
~.
n n
i i
[ D P,(D ) =
D P (D )
g 9
g 1=1 1=
h
_