Measurement Diagnostics by Analysis of Last Digits

Though measurement instruments and methods have their natural precision limits, long term monitoring, which is so importantfor environmental studies, requires to set up and maintaincertain standards regarding precision. The elementaryrequirement is to produce measurements with a prescribed numberof valid digits. If the technology further develops and allowsto increase the number of valid digits, the change should bedocumented. In this paper we discuss a simple procedure for validity assessment of the last digits of data in a data base.It may be useful for both calibration control of an instrumentas well as for preliminary data analysis in scientific studies.The procedure utilizes the assumption that, under normalconditions, the last digits of the observed data can usually beconsidered as uniformly distributed random numbers. Failures ordeliberate changes in the observation mechanism can be detectedif particular digits occur more frequently than the others. Thedecision about proper representations of the last digits is doneby means of the familiar chi-square statistics.