Analysis of rice sample size variability due to development stage, nitrogen fertilization, sowing technique and variety using the visual jackknife.

The determination of sample size before collecting experimental data is fundamental to obtain reliable estimates of variables describing agroecosystem development. In order to analyze the influence of experimental factors (artificially-induced variability) on rice sample size, an
experiment was carried out in 2004 in northern Italy. In particular, different sample size determinations were carried out for different fertilization levels, varieties (Indica and Japonica type), development stages, sowing techniques and typologies of the sampling unit. The obtained sample sizes were compared to investigate the influence of each factor, keeping the others constant (for example, we have compared the sample sizes computed for different fertilization levels within the same variety, the same phenological stage and the same sampling unit). Since original data were often not normally distributed and the variances of the original samples were not homogeneous, a new approach for sample size determination based on a visual evolution of the jackknife was preferred to classical techniques.
Results (expressed as number of plants) showed that (i) sample sizes computed in an early phenological stage (between 21 and 27) are higher than those calculated for later stages (15–21); (ii) fertilization hides soil N content variability with the consequence that larger sample sizes are required for unfertilized plots (21–27) compared to fertilized plots (15–27) and (iii) for the early sampling, the Indica type variety required larger sample size (always 27) with respect to the Japonica type variety (21–24). For row-seeded rice, the number of plants instead of linear centimeters as the sampling unit led to lower sample sizes (18–27 versus 30–33). These results highlight the influence of experimental factors and development stage on within-plot variability, and therefore the importance of preliminary samplings for sample size determination.