Environ. Sci. Technol. 2002, 36, 2116-2121
Statistical Results and Implications of the Enchytraeid Reproduction Ringtest A R N D W E Y E R S , * ,† J O ¨ RG RO ¨ MBKE,‡ THOMAS MOSER,‡ AND HANS T. RATTE§ European Commission, European Chemicals Bureau, 21020 Ispra (VA), Italy, ECT Oekotoxikologie GmbH, Bo¨ttgerstr. 2-14, 65439 Flo¨rsheim/Main, Germany, and Department of Biology V (Ecology, Ecotoxicology and Ecochemistry), Aachen University of Technology, Worringerweg 1, 52056 Aachen, Germany
An international ringtest of the enchytraeid reproduction test (ERT) was performed by 29 laboratories, with two substances (carbendazim and 4-nitrophenol) and two test designs (NOEC and ECx). Although many participants had no previous experience in enchytraeid ecotoxicology, the majority of tests were performed accurately and fulfilled the validity criteria, demonstrating method feasibility. However, variability in control reproduction was fairly high and exceeded the proposed validity criterion of a 50% coefficient of variation in 24% of the tests. Variability may be lowered with increasing experience and by using better defined animals, either through synchronized culture or by weighing animals before the test. In all tests, NOECs were higher than the EC10 values, sometimes by more than 1 order of magnitude and usually closer to the EC50. This shows that NOECs, due to their dependence on test design and variability, are an unsatisfactory measure of “no toxicity” for regulatory purposes. Preferably, differences to the control at NOEC concentrations should always be given. To achieve appropriate power and to avoid frequent false negatives, in the ERT, high numbers of replicates would be necessary. We favor using regression approaches with means of replicates whenever possible.
Introduction For the development and update of international guidelines, ringtests are carried out to demonstrate method feasibility (1, 2). To complement existing terrestrial toxicity tests, a ringtest with the enchytraeid reproduction test (ERT), sponsored by the German Federal Environmental Agency (UBA) and the European Commission, was performed between 1996 and 1998. This paper describes the initial test designs, presents summary statistics of the ringtest, compares the statistical methods used, and recommends test designs and statistical methods for the guideline. The ringtest was based on a draft protocol of the method, which was discussed and revised at two international workshops held in Bad Soden, Germany, and Ispra, Italy (3). * Corresponding author phone: +49 (0)351 463 36130; fax: +49 (0)351 463 32670; e-mail:
[email protected]. Current address: BUA-Bu ¨ ro O ¨ kotoxikologie, 01062 Dresden, Germany. † European Chemicals Bureau. ‡ ECT Oekotoxikologie GmbH. § Aachen University of Technology. 2116
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 10, 2002
Enchytraeidae are a family of small oligochaete worms distributed worldwide in a variety of habitats (4). Especially, soil-inhabiting species are seen as ecologically relevant, because they influence soil structure and contribute significantly to organic matter breakdown. In addition, enchytraeids are often abundant in soils where earthworms are scarce (the opposite is rare) and are sensitive to chemicals. Species of the genus Enchytraeus offer practical advantages as test organisms as they are easy to handle and breed. The test duration of 4-6 weeks, depending on the Enchytraeus species used, is shorter than 8 weeks (20 weeks, including synchronization time) used in the earthworm test (5), smaller amounts of soil are necessary, and there is minimal experimentation between start and evaluation of the test, making the test less expensive. Consequently, enchytraeids were selected as organisms for a new chronic ecotoxicological test, known as the enchytraeid reproduction test (ERT), which is currently finalized as an international guideline for the Organization for Economic and Cooperation and Development (OECD), ISO, and American Society for Testing and Materials (ASTM). The ringtest was one of the largest ever with 29 institutions from 15 countries, including the United States and Canada. Most participants tested two substances (4-nitrophenol and carbendazim (chemical name: methyl benzimidazol-2ylcarbamate)) in two test designs (NOEC and ECx). Test substances were chosen to differ in use (pesticide and industrial chemical) and environmental fate, with carbendazim being poorly degradable and 4-nitrophenol being degradable. Also, toxicity to oligochaete worms was known for both substances. Following OECD recommendations (1), the ringtest was conducted to refine the draft guideline and to determine if the proposed validity criteria can be met, if the proposed reference substance (the fungicide Carbendazim) is suitable, and if the interlaboratory variability is acceptable and to recommend suitable statistical methods. Recommendations on improving the quality of statistics in regulatory ecotoxicity tests, which were the outcome of two international workshops on statistics in ecotoxicology, were followed. One was organized by the Society for Environmental Toxicology and Chemistry (SETAC) in 1995 (6) and the other