Making the Most of Uncertain Low-Level Measurements Presented to the Savannah River Chapter of the Health Physics Society Aiken, South Carolina, 2011 April 15 Daniel J. Strom, Kevin E. Joyce, Jay A. MacLellan, David J. Watson, Timothy P. Lynch, Cheryl. L. Antonio, Alan Birchall, Kevin K. Anderson, Peter A. Zharov Pacific Northwest National Laboratory strom@pnl.gov +1 509 375 2626 PNNL-SA-75679 Prologue • Uncertainty is different for sets of sets of data than it is for single data points • If you have more than one uncertain measurement, you need to learn about measurement error models • HPs generally do not speak the language of statisticians well enough to be comprehended – σ is not a synonym for standard deviation – s is not σ is not ˆ • We have to get smarter! – Or some biostatistician will commit regression calibration on our numbers! Carroll RJ, D Ruppert, LA Stefanski, and CM Crainiceanu. 2006. Measurement Error in Nonlinear Models: A Modern Perspective. Chapman & Hall/CRC, Boca Raton. 2 Outline • • • • • • • • Censoring The lognormal distribution Measurements and measurands Requirements and assumptions for this novel method Population variability and measurement uncertainty Disaggregating the variance Distribution of measurands The “everybody” prior 3 Outline 2 • • • • • • • Probability distributions for individual measurands The Bayesian approach The “everybody else” prior Applications to real radiobioassay data The importance of accurate uncertainty Bohr’s correspondence principle Conclusions 4 Censoring • Changing a measurement result • Common practices – Set negative values to 0 – Set all results less than some value to • 0 • ½ the value • The value • A non-numeric character like “M” • Changing measurement results causes great problems in statistical inference – DR Helsel. 2005. Nondetects and data analysis. Statistics for censored environmental data. John Wiley & Sons. • This method requires uncensored data 5 The Lognormal Distribution • Frequently observed in Nature • Multiplication of arbitrary distributions results in lognormals Ott WR. 1990. A Physical Explanation of the Lognormality of Pollutant Concentrations. J.Air Waste Mgt.Assoc. 40 (10):1378-1383 6 Measurand, Measurement, Error, and Uncertainty (ISO) • measurand: particular quantity subject to measurement – also, the “true value of the quantity subject to measurement” • result of a measurement: value attributed to a measurand, obtained by measurement • error: the unknown difference between the measurand and the measurement – this is a different meaning from the theoretical concept in statistics! • uncertainty: a quantitative estimate of the magnitude of the error – statisticians often do not distinguish between error and uncertainty and may use them synonymously 7 Requirements and Assumptions • This method requires uncensored data – small values are reported as they are calculated, with no rounding, setting negative values to zero, or otherwise changing • Assume measurands are lognormally distributed – Many populations in nature are lognormally distributed – Lognormal common in radiological and environmental measurements – Other functions could be used as long as they have a mean 8 Population Variability and Measurement Uncertainty • The sample variance of a set of measurements on a population arises from two sources: – population variability – measurement error • If measurements have no error, then all observed sample variance is due to variability in the population 9 Measurement Error Model • True values (measurands) ti give rise to measured values mi • We have good independent estimates of the combined standard uncertainty ui of each measurement mi mi = ti + ui ui ~ N(0, ui2) • We calculate the sample variance of mi • We use sample variance and a summary measure of the ui to estimate the variance due to population variability of ti 10 Spread of Measurement Results (Sample Variance) Is Due to 2 Causes “Average” Measurement Uncertainty Variability within Population 11 Spread of Measurement Results (Sample Variance) Is Due to 2 Causes uRMS Variability within Population 12 Spread of Measurement Results (Sample Variance) Is Due to 2 Causes uRMS θ ˆ (ti ) 2 s 2 (mi ) ˆ 2 (ti ) uRMS 13 Estimating the Variance of the Distribution of Measurands ˆ 2 i Estimated Variance of the Measurands ☑ Calculated s 2 xi Sample Variance of the Measurements ☑ Known - 2 uRMS Mean Square Measurement Uncertainty ☑ Known • The “reliability” or “attenuation” or “variability fraction” 2 2 is ˆ ˆ ( t ) (ti ) 2 i r 2 2 2 ˆ (ti ) uRMS s (ti ) • Analogous to a correlation coefficient – r2: fraction of variance explained by model – r′ 2: fraction of variance due to measurand variability 14 Distribution of Measurands • The estimated variance of the measurands is ˆ 2 (ti ) • Assume measurands are lognormally distributed • Assume the expectation of the measurands equals the mean of the measurements: E (t ) m – measurements are unbiased – this assumption respects the data • Calculate the parameters of the lognormal – geometric mean – geometric standard deviation sG • This is the distribution of “possibly true values” 15 Analysis of Baseline Radiobioassay Data • 90Sr: 128 baseline urine bioassays – Everyone is exposed to global fallout – gas proportional counter – 100-minute counts • 137Cs: 5,337 baseline in vivo bioassays – Everyone is exposed to global fallout & Chernobyl – coaxial high-purity germanium (HPGe) scanning system – 10-minute scans • 239+240Pu: 3,270 baseline urine bioassays – All exposure is occupational; essentially no environmental exposure in North America – α-spectrometry – ~2,520 minute counts 16 probability density The “Everybody” Probability Density Function (PDF): A Distribution of Possibly True Values Histogram of data 90Sr (mBq/day) probability density probability density • Histogram and PDF have identical arithmetic means PDF of measurands 239Pu (µBq/sample) 137Cs (mBq/kg) Probability Distributions for Individual Measurands • Now that we have the lognormal PDF of all measurands, what can we say about individual measurands? • Each individual’s measurand is somewhere within the population of measurands • We now assume that each mi, ui pair is the mean and standard deviation of the Normal “likelihood” PDF for individual i • Assume the ith measurement was the last one made in the population – When the ith measurement was made, the other M1 m and u values were known • Use this with Bayes’s theorem 18 The Bayesian Approach to Assigning Possibly True Results to Individuals (Likelihoo d PDF)(Prior PDF) Posterior PDF Normalizing Factor Thomas Bayes 1702 – 1761 19 Bayesian Method for Individuals • Instead of the “everybody” PDF, the “everybody else” PDF is used as the prior for each individual • Each individual’s likelihood is a normal distribution with mean mi and standard deviation ui • Using Bayes’s theorem, we developed a method to derive a posterior probability density function (PDF) for each individual’s measurand ti p (ti | mi , ui ) p (mi | ti , ui ) p (ti | {mk i , uk i }) p(m | t , u ) p(t | {m i i i 0 20 i k i , uk i }) d ti Applications to Real Radiobioassay Data Data Analyte 90 Sr (mBq/day) 137 Cs (mBq/kg) m x N ss(x (x ii)) r '2 128 3.61 5.08 0.151 5,337 50.6 112.0 0.350 239-240 Pu (µBq) 3,270 1.24 61 0.471 239-240 Pu (µBq) 3,268 0.040 34 -0.67 Impossible! For Pu measurements, either the uncertainties ui are overestimated, or a covariance term has been neglected. 