2. Topics Covered
Brief overview of concepts and terms of
probability and statistics
Measurement uncertainty
Detection and quantification limits
Miscellaneous
3. Target Audience(s)
Project planners and managers
Radiochemists and technicians
Computer programmers
Data validators and assessors
Metrologists?
4. Measurement Uncertainty
We use the Guide to the Expression of
Uncertainty in Measurement (ISO-
GUM).
International guidance – years of
development and review by seven
international organizations
Strongly recommended by NIST
Best way to ensure consistency among
labs in the U.S. and the rest of the world
5. Measurement Model
Define the measurand – the quantity subject
to measurement
Determine a mathematical model, with input
quantities, X1,X2,…,XN, and (at least) one
output quantity,Y.
The values determined for the input quantities
are called input estimates and are denoted by
x1,x2,…,xN.
The value calculated for the output quantity is
called the output estimate and denoted by y.
6. Standard Uncertainty
The standard uncertainty of a measured
value is the uncertainty expressed as an
estimated standard deviation – i.e., the one-
sigma uncertainty.
The standard uncertainty of an input
estimate, xi, is denoted by u(xi).
The standard uncertainty of the output
estimate, y, determined by uncertainty
propagation, is called the combined standard
uncertainty, and is denoted by uc(y).
7. Type A Evaluation
Statistical evaluation of uncertainty
involving a series of observations
Always has an associated number of
degrees of freedom
Examples include simple averages and
least-squares estimates
Not “random uncertainty”
8. Type B Evaluation
Any evaluation that is not a Type A evaluation
is a Type B evaluation.
Not “systematic uncertainty”
Examples:
Calculating Poisson counting uncertainty (error) as
the square root of the observed count
Using professional judgment combined with
assumed rectangular or triangular distributions
Obtaining standard uncertainties in any manner
from standard certificates or reference books
9. Covariance
Correlations among input estimates
affect the combined standard
uncertainty of the output estimate.
The estimated covariance of two input
estimates, xi and xj, is denoted by
u(xi,xj).
10. Uncertainty Propagation
“Law of Propagation of Uncertainty,” or,
more simply, the “uncertainty
propagation formula”
Standard uncertainties and covariances
of input estimates are combined
mathematically to produce the
combined standard uncertainty of the
output quantity.
11. Expanded Uncertainty
Multiply the combined standard uncertainty,
uc(y), by a number k, called the coverage
factor to obtain the expanded uncertainty, U.
The probability (or one’s degree of belief) that
the interval y +- U will contain the value of the
measurand is called either the coverage
probability or the level of confidence.
12. Recommendations
Follow ISO-GUM in terminology and
methods.
Consider all sources of uncertainty and
evaluate and propagate all that are
considered to be potentially significant
in the final result.
Do not ignore subsampling uncertainty
just because it may be hard to evaluate.
13. Recommendations
- Continued
Report all results – even if zero or negative
Report either the combined standard
uncertainty or the expanded uncertainty.
Explain the uncertainty – in particular state
the coverage factor for an expanded
uncertainty.
Round the reported uncertainty to either 1 or
2 figures (suggest 2) and round the result to
match.
14. Detection and Quantification
There are several standards on the
subject of detection limits.
We try to follow the principles that are
common to all.
We follow IUPAC (more or less) for
quantification limits.
15. Detection
A detection decision is based on the
critical value (critical level, decision
level) of the response variable (e.g.,
instrument signal, either gross or net).
The minimum detectable concentration
(MDC) is the smallest (true) analyte
concentration that ensures a specified
high probability of detection.
16. “A Priori” vs. “A Posteriori”
Avoid the “a priori” vs. “a posteriori”
distinction.
We recognize:
Many labs report a sample-specific estimate of the
MDC
Many experts insist it should not be done
We take no firm position except to state that
the sample-specific MDC has few valid uses
and is often misused.
17. Misuse of the MDC
We state that no version of the MDC
should be used in deciding whether an
analyte is present in a laboratory
sample.
The MDC cannot be determined unless
the detection criterion has already been
specified.
18. Quantification Limits
We cite IUPAC’s guidance for defining
quantification limits.
The minimum quantifiable concentration
(MQC) is the analyte concentration that
gives a relative standard deviation of
1/k, for some specified number k
(usually 10).
19. The MQC
We hoped to unify the approaches to
uncertainty and to detection and
quantification limits.
ISO-GUM in effect treats all error components
as random variables.
Is this approach consistent with IUPAC’s
approach to quantification limits? We
proceeded as if the answer were yes.
20. The MQC - Continued
Our MQC is based on an overall
standard deviation that represents all
sources of measurement error – not just
“random errors.”
This standard deviation differs from the
combined standard uncertainty, a
random variable whose value changes
with each measurement.
21. Use of the MQC
The MQC is almost unknown among
radiochemists but should be a useful
performance characteristic.
The MDC is well-known and is
sometimes used for purposes that
would be better served by the MQC.
E.g., choosing a procedure to measure
Ra-226 in soil.
22. Other Topics
Effects of nonlinearity on uncertainty
propagation
Laboratory subsampling – based on
Pierre Gy’s sampling theory
Tests for normality
Example calculations
23. Other Topics - Continued
Detection decisions based on low-
background Poisson counting or few
degrees of freedom
Expressions for the critical net count in
the pure Poisson case
Well-known (so-called “Currie’s equation”)
Not so well-known (Nicholson, Stapleton)
24. Concerns & Questions
Overkill?
Is anything important missing?
E.g., a table of “typical” uncertainties
More real-world examples of good
uncertainty evaluation
How can the examples be improved?
Contradictory standards on detection
limits