by OECD in 1996 (7). Both concluded that NOECs, for various reasons, are not a good summary statistic in regulatory ecotoxicology and should be phased out. Currently, ECx estimates are mainly used for acute tests (e.g., Daphnia and algae), while NOECs, despite the criticism of their appropriateness (8-10), are still used in higher tier reproduction tests (e.g., fish or earthworm) and thus possibly also the ERT. Therefore, two test designs were performed by the participants: the first (“ECx design”) with fewer replicates per treatment and more concentrations for the calculation of dose-response curves and the second (“NOEC design”) with more replicates and fewer concentrations for the calculation of NOECs. From both designs, NOEC and ECx were calculated to assess the effect of test design on test power and dose-response curves. Normal sigmoid dose/response functions using logtransformed concentrations were used to calculate point estimates for effect levels (EC10 and EC50). To determine the NOEC, variance homogeneity and normal distribution was tested using Bartlett’s and Kolmogoroff-Smirnov’s test procedures to select an appropriate test. Depending on the result, the NOEC was determined by two parametric (Dunnett and Williams’) and the nonparametric Bonferroni-Holm test 10.1021/es000259h CCC: $22.00
2002 American Chemical Society Published on Web 04/02/2002
TABLE 1. Test Regimes for Tests with Carbendazima concentration [mg/kg]
0
1
1.8
2.4
2.7
3.2
3.7
4.2
4.9
5.6
6.5
7.5
10
total n
ECx design NOEC design
6 8
2 4
2 4
2
2
2 4
2
2
2
2 4
2
2
2 4
30 28
a
Concentrations given in mg/kg Derosal (carbendazim content 32 wt %). n ) number of treatment and control replicates.
(11). To assess the interlaboratory variability, h values were computed according to Burton et al. (12).
Materials and Methods Tests were performed with Enchytreaus albidus (3); results of additional tests with a new species of the genus, E. luxuriosus, which was discovered early in the ringtest, are reported elsewhere (13). Test chemicals were carbendazim (a fungicide) using the liquid formulation Derosal, and 4-nitrophenol (an industrial chemical and a metabolite of the insecticide parathion), the former being also the envisaged reference substance. Most participants performed both the NOEC and the ECx design with both test substances. Test duration was 42 days, using 20 g (dry weight) of OECD artificial soil (5) and 10 adult worms per test vessel. Reproduction was determined counting juveniles at the end of the experiment either by wet extraction (14) or using a staining method with bengal red/ethanol (15) to facilitate counting. Validity Criteria. The following validity criteria for control data were proposed: (1) less than 20% mortality of adult worms after 21 days; (2) more than 25 juveniles per test vessel at the end of the definitive test; and (3) less than a 50% coefficient of variation at the end of the definitive test (endpoint: number of juveniles). Previously, a mortality criterion of 10% had been proposed but was increased to 20% after initial studies indicated that 10% could sometimes not be met, leading to the exclusion of studies that were valid regarding the reproduction criterion. The coefficient of variation (reproduction) was not used as a cutoff criterion during the ringtest because of the limited experience with this criterion. Test Design. Because effect thresholds of carbendazim to enchytraeids were known before the start of the ringtest (16), test concentrations were identical for all participants, both for the NOEC and ECx design (Table 1). For the tests with 4-nitrophenol, the same number of replicates and concentrations was used, but participants selected the spacing of concentrations according to the outcome of their own range finding tests. Statistical Methods. The statistical evaluation was performed with the PC software EASY ASSAY multiple testing and EASY ASSAY critical values. Pretesting. Kolmogoroff-Smirnov Test (KS Test). This test (17) assesses how the data distribution fits to the theoretical normal distribution. Because the parameters of this normal distribution, µ and σ, are unknown and thus had to be estimated using the raw data, the modification of Lilliefors (18) was applied. The KS test uses the tabulated critical margins for the maximum difference published by Lilliefors. It reacts sensitively to irregularities in the distribution shape. Bartlett’s Test. This test is most suitable if the data are close to the normal distribution (17), which was often the case with enchytraeid data. From the single sample variances, a χ2-distributed test statistic is computed. On the basis of χ2 and the degrees of freedom, the corresponding probability p (χ2) is calculated. In case this is lower or equal to the chosen significance level, the homogeneity hypothesis is rejected. Such a result points to a marked inhomogeneity of variances. NOEC/LOEC Testing. Dunnett’s test. Dunnett (19, 20) developed this sort of t test for comparing several indepen