21 Variability Fractions r′2 137Cs r´2=0.35 uRMS s(mi ) ˆ (ti ) s(mi ) 22 137Cs Variability Fractions r′2 90Sr 90Sr r´2=0.15 137Cs r´2=0.35 uRMS s(mi ) ˆ (ti ) s(mi ) 23 137Cs Variability Fractions r′2 239Pu 239Pu r´2~0 90Sr 90Sr r´2=0.15 137Cs r´2=0.35 uRMS s(mi ) ˆ (ti ) s(mi ) 24 137Cs 90Sr Results for 4 Individuals Uncensored Data Are Critical! Measurand Negative Result Result ≈ 0 Measurement Likelihood PDF Prior Result ≈ Average Result = Large Positive 25 A Movie of 128 90Sr Results • Short Dashes (Green): Likelihood (Data) • Long Dashes (Red): Everybody Else Prior • Solid (Blue): Posterior 26 90Sr Measurand Mean of90 Arithmetic 90 90 of Arithmetic etic Mean Measurand Sr of Mean Measurand Arithmetic Sr of (mBq/day) Mean Sr Measurand 90Sr of(mBq/day) (mBq/day)(mB Mean Measurand Arithmetic (mBq/day) 25 90 / 4242..68 68 33;3..31 uu RMS uu RMS uRMS .31 31;; RMS u 31 RMS //// 2 uuu 2 3 . 31 RMS 2 3 . RMS / 2 3 . 31 RMS /// 22 .3.34 ;; ; RMS 2 .31 uRMS 2 3 . 31 RMS 0.707u RMS uuuuu / 2 2 . 34 / 2 2 34 RMS / 2 3 . 31 RMS 222 .31 //22 RMS 2...34 .334 34 ;; RMS uuu 34 RMS RMS ///2 2 2 34 RMS u 2 2 . ; 0.5u RMS 242 212 .17 RMS////4 ..34 17 ;;; RMS uRMS . 34 RMS uuuu 1 . / 2 2 . 34 / 4 1 17 RMS 4 1 . 17 RMS RMS 1112..17 .17 34; uuuRMS //42 .25u 4 RMS 17 RMS RMS ///4 RMS u 4 1 . 17 u 4 1 . 17 ; RMS s ( data ) 5 . 08 ; RMS s ( data) 5.08 u 4 . 68 ; / 4 1 . 17 u / 4 1 . 17 ;; s ( data) 5.08 Measurand RMS u=0 ss(((RMS data )) 55..;17 08 RMS 68 . 4 data) 5.08 / 4 1 ; ussu 4 . 68 data 08 RMS ;08; 68 4...68 )4 usyRMS ((RMS data) 5.08 data 5 . ux udata) x y x ; 68 4 u yu = RMS s ( 5.08 y x RMS RMS u //)44422...68 ..31 (RMS data 5..;3308 08 ; ;; xx uyyssyy=RMS 68 uyuu 68 RMS xx RMS 31 ( data ) 5 ; Measurand x u / 2 3 . 31 4 . 68 u 4 . 68 ; RMS uuuyRMS RMS 31;; .31 3333..31 RMS xx///// 22222 RMS uuuyyRMS 31 RMS . ..31 xx/// 22 RMS .33.34 ;; ; RMS uu 2 31 uyRMS 2 3 . 31 RMS 0.707u RMS u / 2 2 . 34 34 2 2 / RMS y = x y = x u / 2 3 . 31 RMS 222 .31 //22 y =u xRMS RMS ;; 34 .334 2...34 RMS uuu 34 RMS ///2 2 2 34 RMS ; . 2 2 u 0.5u RMS 242 212 .17 RMS////4 ..34 RMS uRMS .17 34;;; RMS uuuu 1 . 2 2 . 34 17 1 4 / RMS / 4 1 . 17 RMS RMS u / 2 2 . 34 ; uuRMS // 444 111..17 .25u 17 RMS 17 RMS RMS RMS u / 4 1 . 17 17;; /44)5.08 (RMS data 511.17 ...08 RMS sussu ((RMS data) / 1 u / 4 17 ;; data) 5.08 u=0 08 . 5 ) data ( s RMS ssus(((RMS data) 5.08 / 4 1 . 17 ; 08; 5.08 data ))5.08 data) 5 data x x yysyssy =(((x data) 5.08 data x ) 5.08; rr2 00..15 58 22 0.58 r rrr222 000...79 58 r22 0.79 58 79 00..95 79 rrrr2222 79 000...79 95 r 22 95 22 0.95 r r 0 rrrr22222 000 ...15 95 0 15 95 r 0 2 15 000..15 rrrrr222222 00.58 15 58 000...15 rrr222 58 22 0.58 r rrr222 000...79 58 r22 0.79 58 79 79 00..95 rr 22 r 79 95 000...79 rr2222 95 00.95 rr22 r 95 000..95 rr222 r2 0 r 0 r 2 0 Sr Measurands v Measurements 20 25 2525 20 1520 20 20 15 15 15 10 10 15 10 10 yysyy( xxxx ) 5.08; data x yyy xx yx 5 5 1055 0 -10 -5 0 5 10 15 20 000 -15 Arithmetic Mean of 90Sr Likelihood (mBq/day) -10 -5-5 00 5 5 10 10 15 15 20 20 25 -15 -10 -15 -10 5 -15 90Sr Likelihood (mBq/day) Arithmetic Mean of 90Sr Likelihood (mBq/day) Arithmetic Mean of 90 Arithmetic Mean of Sr 27 Likelihood (mBq/day) 25 25 90Sr Measurand Mean90of90 Arithmetic 90 90 90 of ofSrArithmetic (mBq/day) Mean Mean ic etic Measurand Measurand Sr of of (mBq/ (mB Mean Mean Measurand Measurand Arithmetic Arithmetic SrSr of (mBq/day) Mean Sr Measurand 90Sr of(mBq/day) (mBq/day) Mean Measurand Arithmetic (mBq/day) 25 90 20 2025 25 25 25 20 1520 20 15 20 15 20 15 15 // 68 33;3..31 uu 4422422...68 RMS uu RMS uRMS / .31 31;;; RMS RMS 68 u 31 / 3 . RMS uuu / 2 3 . 31 u 4 . 68 ; RMS 4 . 68 / 2 3 . 31 RMS RMS RMS / 2 3 . 31 22..68 33.31 ..31 2242 2 ..;34 ;; ;; ///// 68 334 RMS uRMS 4 RMS 31 RMS uu 0.707u RMS RMS uuuuu 2 2 . 2 34 2 3 . 31 RMS / RMS ..31 2 334 .31 RMS //2222 2 2....33334 .34 ; ;; RMS 2 31 uuu 2 34 RMS 2 ; RMS ////2 2 2 34 RMS 2 . 31 2 3 . 31 u 2 2 . 34 ; RMS 0.707u 0.5u RMS uuuuRMS 2 22...17 ..17 34 RMS/////2 42 11 .34 ;;;; ; 2 334 .31 RMS RMS 2 34 RMS 4 1 2 3 . 31 2 2 RMS u / RMS u / 4 . 17 RMS 2 2 . 34 RMS 4 1 . 17 RMS RMS 1122112.....17 .17 34;; uuRMS //2 34 44242 .25u 0.5u 4 17 RMS RMS 17 34 RMS////4 RMS 2 2 . 34;;; RMS usuu 1 . 17 u 4 1 . 17 RMS ussu / 24 .5.08 25 .17 34 RMS / 1 17 ( data ) . 08 RMS ( data) 4 1 . 4 68 ; RMS RMS 4 1 . 17 ;; u / 2 2 . 34 RMS ( data) 5.08 Measurand u / 4 1 . 17 RMS u=0 .25u ((RMS data )) 55..17 .;17 08 RMS ///4 1 68 . 4 ((RMS data) 5.08 RMS 4 1 . ; RMS uuussuussRMS 4 . 68 4 1 17 data 08 RMS 68 4 44))44 RMS data) 5.08 (RMS data ..;;;08 ;; RMS RMS ....68 68 uysu . / 15 .17 sy=(RMS (RMS data 5 08 // 1 17 ux udata) 5.08 x x 68 4 u yu=0 s ( data ) 5 . 08 ;; y x RMS ; 68 4 2.268 317 ..31 ssyRMS ((RMS data ) 5 08 / 4 1 . ; ssu ((RMS data) 5.08 u 4 x u 4 . 68 ; y x RMS RMS y = x data) 5.08 RMS 31 3 / data 5 ; ;; RMS Uncertainty Assigned Measurand yy=yyRMS xxx/xx/)424 5.08 33.;08 ..31 ...68 usu 68 RMS uyu x RMS 31 . 3 2 u ( data) RMS RMS 4 68 u / 2 31 RMS ///2)42242 ...31 data xxx// (RMS 5.33.;334 08 ; ;;; uuu ..31 uyyyssyyRMS 68 RMS 31 308 2....68 RMS x 31 RMS 31 3 2 2 ; ( data ) 5 x RMS u / 2 3 . 31 u 4 68 u 4 68 ; RMS / 2 3 . 31 ; RMS 0.707u uuuuyRMS == xxx/x////222 RMS 2 . 34 RMS ; 34 . 2 2 3 . 31 yy RMS y RMS 2 3 . 31 u 2 3 . 31 ; RMS 2 . 34 y =u x ;; 34 .34 2...3334 RMS uuyRMS ..31 xx////2222222 222 ; RMS 34 RMS 31 ; 34 . 2 2 3 . 31 RMS 0.5u 0.707u RMS y uuuuu / 2 2 . 34 ; 34 . 2 2 / RMS ..334 ;;; ; .31 RMS RMS RMS 222 112211 2....17 .17 34 RMS /////424244 u 334 .31 RMS RMS 17 u RMS 34 RMS u 17 RMS u / 2 2 . ; 34 uuRMS ///424244 112211.....17 .25u 17 0.5u RMS 17 RMS 34 RMS 17 ; RMS u / 4 1 . 17 17 42)5.08 u(RMS .08 34;;;; RMS RMS 24 25 (RMS data ...34 RMS 17 111522...17 suussu data) /////4 1 RMS 4 17 u 2 . 34 RMS ( data) 5.08 4 1 17 u=0 .25u RMS ; 08 . ) data ( s RMS u ///4 1 ..17 data) 5.08 RMS 4 1 . 17 ; RMS ussus((RMS 4 1 17 08 5 ) data ( RMS RMS data) 5.08 data ..08 uyssyussy=(((RMS //44))) 15 08;;; 5.17 5.08 data (RMS 1 17 x x data) yu=0 x 08 . 5 data ( RMS x data) 5.08 s(data ) 5.08; 000...58 15 58 rrrrr222222 15 0 . 58 58 rrr222 15 000...79 58 79 2 r 0 . 58 79 rr222 00..79 58 79 79 rrr222 0 . 95 0 . 79 58 00..79 95 95 rrr2222222 0 . 95 0 . 79 95 . 95 r 0 rrrr22222 000 ...15 95 79 0 15 2 95 222 r 0 . 15 0000..15 rrrrr2222222 95 15 15 . 0 58 0 95 rrr22222 0 . 15 58 0000...15 rrrr2222 58 0 . 58 r 0 222 15 58 . 58 rrrr2222 0000...79 15 58 79 22 0.58 r 79 rrr222 000...79 79 58 79 95 rr2222 0 . 79 58 95 00..79 95 rrrr222 0 . 95 95 0 . 79 0..79 95 95 rrr22222 0 0 0 . 