Certification of a laboratory by the National Association of Assessment of the comparative results for these pairs of
Testing Authorities (NATA) provides some surety that the pulps allows the average differences in grade to be
written procedures adopted by the laboratory are followed, quantified for various grade ranges. Inclusion of standard
and that an audit trail is established for all work carried and blank samples enables the accuracy to be monitored at
out. However as with all quality assurance schemes, such the same time.
certification provides no guarantee that the documented
procedures are appropriate, just that these procedures are Common Problems With Sampling And Assaying
followed. Studies
Many of the accuracy and repeatability studies seen to date
While quality control systems are now uniformly adopted, suffer from design problems, either because of a lack of
these are designed to meet the production needs of the clear understanding of the principles discussed above, or
laboratory, rather than the specific needs of the client. because the experimental work is trying to achieve too
Accordingly they can suffer from two specific drawbacks. much. Common failings are:
The laboratory carries out repeat assays on the same pulp • reliance solely on the quality control procedures of the
using the same method, but it knows the final result. Where laboratory, with no independent validation
the original and the repeat assays do not agree sufficiently, (consequently standards used to check instrument
the results may be abandoned and the work repeated. This calibration are claimed to demonstrate accuracy, and
is appropriate from the point of view of the laboratory, but repeats used to monitor and eliminate between batch
may mask the true variability of the results being produced. variability are used to demonstrate precision)
Secondly, from a theoretical point of view, all repeat • the systematic re-submission of a proportion of samples
assays are open to question if they are not blind and (e.g. 10%) so that most of the repeats are of irrelevant
randomised. All the pulps should be submitted as blind very low grades
randomised samples so that they cannot be matched by the • re-submission of samples using the same sample
laboratory and to remove the effect of any bias due to numbers, or in the same order, so that the reason for
procedural or instrument errors. systematic differences (bias) is not clearly defined
• confusion between the comparison of duplicate field
Definition of the assaying accuracy is frequently addressed samples and replicate assays on pulps, so that it is
by assaying standard samples. Such standards are obtained unclear which errors are due to assaying, which are due
from commercial sources and themselves suffer from a to sample preparation and which are due to the initial
number of problems: sample splitting
• some are very expensive, and differences in costs may • submission of all the duplicates and repeats at the end
reflect differences in quality, of the resource drilling when it is too late to change the
• the expected accuracy and repeatability characteristics protocols if significant precision or bias errors are
may differ for different assaying techniques (these identified
• submission of duplicates and repeats without
monitoring (and acting on) the results. The methodology adopted by Pitard (1993) for
determining precision using the Fundamental Sampling
Error as the relative variance of the sample data set is
ANALYSIS OF RESULTS extremely sensitive to single outlier values. Similar
Tests For Accuracy difficulties occur with other univariate statistics (e.g. the
The most poorly utilised tool for interpreting sampling data standard deviation and the geometric mean) and with
is the scatterplot. Plots with different scales on the abscissa bivariate statistics (e.g. the covariance, the correlation
and ordinate, and plots showing regression lines between coefficient and regression). Although it may be argued that
the paired data, seem designed to confuse. Rather, square it is the outlier values that are of interest, problems with the
format plots with equal scales should be produced, with a statistical analysis can make the results of the sampling
45o line defining the expected perfect correlation. experiment apparently unrepeatable.
Systematic differences between pairs of assays (bias) are
then clearly seen as deviations of the trend of paired data To enable comparison between various assay methods in
from the 45o line. gold deposits, a method of using paired samples is
Accuracy is frequently defined as the difference between required, with the results standardised to enable statistical
the means of the paired data. This is a valid measure but comparison. Techniques for quantifying assaying precision
the sensitivity of this statistic to outliers must be tested by have been discussed by Thompson and Howarth (1973)
trimming the data and by examining subsets of the data4. A and Bumstead (1984). These have been adopted by the
non-linear bias may be identified between two assay data author to define a robust estimate of error termed the Half
sets (called here the “standard” and the “check” assay Absolute Relative Difference (HARD) for paired data.
