Analyst, November 1997, Vol.
122 (1283–1288) 1283
Analytical Bias: the Neglected Component of
Measurement Uncertainty†
Jean S. Kane
Robert J. Kane Associates Inc., Brightwood, VA 22715, USA
All analysts are accustomed to reporting measurement
results accompanied by either the standard deviation of
individual results from the mean or the standard error of
the mean. These statements indicate repeatability of
measurement under unchanged conditions, or
reproducibility, where the time period over which
measurements are taken is the source of changed
conditions. Repeatability generally produces a smaller
deviation between replicates than does reproducibility.
However, most of the variability of measurements made
by different laboratories, or using different methods in a
single laboratory, is not accounted for by either
repeatability or reproducibility attributable to time
period. This is evident in all interlaboratory data sets;
these frequently contain between-laboratory and
between-method discrepancies that are very large in
comparison with the uncertainty of measurements
typically reported by individual laboratories. Recently
issued ISO guidelines and related documents address
these discrepancies as a legitimate component of
measurement uncertainty, and recommend expanding the
concept to include the deviation of a measurement from
the true value of the measurand, so long as this deviation
is small relative to fitness for purpose requirements.
When the bias of measurement renders the result
unsuitable for purpose, however, that bias is a significant
effect and must be removed by use of a correction factor.
To correct for significant bias, or to include smaller bias
as an uncertainty component, laboratories must evaluate
and quantify the bias in their measurements to the fullest
extent possible. This paper presents uncertainty
statements developed in accordance with the ISO
guidelines for several reference sample measurements.
Some are very complex, drawn from published work of
metrology laboratories. Others, drawn from the author’s
data for the Japanese Sedimentary rock reference
materials, are more suitable for routine laboratory use.
Keywords: Analytical bias; measurement uncertainty;
reference materials
The Guide to the Expression of Uncertainty in Measurement1
established general rules for evaluating and stating measure-
ment uncertainty across a broad spectrum of measurements.
This guide has particular relevance to laboratories whose data
will be used to develop reference values for geological
reference materials. These laboratories will need to establish
uncertainties for all contributed data in keeping with the
principles of this important new ISO guide.
A detailed study of measurement uncertainty requires
extensive effort, and is typically undertaken only in national
metrology laboratories. The ISO guide1 provides guidance on
the process, and the Eurachem Working Group on Uncertainty
in Chemical Measurement2 has published a number of examples
† Presented at Geoanalysis 97: 3rd International Conference on the Analysis of
Geological and Environmental Materials, Vail, CO, USA, June 1–5, 1997.
to illustrate what is required. Two examples from the literature
describing NIST certifications3,4 provide similar illustration.
Most routine laboratories will not be able to develop uncertainty
statements for their data in this manner; they will instead be
drawn largely on data from their routine quality control and
quality assurance programs and the more in-depth knowledge of
measurement uncertainty developed by metrology laborato-
ries.
Uncertainty in the broadest terms means the doubt about the
validity of any measurement results. While doubt about validity
arises from error, error and uncertainty are not the same. The
doubt may exist about the exactness of the results, i.e., as a
result of repeatability or reproducibility attributable to time
period factors. Alternatively, it may exist about the validity of a
result attributable to inappropriate sample collection or han-
dling, to heterogeneity leading to subsampling error or to the
presence of analytical bias.
In order to estimate uncertainty reliably for a measurement, it
is essential that the measurement process be well defined and
that the process be under statistical control when the measure-
ment is made. In defining the measurement process, the
laboratory will determine the degree of accuracy necessary for
‘fitness of purpose’ and will identify those factors in the process
which have greatest influence on the measurement result.
Influences may be either random, arising from unpredictable
variations, or systematic, arising from an influence quantity
whose effect is either constant or varies in a predictable way. If
an influence quantity produces a systematic effect that is
significant from a fitness-for-purpose perspective, correction is
essential, and the uncertainty associated with the correction
must be included in the overall uncertainty of measure-
ment.1,2,5
If, instead, the systematic effect is small relative to acceptable
uncertainty for the purpose, the bias itself is the source of
uncertainty, and its magnitude may be included as an un-
certainty component, as illustrated in example 4 of Appendix A
in the Eurachem document.2 A NIST document6 illustrating
tests for bias in comparison with reference values is instructive
in this regard. According to this reference,6 not all deviations of
a measurement result represent bias: only those which are
significant in comparison with the uncertainty of the reference
value. However, NIST includes any deviations of reference
material (RM) results from reference values in its uncertainty
estimates,3 when an existing NIST Standard Reference Material
(SRM) is analyzed concurrently with a new SRM undergoing
certification analysis. This uncertainty is termed ‘recovery of
standard’ ( = measured/certified) in ref. 3 and in the text and
Tables 2 and 5 in this paper.
Many potential soruces of uncertainty are listed in the
Eurachem document.2 The list is intended to be illustrative,
rather than definitive. These have been broadly categorized as
sampling, random variation and analytical bias, and are shown
in Table 1. It should be noted that spurious error, also included
in Table 1, is an extreme case of random error; uncertainty
estimates do not allow for the possibility of spurious error.
Sampling as a source of uncertainty is the subject of the other
papers in this issue.7–9 The focus of all that follows will be on
analytical bias alone among these potential soruces of un-
1284 Analyst, November 1997, Vol. 122