95 00.95 rr2222 0.95 rr 22 0 r2 0 r 0 r 2 0 Sr Measurands v Measurements 10 5.08 155..17 ssyussyy((((RMS data) xxxx/ 4)) 08;; data 10 10 data) 5.08 data 08 y y x y x ysyy = x 15 xxxxx )5.08 data 5..08 08;; yyssyyy(((data) x data ) 5 x 10 10 15 yyy xxx yx 5 5 555 10 10 0 -10 -5 0 5 10 15 20 000 -15 Arithmetic Mean of 90Sr Likelihood (mBq/day) 0 -10 -5-5 00 5 5 10 10 15 15 20 20 25 -15 -10 -15 -10 5 -15 90Sr Likelihood (mBq/day) Mean of 90 -10 Arithmetic -5 0 5 Sr Likelihood 10 15 20 25 5 -15 Arithmetic Mean of (mBq/day) 90 Arithmetic Mean of Sr 28 Likelihood (mBq/day) 25 25 90Sr 90Sr 90Sr 90Sr of of (mBq/day) (mBq/day) Mean Mean ic etic Measurand Measurand of of (mBq/ (mB Mean Mean Measurand Measurand Arithmetic Arithmetic 90 Arithmetic Mean of Sr Measurand (mBq/day) 20 2025 25 25 000...58 15 58 rrrrr222222 15 0 . 58 58 rrr222 15 000...79 58 79 2 r 0 . 58 79 rr222 00..79 58 79 79 rrr222 0 . 95 0 . 79 58 00..79 95 95 rrr2222222 0 . 95 0 . 79 95 . 95 r 0 rrrr22222 000 ...15 95 79 0 15 2 95 222 r 0 . 15 0000..15 rrrrr2222222 95 15 15 . 0 58 0 95 rrr22222 0 . 15 58 0000...15 rrrr2222 58 0 . 58 r 0 222 15 58 . 58 rrrr2222 0000...79 15 58 79 22 0.58 r 79 rrr222 000...79 79 58 79 95 rr2222 0 . 79 58 95 00..79 95 rrrr222 0 . 95 95 0 . 79 0..79 95 95 rrr22222 0 0 0 . 95 00.95 rr2222 0.95 rr 22 0 r2 0 r 0 r 2 0 Effect of Reducing Uncertainty 1520 15 20 20 15 // 68 33;3..31 uu 4422422...68 RMS uu RMS uRMS / .31 31;;; RMS RMS 68 u 31 / 3 . RMS uuu / 2 3 . 31 u 4 . 68 ; RMS 4 . 68 / 2 3 . 31 RMS RMS RMS / 2 3 . 31 22..68 33.31 ..31 2242 2 ..;34 ;; ;; ///// 68 334 RMS uRMS 4 RMS 31 RMS uu 0.707u RMS RMS uuuuu 2 2 . 2 34 2 3 . 31 RMS / RMS ..31 2 334 .31 RMS //2222 2 2....33334 .34 ; ;; RMS 2 31 uuu 2 34 RMS 2 ; RMS ////2 2 2 34 RMS 2 . 31 2 3 . 31 u 2 2 . 34 ; RMS 0.707u 0.5u RMS uuuuRMS 2 22...17 ..17 34 RMS/////2 42 11 .34 ;;;; ; 2 334 .31 RMS RMS 2 34 RMS 4 1 2 3 . 31 2 2 RMS u / RMS u / 4 . 17 RMS 2 2 . 34 RMS 4 1 . 17 RMS RMS 1122112.....17 .17 34;; uuRMS //2 34 44242 .25u 0.5u 4 17 RMS RMS 17 34 RMS////4 RMS 2 2 . 34;;; RMS usuu 1 . 17 u 4 1 . 17 RMS ussu / 24 .5.08 25 .17 34 RMS / 1 17 ( data ) . 08 RMS ( data) 4 1 . 4 68 ; RMS RMS 4 1 . 17 ;; u / 2 2 . 34 RMS ( data) 5.08 u / 4 1 . 17 RMS u=0 .25u ((RMS data )) 55..17 .;17 08 RMS ///4 1 68 . 4 ((RMS data) 5.08 RMS 4 1 . ; RMS uuussuussRMS 4 . 68 4 1 17 data 08 RMS 68 4 44))44 RMS data) 5.08 (RMS data ..;;;08 ;; RMS RMS ....68 68 uysu . / 15 .17 sy=(RMS (RMS data 5 08 // 1 17 udata) 5.08 x x 68 4 u yu=0 x s ( data ) 5 . 08 ;; y x RMS ; 68 4 2.268 317 ..31 ssyRMS ((RMS data ) 5 08 / 4 1 . ; ssu ((RMS data) 5.08 u 4 x u 4 . 68 ; y x RMS RMS data) 5.08 RMS 31 3 / data 5 ; ;; RMS Uncertainty Assigned yy=yyRMS xxx/xx/)424 5.08 33.;08 ..31 ...68 usu 68 RMS uyu x RMS 31 . 3 2 u ( data) RMS RMS 4 68 u / 2 31 RMS ///2)42242 ...31 data xxx// (RMS 5.33.;334 08 ; ;;; uuu ..31 uyyyssyyRMS 68 RMS 31 308 2....68 RMS x 31 RMS 31 3 2 2 ; ( data ) 5 x RMS u / 2 3 . 31 u 4 68 u 4 68 ; RMS / 2 3 . 31 ; RMS 0.707u uuuuyRMS = xx/x////222 RMS 2 . 34 RMS ; 34 . 2 2 3 . 31 y RMS y RMS 2 3 . 31 u 2 3 . 31 ; RMS u 2 . 34 ;; 34 .34 2...3334 RMS uuyRMS ..31 xx////2222222 222 ; RMS 34 RMS 31 ; 34 . 2 2 3 . 31 RMS 0.5u 0.707u RMS y uuuuu / 2 2 . 34 ; 34 . 2 2 / RMS ..334 ;;; ; .31 RMS RMS RMS 222 112211 2....17 .17 34 RMS /////424244 u 334 .31 RMS RMS 17 u RMS 34 RMS u 17 RMS u / 2 2 . ; 34 uuRMS ///424244 112211.....17 .25u 17 0.5u RMS 17 RMS 34 RMS 17 ; RMS u / 4 1 . 17 17 42)5.08 u(RMS .08 34;;;; RMS RMS 24 25 (RMS data ...34 RMS 17 111522...17 suussu data) /////4 1 RMS 4 17 u 2 . 34 RMS ( data) 5.08 4 1 17 u=0 .25u RMS ; 08 . ) data ( s RMS u ///4 1 ..17 data) 5.08 RMS 4 1 . 17 ; RMS ussus((RMS 4 1 17 08 5 ) data ( RMS RMS data) 5.08 data ..08 uyssyussy=(((RMS //44))) 15 08;;; 5.17 5.08 data (RMS 1 17 x x data) yu=0 x 08 . 5 data ( RMS x data) 5.08 s(data ) 5.08; 10 10 15 1510 5 55 10 10 00 0 -15 5 -15 5 -15 5.08 155..17 ssyussyy((((RMS data) xxxx/ 4)) 08;; data data) 5.08 data 08 y y x y x ysyy = x xxxxx )5.08 data 5..08 08;; yyssyyy(((data) x data ) 5 x yyy xxx yx uRMS 4.68; 2 r 2 0.15 uRMS 4.68; r 0.15 uRMS / 2 3.312; r2 0.58 uRMS / 2 3.31; r 0.258 uRMS / 2 2.34; 2 r 0.79 uRMS / 2 2.34; r 0.79 uRMS / 4 1.17; 2 r 2 0.95 uRMS / 4 1.17; r 0.95 -10-10 -5-5 00 5 5 10 10 s(data 15 )15 5.08 20; 220 r2 25 025 90Sr Likelihood s(data ) 5.08; 20 r 0 25 Mean of (mBq/day) -10 Arithmetic -5 0 5 10 15 yx Arithmetic Mean of 90Sr Likelihood (mBq/day) 29 yx 90Sr 90Sr 90Sr 90Sr of of (mBq/day) (mBq/day) Mean Mean ic etic Measurand Measurand of of (mBq/ (mB Mean Mean Measurand Measurand Arithmetic Arithmetic 90 Arithmetic Mean of Sr Measurand (mBq/day) 20 2025 25 25 // 68 33;3..31 uu 4422422...68 RMS uu RMS uRMS / .31 31;;; RMS RMS 68 u 31 / 3 . RMS uuu / 2 3 . 31 u 4 . 68 ; RMS / 2 3 . 31 4 . 68 RMS RMS / 2 3 . 31 RMS RMS 22..68 33.31 ..31 2242 2 ..;34 ;; ;; RMS ///// 68 334 uRMS 4 RMS 31 RMS 0.707u uu RMS uuuuu 2 2 . RMS 2 34 2 3 . 31 / RMS RMS ..31 2 334 .31 RMS //2222 2 2....33334 .34 ; ;; RMS 2 31 uuu 2 34 RMS 2 ; RMS ////2 2 2 34 RMS 2 . 31 2 3 . 31 u 2 2 . 34 ; RMS 0.5u 0.707u RMS uuuuRMS 2 22...17 ..17 34 RMS/////2 42 11 .34 ;;;; ; 2 334 .31 RMS RMS 2 34 RMS 4 1 2 3 . 31 2 2 RMS u / RMS u / 4 . 17 RMS 2 2 . 34 RMS 4 1 . 17 RMS RMS 1122112.....17 .17 34;; uuRMS //2 34 44242 .25u 4 17 0.5u RMS RMS 17 34 RMS////4 RMS 2 2 . 34;;; u 4 1 . 17 RMS usuu 1 . 17 RMS ussu / 24 .5.08 25 .17 34 RMS / 1 17 ( data ) . 08 RMS ( data) 4 1 . 4 68 ; RMS RMS 4 1 . 17 ;; u / 2 2 . 34 RMS ( data) 5.08 u / 4 1 . 17 RMS u=0 .25u ((RMS data )) 55..17 .;17 08 RMS / 4 1 68 . 4 ((RMS data) 5.08 RMS / 4 1 . ; RMS uuussuussRMS 4 . 68 / 4 1 17 data 08 u RMS 68 4 44))44 RMS data) 5.08 (RMS data ..;;;08 ;; RMS RMS ....68 68 uysu . / 15 .17 sy=(RMS (RMS data 5 08 / 1 17 data) 5.08 x x 68 4 u yu=0 x s ( data ) 5 . 08 ;; y x RMS ; 68 4 ///44) 2.268 317 ..31 ssyRMS ((RMS data 5 08 1 . ; ssu ((RMS data) 5.08 u x u 4 . 68 ; y x RMS RMS data) 5.08 RMS 31 3 data 5 ; ;; RMS Uncertainty Assigned yy=yyRMS xxx/xx/)424 5.08 33.;08 ..31 ...68 usu 68 RMS uyu x RMS 31 . 3 2 u 0.707u ( data) RMS RMS 4 68 u / 2 31 RMS ///2)42242 ...31 data xxx// (RMS 5.33.;334 08 ; ;;; uuu ..31 uyyyssyyRMS 68 RMS 31 308 2....68 RMS x 31 RMS 31 3 2 2 ; ( data ) 5 x RMS u / 2 3 . 31 u 4 68 u 4 68 ; RMS / 2 3 . 31 ; RMS 0.707u uuuuyRMS x/x////222 RMS 2 . 34 RMS ; 34 . 2 2 3 . 31 RMS y RMS 2 3 . 31 u 2 3 . 31 ; RMS u 2 . 34 ;; 34 .34 2...3334 222 uuyRMS ////2 ..31 yRMS = xx 222 ; RMS 22 34 RMS 31 ; 34 . 2 2 2 3 . 31 RMS 0.5u 0.707u RMS y x uuuuu / 2 2 . 34 ; 34 . 2 2 / RMS ..334 ;;; ; .31 RMS RMS RMS 222 112211 2....17 .17 34 RMS /////424244 u 334 .31 RMS RMS 17 u RMS 34 RMS u 17 RMS u / 2 2 . ; 34 uuRMS ///424244 112211.....17 .25u 17 0.5u RMS 17 RMS 34 RMS 17 ; RMS u / 4 1 . 17 17 42)5.08 u(RMS .08 34;;;; RMS RMS 24 25 (RMS data ...34 17 111522...17 RMS suussu data) /////4 1 RMS 4 17 u 2 . 34 RMS ( data) 5.08 4 1 17 u=0 .25u RMS ; 08 . ) data ( s RMS u ///4 1 ..17 data) 5.08 RMS 4 1 . 17 ; RMS ussus((RMS 4 1 17 08 5 ) data ( RMS RMS data) 5.08 data ..08 uyssyussy=(((RMS //44))) 15 08;;; 5.17 5.08 data (RMS 1 17 x data) x yu=0 x 08 . 5 data ( RMS x data) 5.08 data 5.08 08;; 155..17 yyussyy(((RMS xxxx/ 4)) s data) data 08 y x y x ysyy xxx )5.08 data s=y((xdata) 5.08; 00..15 58 rrrrr222222 0 . 58 15 0 . 58 58 rrr222 15 000...79 58 79 2 r 0 . 58 79 rrr222 000...79 58 79 79 95 2 rr222 0 . 79 58 00..79 95 95 rrrr222222 0 . 95 0 . 79 95 . 95 0 rrrr22222 000 ...15 95 79 0 15 2 95 222 r 0 . 15 0000..15 rrrrr2222222 95 15 15 . 0 58 0 95 rrr22222 0 . 