methods). For example the “standard” method may This is produced by dividing half the absolute difference
overestimate gold grades below 4 g/t Au and underestimate between the two values by the mean of the two values. A
them above that value. At a mining cut-off grade of similar measure termed the Half Relative Difference
1 g/t Au this creates a problem. Use of the “standard” (HRD) may be used where the sign of the differences is
method results in overestimation of tonnages above the significant.
cut-off grade, but underestimation of their grade. Use of
the “check” method would produce the opposite result. The Half Absolute Relative Difference (HARD) is
calculated as follows (note that the factors of two cancel
Tests For Precision out):
Precision can be determined by carrying out a large
number of assays on a single homogeneous sample, to 1 ABS( Assay1 − Assay2)
define the distribution of errors about the mean result. The HARD = * *100
residuals can be found by subtracting the mean from each 2 Assay1 + Assay2
assay result and examined to ensure that they are 2
distributed normally. The standard deviation of the
residuals defines the precision. Thus, for example, for a The HARD and HRD values may be used to produce
confidence level of 95 percent, the precision would be the robust measures of the relative bias and the relative
standard deviation times 1.96. This can then be used to variance of the Fundamental Sampling Error for the
determine the precision as the absolute error of the assays purpose of defining sampling protocols5.
(in the case of a gold deposit as ± g/t Au).
Three examples follow:
In dealing with real gold bearing samples it is necessary to • If the original assay was 0.33 g/t Au and the repeat
recognise that each sample is inherently different. It would assay was 0.99 g/t Au, the average of the two grades
not be appropriate to compare the means of 40 different would be 0.66 g/t Au, and the average of the two
samples, it would be too expensive to carry out a large residuals would be 0.33 g/t Au. Thus the relative error
number of replicates (repeat assays) on many different as measured by the HARD would be 50%.
samples, and generally there is insufficient homogeneous • The relative error between an original assay of 1 g/t Au
material available to carry out a large number of replicates. and a repeat of 2 g/t Au is 33.3% (the same as if the
Repea
Sampling nomograms, in the format adopted by Pitard 10
0.1
1 mm 5 mm
Variance
20% error
0.01
1 0% error
5% error
0.001
75 micron
0.0001
0.00001
10 1 00 M ass (g) 1 000 1 0000
Figure 3 Sampling nomogram for Case 1, a gold deposit with moderate visible gold
0.1
Variance
20% error
0.01 5 mm
1 0% error
1 mm
5% error
0.001
0.0001 75 micron
0.00001
10 1 00 1 000 1 0000
M ass (g)
Figure 4 Sampling nomogram for Case 2, a gold deposit with very little visible gold
Field Techniques Sensitivity Studies
The drilling or other initial sampling method, and all stages A resource estimate should be carried out on a set of data
of subsampling and assaying, impact on both the precision for which there is information on the sampling and
and accuracy of the data. Correctly designed experiments assaying quality. If a bias or a precision problem has been
in conjunction with the analytical methods presented here identified it would thus be reasonable, as part of a bankable
can indicate where expenditure on improved procedures feasibility study, to examine the sensitivity of the study to
can significantly reduce the errors. However in many cases, the data quality. It may well be that differences in the Net
there is no need for an experiment. Poor sampling Present Value of a project would be identified, depending
techniques, such as grabbing specimens rather than on the change of average grade (in the case of bias) or the
splitting samples, can be avoided at very little cost. change in recovered tonnes and grade (in the case of
misclassification).
Statistical analysis of the data can quickly indicate whether
sampling problems are a characteristic of the particular The effect on a project of the sampling and assaying error
deposit or whether the sampling has been done poorly. should be kept in perspective. Cases have been seen where
Significant advances in this area can be expected as the the effect on the average grade was less than the error
techniques presented here are further refined to handle associated with the estimation method; in other cases the
other problems, for example the quantitative analysis of viability of the project was cast into doubt by the errors
downhole smearing enabling comparisons between drilling associated with the sampling or differences between the
methods and across deposits. sampling methods.
