Field replicate water samples (“field replicates”) collected for the U.S. Geological Survey National Water-Quality Assessment (NAWQA) Program during 1992 to 1997 were used to assess the variability of pesticide detections and concentrations in environmental water samples collected from the surface-and ground-water-quality networks of the NAWQA Program. Field replicates are two or more identically collected, processed, and analyzed environmental water samples that are used to assess the overall variability of field and laboratory procedures. Variability is the degree of random error in independent measurements of the same quantity and is the opposite of precision—the degree of mutual agreement. Information on variability can be used to estimate the reproducibility of individual measurements, the concentration needed to be assured of exceeding a water-quality standard, and the likelihood that two measurements of water quality are different.

Variability of pesticide detections was assessed by calculating the mean percentage detection of a pesticide and the percentage of inconsistent replicate sets. Variability of pesticide concentrations was assessed by pooling estimates of the standard deviation and relative standard deviation in replicate sets. Variability of pesticide detections and concentrations was a function of concentration, and estimates of variability were developed for discrete ranges of concentration. Reliability of estimates of variability was assessed by calculating 90-percent upper confidence bounds for the percentage of inconsistent replicate sets and for the pooled estimates.

The variability of detection for most pesticides is high at concentrations less than the minimum reporting level, but the variability of detection decreases dramatically at higher concentrations. In view of the highly diverse sources of water submitted as field replicates for the NAWQA Program and the generally low concentrations (concentrations in 79 percent of replicate sets were less than 0.1 microgram per liter) of pesticides in most replicates, inconsistent detections in replicate sets likely were caused by variability in the analytical method and by water-matrix interferences (or other loss processes) that result in false-negative errors. Consequently, estimates of the frequency of detection of pesticides in environmental water samples collected for the NAWQA Program probably are biased low because of false-negative errors at concentrations near the minimum reporting level.

Correlation analysis indicates that for most pesticides and concentrations, pooled estimates of relative standard deviation rather than pooled estimates of standard deviation should be used to estimate variability because pooled estimates of relative standard deviation are less affected by heteroscedasticity. The 2 Variability of Pesticide Detections and Concentrations in Field Replicate Water Samples, 1992–97 median pooled relative standard deviation was calculated for all pesticides to summarize the typical variability for pesticide data collected for the NAWQA Program. The median pooled relative standard deviation was 15 percent at concentrations less than 0.01 micrograms per liter (µg/L), 13 percent at concentrations near 0.01 µg/L, 12 percent at concentrations near 0.1 µg/L, 7.9 percent at concentrations near 1 µg/L, and 2.7 percent at concentrations greater than 5 µg/L. Pooled estimates of standard deviation or relative standard deviation presented in this report are larger than estimates based on averages, medians, smooths, or regression of the individual measurements of standard deviation or relative standard deviation from field replicates. Pooled estimates, however, are the preferred method for characterizing variability because they provide unbiased estimates of the variability of the population. Assessments of variability based on standard deviation (rather than variance) underestimate the true variability of the population. Because pooled estimates of variability are larger than estimates based on other approaches, users of estimates of variability must be cognizant of the approach used to obtain the estimate and must use caution in the comparison of estimates based on different approaches.