15 58 0000...15 rrrr2222 58 0 . 58 r 0 222 15 58 . 58 rrrr2222 0000...79 15 58 79 22 0.58 r 79 rrr222 000...79 79 58 79 95 rr2222 0 . 79 58 95 00..79 95 rrrr2222 0 . 95 95 0 . 79 0..79 95 95 rrr22222 0 0 0 . 95 00.95 rr2222 0.95 rr 22 0 r2 0 r 0 r 2 0 Effect of Reducing Uncertainty 1520 15 20 20 15 10 10 15 1510 5 55 10 10 00 0 -15 5 -15 5 -15 yysyy( xxxx ) 5.08; data x yyy xx yx uRMS 4.68; r 2 0.15 uRMS / 2 3.31; r2 0.58 uRMS / 2 2.34; r 2 0.79 uRMS / 4 1.17; r 2 0.95 s(data ) 5.08; r 2 0 yx uRMS 4.68; 2 r 2 0.15 uRMS 4.68; r 0.15 uRMS / 2 3.312; r2 0.58 uRMS / 2 3.31; r 0.258 uRMS / 2 2.34; 2 r 0.79 uRMS / 2 2.34; r 0.79 uRMS / 4 1.17; 2 r 2 0.95 uRMS / 4 1.17; r 0.95 -10-10 -5-5 00 5 5 10 10 s(data 15 )15 5.08 20; 220 r2 25 025 90Sr Likelihood s(data ) 5.08; 20 r 0 25 Mean of (mBq/day) -10 Arithmetic -5 0 5 10 15 yx Arithmetic Mean of 90Sr Likelihood (mBq/day) 30 yx 90Sr 90Sr 90Sr 90Sr of of (mBq/day) (mBq/day) Mean Mean ic etic Measurand Measurand of of (mBq/ (mB Mean Mean Measurand Measurand Arithmetic Arithmetic 90 Arithmetic Mean of Sr Measurand (mBq/day) 20 2025 25 25 // 68 33;3..31 uu 4422422...68 RMS uu RMS uRMS / .31 31;;; RMS RMS 68 u 31 / 3 . RMS uuu / 2 3 . 31 u 4 . 68 ; RMS / 2 3 . 31 4 . 68 RMS RMS / 2 3 . 31 RMS RMS 22..68 33.31 ..31 2242 2 ..;34 ;; ;; RMS ///// 68 334 uRMS 4 RMS 31 RMS 0.707u uu RMS uuuuu 2 2 . RMS 2 34 2 3 . 31 / RMS RMS ..31 2 334 .31 RMS //2222 2 2....33334 .34 ; ;; RMS 2 31 uuu 2 34 RMS 2 ; RMS ////2 2 2 34 RMS 2 . 31 u 2 2 . 34 ; 2 3 . 31 RMS 0.5u 0.707u RMS uuuuRMS 22...17 ..17 34 RMS/////2 11 .34 ;;;; ; 2 334 .31 RMS RMS 2 34 RMS 424242 112 2 3 . 31 2 RMS u / RMS u / 4 . 17 RMS 2 . 34 RMS . 17 RMS 1122112.....17 .17 34;; uuRMS //2 34 44242 .25u 4 17 0.5u RMS RMS 17 34 RMS////4 RMS u 4 1 . 17 2 2 . 34;;; RMS usuu 1 . 17 RMS u / 2 2 . 34 RMS u / 4 1 17 s ( data ) 5 . 08 RMS ( data) 5.08 4 1 . 17 4 . 68 ; RMS RMS 4 1 . 17 ;; u / 2 2 . 34 RMS s ( data) 5.08 u / 4 1 . 17 RMS u=0 .25u data )) 55..17 .;17 08 RMS / 4 1 68 . 4 ((RMS data) 5.08 RMS u( / 4 1 . ; RMS uuussuussRMS 4 . 68 RMS / 4 1 17 ( data 08 RMS 68 4 44))44 RMS data) 5.08 (RMS data ..;;;08 ;; RMS RMS ....68 68 uysu . / 15 .17 sy=(RMS (RMS data 5 08 / 1 17 data) 5.08 x x 68 4 u yu=0 x s ( data ) 5 . 08 ;; y x RMS ; 68 4 ///44) 2.268 317 ..31 ssyRMS ((RMS data 5 08 1 . ; ssu ((RMS data) 5.08 u x u 4 . 68 ; y x 0.707u RMS RMS data) 5.08 RMS 31 3 data 5 ; ;; RMS Uncertainty Assigned yy=yyRMS xxx/xx/)424 5.08 33.;08 ..31 ...68 usu 68 RMS uyu x RMS 31 . 3 2 u ( data) RMS RMS 4 68 u / 2 31 RMS ///2)42242 ...31 data xxx// (RMS 5.33.;334 08 ; ;;; uuu ..31 uyyyssyyRMS 68 RMS 31 308 2....68 RMS x 31 0.5u RMS 31 3 2 2 ; ( data ) 5 x RMS u / 2 3 . 31 u 4 68 u 4 68 ; RMS / 2 3 . 31 ; RMS 0.707u uuuuyRMS x/x////222 RMS 2 . 34 RMS ; 34 . 2 2 3 . 31 RMS y RMS 2 3 . 31 u 2 3 . 31 ; RMS u 2 . 34 ;; 34 .34 2...3334 uuyRMS ..31 x////2222222 222 ; RMS yRMS = xx 34 RMS 31 ; 34 . 2 2 3 . 31 RMS 0.5u 0.707u RMS y uuuuu / 2 2 . 34 ; 34 . 2 2 / RMS ..334 ;;; ; .31 RMS RMS RMS 222 112211 2....17 .17 34 RMS /////424244 u 334 .31 RMS RMS 17 u RMS 34 RMS u 17 RMS u / 2 2 . ; 34 uuRMS ///424244 112211.....17 .25u 17 0.5u RMS 17 34 RMS RMS 17 ; RMS u / 4 1 . 17 17 42)5.08 u(RMS .08 34;;; RMS RMS 24 25 (RMS data ...34 17 111522...17 RMS suussu data) /////4 1 RMS 4 17 ;; u 2 . 34 RMS ( data) 5.08 4 1 17 u=0 .25u RMS 08 . ) data ( s RMS u / 4 1 . 17 data) 5.08 RMS / 4 1 . 17 ; RMS ussus((RMS / 4 1 . 17 08;; data data) (RMS data ))) 5.08 5155.17 ..08 RMS uyssyussy=((RMS / 4 1 08 data ((RMS / 4 17 x data) 5.08 x yu=0 x ; 08 . 5 ) data ( RMS xxx/ 4)5.08 data) data 5 08 1 . 17 ; yyussyy(((RMS s data) 5.08 data 5.08; xxxxx )5.08 ysyy x ) 5.08; s=y((xdata) data 00..15 58 rrrrrr222222 0 . 58 15 0 . 58 58 15 000...79 58 rrr22222 79 0 . 58 79 rrr22 000...79 79 58 79 95 2 rr222 0 . 79 58 00..79 95 95 rrrr222222 0 . 95 95 0 . 79 . 95 0 rrrr22222 000 ...15 95 79 0 15 2 95 222 r 0 . 15 0000..15 rrrrr2222222 95 15 15 . 0 58 0 95 rrr22222 0 . 15 58 0000...15 rrrr2222 58 0 . 58 r 0 222 15 58 . 58 rrrr2222 0000...79 15 58 79 22 0.58 r 79 rrr222 000...79 79 58 79 95 rr2222 0 . 79 58 95 00..79 95 rrrr2222 0 . 95 95 0 . 79 0..79 95 95 rrr22222 0 0 0 . 95 00.95 rr2222 0.95 rr 22 0 r2 0 r 0 r 2 0 Effect of Reducing Uncertainty 1520 15 20 20 15 10 10 15 1510 5 55 10 10 00 0 -15 5 -15 5 -15 uRMS 4.68; r 2 0.15 uRMS / 2 3.31; r22 0.58 uRMS 4.68; r 2 0.15 r 0.79 uRMS / 2 2.34; uRMS / 2 3.31; r22 0.58 r 0.95 uRMS / 4 1.17; uRMS / 2 2.34; r 22 0.79 r 0 s(data ) 5.08; uRMS / 4 1.17; r 2 0.95 yx 2 s ( data ) 5 . 08 ; r 0 x yysyy( x x data ) 5 . 08 ; x yyy xxx yx yx uRMS 4.68; 2 r 2 0.15 uRMS 4.68; r 0.15 uRMS / 2 3.312; r2 0.58 uRMS / 2 3.31; r 0.258 uRMS / 2 2.34; 2 r 0.79 uRMS / 2 2.34; r 0.79 uRMS / 4 1.17; 2 r 2 0.95 uRMS / 4 1.17; r 0.95 -10-10 -5-5 00 5 5 10 10 s(data 15 )15 5.08 20; 220 r2 25 025 90Sr Likelihood s(data ) 5.08; 20 r 0 25 Mean of (mBq/day) -10 Arithmetic -5 0 5 10 15 yx Arithmetic Mean of 90Sr Likelihood (mBq/day) 31 yx 90Sr 90Sr 90Sr 90Sr of of (mBq/day) (mBq/day) Mean Mean ic etic Measurand Measurand of of (mBq/ (mB Mean Mean Measurand Measurand Arithmetic Arithmetic 90 Arithmetic Mean of Sr Measurand (mBq/day) 20 2025 25 25 00..15 58 rrrrr222222 0 . 58 15 0 . 58 58 rrr222 15 000...79 58 79 2 r 0 . 58 79 rr222 00..79 58 79 79 rrr222 0 . 95 0 . 79 58 00..79 95 95 rrrr2222222 0 . 95 0 . 79 95 . 95 0 rrrr22222 000 ...15 95 79 0 15 2 95 222 r 0 . 15 0000..15 rrrrr2222222 95 15 15 . 0 58 0 95 rrr22222 0 . 15 58 0000...15 rrrr2222 58 0 . 58 r 0 222 15 58 . 58 rrrr2222 0000...79 15 58 79 22 0.58 r 79 rrr222 000...79 79 58 79 95 rr2222 0 . 79 58 95 00..79 95 rrrr2222 0 . 95 95 0 . 79 0..79 95 95 rrr22222 0 0 0 . 95 00.95 rr2222 0.95 rr 22 0 r2 0 r 0 r 2 0 uRMS 4.68; r 2 0.15 Effect of Reducing Uncertainty 1520 15 20 20 15 // 68 33;3..31 uu 4422422...68 RMS uu RMS uRMS / .31 31;;; RMS RMS 68 u 31 / 3 . RMS uuu / 2 3 . 31 u 4 . 68 ; RMS 4 . 68 / 2 3 . 31 RMS RMS / 2 3 . 31 RMS RMS 22..68 33.31 ..31 2242 2 ..;34 ;; ;; RMS ///// 68 334 uRMS 4 RMS 31 RMS 0.707u uu RMS uuuuu 2 2 . RMS 2 34 2 3 . 31 RMS / RMS ..31 2 334 .31 RMS //2222 2 2....33334 .34 ; ;; RMS 2 31 uuu 2 34 RMS 2 ; RMS ////2 2 2 34 RMS 2 . 31 2 3 . 31 u 2 2 . 34 ; RMS 0.5u 0.707u RMS uuuuRMS 2 22...17 ..17 34 RMS/////2 42 11 .34 ;;;; ; 2 334 .31 RMS RMS 2 34 RMS 4 1 2 3 . 31 2 2 RMS u / RMS u / 4 . 17 RMS 2 2 . 34 RMS 4 1 . 17 RMS RMS 1122112.....17 .17 34;; uuRMS //2 34 44242 .25u 4 17 0.5u RMS RMS 17 34 RMS////4 RMS 2 2 . 34;;; RMS u 4 1 . 17 usuu 1 . 17 RMS ussu / 24 .5.08 25 .17 34 RMS / 1 17 ( data ) . 08 RMS ( data) 4 1 . 4 68 ; RMS RMS 4 1 . 17 ;; u / 2 2 . 34 RMS ( data) 5.08 u / 4 1 . 17 RMS u=0 .25u data )) 55..17 .;17 08 RMS / 4 1 68 . 4 u( ((RMS data) 5.08 RMS / 4 1 . ; RMS uuussuussRMS 4 . 68 RMS / 4 1 17 ( data 08 RMS 68 4 44))44 RMS data) 5.08 (RMS data ..;;;08 ;; RMS RMS ....68 68 uysu . / 15 .17 sy=(RMS (RMS data 5 08 // 1 17 data) 5.08 x x 68 4 u yu=0 x s ( data ) 5 . 08 ;; 0.707u y x RMS ; 68 4 2.268 317 ..31 ssyRMS ((RMS data ) 5 08 / 4 1 . ; ssu ((RMS data) 5.08 u 4 x u 4 . 68 ; y x RMS RMS data) 5.08 RMS 31 3 / data 5 ; ;; RMS Uncertainty Assigned yy=yyRMS xxx/xx/)424 5.08 33.;08 ..31 ...68 usu 68 RMS uyu x RMS 31 . 