A Statistical Methodology to Analyze the Effect of Changes in Testing Technology on Measurement Results
In the field of soil mechanics and highway engineering, there is great interest in the comparison of two different instruments or two different measuring technologies that measure the same characteristics of given items (or samples). This interest holds also for comparison of semi-identical instruments having minor changes in a given testing technology. All such comparisons involve statistical methods which permit the researcher to estimate separately the natural variability of the characteristic and the variability caused by measurement errors. However, previous studies consider only two components of measurement error: (a) the random error of measurement and (b) the instrumental systematic additive bias, thus skipping over the important third component, i.e., the instrumental systematic multiplicative bias. This paper deals with the above-mentioned modifications for the two-instrument case of the randomized-pair comparison sets. It is shown that this case necessitates a preliminary assumption in order to identify the different errors in the statistical analysis. The appropriate assumption depends on the nature of the testing mechanisms as demonstrated in this paper by three examples involving testing with the dynamic cone penetrometer. Finally, it should be mentioned that the nonnormally distributed case is very frequent, especially when the pool of results includes tests performed on different types of soils, even when the results of each type are regarded as normally distributed with different means and variances. Thus, the paper suggests sensitivity tests for the calculated parameters which include the above-mentioned case. Also, in the paper´s appendix, the known t and F tests are modified for the same purpose.