certainty. The goal is to illustrate the use of control data and the
like in estimating overall uncertainty. This process is included
as appropriate in the ISO guide and related documents,1,2,5 but
illustrative examples are very limited in comparison with those
showing the more exhaustive metrological approaches.
Analytical Bias
All analysts are very familiar with instances of disagreement
between results that are nominally for the same measurand.
These discrepancies have been the topic of much discussion
among geoanalysts since Fairbairn et al.’s study10 first made it
clear how common they are. Data ranges for all major rock-
forming constituents other than SiO2 were completely un-
Table 1 Potential sources of uncertainty in analytical measurements
acceptable, and outlier rejection before deriving recommended
values was greater than 50% in some cases. This is seen in Figs.
1 and 2, presenting the G-1 and W-1 data from Fairbairn et al.’s
report.10 Youden11 attributed most discrepancies of this type to
systematic error or analytical bias inherent in methods being
used.
It is these systematic errors leading to analytical bias, and the
doubt about the validity of the result due to them, that will be
addressed in all that follows. Some of these errors and resulting
uncertainties can be evaluated from the statistical distribution of
a series of measurement results, and can therefore be charac-
terized by a standard deviation, in just the same way that
random variations are. Other components are evaluated from
assumed probability distributions based on experience or other
information. These two categories for evaluating uncertainties
are referred to as Type A and Type B in the ISO guide.1

