Abstract
Results from recently conducted collaborative trials on the determination of aflatoxin B1 in various matrices have been evaluated to establish whether the use of recovery data would result in a distinct change of the relative between-laboratory standard deviation (RSDR) of the corrected data compared with the uncorrected data. In addition, we applied conventional and robust statistics to evaluate whether the impact of the use of recovery data on the estimation of RSDR depended on the statistical method applied for data analysis. This investigation was based on means before and after correction for recovery. The method performance characteristics were calculated using results from naturally contaminated test materials, while the results from test materials fortified with the target analytes were used to estimate the recovery. The study revealed that applying conventional and robust statistics in general led to comparable estimates for RSDR. The comparison about the use of recovery data showed that in most cases, the RSDR obtained from the analysis of aflatoxin B1 decreased after correction of the results for recovery. This tendency was similar when the comparison was done using robust or conventional statistics. However, in three cases, conventional statistics yielded a higher RSDR for the corrected data, whereas robust statistics showed the opposite. Looking carefully at the data, the treatment of conventional statistics indicated that the way outliers are detected and removed could result in an under- or overestimation of RSDR. Applying the law of error propagation revealed that most likely the correlation between the uncorrected data and the recovery rate led to a reduced variability of the data corrected for recovery.