dent treatments with a control; it requires the normality and homogeneity prerequisite. If this was not fulfilled, the Bonferroni U test was performed (see the following discussion). Dunnett’s test is often recommended by standards and guidelines and is available through most statistical program packages. The sample sizes should be equal except for the control, the replicate number of which should be nxp replicates (p ) number of treatments without the control, n ) number of replicates per treatment) to optimize the test power. It aids in the interpretation of the results if the effects occur in a monotonous order, either steadily increasing or decreasing. If this does not apply, the NOEC often cannot be determined unambiguously. In this case, the Williams’ test procedure often leads to a reasonable estimation of the NOEC. Williams’ test. This multiple test (21) is also a t test and is closely related to the Dunnett test, comparing all of the treatments with the control. The Williams’ test can easily be conducted with unequal sample sizes, using corrected critical t margins (22). Because of a sequential approach, this test is more powerful than Dunnett’s test. The NOEC can be detected, even if the treatment means do behave nonmonotonously. The procedure computes a maximum-likelihood estimate of a monotonous order of treatment means prior to conducting the multiple t test. Bonferroni-Holm Test (Multiple U Test). The BonferroniHolm test comes as a multiple sequential U test, comparable to Williams’ test but nonparametric. It was used for NOEC computation whenever Dunnett and Williams’ test could not be performed because of inhomogeneous variances or missing normal distribution. Also here, all of the treatments are compared with the control. The test is based on pairwise U tests, from which the result (probabilities of treatment U values) are further assessed by a sequential Bonferroni procedure (i.e., each comparison of the U probabilities with the significance level is done using an adjusted significance level R′ (R′ ) R/number of comparisons being left; R, overall significance level)). Using the same data set, this test leads to fewer significant decisions than is the case with Dunnett’s and Williams’ test, because rank-based tests are somewhat less powerful than tests based on metric distances. An overview of the decision tree leading to methods for determining NOECs is given in Figure 1. With the present ringtest, however, no transformation was applied, and the general scheme of Figure 1 was used to decide which of the results obtained from all of the three NOEC tests had to be selected for further evaluation and discussion. Dose-Response Function and ECx Determination. For every ringtest sample, the EC10 and EC50 were derived from a normal sigmoid dose-response function. Treatment means were transformed into inhibition values relative to the control using eq 1.
I)1-
PT PC
(1)
with PC ) the arithmetic mean of the offspring number in the control and PT ) the arithmetic mean of the offspring number in a treatment The relation between these inhibitions and the concentrations often described a specific concentration-response VOL. 36, NO. 10, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2117
FIGURE 1. Decision tree for parametric and nonparametric tests for NOEC determination.
TABLE 2. Meeting Validity Criteria of All Tests with Both Substances and Both Test Designs for the Controlsa carbendazim
a
4-nitrophenol
test design
NOEC
ECx
NOEC
ECx
no. of tests valid test: general conditions valid tests: mortality valid tests: control reproduction [valid tests: CV of control reproduction] valid tests (all criteria)
32 24 (75%) 23/24 16/23 [12] 16 (67%)
30 24 (80%) 23/24 17/23 [14] 17 (71%)
23 21 (91%) 17/21 15/17 [11] 15 (71%)
25 23 (92%) 20/23 17/20 [13] 17 (74%)
Coefficient of variation (CV) higher than 50% was not used as an exclusion criterion.
curve, to which a normal sigmoid function could be fitted using probit analysis, based on a maximum likelihood regression analysis (23, 24). Although, initially, probit analysis was “invented” for quantal responses, there exists a modification for quantitative parameters, such as in the present test. The modification (25, 26) was done for the weighting with regression analysis as well as for the computation of the residual variance and confidence limits (“modified probit”). Interlaboratory Variability. According to a recommendation by ASTM (27), the reproducibility of data derived from various laboratories can be determined by using the h index, which is the standardized measure of the distance between the result (e.g., the NOEC or ECx) of a single laboratory and the mean of the results from all laboratories. To our knowledge, this index has not been used in terrestrial ecotoxicology so far; hence, it is briefly described. The h value equals the standard normal variable z and is determined (12) by dividing the sample deviation (i.e., the difference between a laboratory NOEC or ECx from the mean NOEC or ECx obtained from all valid tests) by the standard deviation of the considered sample statistic (i.e., the standard deviation of the laboratory NOECs from the mean NOEC of all laboratories or the standard deviation of the laboratory ECx 2118
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 10, 2002
from the mean ECx of all laboratories from which valid results were obtained; eq 2).