Sampling uncertainty—
Metrology Laboratory Approach
Sample was inappropriate to the measured purpose The isotope dilution mass spectrometric method is a definitive
Sample was suitable, but has degraded in handling and/or storage method, commonly used by metrology laboratories for certifica-
between collection and measurement (see ref. 7) tion analyses. When properly applied and fully under control,
Field sampling uncertainty was too large to be ‘fit for purpose’ (see ref.
8)
results will be obtained that are essentially free from analytical
Measurand is inhomogeneously distributed in the sample bias. The uncertainty in result for the homogeneous sample will
be determined entirely by repeatability of measurement, with
Note: Sampling error due to heterogeneity of sample is more properly
described as sub-sampling error, leading to between-unit sub-sampling
RSDs of about 0.2%. This small uncertainty of measurement is
variance. The uncertainty due to sub-sampling error may be quantified
using analysis of variance where the unit of issue is the classification
variable (Type A evaluation) or by modeling mineralogy and using the
theoretical Poisson distribution for the count of mineralized grains in the
subsample (Type B evaluation) (see ref. 9)
Random error—
Measurement repeatability for sample measured repetitively
Measurement repeatability for replicate subsamples of a single sample
Note: Random error can never be assigned a definitive value. The
standard deviation of replicate measurements from the arithmetic mean is
a measure of uncertainty due to some random effects, but is not the
random error2
Note: The standard deviation of the measured values from the arithmetic
mean can be appropriately estimated from 15 or fewer measurements,
unless very high precision is required, in which case a larger number of
replicates is needed. This estimation of uncertainty is Type A
Spurious error—
Transcription error in recording analytical mass and/or signal
Misreading of analog signal output
Undetected instrument malfunction during measurement Fig. 1 Interlaboratory ranges relative to the recommended values for
Sample loss due to poor technique, spillage, etc., that did not result in major oxides in G-1. Data from Fairbairn et al.10 Open bar, full range; solid
discarding sample bar, 1s range for recommended value.
Note: Results affected by spurious error are invalid. They should be
discarded and there should be no attempt to incorporate the spurious error
into statistical analysis of error sources1,2,5
Analytical bias—
Incomplete knowledge of factors having critical impact on measurement
accuracy
No correction or incomplete correction for matrix effects (physical
interferences)
No correction or incomplete correction for spectral interferences
Incomplete separation and/or preconcentration of measurand from
matrix
Calibration error due to uncertainty in reference values used for
calibration
Approximations and assumptions incorporated in measurement proce-
dure
Uncertainty in masses, volumetric apparatus, etc.
Uncertainties in values of parameters used in data reduction (e.g., atomic
mass and isotopic abundances of the elements for mass spectrome-
try)
Instrument resolution or discrimination threshold
Inadequate control of environmental conditions needed for measurement Fig. 2 Interlaboratory ranges relative to the recommended values for
accuracy major oxides in W-1. Data from Fairbairn et al.10 Open bar, full range; solid
bar, 1s range for recommended value.
 Analyst, November 1997, Vol. 122 1285

obtained at great effort and cost, however, and should not be the
analytical goal except when necessary for making valid
decisions based on the measurement results.
Similarly, gravimetric analysis when carried out under full
control can produce results that have random uncertainties
similar to the overall uncertainty of the isotope dilution
measurement. Freedom from systematic error in the gravimetric
analysis, however, is less assured than with isotope dilution
measurement. Two examples taken from NIST certifications are
illustrative.3,4 The NaCl SRM 919a is the starting material for
the Na spectrometric solution SRM 3152 and the Cl anion
chromatography solution SRM 3182. The solution SRM 3181 is
the sulfate anion chromatography solution standard.
For the certification of both, NIST employed gravimetry.
Error evaluation examined incomplete precipitation, con-
tamination of the precipitate and losses in sample handling at
every step in the process, among other things.
The method used for SRM 919a certification3 is a modifica-
tion and improvement of the definitive method first developed
as a National Committee for Clinical Laboratory Standards
method. The sample was first decomposed and then separated
on an AG 50W-X8 ion-exchange column. The sodium fraction
was collected and converted into the sulfate using H2SO4;
evaporation produced the salt, which was then ignited and
weighed to complete the gravimetric determination. The ion-
exchange separation of sodium from any Li, K, Mg and Ca also
in the sample avoided an error due to the co-existence of other
sulfates in the ignited residue.
To summarize briefly the error sources examined, Moody
and Vetter3 considered the sample handling and transfer steps
(sample vial to digestion beaker, beaker to column, beaker after
evaporation to ignition crucible) to be very important. They also
considered ignition losses due to evaporation and/or splattering.
Ability to achieve the certified result for concurrent analysis of
previously certified materials (‘recovery of standard’) was
evaluated. Additionally, incompleteness of the chromato-
graphic separation or failure to collect 100% of the sodium
fraction introduced uncertainty. Finally, the chemical blanks
and any weighing or volumetric errors were evaluated.
Only measurement replication (random error) and ‘recovery
of standard’ were evaluated by statistical means (Type A). All
others were evaluated in other ways (Type B). The individual
uncertainties were first combined within each type following
the law of propagation of errors, and then the uncertainties
evaluated by Type A and Type B means were combined and
expanded to provide the final uncertainty for the certification. A
coverage factor of 2 for the expansion gives a level of
confidence of approximately 95%, and a factor of 3 gives
approximately 99% confidence. The recommended coverage
factor1 for general use is 2. The detailed evaluation taken
directly from Moody and Vetter3 is given in Table 2.
In the case of the gravimetric analysis of the sulfate solution
SRM 3181,4 classical gravimetry was coupled with in-
strumental techniques; the latter allowed corrections to be made
for any Ba that remained unprecipitated, for any BaCl2 or
K2SO4 occluded in the BaSO4 precipitate and for any sulfate
volatilized during ignition. Uncertainties in the contribution of
the blank and from measurement replication were determined
using a Type A evaluation. Mechanical losses and all in-
strumentally determined corrections were estimated using Type

