ABSTRACT
The monitoring of quality of soil analysis suppresses errors into the recommendations of liming and fertilizers. The program of quality control from ROLAS-RS/SC network evaluates the analysis accuracy by testing four soil samples monthly during a year. However, the data sets must have normal distribution and absence of outliers to insure that median can be consider an estimation of the true value from the soil samples. Therefore, the objective of this work was to check the normal distribution, identify outliers and evaluate procedures of accuracy calculations, such as how these aspects may affect laboratories accuracy. The Lilliefors test was run to check the normality and outliers were identified through the quartile test. Elimination of these ones from data population was test, such as the substitution of the median by average as criteria of central reference and the procedure of calculation the accuracy per attribute instead of annual average accuracy was tested. Only 48% of data followed normal distribution. Outliers exclusions of data sets improved analyzes with normal distribution up to 65%, which decreased laboratories with minimal accuracy required by the network. When data sets have normal distribution, the average shows a better estimation then the median and the procedure of calculate annual average accuracy may hide attributes of analyzes less accurate then the minimal average required.
Key words:
normality; outliers; accuracy; laboratories; ROLAS-RS/SC network
