ESTABLISHING A REJECTION PROCEDURE FOR CROP PERFORMANCE DATA

Crop performance trials are conducted to provide unbiased relative performance data for breeders, growers, and extension personnel, This report presents a method for establishing a criterion level of precision for rejection of questionable data, For a given crop, historical data (a minimum of 30 data points) are used to calculate average error variance and to determine any relationship between productivity level and error variance by regressing log(variance) on log(mean) for the variable of primary interest, usually yield, A nonsignificant relationship dictates using the pooled error variance over all trials as the reference or expected variance; a significant relationship dictates using the fitted variance predicted from the regression at the observed mean yield of each trial, An upper bound on error variance (a multiple k of the expected error variance) is calculated, Any trial with an observed error variance above the upper bound would be discarded as having unusually low precision, Historical data from five crop species tested in North Carolina were examined as working examples. Of eight data sets, four showed no relationship between variance and productivity level on a log-log scale, Compared with corn and soybean, more small-grain (barley, oat, and wheat) trials were above the criterion level when k = 2. Thus, k-values may need to be adjusted among crop species, Most trials with unacceptably low precision showed a record of some sort of problem, although it could not be confirmed that the noted problem caused the high variability, Researchers using the proposed procedure are cautioned to examine their data closely to ensure an appropriate k-value and to note test sites of low precision or methods causing low precision.