3 2 u 0.5u ( data) RMS RMS 4 68 u / 2 31 RMS ///2)42242 ...31 data xxx// (RMS 5.33.;334 08 ; ;;; uuu ..31 uyyyssyyRMS 68 RMS 31 308 2....68 RMS x 31 RMS 31 3 2 2 ; ( data ) 5 x RMS u / 2 3 . 31 u 4 68 u 4 68 ; RMS / 2 3 . 31 ; RMS 0.707u uuuuyRMS .25u x/x////222 RMS 2 . 34 RMS ; 34 . 2 2 3 . 31 RMS y RMS 2 3 . 31 u 2 3 . 31 ; RMS u 2 . 34 ;; 34 .34 2...3334 RMS uuyRMS ..31 x////2222222 222 ; RMS 34 RMS 31 ; 34 . 2 2 3 . 31 RMS yRMS = x/x 0.5u 0.707u RMS y uuuuu 2 2 . 34 ; 34 . 2 2 / ..334 ;;; ; .31 RMS RMS RMS 222 112211 2....17 .17 34 RMS /////424244 u 334 .31 RMS RMS 17 u RMS 34 RMS u 17 RMS u / 2 2 . ; 34 uuRMS ///424244 112211.....17 .25u 17 0.5u RMS 17 RMS 34 RMS 17 ; RMS u / 4 1 . 17 17 42)5.08 u(RMS .08 34;;;; RMS RMS 24 25 (RMS data ...34 17 111522...17 RMS suussu data) /////4 1 RMS 4 17 u 2 . 34 RMS ( data) 5.08 4 1 17 u=0 .25u RMS ; 08 . ) data ( s RMS u ///4 1 ..17 data) 5.08 RMS 4 1 . 17 ; RMS ussus((RMS 4 1 17 08 5 ) data ( RMS RMS data) 5.08 data ..08 uyssyussy=(((RMS //44))) 15 08;;; 5.17 5.08 data (RMS 1 17 x data) x yu=0 x 08 . 5 data ( RMS x data) 5.08 08; us(data / 4) 15.17 10 10 15 1510 5 55 10 10 00 0 -15 5 -15 5 -15 uRMS / 2 3.31; r22 0.58 r 2 0.15 uRMS 4.68; r 2 0.79 uRMS / 2 2.34; uRMS / 2 3.31; r2 0.58 95 r 2 0.15 1.;17; uRMS /44.68 r 2 0.79 uRMS / 2 2.34; 0.58 ; ; r22 data us(RMS / ) 2 5.08 3.31 r 0.95 uRMS / 4 1.17; x/ 2 2.34; uyRMS r 22 0.79 r 0 s(data ) 5.08; uRMS / 4 1.17; r 2 0.95 yx ssyssy((((RMS data) xxxx )) 08;; 5..08 5.08 data data) 5.08 data 5 y y 2 xxxxx )5.08 yysyy s ( data ) 5 . 08 ; r 0 sysy=yyy(((xdata) data 5 . 08 ; x xxx ) 5.08; yy data yy xxx yx yx uRMS 4.68; 2 r 2 0.15 uRMS 4.68; r 0.15 uRMS / 2 3.312; r2 0.58 uRMS / 2 3.31; r 0.258 uRMS / 2 2.34; 2 r 0.79 uRMS / 2 2.34; r 0.79 uRMS / 4 1.17; 2 r 2 0.95 uRMS / 4 1.17; r 0.95 -10-10 -5-5 00 5 5 10 10 s(data 15 )15 5.08 20; 220 r2 25 025 90Sr Likelihood s(data ) 5.08; 20 r 0 25 Mean of (mBq/day) -10 Arithmetic -5 0 5 10 15 yx Arithmetic Mean of 90Sr Likelihood (mBq/day) 32 yx 90Sr 90Sr 90Sr 90Sr of of (mBq/day) (mBq/day) Mean Mean ic etic Measurand Measurand of of (mBq/ (mB Mean Mean Measurand Measurand Arithmetic Arithmetic 90 Arithmetic Mean of Sr Measurand (mBq/day) 20 2025 25 25 // 68 33;3..31 uu 4422422...68 RMS uu RMS uRMS / .31 31;;; RMS RMS 68 u 31 / 3 . RMS uuu / 2 3 . 31 u 4 . 68 ; RMS / 2 3 . 31 4 . 68 RMS RMS / 2 3 . 31 RMS RMS 22..68 33.31 ..31 2242 2 ..;34 ;; ;; RMS ///// 68 334 uRMS 4 RMS 31 RMS 0.707u uu RMS uuuuu 2 2 . RMS 2 34 2 3 . 31 / RMS RMS ..31 2 334 .31 RMS //2222 2 2....33334 .34 ; ;; RMS 2 31 uuu 2 34 RMS 2 ; RMS ////2 2 2 34 RMS 2 . 31 2 3 . 31 u 2 2 . 34 ; RMS 0.5u 0.707u RMS uuuuRMS 2 22...17 ..17 34 RMS/////2 42 11 .34 ;;;; ; 2 334 .31 RMS RMS 2 34 RMS 4 1 2 3 . 31 2 2 RMS u / RMS u / 4 . 17 RMS 2 2 . 34 RMS 4 1 . 17 RMS RMS 1122112.....17 .17 34;; uuRMS //2 34 44242 .25u 4 17 0.5u RMS RMS 17 34 RMS////4 RMS u 4 1 . 17 2 2 . 34;;; RMS usuu 1 . 17 RMS u / 2 2 . 34 RMS u / 4 1 17 s ( data ) 5 . 08 RMS ( data) 5.08 4 1 . 17 4 . 68 ; RMS RMS 4 1 . 17 ;; u / 2 2 . 34 RMS s ( data) 5.08 u / 4 1 . 17 RMS u=0 .25u data )) 55..17 .;17 08 u(RMS / 4 1 68 . 4 ((RMS data) 5.08 RMS / 4 1 . ; RMS uuussuussRMS 4 . 68 RMS / 4 1 17 ( data 08 RMS 68 4 44))44 RMS data) 5.08 (RMS data ..;;;08 ;; RMS RMS ....68 68 uysu . / 15 .17 sy=(RMS (RMS data 5 08 / 1 17 0.707u data) 5.08 x x 68 4 u yu=0 x s ( data ) 5 . 08 ;; y x RMS ; 68 4 ///44) 2.268 317 ..31 ssyRMS ((RMS data 5 08 1 . ; ssu ((RMS data) 5.08 u x u 4 . 68 ; y x RMS RMS data) 5.08 0.5u RMS 31 3 data 5 ; ;; RMS Uncertainty Assigned yy=yyRMS xxx/xx/)424 5.08 33.;08 ..31 ...68 usu 68 RMS uyu x RMS 31 . 3 2 u ( data) RMS RMS 4 68 u / 2 31 RMS ///2)42242 ...31 data xxx// (RMS 5.33.;334 08 ; ;;; uuu ..31 uyyyssyyRMS 68 RMS .25u 31 308 2....68 RMS x 31 RMS 31 3 2 2 ; ( data ) 5 x RMS u / 2 3 . 31 u 4 68 u 4 68 ; RMS / 2 3 . 31 ; RMS 0.707u uuuuyRMS x/x////222 RMS 2 . 34 RMS ; 34 . 2 2 3 . 31 u=0 RMS y RMS 2 3 . 31 u 2 3 . 31 ; RMS u 2 . 34 ;; 34 .34 2...3334 RMS uuyRMS ..31 x////2222222 222 ; RMS 34 RMS 31 ; 34 . 2 2 3 . 31 RMS 0.5u 0.707u yRMS = x/x RMS y uuuuu 2 2 . 34 ; 34 . 2 2 / ..334 ;;; ; .31 RMS RMS RMS 222 112211 2....17 .17 34 RMS /////424244 u 334 .31 RMS RMS 17 u RMS 34 RMS u 17 RMS u / 2 2 . ; 34 uuRMS ///424244 112211.....17 .25u 17 0.5u RMS 17 34 RMS RMS 17 ; RMS u / 4 1 . 17 17 42)5.08 u(RMS .08 34;;; RMS RMS 24 25 (RMS data ...34 17 111522...17 RMS suussu data) /////4 1 RMS 4 17 ;; u 2 . 34 RMS ( data) 5.08 4 1 17 u=0 .25u RMS 08 . ) data ( s RMS u / 4 1 . 17 data) 5.08 RMS / 4 1 . 17 ; RMS ussus((RMS / 4 1 . 17 08;; data data) (RMS data ))) 5.08 5155.17 ..08 RMS uyssyussy=((RMS / 4 1 08 data ((RMS / 4 17 x data) 5.08 x yu=0 x ; 08 . 5 ) data ( RMS xxx/ 4)5.08 data) data 5 08 1 . 17 ; yyussyy(((RMS s data) 5.08 data 5.08; xxxxx )5.08 ysyy x ) 5.08; s=y((xdata) data 00..15 58 rrrrrr222222 0 . 58 15 0 . 58 58 15 000...79 58 rrr22222 79 0 . 58 79 rrr22 000...79 58 79 79 95 2 rr222 0 . 79 58 00..79 95 95 rrrr222222 0 . 95 95 0 . 79 . 95 0 rrrr22222 000 ...15 95 79 0 15 2 95 222 r 0 . 15 0000..15 rrrrr2222222 95 15 15 . 0 58 0 95 rrr22222 0 . 15 58 0000...15 rrrr2222 58 0 . 58 r 0 222 15 58 . 58 rrrr2222 0000...79 15 58 79 22 0.58 r 79 rrr222 000...79 79 58 79 95 rr2222 0 . 79 58 95 00..79 95 rrrr2222 0 . 95 95 0 . 79 0..79 95 95 rrr22222 0 0 0 . 95 00.95 rr2222 0.95 rr 22 0 r2 0 r 0 r 2 0 uRMS 4.68; r 2 0.15 Effect of Reducing Uncertainty 1520 15 20 20 15 10 10 15 1510 5 55 10 10 00 0 -15 5 -15 5 -15 uRMS / 2 3.31; r22 0.58 r 2 0.15 uRMS 4.68; r 2 0.79 uRMS / 2 2.34; uRMS / 2 3.31; r2 0.58 95 r 2 0.15 1.;17; uRMS /44.68 r 2 0.79 uRMS / 2 2.34; 0.58 ; ; r22 data us(RMS / ) 2 5.08 3.31 r 0.95 uRMS / 4 1.17; x/ 2 2.34; uyRMS r 22 0.79 r 0 s(data ) 5.08; uRMS / 4 1.17; r 2 2 0.95 x 4.68; y uRMS r 0.15 2 s ( data ) 5 . 08 ; r 2 0 x yysyy( x x data ) 5 . 08 ; x uRMS / 2 3.31; r 0.58 x yyy x yx xx y uRMS / 2 2.34; r 2 02 .79 uRMS 4.68; 2 r 0.15 uRMS 0.015 uRMS /44.68 1; .17; r r 2 .95 uRMS / 2 3.312; 2 r2 0.58 uRMS / 2) 53.08 31; r r 0.20 58 s(data uRMS / 2 2.34; 2 r 0.79 y / x2 2.34; uRMS r 0.79 uRMS / 4 1.17; 2 r 2 0.95 uRMS / 4 1.17; r 0.95 -10-10 -5-5 00 5 5 10 10 s(data 15 )15 5.08 20; 220 r2 25 025 90Sr Likelihood s(data ) 5.08; 20 r 0 25 Mean of (mBq/day) -10 Arithmetic -5 0 5 10 15 yx Arithmetic Mean of 90Sr Likelihood (mBq/day) 33 yx Visualizing Uncertainty Reduction uRMS r´2 = 0.15 ‹ σ(νi) 34 Visualizing Uncertainty Reduction uRMS r´2 = 0.15 r´2 = 0.57 ‹ σ(νi) 35 Visualizing Uncertainty Reduction uRMS r´2 = 0.15 r´2 = 0.57 r´2 = 0.78 ‹ σ(νi) 36 Visualizing Uncertainty Reduction uRMS r´2 = 0.15 r´2 = 0.57 r´2 = 0.78 r´2 = 0.94 ‹ σ(νi) 37 Visualizing Uncertainty Reduction uRMS r´2 ≈ 0 r´2 = 0.15 r´2 = 0.57 r´2 = 0.78 r´2 = 0.94 ‹ σ(νi) 38 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 The Common View: The Measurement Is the Measurand 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 Oh, no! Results are below some level (DL, DT, LOD, etc.). Might not be real! 500 0 -500 0 500 1000 1500 Oops! Activity < 0 is meaningless. -500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 39 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 The