Correlations between microbial parameters from water samples: expectationsand reality

Data which are collected in order to estimate the correlation between param eters must be analysed with caution. Classical statistics of correlation ar e often inappropriate. The "r" statistic is very easily distorted by non-No rmal data. Non-parametric statistics can be helpful. The interpretation and usefulness of the estimates of correlation will depend on the study plan. If water samples come from disparate sources (e.g. upstream or downstream f rom sewage outlets) then parameters A and B may occur in their highest and lowest numbers according to how close the samples were to contamination sou rces thus correlating closely. However, if all samples come from sources wi th similar pollution levels then plots of A and B will show considerable sc atter and apparently little correlation. So what is the relationship betwee n A and B? An example of "perfect" correlation, as demonstrated by replicat e counts of a single parameter from split samples, gave an r value of only 0.63 (p = 0.62) due to random variation in numbers of organisms between the two halves of the sample. Thus large amounts of data are needed for studyi ng true correlation because relationships between parameters are embedded i n the natural variation. This also illustrated that Standards for a single parameter can be "passed" or "failed" by two halves of the same sample. Stu dy design is clearly of fundamental importance. Consideration must be given to the appropriate way of asking questions about correlation between diffe rent parameters.