Table 2 Uncertainty of measurement of sodium for the certification of SRM 919a3

Source of uncertainty Relative standard uncertainty (%)

Type A evaluation used—
Measurement replication 0.007
‘Recovery of standard’ 0.019
Combined Type A ABBBBBB BB2 = 0.020
(0.007)2 + (0.019)
Type B evaluation used—
Sampling handling 0.036
Sample volume 0.002
Separation—loading/retention 0.005
Separation—loss/contamination 0.001
Weighing and ignition 0.015
Combined Type B ABBBBBBBBBBBBBBBBBBBB BB
(0.036)2 + (0.002)2 + (0.005)2 + (0.001)2 + (0.015) 2 = 0.040

Combined uncertainty ABBBBBB BB2 = 0.0443

(0.020)2 + (0.040)
Coverage factor 2
Expanded uncertainty (relative 0.0089

Table 3 Measurement uncertainty for the certification of sulfate SRM 31814

Source of uncertainty Relative standard uncertainty (%)

Type A evaluation used—
Measurement replication 0.018
Blank 0.009
Combined Type A ABBBBBB BB2 = 0.020
(0.018)2 + (0.009)
Type B evaluation used—
Mechanical loss 0.006
BaCl2 correction 0.060
K2SO4 correction 0.003
Filtrate sulfate correction 0.033
Volatile sulfate correction 0.035
Combined Type B ABBBBBBBBBBBBBBBBBBBBB
(0.006)2 + (0.060)2 + (0.003)2 + (0.033)2 + (0.035)2 = 0.077
Combined uncertainty ABBBBBB BB2 = 0.080
(0.020)2 + (0.077)
Coverage factor 2
Expanded uncertainty (relative) 0.160
1286 Analyst, November 1997, Vol. 122
B methods. The results taken directly from Vetter et al.4 are
given in Table 3.
It can be seen from these examples that carefully applied
gravimetric procedures can produce total uncertainties, includ-
ing analytical bias, as small as those for the isotope dilution
mass spectrometric method. The potential for biased results in
gravimetric methods is also seen clearly in these examples. Note
that separation step errors in the case of the Na certification and
of precipitation errors in the case of Ba account for a
considerable portion of the total uncertainty reported. Laborato-
ries applying either of these methods under less strignent
control will produce results with greater associated un-
certainties.
The Routine Laboratory Approach
The average laboratory cannot devote the time and effort to
method development and to evaluating sources of uncertainty
that a national metrology laboratory such as NIST does. Yet the
average laboratory needs to be able to state uncertainties for its
measurements in a way that allows users of the data to judge
their validity in addition to their repeatability. This is the case
regardless of the use to which the data are put; it is particularly
the case when the data will contribute to a reference sample
characterization. The magnitude of the uncertainty is dictated
by requirements for the end use of the data, by ‘fitness for
purpose’ considerations. For certification purposes, the un-
certainty should be 3–10 times smaller than that of routine
analyses.12
It has been stated previously1,2,5 that many aspects of a
laboratory’s quality control/quality assurance program will
provide data that can be used in the estimation of uncertainty.
Information from calibration certificates for balances and
laboratory glassware provides Type B estimations of un-
certainties due to weighing and volume errors. Information on
reference sample certificates of analysis provides ‘recovery of
standard’ uncertainty estimates and/or calibration error esti-
mates, depending on how the laboratory used the reference
sample. Similarly, information from control charts, collabor-
ative studies and proficiency tests can be used to estimate
laboratory uncertainty. Evaluations could be either Type A or
Type B, depending on whether sufficient data existed to
produce a statistical standard deviation or whether judgment
instead was used. These approaches to uncertainty evaluation
will often provide a single combined uncertainty of several
related error sources that the metrology laboratory might
evaluate individually.
For example, the laboratory measuring Ba or S gravimet-
rically as the BaSO4 precipitate might analyze a number of
reference samples, and determine their precipitation losses and
contaminants (analytical bias) collectively with a Type A
evaluation from the average ‘recovery of standard’ for the
several RMs, rather than following the NIST process. This is a