Bayesian View: The Measurement and the Prior Give the Measurand 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 40 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 The Bayesian View: The Measurement and the Prior Give the Measurand 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 41 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 42 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 43 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 44 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 45 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 46 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 47 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 48 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 49 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 50 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 51 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 52 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 53 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 54 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 55 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 56 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 57 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 58 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 59 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 60 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 61 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 62 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 63 2000 1500 500 1000 000 500 500 Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 000 The Bayesian View: The Measurement and the Prior Give the Measurand 2000 Measurand Measurement y= x Measurand Measurand Measurand yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) 64 2000 000 1500 500 1000 000 500 500 137137 of of Mean Measurand Arithmetic CsCs (mBq/kg) Mean Measurand(mBq/kg) Arithmetic 5,337 137Cs Measurements Showing 2000 Uncertainty in Measurements & Measurands 2000 Measurand Measurand Measurand Measurement Measurand Measurand Measurand y= y =x x yyy= ==xxx 1500 y= x 1000 500 0 -500 -500 0 500 1000 1500 Arithmetic Mean of 137Cs Likelihood (mBq/kg) (mBq/kg) 65 2000 90Sr 90Sr 90Sr 90Sr 90 of of (mBq/day) Mean Mean ic etic Measurand Measurand of of (mBq/ (mB Mean Mean Measurand Measurand Arithmetic Arithmetic of (mBq/day) Mean Arithmetic Sr Measurand (mBq/day) // 68 33;3..31 uu 4422422...68 RMS uu RMS uRMS / .31 31;;; RMS RMS 68 u 31 / 3 . RMS uuu / 2 3 . 31 u 4 . 68 ; RMS 4 . 68 / 2 3 . 31 RMS RMS RMS / 2 3 . 31 22..68 33.31 ..31 2242 2 ..;34 ;; ;; ///// 68 334 RMS uRMS 4 RMS 31 RMS uu 0.707u RMS RMS uuuuu 2 2 . 2 34 2 3 . 31 RMS / RMS ..31 2 334 .31 RMS //2222 2 2....33334 .34 ; ;; RMS 2 31 uuu 2 34 RMS 2 ; RMS ////2 2 2 34 RMS 2 . 31 2 3 . 31 u 2 2 . 34 ; RMS 0.707u 0.5u RMS uuuuRMS 2 22...17 ..17 34 RMS/////2 42 11 .34 ;;;; ; 2 334 .31 RMS RMS 2 34 RMS 4 1 2 3 . 31 2 2 RMS u / RMS u / 4 . 17 RMS 2 2 . 34 RMS 4 1 . 17 RMS RMS 1122112.....17 .17 34;; uuRMS //2 34 44242 .25u 0.5u 4 17 RMS RMS 17 34 RMS////4 RMS 2 2 . 34;;; RMS usuu 1 . 17 u 4 1 . 17 RMS ussu / 24 .5.08 25 .17 34 RMS / 1 17 ( data ) . 08 RMS ( data) 4 1 . 4 68 ; RMS RMS 4 1 . 17 ;; u / 2 2 . 34 RMS ( data) 5.08 Measurand u / 4 1 . 17 RMS u=0 .25u ((RMS data )) 55..17 .;17 08 RMS ///4 1 68 . 4 ((RMS data) 5.08 RMS 4 1 . ; RMS uuussuussRMS 4 . 68 4 1 17 data 08 RMS 68 4 44))44 RMS data) 5.08 (RMS data ..;;;08 ;; RMS RMS ....68 68 uysu . / 15 .17 sy=(RMS (RMS data 5 08 // 1 17 data) 5.08 x x 68 4 u yu=0 x s ( data ) 5 . 08 ;; y x RMS ; 68 4 2.268 317 ..31 ssyRMS ((RMS data ) 5 08 / 4 1 . ; ssu ((RMS data) 5.08 u 4 x u 4 . 68 ; y x RMS RMS y = x data) 5.08 RMS 31 3 / data 5 ; ;; RMS Uncertainty Assigned yy=yyRMS xxx/xx/)424 5.08 33.;08 ..31 ...68 usu 68 RMS uyu x RMS Measurand 31 . 3 2 u ( data) RMS RMS 4 68 u / 2 31 RMS ///2)42242 ...31 data xxx// (RMS 5.33.;334 08 ; ;;; uuu ..31 uyyyssyyRMS 68 RMS 31 308 2....68 RMS x 31 RMS 31 3 2 2 ; ( data ) 5 x RMS u / 2 3 . 31 u 4 68 u 4 68 ; RMS / 2 3 . 31 ; RMS 0.707u uuuuyRMS x/x////222 RMS 2 . 34 RMS ; 34 . 2 2 3 . 31 RMS y RMS 2 3 . 31 u 2 3 . 31 ; RMS u 2 . 34 ;; 34 .34 2...3334 y= x RMS uuyRMS ..31 xx////2222222 222 ; RMS 34 RMS 31 ; 34 . 2 2 3 . 31 RMS 0.5u 0.707u RMS y uuuuu / 2 2 . 34 ; 34 . 2 2 / RMS ..334 ;;; ; .31 RMS RMS RMS 222 112211 2....17 .17 34 RMS /////424244 u 334 .31 RMS RMS 17 u RMS 34 RMS u 17 RMS u / 2 2 . ; 34 uuRMS ///424244 112211.....17 .25u 17 0.5u RMS 17 RMS 34 RMS 17 ; RMS u / 4 1 . 17 17 42)5.08 u(RMS .08 34;;;; RMS RMS 24 25 (RMS data ...34 RMS 17 111522...17 suussu data) /////4 1 RMS 4 17 u 2 . 34 RMS ( data) 5.08 4 1 17 u=0 .25u RMS ; 08 . ) data ( s RMS u ///4 1 ..17 data) 5.08 RMS 4 1 . 17 ; RMS ussus((RMS 4 1 17 08 5 ) data ( RMS RMS data) 5.08 data ..08 uyssyussy=(((RMS //44))) 15 08;;; 5.17 5.08 data (RMS 1 17 x x data) yu=0 x 08 . 5 data ( RMS x data) 5.08 s(data ) 5.08; 000...58 15 58 rrrrr222222 15 0 . 58 58 rrr222 15 000...79 58 79 2 r 0 . 58 79 rr222 00..79 58 79 79 rrr222 0 . 95 0 . 79 58 00..79 95 95 rrr2222222 0 . 95 0 . 79 95 . 95 r 0 rrrr22222 000 ...15 95 79 0 15 2 95 222 r 0 . 15 0000..15 rrrrr2222222 95 15 15 . 0 58 0 95 rrr22222 0 . 15 58 0000...15 rrrr2222 58 0 . 58 r 0 222 15 58 . 58 rrrr2222 0000...79 15 58 79 22 0.58 r 79 rrr222 000...79 79 58 79 95 rr2222 0 . 79 58 95 00..79 95 rrrr222 0 . 95 95 0 . 79 0..79 95 95 rrr22222 0 0 0 . 95 00.95 rr2222 0.95 rr 22 0 r2 0 r 0 r 2 0 2000 2000 20 25 25 20 u 1500 Arithmetic Mean of 137 Cs Measurand (mBq/kg) Arithmetic Mean of 137Cs Measurand (mBq/kg) 25 5,337 137Cs20 Results with Same Uncertainty 1500 1000 1000 500500 00 -500 -500 15 15 20 y = x 15 20 10 5.08 155..17 ssyussyy((((RMS data) xxxx/ 4)) 08;; data 10 10 data) 5.08 data 08 y y x y x ysyy = x 15 xxxxx )5.08 data 5..08 08;; yyssyyy(((data) x data ) 5 x 15 yyy xxx yx 5 5 5 10 10 0 -10 -5 0 5 10 15 0 -15 90Sr Likelihood (mBq/d Arithmetic Mean of 0 0 500 1000 1500 -10 -5 0 5 10 2000 15 5 -15 Arithmetic Mean of 137 (mBq/kg) -10 -5 Cs Likelihood 0 5 10 15 5 -15 90 Arithmetic Mean of Sr Likelihood (mBq/day) 66 0 500 1000 Arithmetic Mean of 137 Cs Likelihood (mBq/kg) 1500 2000 90Sr 90Sr 90Sr 90Sr 90 of of (mBq/day) Mean Mean ic etic Measurand Measurand of of (mBq/ (mB Mean Mean Measurand Measurand Arithmetic Arithmetic of (mBq/day) Mean Arithmetic Sr Measurand (mBq/day) 1500 1500 1000 1000 500500 00 -500 -500 000...58 15 58 rrrrr222222 15 0 . 58 58 rrr222 15 000...79 58 79 2 r 0 . 58 79 rr222 00..79 58 79 79 rrr222 0 . 95 0 . 79 58 00..79 95 95 rrr2222222 0 . 95 0 . 79 95 . 95 r 0 rrrr22222 000 ...15 95 79 0 15 2 95 222 r 0 . 15 0000..15 rrrrr2222222 95 15 15 . 0 58 0 95 rrr22222 0 . 15 58 0000...15 rrrr2222 58 0 . 58 r 0 222 15 58 . 58 rrrr2222 0000...79 15 58 79 22 0.58 r 79 rrr222 000...79 79 58 79 95 rr2222 0 . 79 58 95 00..79 95 rrrr222 0 . 95 95 0 . 79 0..79 95 95 rrr22222 0 0 0 . 95 00.95 rr2222 0.95 rr 22 0 r2 0 r 0 r 2 0 137Cs Results 25 5,331 20 20 25 25 20 u Arithmetic Mean of 137 Cs Measurand (mBq/kg) Arithmetic Mean of 137Cs Measurand (mBq/kg) 2000 2000 // 68 33;3..31 uu 4422422...68 RMS uu RMS uRMS / .31 31;;; RMS RMS 68 u 31 / 3 . RMS uuu / 2 3 . 