In chilli pepper, the calculation of the effective or minimum sample size can minimize costs with characterization. In order to determine the effective sample size, a general multivariate statistical method consisting of resampling subsamples from a reference sample is presented. Data from a field experiment involving eight accessions of Capsicum pepper are used to illustrate the method. Six response variables relating to morphological characterization of fruits were analyzed: mean weight, peduncle length, fruit length, largest diameter, lowest diameter, pericarp thickness. The reference sample consisted of the vector of scores of the first principal component, thus representing 30 observations on the 6 morphological variables. Through the percentile bootstrap method, a 99% confidence interval was created for two parameters: mean and standard deviation of the reference sample, which was then resampled with replacement, creating 500 subsamples of sizes ranging from 2 to 29. Afterwards, we estimated both mean and standard deviation for each subsample of each size. The proportion of estimates outside their respective confidence interval was computed. We also compared the results of the multivariate approach with its univariate form. The multivariate approach has taken into account the correlations among the response variables and was more efficient than the univariate form. A sample containing 22 fruits is considered suitable for estimating the mean of pepper fruit traits, whereas 24 fruits should be enough to estimate the standard deviation.
展开▼
