Introduction
A laboratory inter-comparison presents an invaluable opportunity to individual laboratories for checking procedures and the outcome provides an objective means of demonstrating comparability of results for the benefit of the wider user community. The C-14 community has already subscribed to a number of such inter-comparisons, but quality assurance is not static, and it seems appropriate at this time to consider a further, large-scale study open to all laboratories.
Objectives
Such a study would have the following objectives:
The study will be jointly organised by Marian Scott, Doug Harkness and Gordon Cook with the help of Philip Naysmith. A scientific advisory committee of Steinar Gulliksen, Tim Jull, Alan Hogg and Ellen Druffel has also been formed.
The timetable
| Timepoint | Objective |
| end of 1998 | samples distributed to participants |
| early 2000 | results returned by laboratories
draft report circulated to all participants |
| 2000 | workshop at 17th C-14 conference |
The samples
Sample material
Number of analyses
Each participating radiometric laboratory will be expected to complete a maximum of 10 analyses (there will be several duplicate analyses). For AMS laboratories, it is hoped that a maximum of 15 analyses will be completed (again including a number of replicate analyses).
Our reasoning for this approach is based on our objective to quantify natural variability and to compare its quantification in the radiometric and AMS labs
Further details can be obtained by e-mail from marian@stats.gla.ac.uk
New information: April 2001
The experimental phase of FIRI has now been completed, 10 samples were distributed to over 100 laboratories world-wide.
The report below summarises the information on the samples and a preliminary analysis of the results.
Report on the FIRI workshop, Edinburgh March 2001.
At the conclusion of the experimental stage of the Fourth International Radiocarbon Inter-comparison (FIRI), a small workshop was convened for study participants, to allow them to receive a report on the preliminary analysis of the results from the inter-comparison. 49 participants from over 20 countries participated in the two-day workshop held at the Edinburgh Conference Centre.
The FIRI results
92 sets of results were returned from FIRI (representing
laboratories from 37 different countries and made up of 25 AMS, 18 GPC
and 49 LSC sets).
The presentations and discussion are summarised in this short report. Sessions in the two-day programme included an overview of the FIRI programme, focussing on the samples selected and the testing they had undergone. This was then followed by a descriptive summary of the results for each of the samples (FIRI A to FIRI J). Then, the results of an analysis that explored the effects of modern standard and background was reported, before finally discussing some special studies: duplicate samples (A and B, D and F and G and J) and the dendro-dated samples (D and F, H and I).
The workshop analysis focussed primarily on a preliminary
and exploratory evaluation of the results for each sample, firstly identifying
and removing any gross outliers and secondly providing simple numerical
(average and median age or pmC) and graphical summaries.
Extreme values or outliers were observed in all samples,
often from the same laboratory. A tendency for the majority of the outliers
to be from LSC laboratories was also apparent. An ongoing investigation
into any distinguishing features of outliers will be summarised in the
final report. Gross outliers (defined as values lying more than twice the
inter-quartile range beyond the quartiles) were identified and omitted
for the preliminary summaries.
Table 1 below shows the initial summaries for the samples.
Table 1: Preliminary descriptive
values (gross outliers omitted)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Samples A and B, the Kauri wood samples are not included in the above table due to the additional complexity in their analysis because of the finite/non-finite convention in reporting of results.
Preliminary consensus values
The next analysis presented preliminary consensus values
and their estimated standard errors (a measure of the precision of the
estimate). The procedure used was iterative and identical to that used
in the analysis of the IAEA reference materials (Rozanski et al, 1992).
Table 2 below shows the preliminary consensus values for each sample and
also by the three different laboratory types.
Note: these values may be subject
to change as further analysis is completed.
Table 2: Preliminary consensus values
and estimated standard error
| FIRI sample | Mean | AMS | GPC | LSC |
| AB (pmC) | 0.33 (0.01) | 0.17 (0.007) | 0.25(0.002) | 0.55 (0.012) |
| C | 18173 (11) | 18183 (13) | 18229 (28) | 18140 (25) |
| DF | 4508 (3) | 4519 (4) | 4484 (5) | 4507 (6) |
| E | 11778 (7) | 11805 (9) | 11738 (19) | 11707 (17) |
| GJ (pmC) | 110.69 (0.09) | 110.52 (0.05) | 110.85 (0.07) | 110.82 (0.08) |
| H | 2232 (5) | 2238 (6) | 2198 (9) | 2233 (8) |
| I | 4485 (5) | 4483 (7) | 4456 (10) | 4499 (11) |
Finally, FIRI A and B (in years BP) were analysed using
an approach which combines both the finite and non-finite results and whose
output is summarised below in Table 3.