valid approach, and will result in an uncertainty statement that
suits the accuracy requirements of the laboratory measure-
ments.
The approach will be illustrated with an evaluation of
uncertainty for major oxide data that the author contributed for
six of the Japanese Geological Survey sedimentary RMs, three
sediments, two slates and one chert, having analyzed the
samples in replicate (three splits, digested in duplicate with each
digestion solution analyzed on three different days) by ICP-
AES.13 The decomposition was by lithium metaborate–tetra-
borate fusion, and a two-point calibration using similarly
prepared USGS rock RMs was employed. Two additional
USGS rock RMs were included in each experimental run for
control purposes.
The control data14 given in Table 4 provide reproducibility
data based on replicate measurements (n = 26 for AGV-1, n =
22 for BHVO-1, n = 24 for BIR-1, n = 24 for G-2) that were
made at widely spaced times over several years between 1984
and 1990. The changed conditions resulting from the time
intervals included such things as replacement torches and
mirrors in the spectrometer, different decomposition solutions
for both the control sample RMs and the calibration RMs and
variations in instrument set-up, including manual wavelength
peaking and torch height adjustments. However, the same
analytical procedure and instrument were used in all measure-
ments and the same analyst performed the measurements, over
the entire time period involved. The detailed within-run
repeatability data are no longer available, but within-run RSDs
were typically only a fraction (less than one fifth) of the
reproducibility data shown in Table 4.
Table 4 also provides bias estimations or ‘recovery of
standard’ data, both as the difference between the mean of the
replicate control sample measurements and the reference value
and as the maximum deviation encountered during the time
period involved. There is no case where the bias estimated from
the mean deviation is statistically significant6 in comparison
with either the certified value uncertainties (see below) or the
measurement reproducibility. Additionally, it is acceptably
small from a fitness for purpose perspective to remain
uncorrected. Acceptable errors for the method are a function of
concentration13 and for SiO2 are 1% relative to the reference
value.
It is presumed that sub-sampling error is not a factor, based
on prior homogeneity studies conducted in establishing refer-
ence values for the USGS rock reference materials. Data
affected by spurious error are absent from the control chart data
summarized in Table 4. Weighing and volume errors are
estimated from balance and glassware calibrations, in the same
manner as reported by NIST3,4 and summarized in Tables 2 and
3.
In using the control chart data to establish uncertainties in
results for the Japanese Geological Survey sedimentary rocks,
the principal factors to consider are how well known the USGS
rock references values are and how the difference between the
matrix of the sedimentary versus the igneous rocks might affect
the ‘recovery of standard’ value for the analytical samples in
comparison with the controls.
The uncertainties in reference values for the USGS rock
standards are comparable to those for NIST SRM certified
values established using two or more reference independent
methods of analysis, but slightly larger than those based on
definitive method certification.15–17 For example, uncertainties
for reference values of SiO2 as stated on the certificates of
analysis18,19 are 0.20% relative for the NIST obsidian and basalt
SRMs 278 and 688. For the USGS rock RMs, the uncertainties
are estimated as twice the stanard error of compilation means,
and range from 0.1% relative for G-1 and W-1 to 0.30% for
DTS-1 and PCC-1.15 The uncertainty in the reference values
Table 4 ICP-AES control data14
Deviation (relative) from reference
value (%)15
Reproducibility
over multi-year of control data
Oxide RM period (as RSD) mean of extreme value
SiO2 AGV-1 1.58 20.12 +1.7
BHVO-1 1.68 +0.60 21.64
Al2O3 BIR-1 1.9 +0.20 +2.6
G-2 2.0 +0.5 +3.2
Fe2O3 AGV-1 1.8 20.20 +3.3
BIR-1 1.7 +1.9 22.4
Na2O G-2 2.9 21.7 22.7
BHVO-1 4.0 +0.4 +2.2
 Analyst, November 1997, Vol. 122 1287