hx )
x-µ s
(2)
where hx ) h value for summary statistics x; x ) individual summary statistic (e.g., NOEC, ECx); µ ) average of the statistic (e.g., mean NOEC, ECx from all laboratories with valid tests); s ) standard deviation of the considered sample statistic (e.g., from the NOEC or ECx). If the h value is higher than the critical t value (p ) 0.01, two-sided), the summary statistic is considered an outlier (27). Thus, this sort of outlier test leads to less-conservative results other than the commonly used 4σ range or the DixonGrubbs test (17) (i.e., results of single laboratories are considered as outliers at a smaller difference to the mean than the commonly used outlier tests). The h value decreases with increasing standard deviation (i.e., if overall variability is high) while the critical t value decreases with an increasing number of observations.
Results and Discussion Validity Criteria. Out of 110 tests, 92 were performed correctly (Table 2) concerning test conditions, such as test species,
FIGURE 2. Control survival (A) and reproduction (B) as validity criterion. Mean and s given. Limit values (80% and 25 respectively) shown as dashed line: Carb ) carbendazim; 4-NP ) 4-nitrophenol.
TABLE 3. Meeting the Prerequisites for Parametric Tests (Dunnett’s or Williams’)
FIGURE 3. Summary statistics for carbendazim (A) using the formulation Derosal (carbendazim content 32 wt %) and 4-nitrophenol (B), each with NOEC and ECx design. Mean (9) and median (0) given. For calculating the mean and its 95% confidence interval (error bars), > < indications were ignored and values treated as if accurately determined: D ) Dunnett’s test, B ) Bonferoni test.
carbendazim 4-nitrophenol test design
NOEC
ECx
NOEC
ECx
no. of tests control groups with normal distribution (%) (Kolmogoroff-Smirnov test) tests with homogeneous variance (%) (Bartlett’s test)
16 89
17 78
15 99
17 100
44
72
87
94
substrate pH, and temperature. Of these, 90% passed the mortality criterion ()20%; see also Figure 2A) of which 80% also passed the reproduction criterion ()25 juveniles per 10 adults; see also Figure 2B). Achieving a low coefficient of variation (CV) of control reproduction ()50%) was more difficult because overall variability was high (Figure 2B). Because many participants were inexperienced with enchytraeids, this criterion was not used to exclude otherwise correctly performed tests, leading to 71% of the tests that met the validity criteria. The higher number of well-conducted tests with 4-nitrophenol probably reflects increasing participants’ experience with the test animals and also with the artificial soil, as carbendazim tests were carried out first. Future attempts should be made to reduce the variability of reproduction in the controls. Variability may be lowered with increasing experience and by using better defined animals, either through synchronized culture or by weighing animals before the test. Pretesting. The results of pretesting are given in Table 3. Whereas most treatment and all control samples of both substances showed a normal distribution, variance homo-
geneity was high only in tests with 4-nitrophenol but low in carbendazim tests. Obviously, the reason for this is that more often, with carbendazim, pronounced effects were observed in the higher treatments, leading to both lower means and lower standard deviations. Summary Statistics for Carbendazim. Figure 3A shows summary statistics of all valid tests with carbendazim. Original data are given in Table 1 in the Supporting Information. EC10 and EC50 values were not much influenced by the test design, whereas the NOEC increased considerably when determined from an ECx design and often was above the tested range. Note that in Figure 3 NOECs outside the tested range (with < or > indications) have been treated as if accurately determined. When obtained from the ECx design, most of these values where higher than the tested range; thus, the “true” mean NOEC would be even higher. The same applies to NOECs for 4-nitrophenol (Figure 3B) for both NOEC and ECx design. Despite this underestimation of the mean NOECs, they were always well above the EC10, even when determined from an NOEC design, and usually closer to, or above, the EC50. This is in line with recent findings (28) on fish growth data and is an indication that the NOEC is an unsatisfactory measure of “no toxicity”. The results correspond to the toxicity range known from effects of carbendazim on enchytraeids from published laboratory and field studies (16, 29) which report LC50s of 5.5-7.7 (mg/kg), NOECs between 0.24 and 4 (mg/kg) and field effects between 0.27 and 2.4 (mg/kg). Test Power. The relative test power of the three tests used to obtain NOECs is illustrated in Figure 4. This comparison VOL. 36, NO. 10, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2119
