TRIAL ERROR AND OUTLIER IN AGRICULTURAL EXPERIMENT DATA AND THEIR HANDLING

Abstract

Outlier is a questionable plot value which differs greatly from the mean of all other replications of a specific treatment while trial error may occur as a result of a mistake in measurement or transcription during data collection or entry. The presence of any of the two confounds the inferences made from the data analyses especially the F-statistic whose estimate is based on sample means and variances. This report was intended to show how Genstat statistical software package assists in identifying ailing data and consequently suggest ways of making corrections. Three sets of data comprising a set of data collected on number of leaves (NLVS), data with error (NLVS-Error) and data where the erroneous datum recorded as missing (NLVS-Missing) were subjected to analysis of variance test. The print option for READ data programme for error showed ‘‘skew’’ error message for the NLVS-Error data set, alone. When treated as missing value, NLVS-Missing, the output obtained was not different from that of the actual data, NLVS. The observed error message is a pointer to an outlying value contained in the data that is not conforming to the assumptions of independent and normal distribution of samples. Accordingly, such errant data should be corrected by revisiting the field book and where not possible, the value should be treated as missing value and resubmitted for analysis. Additionally, in plant or animal breeding experiments, if the outlying value is a confirmed true value, the experimental unit is carefully protected for use as a source of improved gene or established as an entirely new variety.

Keywords: Outlier, measurement error, data handling, Genstat statistical software