shown in Table 5 is the pooled uncertainty for the two RMs used
as controls for SiO2 analysis, a Type A evaluation.
While long experience with the method suggests that there is
no correlation between magnitude of bias and matrix type,13,14
this has not specifically been evaluated. To be certain that the
uncertainty is not underestimated, a rectangular distribution is
assumed, and the uncertainty is stated without specifying the
confidence level. In this case, standard uncertainty for the bias
estimate is the worst case control value deviation from the
reference value divided by AB 3.2,3
It is recognized that the calibration range for the measure-
ments does not encompass extreme values, e.g., the 98.8% SiO2
of the Japanese sedimentary chert sample JCh-1. A bias larger
than observed for the two control samples is likely to result from
extrapolation of the calibration curve beyond its linear range.13
This is another reason for assuming a rectangular distribution
for ‘recovery of standard’ values, rather than a normal
distribution, for which uncertainty would be based on the
average deviation of all control results from the reference
value.
Applying the above to the control data, total uncertainties
were calculated as shown in Table 5. The expanded uncertainty
shown can be expected to estimate reliably not only the total
uncertainties in results for the two control samples measured
using the ICP-AES method, but more generally for most rock
samples measured using the same analytical method. They
included not only the random component of uncertainty that is
customarily reported in all of the analytical literature, but also
the analytical bias component that is often neglected.
A laboratory’s participation in collaborative measurement
programs can similarly provide a means of uncertainty
estimation. In this case, rather than using control samples run
Table 5 Estimation of uncertainty for SiO2 in control samples AGV-1 and
BHVO-114
Type Source of uncertainty Relative standard uncertainty (%)
A Repeatability* 0.34
Uncertainty in reference 0.17
value15
B ‘Recovery of standard’ 0.17/AB 3 = 0.98
B Weighing errors 0.1
B Volume errors 0.02
Combined Type A ABBB
(0.34)BBBBB
2 + (0.017)2 = 0.0038
simultaneously with the Japanese sedimentary RMs to provide
the estimates of ‘recovery of standard’, the reference values of
the newly characterized reference samples20 are used for that
purpose (see Table 6).
These data reflect a larger relative deviation of mean
measured results from reference values than the USGS control
sample data had. This may result because uncertainties in the
reference values themselves are larger than for the USGS rocks,
being based on a 8–10-fold smaller database. Alternatively, it
could result from extrapolation of the calibration curve for some
individual analyses, as noted above. Were we to examine data
from the proficiency test program, even larger deviations of
individual laboratory data from the central value would be
noted, as would greater uncertainty in the central value.
However, all of these approaches to estimating measurements
uncertainties follow the principles of the process outlined in the
ISO Guide,1 and are recommended to all geoanalytical
laboratories, particularly those participating in collaborative
programs to establish values for reference materials.
Conclusion
Most laboratories appropriately evaluate and report the repeata-
bility of their laboratory measurements. However, repeatability
is only one component of the total measurement uncertainty, or
doubt about the validity of measurement. Approaches to
evaluating measurement uncertainty more completely have
been presented.
These approaches to evaluating measurement uncertainty,
although less rigorous than the approach taken by metrology
laboratories, will help geoanalytical laboratories to identify any
of their standard operating procedures (SOPs) that do not
produce ‘fit for purpose’ data. This will lead to improved SOPs,
and therefore to overall improvement of the reliability of
laboratory data. Improved data in turn will permit laboratories
participating in proficiency tests to see continuous improve-
ments in their respective z-scores. Alternatively, it will permit
laboratories producing ‘fit-for purpose’ data to refine their
uncertainty estimates to include any acceptably small bias that
contributes to uncertainty. Were this to be done, comparability
or harmony of data from collaborative programs would improve
and the definition of reference values and their uncertainties for
geochemical RMs would improve accordingly.