FIGURE 4. Comparison of results for carbendazim using the formulation Derosal (carbendazim content 32 wt %) obtained from Williams’ (9), Dunnett’s (0), and Bonferoni U (O) test for 16 valid tests from the NOEC design. Two valid tests with NOECs > 10 mg/kg (all three statistical tests) not shown. was only possible for the NOEC design with carbendazim, because replication in the ECx design does not allow for the use of the two parametric tests and because chosen concentrations varied between participants for 4-nitrophenol. Both parametric tests are more powerful than the Bonferoni U test, with Williams’ being slightly more powerful than Dunnett’s test. Summary Statistics for 4-Nitrophenol. Figure 3B shows the summary statistics of all tests with 4-nitrophenol (see Table 2 in the Supporting Information for original data). As for carbendazim, the NOEC was always higher than the EC10 and on average closer to the EC50. However, this comparison is based on fewer data points, because some NOECs outside the tested concentration range or EC10 values could not be determined. The small acute-chronic ratio for the substance, in some cases, led to test concentrations in the ultimate test being chosen too low. Also, as 4-nitrophenol is degraded fairly quickly under ERT test conditions with concentrations close to zero at the end of the test (3), exposure may have been influenced by minor differences in the time between application of test substance and introduction of worms. No data on the toxicity of 4-NP on enchytraeids are available in the literature, but mortality and reproduction of these worms seem to be affected at similar concentrations as in the case of earthworms (3). Interlaboratory Variability. For all endpoints, substances, and designs, the h values of the participants’ results were within the critical limits. However, as the critical value depends partly on the overall variability, which was fairly high in the ERT, critical limits were less likely to be exceeded. Control of β Error. The problem of not considering an existing effect as significant (type II or β error) is well-known and was obvious from the data. Conventionally, R is wellcontrolled and typically set to 0.05 to minimize false positives. On the other hand, the β error depends on chosen replication, and the R level and is usually not sufficiently controlled. We calculated test power for Dunnetts’ test according to Horn and Vollandt (30). Given the observed coefficient of variation of 43%, we assumed that a 30% effect should be detected as significantly different in 80% of the tests. To achieve this test power, 49 control replicates and 22 replicates in each of the five concentrations would be necessary, which is beyond the resources typically available for a test. However, the problem is not specific to the enchytraeid reproduction test. It would also be a problem, for example, in the earthworm reproduction test, which requires a 1.8 for the higher and lower concentrations. (3) A combined approach allows for the determination of both the NOEC and ECx. Eight treatment concentrations in a geometric series should be used. Four replicates for each treatment plus eight controls are recommended. The concentrations should be spaced by a factor not exceeding 1.8. It should be noted that NOEC testing with these numbers of replicates has a risk of failure to detect effects and that more replicates are desirable. To detect 30% effect in more than 80% of the tests, a reduction of the coefficient of variation to 18% would be necessary and less effect may still remain undetected. Thus agreed quality criteria for NOECs used for regulatory purposes are needed, and preferably, differences to the control at NOEC concentrations should be given in test report summaries, if NOECs are not to be completely substituted by ECx estimates.
Acknowledgments We thank the German Federal Environmental Agency (UBA) and the European Commission-European Chemicals Bureau (ECB) for financial support of the ringtest and workshops. Thanks to all ringtest participants for their dedication. Three anonymous reviewers made valuable comments on the manuscript.
Supporting Information Available Summary Statistics (NOEC and ECx values) of all tests in the ERT ringtest. This material is available free of charge via the Internet at http://pubs.acs.org.
