Page 147 - Freshwater-Biology-and-Ecology-Handbook

2.5 Correcting bias caused by analytical error

Analytical quality affects our assessment of river deduced. To ensure that differences in analytical quality
invertebrate quality, particularly when we use taxonomic are not mistaken for differences in environmental quality,
richness as a measure of quality, such as WHPT NTaxa (the we make a ‘bias correction’ to the observed data, based on
number of WHPT-scoring taxa). We have measured these the result of audits in which sorting errors were measured
errors in independent audits. Audit ‘gains’ caused by taxa independently, before using it to determine WFD status.
present in a sample in low numbers or as a single specimen
that are not noticed in the sorting tray by the laboratory The bias correction for WHPT NTaxa is relatively simple.
analyst, are much more common than ‘losses’, errors We add the net gains (gains minus losses) recorded in
where a taxon is recorded that is not actually present in the audits to WHPT NTaxa that the laboratory analysts record
sample. The effect of poorer analytical quality is therefore for every sample. The net gains are based on the annual
not random but biased, almost always resulting in fewer average for a laboratory.
taxa being recorded and therefore poorer quality being

Observed WHPT NTaxa Bias (= net gains) Corrected WHPT NTaxa

20 + 1.68 = 21.68

For WHPT:
Bias = net gains (net additional taxa revealed by the audit)
= mean net effect of errors on WHPT NTaxa
= (mean gains of WHPT-scoring taxa) – (mean losses of WHPT-scoring taxa)

The status class boundaries for WHPT NTaxa (Table 3.7) assume that bias correction has been applied to the observed
index values.

WHPT ASPT is also corrected for analytical error but the Correcting for bias does not completely remove the effect
correction is far smaller because the effect of errors on of laboratory errors but it ensures that the bias in observed
this metric are less biased. The mean net effect of error (= values match the bias in values predicted by RIVPACS. They
bias) on ASPT arises because of the WHPT values of the cancel out in EQRs (Chapter 1, Section 5.6 Biological
taxa that are most prone to error. If only sensitive taxa that Quality Elements) because the audit was undertaken by the
have high WHPT values were more prone to error than same team that analysed the RIVPACS reference samples,
low value taxa, greater analytical error would lead to lower so after bias correction, the remaining bias in observed
ASPT values, and vice versa. Fortunately, this is not the results caused by the auditors’ error equals the bias in
case and the taxa causing most errors do not all have high the reference values. RICT2 also uses bias results with
or low values. The bias correction for ASPT is calculated by information about sampling error to determine the overall
RICT2 (River Invertebrate Classification Tool) software (see imprecision in EQRs that are used in Monte Carlo simulations
Section 4.4), using WHPT NTaxa and WHPT NTaxa bias for to determine probabilities of class (Section 4.3).
the samples.

Freshwater Biology and Ecology Handbook | 147
–