Combined Type B ABBBBBBBBBB

(098)2 + (0.1)2 + (0.02)2 = 0.985
References
Combined uncertainty ABBBBBB BB
(0.38)2 + (0.985) 2 = 1.06

Coverage factor 2 1 ISO, Guide to the Expression of Uncertainty in Measurement, ISO,

Expanded uncertainty (rela- 2.12 Geneva, 1993.
tive) 2 Eurachem Working Group, Quantifying Uncertainty in Analytical
Measurements, Laboratory of the Government Chemist, Teddington,
* Maximum repeatability value = reproducibility value (Table 4) 3 UK, 1995.
0.2. 3 Moody, J. R., and Vetter, T. W., J. Res. Natl. Inst. Stand. Technol.,
1996, 101, 155.
4 Vetter, T. W., Pratt, K. W., Turk, G. C., Beck, C. M., and Butler, T.
A., Analyst, 1995, 120, 2025.
Table 6 ‘Recovery of standard’ uncertainty estimate for SiO2 in rocks based 5 Taylor, B. N., and Kuyatt, C. E., Guidelines for Evaluating and
on participation in collaborative certification of Japanese sedimentary rock Expressing the Uncertainty of NIST Measurement Results, NIST
standards Technical Note 1297, US Government Printing Office, Washington,
DC, 1993.
Relative deviation 6 Becker, D., Christiansen, R., Currie, L., Diamondstone, B., Eber-
Contributed data Reference value of contributed data hardt, K., Gills, T., Hertz, H., Klouda, G., Moody, J., Parris, R.,
Standard (mass (mass fraction as %) from reference Schaffer, R., Steel, E., Taylor, J., Watters, R., and Zeisler, R., NIST
Standard fraction as %) (mean ± s) value (%) Spec. Publ., 1992, No. 829.
JSd-1 66.17 66.55 ± 0.354 20.57 7 Horowitz, A. J., Analyst, 1997, 122, 1193.
JSd-2 60.82 60.78± 0.411 +0.066 8 Ramsey, M. J., Analyst, 1997, 122, 1255.
JSd-3 76.03 76.00± 0.606 +0.039 9 Kane, J. S., Analyst, 1997, 122, 1289.
JSI-1 59.20 59.47± 0.263 20.46 10 Fairbairn, H. W., Schlecht, W. G., Stevens, R. E., Dennen, W. H.,
JSI-2 59.8 59.45± 0.223 +0.58 Ahrens, L. H., and Cheyes, F., US Geol. Surv. Bull., 1951, No.
JCh-1 98.15 97.81± 0.484 +0.35 980.
11 Youden, W. J., Anal. Chem., 1960, 32, 23A.
1288 Analyst, November 1997, Vol. 122

12 Uriano, G. A., and Gravatt, C. A., CRC Crit. Rev. Anal. Chem., 1977, 18 Uriano, G. A., NBS Certificate for Analysis for SRM 278, National
6, 361. Bureau of Standards, Washington, DC, 1981.
13 Kane, J. S., and Dorrzapf, A. F., Jr., in GeoExpo/86: Exploration of 19 Uriano, G. A., NBS Certificate for Analysis for SRM 278, National
the North American Cordillera, ed. Elliott, I. L., and Smee, B. M., Bureau of Standards, Washington, DC, 1981.
Association of Exploration Geochemists, Vancouver, BC, Canada, 20 Imai, N., Sakuramachi, H., Terashima, S., Itoh, S., and Ando, A.,
1987, pp. 184–188. Geostand. Newsl., 1996, 20, 165.
14 Kane, J. S., unpublished control sample data.
15 Gladney, E. S., Burns, C. E., and Roelandts, I., Geostand. Newsl.,
1983, 7, 1988, 12, 253. Paper 7/04789D
16 Kane, J. S., J. Geochem. Explor., 1992, 44, 37. Received July 7, 1997
17 Kane, J. S., Spectrochim. Acta, Part B, 1991, 46, 1623. Accepted September 22, 1997