31 u 4 . 68 ; RMS 4 . 68 / 2 3 . 31 RMS RMS RMS / 2 3 . 31 22..68 33.31 ..31 2242 2 ..;34 ;; ;; ///// 68 334 RMS uRMS 4 RMS 31 RMS uu 0.707u RMS RMS uuuuu 2 2 . 2 34 2 3 . 31 RMS / RMS ..31 2 334 .31 RMS //2222 2 2....33334 .34 ; ;; RMS 2 31 uuu 2 34 RMS 2 ; RMS ////2 2 2 34 RMS 2 . 31 2 3 . 31 u 2 2 . 34 ; RMS 0.707u 0.5u RMS uuuuRMS 2 22...17 ..17 34 RMS/////2 42 11 .34 ;;;; ; 2 334 .31 RMS RMS 2 34 RMS 4 1 2 3 . 31 2 2 RMS u / RMS u / 4 . 17 RMS 2 2 . 34 RMS 4 1 . 17 RMS RMS 1122112.....17 .17 34;; uuRMS //2 34 44242 .25u 0.5u 4 17 RMS RMS 17 34 RMS////4 RMS 2 2 . 34;;; RMS usuu 1 . 17 u 4 1 . 17 RMS ussu / 24 .5.08 25 .17 34 RMS / 1 17 ( data ) . 08 RMS ( data) 4 1 . 4 68 ; RMS RMS 4 1 . 17 ;; u / 2 2 . 34 RMS ( data) 5.08 Measurand u / 4 1 . 17 RMS u=0 .25u ((RMS data )) 55..17 .;17 08 RMS ///4 1 68 . 4 ((RMS data) 5.08 RMS 4 1 . ; RMS uuussuussRMS 4 . 68 4 1 17 data 08 RMS 68 4 44))44 RMS data) 5.08 (RMS data ..;;;08 ;; RMS RMS ....68 68 uysu . / 15 .17 sy=(RMS (RMS data 5 08 // 1 17 data) 5.08 x x 68 4 u yu=0 x s ( data ) 5 . 08 ;; y x RMS ; 68 4 2.268 317 ..31 ssyRMS ((RMS data ) 5 08 / 4 1 . ; ssu ((RMS data) 5.08 u 4 x u 4 . 68 ; y x RMS RMS y = x data) 5.08 RMS 31 3 / data 5 ; ;; RMS Uncertainty Assigned yy=yyRMS xxx/xx/)424 5.08 33.;08 ..31 ...68 usu 68 RMS uyu x RMS Measurand 31 . 3 2 u ( data) RMS RMS 4 68 u / 2 31 RMS ///2)42242 ...31 data xxx// (RMS 5.33.;334 08 ; ;;; uuu ..31 uyyyssyyRMS 68 RMS 31 308 2....68 RMS x 31 RMS 31 3 2 2 ; ( data ) 5 x RMS u / 2 3 . 31 u 4 68 u 4 68 ; RMS / 2 3 . 31 ; RMS 0.707u uuuuyRMS x/x////222 RMS 2 . 34 RMS ; 34 . 2 2 3 . 31 RMS y RMS 2 3 . 31 u 2 3 . 31 ; RMS u 2 . 34 ;; 34 .34 2...3334 y= x RMS uuyRMS ..31 xx////2222222 222 ; RMS 34 RMS 31 ; 34 . 2 2 3 . 31 RMS 0.5u 0.707u RMS y uuuuu / 2 2 . 34 ; 34 . 2 2 / RMS ..334 ;;; ; .31 RMS RMS RMS 222 112211 2....17 .17 34 RMS /////424244 u 334 .31 RMS RMS 17 u RMS 34 RMS u 17 RMS u / 2 2 . ; 34 uuRMS ///424244 112211.....17 .25u 17 0.5u RMS 17 RMS 34 RMS 17 ; RMS u / 4 1 . 17 17 42)5.08 u(RMS .08 34;;;; RMS RMS 24 25 (RMS data ...34 RMS 17 111522...17 suussu data) /////4 1 RMS 4 17 u 2 . 34 RMS ( data) 5.08 4 1 17 u=0 .25u RMS ; 08 . ) data ( s RMS u ///4 1 ..17 data) 5.08 RMS 4 1 . 17 ; RMS ussus((RMS 4 1 17 08 5 ) data ( RMS RMS data) 5.08 data ..08 uyssyussy=(((RMS //44))) 15 08;;; 5.17 5.08 data (RMS 1 17 x x data) yu=0 x 08 . 5 data ( RMS x data) 5.08 s(data ) 5.08; 15 15 20 y = x 15 20 10 5.08 155..17 ssyussyy((((RMS data) xxxx/ 4)) 08;; data 10 10 data) 5.08 data 08 y y x y x ysyy = x 15 xxxxx )5.08 data 5..08 08;; yyssyyy(((data) x data ) 5 x 15 yyy xxx yx 5 5 5 10 10 0 -10 -5 0 5 10 15 0 -15 90Sr Likelihood (mBq/d Arithmetic Mean of 0 0 500 1000 1500 -10 -5 0 5 10 2000 15 5 -15 Arithmetic Mean of 137 (mBq/kg) -10 -5 Cs Likelihood 0 5 10 15 5 -15 90 Arithmetic Mean of Sr Likelihood (mBq/day) 67 0 500 1000 Arithmetic Mean of 137 Cs Likelihood (mBq/kg) 1500 2000 The Importance of Accurate Uncertainty Upper (red) point everybody else prior likelihood (data) posterior (measurand) • Nearly the same measurement result • s(lower) ≈ 2.5s(upper) • Upper posterior resembles likelihood (i.e., measurement) • Lower posterior resembles prior 68 Lower (yellow) point everybody else prior posterior (measurand) likelihood (data) n of 137 of (mBq/ Mean Arithmetic Cs Measurand 137 (mBq/kg) Arithmetic Mean of Cs Measurand 500 137Cs 5,331 2000 100 1000 1500 u Likelihood (mBq/kg) 1500 Measurand 137 y=x zero line Cs Results 2000 25 90Sr (mBq/day) easurand of ean Measurand of 90 Mean metic Sr Measurand(mBq/day) (mBq/day) Arithmetic Mean of 137 Cs Measurand (mBq/kg) 222 0.15 uuRMS 44..68 ; r 68 ;; rr 2 00..15 uuRMS .68 RMS 44 . 68 15 RMS u 4 . 68 22 0.15 y= x u 4 . 68 ; r RMS RMS RMS u // 44422...68 ;33..31 ;; rr222 00..58 uRMS 68 15 uuu 68 RMS RMS/ RMS 31 58 u 2 3 . 31 4 . 68 u 4 . 68 ; r 0 . 15 RMS uuuRMS RMS / 2 3 . 31 ; r 0 . 58 RMS 3 . 31 RMS /// 2 222 0.58 uuuRMS 2 3 . 31 RMS 2 3 . 31 ; r ..31 RMS //// 22 .33.34 ;; ; rrr222 000...79 RMS 2 2 31 uRMS 2 3 . 31 58 RMS 0.707u RMS uuuuu / 2 2 . 34 25 / 2 2 34 79 RMS / 2 3 . 31 RMS u / 2 3 . 31 ; r 0 . 58 / 2 2 . 34 RMS 2 2222..34 .34 34;; r 22 0.79 79 RMS uuuRMS RMS ///2 20 2 34 RMS 2 . 0.5u 1000 2 RMS u / 2 2 . 34 RMS / 2442 ..34 000...95 RMS uRMS 112112...17 .17 34;;; rrr222 79 RMS uuu 17 95 RMS/////4 4 17 RMS RMS u 2 2 . 34 ; r 0 . 79 RMS u / 4 1 . 17 .25u 4 1 17 95 2 RMS u / 4 1 . 17 500 1000 1500 2000 10 RMS RMS 22 0.95 RMS //44 25 2 u 1 . 17 u 1 . 17 ; r RMS data ..;08 ;; RMS sMeasurand data) ususus(((RMS 4 68 rrrrr22222 00000...15 / 15 /44)) ..5.08 15.17 17 95 data) 5.08 u=0 ( data . 08 ; RMS 15 ; 68 4 137 s ( data) 5.08 / 4 1 . 17 ; 95 u 4 . 68 RMS an of Cs Likelihood (mBq/kg) data 5.;08 08; 000.15 20 RMS rrr2222 68 4...68 ))4 ussyRMS su (((RMS data) 5.08 u data 5 x y x 15 00..58 r ; 68 4 y = x 2 RMS s ( data) 5.08 y x RMS RMS u 4 . 68 / 2 3 . 31 ; r 0 2 s ( data ) 5 . 08 ; r Measurand 2 y x u 4 . 68 ; 15 RMS y x u 4 . 68 22 0.58 RMS yu = x x/x RMS RMS r ; 31 . 3 2 / uuuysyRMS ( data ) 5 . 08 ; u 2 3 . 31 4 . 68 4 . 68 ; r 0 . 15 RMS u RMS 58 00..58 31;; rr22 .31 3333..31 RMS 15 500 xx///// 22222 RMS uuuyyyRMS 31 y= x u . 2 RMS 3 . 31 x RMS /// 22 .3.34 ;; ; rrr222 000...79 RMS .31 uyRMS 222 334 .31 58 RMS 0.707u x RMS uuuuu / 2 . 34 79 2 2 / RMS / 2 3 . 31 RMS u / 2 3 . 31 ; 0 . 58 / 2 2 . 34 RMS 79 r ; 34 . 2 2 2 RMS uuuRMS 222..34 RMS ///2 20 15 2 34 RMS 79 00..95 34;; rr222 .17 2 0.5u RMS u / 2 212 .17 34 RMS / 4 . RMS uRMS / 2 . 34 ; r 0 . 79 RMS uuu / 4 1 . 2 / 2 2 . 34 2 95 . 0 r ; 17 . 1 RMS //44 1 . 17 RMS RMS u 2 2 . 34 ; r 0 . 79 RMS u / 4 1 . 17 .25u 95 17 1 4 2 RMS u / 4 1 . 17 RMS RMS 2 RMS //44 u 1 . 17 95 . 0 r ; 17 . 1 u 2 RMS s ( data ) 5 . 08 ; r 0 RMS susus((RMS data) //44) 115.17 5.08 ..17 ;; rr222 95 data) 5.08 u=0 000..95 08 data RMS ssus((((RMS data) 5.08 1010-500 / 4 1 . 17 ; r 500 1000 15000 2000500 10 08 5 ) data 2 data) 1000 1500 5.08; 2000 r2 0 )5.08 data x x yysyssy =(((x data) 5.08 x 10 data ) 50..08 082000 ; r100 00 yysyy(1500 xxxx ) 5 -500 0 500 1000 Arithmetic Mean of Cs Likelihood (mBq/kg) ean of 137Cs Likelihood (mBq/kg) -500 -400 -300 -200 -100 2 data ; r 15 yyy xxx 137 Arithmetic Mean of Arithmetic Cs Likelihood y(mBq/kg) of x 137Cs Likelihood (m Mean 137 69 n of 137 of (mBq/ Mean Arithmetic Cs Measurand 137 (mBq/kg) Arithmetic Mean of Cs Measurand 500 137Cs Measurand y=x zero line 5,331 137Cs Results – Log Scale 2000 100 1000 1500 u Likelihood (mBq/kg) 1500 2000 25 25 90Sr 90Sr 90 (mBq/day) asurand easurand ofof (mBq/day) (mBq/day) an ean Measurand Measurand of (mBq/day) Mean metic Sr Measurand (mBq/day) Arithmetic Mean of 137 Cs Measurand (mBq/kg) 222 0.15 uuRMS 44..68 ; r 68 ;; rr 22 00..15 uuRMS ..68 RMS 44 . 68 15 2222 00..15 RMS 4 68 ; r u 4 . 68 y= x u 4 . 68 ; r 15 RMS RMS RMS u 4 . 68 ; r 0 . 15 RMS //4442.2.68 ;33..31 ;; rr222 uuu RMS uRMS .68 68 00..58 15 RMS RMS RMS RMS 31 58 RMS u / 2 3 . 31 4 . 68 u u 4 . 68 ; r 0 . 15 2 RMS uuuRMS 2 RMS / 2 3 . 31 ; r 0 . 58 RMS RMS 4 . 68 2 3 . 31 RMS // / 2 3 . 31 ; r 0 . 58 222 uuu 2 3 . 31 u 4 . 68 ; 15 RMS / 2 3 . 31 ; r 0 . 58 4 . 