Potential Inaccuracy of Microbiological Counts from Routine Water Samples

The procedures for estimating numbers of organisms in water lead potentially to large inaccuracies. Routine microbiological sampling is an important component of monitoring source waters for drinking supplies and recreational waters. The results should be interpreted with awareness that each result is liable to two sorts of error, even when correct procedures are followed. Firstly, there is sampling error due to variation over time in microbial density at the place of sampling and, secondly, statistical inaccuracies introduced by laboratory methods. Some of the errors can be measured or estimated, especially those in the second category.

Pre-dilution of a sample, which is unavoidable with heavily contaminated waters, can affect results by chance. For example, if 10 organisms are found after a 100-fold dilution then the estimated count for the original sample would be 1000 with 95% confidence intervals of 482 - 1834. If a presumptive count is followed up with confirmatory procedures which involve sampling then another imprecision is introduced. For example, if a presumptive colony count of 25 is checked by confirmatory tests on only 5 colonies then, if 4 of the 5 confirm, the 95% confidence for the final count would be 8 - 24. Although statistical confidence intervals from the combination of such laboratory procedures could be calculated they have little practical value and could cause confusion. An investigator who receives a report in the format of an organism count which is qualified by a 95% confidence interval might wrongly attribute that range to the source water.

Variation at the water source is usually much greater than the errors introduced by proper laboratory procedures. This sample ‘error' can be estimated only by replicate sampling. All sources of error should be considered when the quality of the water is assessed. Decisions about whether the water meets a particular Standard might be better made, from a statistical point of view, by assessing multiple samples rather than concentrating on single counts from very small volumes of water.