Journal of Taiwan Agricultural Research

Author:Chih-Hao Chiu and Dah-Jing Liao*

Abstract:

    This study used the data collected from Xikou Farm of Chiayi Agricultural Experiment Branch of Taiwan Agricultural Research Institute in 2017 to estimate the crop cutting surveys of rice grain yield. Monte simulation of sampling was adopted to realize the performance between the point estimation and interval estimation, while bootstrap method was in the conditions of different group distributions and different samplings, then it was compared to conventional parametric estimation. The results showed that point estimation from bootstrap and parametric estimation was nearly identical, no matter the group distribution was normal or skewed distribution, or the sampling size was small (n < 30), or large sample. The bootstrap method of point estimation did not add new sampling information, so it could not increase precision of point estimation a lot. In the interval estimation, compared to the performances of 4 interval estimations of conventional parametric estimation, bootstrap normal (BN), bootstrap percentile (BP) and bootstrap-t (Bt) in 0.95 confidence coefficient, the coverage probability of Bt reached 0.95 or higher in the conditions of any group distributions and samplings. BN and BP reached nearly 0.95, while its group was normal distribution and its sample size was 100. Expected length of BN and BP were narrower than others, but Bt was the widest one. In reality, both coverage probability and expected length varied alternatively while confidence coefficient was fixed. This study showed that in the conditions of various group distributions and samplings, the coverage probability and expected length from bootstrap methods got the best performances, implying that bootstrap method was relatively robust to estimate the performance of the interval estimation precisely.

Key words:Rice, Crop cutting surveys, Bootstrapping, Monte Carlo simulation

Download: PDF Links