68 RMS RMS 2 2 31 RMS 22..68 33..31 ..31 58 RMS/// 2242 .3.;34 ;; ;; rrrr2222 000...79 RMS 68 334 4 15 RMS / 31 58 RMS 0.707u uu RMS u RMS uuuuu / 2 2 . RMS 25 / 2 34 79 2 3 . 31 / RMS 2 RMS 2 3 . 31 u / 2 3 . 31 ; r 0 . 58 RMS / 2 2 . 34 222 2...3334 .34 34 ; ; rr22 00..79 79 RMS uuRMS ..31 222 ; RMS ////2 20 22 34 RMS 31 2 2 . 34 ; 79 2 3 . 31 58 RMS 0.5u 1000 0.707u 2 RMS uuuuu / 2 2 . 34 2 / 2 2 . 34 ; r 0 . 79 RMS 4 1 . 17 ; 95 / 2 3 . 31 RMS RMS RMS 2 2 . 34 ; r 0 . 79 RMS / 4 1 . 17 u / 2 3 . 31 ; 58 2 2 2 . 34 RMS u .34 17 00..79 95 RMS ////4 2244 RMS u 20 RMS RMS uRMS 24 1112221112.......17 .17 34;;; rr2222 u 34 RMS 17 .25u 17 95 0.5u RMS u / 4 17 500 1000 1500 2000 10 2 34 RMS RMS / 4 r 0 . 95 RMS 25 2 u / 4 1 . 17 2 2 . 34 ; r 0 . 79 RMS u / 4 1 . 17 ; 95 2 2 RMS 25 //442424) 112...17 17 95 su (RMS data .5.08 ..34 RMS sMeasurand data) u // 1 4 68 rrrrr222222 00000....15 RMS 17 95 us(RMS .;;08 34;;; 79 ssu ((RMS data) 5.08 u 1 17 u=0 .25u RMS ( data ) 5 . 08 RMS u / 4 1 . 17 15 68 . 4 RMS 137 data) 5.08 RMS 2 / 4 1 . 17 ; 95 RMS 25 u 4 . 68 2 RMS u / 4 1 . 17 u r 0 . an of Cs Likelihood (mBq/kg) sRMS (RMS data 08;; 000.15 20 RMS 15 68 .68 4 44)))44 ((RMS data) 5.08 (RMS data 5155.17 ..;;;08 rr22222222 RMS uyssu . 68 r / 1 / 17 95 s ( data 08 data) 5.08 u . y x x 15 . 0 r 68 . 4 u y = x u=0 RMS s ( data ) 5 . 08 ; RMS y x 22 15 ; 68 . 4 u 4 . 68 sysuysyRMS (=(RMS data) 5.08 / 2 3 . 31 ; r 0 . 58 2 ( data ) 5 08 Measurand / 4 1 . 17 ; r 0 . 95 RMS x 4 . 68 ; 15 RMS x u 4 . 68 2 RMS 000..15 data) 5.08 x Uncertainty RMS RMS 58 ..31 22..68 //)42 uuuys=yRMS (RMS data 5 .;308 ; ;; rrrr22222 RMS x Assigned x / 3 . 31 u 4 68 uyu x RMS 58 . 0 31 3 s ( data) 5.08 RMS y x RMS 15 500 y x 4 . 68 u / 2 3 . 31 / 2 3 . 31 ; r 0 . 58 RMS 2 s ( data ) 5 . 08 ; r 0 y x 2 2 .15 / 2 3 . 31 y x 4 . 68 ; y= x u RMS 4 . 68 58 ; 31 . 3 2 / u 2 RMS y x RMS RMS u / 2 3 . 31 y x 2 RMS 58 . 0 r ; 31 . 3 2 / u 2 ..;34 ;; ; rrr222 (RMS 5 000...79 data x//x 68 334 .31 RMS ..68 uyysyRMS 15 /222)424 2 308 .31 58 RMS uu 0.707u RMS RMS uuuu 2 . 79 34 2 / 15 RMS x 2 RMS / 2 3 . 31 u / 2 3 . 31 ; 0 . 58 RMS 2 2 . 34 79 r ; 34 . 2 2 2 RMS ..31 xx////222222 uuyyRMS 222 ...3334 34 ;;; ; rr22 00..79 RMS 20 15 34 RMS 31 2 3 . 31 58 79 34 . 2 RMS 0.707u 0.5u RMS uuuuu / 2 2 . 34 79 34 . 2 2 RMS / 4 1 . 17 ; r 0 . 95 2 / 2 3 . 31 RMS RMS RMS 2 2 . 34 ; r 0 . 79 RMS / 4 1 . 17 / 2 3 . 31 ; 58 2 2 . 34 2 RMS u / 2 95 00..79 17 .34 RMS 2424 RMS RMS ///4 20 RMS uRMS 24 1112221112.......17 .17 34;;; rrr222 34 RMS uuu 17 0.5u .25u 95 17 RMS RMS / 4 17 2 34 / 4 0 . 95 RMS RMS 2 2 . 34 r 0 . 79 RMS ususu / 4 1 . 17 95 ; 17 . 1 4 / u(RMS 2 / 2 2 . 34 RMS 2 95 . 17 1 4 / ( data ) 5 . 08 ; r 0 RMS data) 1 RMS RMS 5.08 34;; rr222 00.79 4) 152...17 ...17 95 ((RMS data) 5.08 uussu ////4442 111 17 .25u u=0 RMS 08 data ( s RMS 17 RMS data) 5.08 RMS 1010-500 u 4 . 17 ; r 0 . 95 RMS 500 1000 15000 2000500 10 u / 4 1 . 17 08 5 ) data ( s 2 RMS ..17 ;; 2000 rr222 00.95 suyssussy=(((RMS data) 1000 1500 //44)) 115 .17 5.08 08 5 data ( data) 5.08 x x yu=0 x RMS r ; 08 . data ( RMS data) 5.08 y x 10 22 / 4) 150.17 ; r100 00.95 data 08 ssyusy(((RMS data) 5.08 x x -500 0 500 1000 1500 2000 Arithmetic Mean of 10 Cs Likelihood (mBq/kg) data) 5.08 ean of 137Cs Likelihood (mBq/kg) -500 -400 -300 -200 -100 data 5.08; r 2 0 xxxx )5.08 ysy yyss=yyy(((xdata) 15 data ) 5 . 08 ; r2 00 x x yysy(mBq/kg) xxx ) 5.08; ( data r Arithmetic Mean of 13715 Cs Likelihood x 137Cs Likelihood (m Arithmetic Mean of yyy x xx 70 y 137 Innovation 1 • this work addresses the situation in which each measurement is accompanied by a very good estimate of its uncertainty – not described in the literature reviewed – which occurs routinely in radiochemical and radiobioassay measurements 71 Innovation 2 • This work provides a solution to the vexing problem of making sense of negative measurement results for a quantity, such as activity in becquerels, which physically must be nonnegative – none of the literature addresses negative values • The method makes sense of uncertain low-level measurements – without injecting a bias into the dataset by left-censoring – by implicitly recognizing that spurious negative results are accompanied by an equal amount of spurious positive signal 72 Innovation 3 • This work provides posterior estimates, in the form of probability distributions, of the true value of each measurand – while the literature is concerned with correcting estimates of slopes of doseresponse relationships for the effects of classical measurement error 73 Innovation 4 • This work shows that accurate estimates of uncertainty are as important as the values of the measurement results – overestimates of uncertainty can lead to nonsense results 74 Innovation 5 • This work provides the ability to explore the impact of the magnitude of uncertainty on the posterior distribution of measurands by thought experiments involving substitution of the mean square measurement uncertainty, or some multiple or submultiple of it, for the individual uncertainties 75 Innovation 6 • The method is shown to closely correspond to classical (frequentist) methods when uncertainty is relatively small 76 Innovation 7 • This work answers the questions, conditional on plausible assumptions, – “What true state of nature gave rise to this set of observations?” – “For each individual measurement result, what are the probable values of the measurand that led to this measurement result?” • The authors believe that the method represents a significant step forward in the making sense of groups of uncertain, low-level radioactivity measurements 77 Conclusions • Sample variance of a set of measurements is disaggregated into – measurement uncertainty – population variability • A reasonable, possible distribution of measurands for a population is the result • When x 0 , positive posterior PDFs of the measurand are computed using “everybody else” priors for each individual – negative values are eliminated – mean of measurements is preserved • When there is essentially no variance in the data due to population variability, the method cannot be expected to work, and it does not work 78 Conclusions: Utility and Correspondence • The method eliminates negative measurement results in an uncensored data set and preserves the arithmetic mean of the data set – If measurement results have a relatively large uncertainty, the posterior PDF of the measurand resembles the prior – If the measurement results have a relatively small uncertainty, the posterior PDF of the measurand resembles the likelihood, that is, it is relatively close to the measurement result before application of the Bayesian methods • Best estimate of uncertainty is just as important as measurement! • As required by Bohr’s correspondence principle, results produced by the methods introduced here correspond to results of traditional statistical inference in the domain in which that inference is known to be correct 79 Authors Strom MacLellan Joyce Watson Lynch Antonio Zharov (Mayak PA) Birchall (UK HPA) (Scherpelz, Vasilenko) Acknowledgments • PNNL: Kevin Anderson, Gene Carbaugh, Michelle Johnson, Bruce Napier, Bob Scherpelz, Paul Stansbury, Rick Traub • SUBI: Vadim Vostrotin 80