Literature Cited (1) Organisation for Economic Cooperation and Development. Final Report of the OECD Workshop on Harmonisation of Validation and Acceptance Criteria for Alternative Toxicological Test Methods; OECD Test Guidelines Programme; Paris, France, 1996. (2) Balls, M.; Blaauboer, B.; Fentem, J.; Bruner, L.; Combes, R.; Ekwall, B.; Fielder, R.; Guillouzo, A.; Lewis, R.; Lovell, D.; Reinhardt, C.; Repetto, G.; Sladowski, D.; Spielmann, H.; Zucco, F. ATLA 1995, 23, 129-147. (3) Ro¨mbke, J.; Moser, T. Organisation and Performance of an International Ringtest for the Validation of the Enchytraeid Reproduction Test; UBA-Texte 4/99, 1999; Vols. I and II. (4) Didden, W. A. M. Pedobiologia 1993, 37 (1), 2-29. (5) Organisation for Economic Cooperation and Development. OECD Guideline for Testing of Chemicals; No. 207; Earthworm Acute Toxicity Test; Paris, France, 1984.
(6) Chapman, P. F.; Crane, M.; Wiles, J.; Noppert, F.; McIndoe, E. Ecotoxicology 1996, 5 (3), 169-186. (7) Organisation for Economic Cooperation and Development. Report of the OECD workshop on statistical analysis of aquatic toxicity data; Series on Testing and Assessment No. 10; OECD Environmental Health and Safety Publications: Paris, France, 1998. (8) Chapman, P. M.; Caldwell, R. S.; Chapman, P. F. Environ. Toxicol. Chem. 1996, 15 (2), 77-79. (9) Hoekstra, J. A.; Van Ewijk, P. H. Environ. Toxicol. Chem. 1993, 12 (1), 187-194. (10) Suter, G. W. Hum. Ecol. Risk Assess. 1996, 2 (2), 331-347. (11) Holm, S. Scand. J. Statist. 1979, 6, 65-70. (12) Burton, G. A.; Norberg-King, T. J.; Ingersoll, C. G.; Benoit, D. A.; Ankley, G. T.; Winger, P. V.; Kubitz, J.; Lazorchak, J. M.; Smith, M. E.; Greer, E.; Dwyer, F. J.; Call, D. J.; Day, K. E.; Kennedy, P.; Stinson, M. Envir. Toxicol. Chem. 1996, 15 (8), 1335-1343. (13) Collado, R.; Schmelz, R. M.; Moser, T.; Ro¨mbke, J. Pedobiologia 1999, 43 (6), 625-629. (14) Ro¨mbke, J. UWSFsZ. Umweltchem. O ¨ kotox. 1995, 7 (4), 246249. (15) Korinkova, J.; Sigmund, J. Vestnik Ceskoslovensko Spolecnosti Zoologicke 1968, 32, 300-305. (16) Ro¨mbke, J.; Moser, T.; Knacker, T In Advances in Earthworm Ecotoxicology; Sheppard, S., et al., Eds.; SETAC Press: Pensacola, FL, 1998; pp 83-97. (17) Sachs, L. Angewandte Statistik; Springer-Verlag: Heidelberg, Germany, 1992.
(18) Lilliefors, H. W. J. Am. Statist. Assoc. 1967, 62, 399-402; Corrigenda 1969, 64, 1702. (19) Dunnett, C. W. J. Am. Statist. Ass 1955, 50 (12), 1096-1121. (20) Dunnett, C. W. Biometrics 1964, 20, 482-491. (21) Williams, D. A. Biometrics 1971, 27 (3), 103-117. (22) Williams, D. A. Biometrics 1972, 28 (6), 519-531. (23) Finney, D. J. Probit Analysis, 3rd ed.; Cambridge University Press: London, U.K., 1971. (24) Finney, D. J. Statistical Method in Biological Assay, 3rd ed.; Cambridge University Press: London, U.K., 1978 (25) Christensen, E. R. Water Res. 1984, 18, 213-221 (26) Christensen, E. R.; Nyholm, N. Environ. Sci. Technol. 1984, 18 (9), 713-718. (27) American Society for Testing and Materials. Standard Practice for conducting an interlaboratory study to determine the precision of a test method; ASTM Guideline No. E 691-92; 1992 (28) Crane, M.; Newman, M. C. Environ. Toxicol. Chem. 2000, 19 (2), 516-519 (29) Van Gestel, C. A. M. Ecotoxicol. Environ. Saf. 1992, 23, 221-236. (30) Horn, M.; Vollandt R. Inf. Biom. Epidemiol. Med. Biol. 2001, 32 (1), 1-28.
Received for review October 27, 2000. Revised manuscript received February 12, 2002. Accepted February 14, 2002. ES000259H
VOL. 36, NO. 10, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2121
