Evaluation of Calculated Reproducibilities and Reaction Errors

Reproducibility of measurements of major dissolved ions can, in theory, be better than 1%. However, in real life poorer reproducibilities are common. In Table 5.5 the reproducibility values of +4.7% for the sodium data and + 4.5% for the chlorine data are on the limit of acceptance. However, 10% for calcium and 12% for HC03 are shaky, and the value of 70% for S04 is totally unacceptable. 

In the Wisdom Spring example, the investigator should discuss the results with the laboratory staff and see whether they can improve. Otherwise, it will be necessary to shift to another laboratory and repeat the check. 

As stated, reaction errors are another way to estimate data quality (section 5.8). The reaction error can (and should) be better than 1%, but they are often larger. Many investigators will accept up to 5%. Using the latter criterion the reaction errors calculated in Table 5.6 (4.3%, 3.3%, 2.1%, 1.2%, and 1.6%) are acceptable. This example brings out a major shortcoming of the reaction error quality test the poor reproducibility of the S04 data, demonstrated above, is not reflected in the reaction error because SO4 is a minor compound of the Wisdom Spring water. Low reaction errors indicate acceptable quality of the data for the major ions alone. 

The reaction errors in repeated measurements are expected to be random, that is, in part of the measurements the total cations will exceed the total anions and in the rest of the cases the total cations will exceed the total anions. However, occasionally a systematic pattern is seen. In Table 5.6 all the measurements reveal total cations to be higher than total anions (check it). Possible explanations are  

Marcel Dekker, Inc. 270 Madison Avenue, New York, New York 10016 

---