Table 3: Reliability analysis results
for FIRI A and B
| sample | Mean | Median | pmC | AMS | GPC | LSC | |||
| A | 45868 | 47520 | 0.26 | 48926 | 50000 | 48019 | 48305 | 41426 | 42211 |
| B | 46504 | 47780 | 0.2 | 50120 | 50980 | 46392 | 53140 | 42615 | 43600 |
| AB | 46102 | 47780 | 0.24 | 49530 | 50380 | 47521 | 48305 | 43682 | 43600 |
These results show evidence of a significant age difference
between the laboratory types, with the LSC results, on average, being younger.
Effects of modern standard, background
materials
The results for each sample were then further analysed
to explore whether there was a significant difference in the average age
for the different laboratory types, or due to the modern standard or to
the background material. Table 4 below shows for which samples statistically
significant differences were found.
Table 4: Preliminary findings of
laboratory type, background and standard material effects.
| FIRI sample | Lab type | Background | Standard |
| A (pmC) | S | S | S |
| B (pmC) | S | S(10%) | S |
| C | NS | NS | S(10%) |
| D | NS | NS | S |
| E | S | NS | NS |
| F | S | NS | S |
| G | NS | NS | S(10%) |
| H | NS | NS | S |
| I | NS | NS | S |
| J | NS | NS | S(10%) |
S: statistically significant (at 5%)
NS: not statistically significant
The results in the table show that there are effects due
to background material for samples A and B and due to modern standard in
all samples except sample E. Indeed, for some samples, up to 30% of the
variability in the results can be explained by these factors.
Dendro-dated samples
Four dendro-dated wood samples from Belfast and Hohenheim
were included in FIRI and their dendro-dates are given in the table below.
Table 5: Dendro-dates and estimated
14C ages for the dendro-dated samples
| Sample identifier | Dendro-date | Master 14C age |
| D, F (Belfast Scots pine) | 3200-3239BC | 4495 BP |
| I (cellulose from Belfast Scots pine) | 3299-3257BC | 4471BP |
| H (Hohenheim oak) | 313-294BC | 2215 BP |
The preliminary consensus values are very close to the
master 14C ages, estimated from the calibration data.
Duplicate samples
Three pairs of duplicates: A and B (Kauri wood), D and
F (Belfast pine) and G and J (barley mash) were included. Duplicates allow
an independent estimate of laboratory repeatability. The summary of the
pair-wise differences is shown in table 6 below.
Table 6: Summary of differences
between duplicate samples
| Duplicate pair | Mean difference | Median difference | Standard deviation of differences |
| A-B | 0.4 pmC | 0.04 | 3.04 |
| G - J | -0.27 pmC | -0.12 | 3.72 |
| D - F | -31.8 years | 17 | 324.9 |
The average difference in the duplicates is very small,
and well within the standard error, implying good agreement.
Discussion
These analyses were presented and discussed in some detail at the workshop and a list of further analyses drawn up. In particular, analysis and summarisation of the ? 13C results, a sensitivity analysis of different approaches to calculation of consensus values and a consideration of the associated errors were identified. In addition, indicators of individual laboratory performance would be computed and summarised.
There was also some discussion about the anonymity of the results: it was suggested that since such comparisons are only snap-shots in time, that publication of a 'league-table of laboratory performance' should be discouraged. Therefore, in the FIRI final report, a list of all participating laboratories would be presented, but these laboratories would not be identified in the data tables. The results would have a footnote suggesting that any user who wished further information on a laboratory performance should contact that laboratory directly.
Finally, laboratories were reminded that FIRI optional
samples (wood, cellulose, whole peat, mammoth tusks, fabric and parchment)
were still available and that material from both TIRI and FIRI has been
archived for future use.