Laboratory Techniques
Any assay method has an associated error. Measures of CONCLUSIONS
precision quantify the expectation that the same result can Monitoring and evaluating the sampling and assaying
be repeated continuously. There are a number of factors precision and bias is necessary for bankable feasibility
that cumulatively affect the precision of any assay result, studies. The statistical methods presented provide a means
including: by which the quality of the data can be quantified.
• instrument calibration and drift Meaningful comparisons are now possible between
• integrity of the chemicals used to dissolve the sampling methods, assaying methods, laboratories and
component of interest deposits. Further refinements and other quantitative tests
• the impact of volumetric determinations using non- can be expected where these provide a mechanism for
calibrated glassware developers of mining projects to reduce their risk.
• matrix effects due to other elements in solution
• the concentration of the component of interest Some words of caution are advised for those who see this
• the mass of material being assayed approach as a panacea: sampling consists of extracting a
• the homogeneity of the material being assayed representative parcel of material from the deposit and then
determining the grade of that parcel through a process of
Establishing An Audit Trail subsampling and assaying. In reality there may be
It is no longer sufficient to expect that a few duplicate and additional “errors” involved in this process which may not
repeat samples can be produced as evidence that the be identified with a sampling study. These errors include
sampling quality has been monitored. ignoring the spatial relationships inherent in geological
data, as well as the non-quantifiable procedural errors that
The assay data set may include original samples, repeats, may be built into a quality assurance scheme. As
duplicates and checks at different laboratories or with previously stated, these errors are best minimised by
different assay techniques. The data to be used for tonnage maintaining good quality control practices (through
and grade estimates should be maintained in a relational vigilance and audits) to reduce their potential impact on
database. The repeats and duplicates should be all stored in profitability.
a manner that allows easy separate extraction and cross
referencing. Averaging should be closely examined during
the data audit stage. It may not be appropriate to include REFERENCES
the data used for monitoring the sample quality in averages Barnett, V. and Lewis, T., 1994,
of assay results. Outliers in statistical data. Third edition, John Wiley &
Sons, 573 pp.
Bumstead, E., 1984, Keogh, D. C. and Hunt, S. J., 1990,
Some comments on the precision and accuracy of gold Statistical determination of a safe sampling protocol: case
analysis in exploration.. AusIMM proceedings 289, pp. 71- studies at Granny Smith and Big Bell, Western Australia.
78, March. In Strategies For Grade Control, Australian Institute of
Geoscientists Bulletin 10, pp 31-36.
Cooper, W. J., 1988,
Sample preparation - gold. Theoretical and practical Pitard, F., 1993,
aspects. In Sample Preparation and Analyses for Gold and Pierre Gy's Sampling Theory and Sampling Practice. CRC
Platinum - Group Elements, Australian Institute of Press Inc. (second edition), 488 pp.
Geoscientists Bulletin 8, pp 31-48.
Radford, N. W., 1987,
François-Bongarçon, D., 1995, Assessment of error in sampling. In Meaningful Sampling
Modern Sampling Theory. Course Notes, Perth. in Gold Exploration, Australian Institute of Geoscientists
Bulletin 7, pp 123-143.
Gy, P. M., 1979,
Sampling of Particulate Materials - Theory and Practice. In Shaw, W. J., 1993,
Developments in Geomathematics, Vol 4, Elsevier, 431 pp. Mining Geology, Grade Control and Reconciliation.
Course Notes, Perth.
Isaacs, E. H. and Srivastava, R. M., 1989,
Applied Geostatistics, Oxford University Press, 561 pp. Taylor, M., 1993,
Grade control review at Boddington gold mine, SW region,
JORC, 1996, WA. In Proceedings of the International Mining Geology
Australasian Code for Reporting of Identified Mineral Conference, Kalgoorlie-Boulder, WA, 5 to 8 July.
Resources and Ore Reserves. Report of the Joint
Committee of the Australasian Institute of Mining and Thompson M., and Howarth R.J., 1973,
Metallurgy, Australian Institute of Geoscientists and The rapid estimation and control of precision by duplicate
Australian Mining Industry Council (JORC), issued July. determination, The Analyst, 98, pp 153-160.