运用Monte Carlo模拟方法评价林木转化分析法

The Statistical Evaluation for Transformation Analysis by Means of Monte Carlo Simulation

使用蒙特卡罗模拟(Monte Carlo simulation)方法来评价了林木非平衡单因素随机区组试验资料转化前后的分析效果.为了减少工作量,并使研究结果具有普遍性,采用了5个试验,单因素RCB设计,将多株小区转化成单株小区,研究转化分析法的统计学基础.评价非平衡试验资料转化前后分析法优劣的指标有,(1)有无负的方差分量;(2)参数的偏性,偏性的显著性和均方误大小;(3)试验资料的转化对家系遗传力和单株遗传力估计值大小和误差的影响.经过比较分析发现t(1)转化分析法可以消灭负的方差分量;(2)在参数的偏性、偏性的显著性和均方误大小方面,试验I至III的结果是一致的,除了未转化的资料Vb偏差达到显著水平外,其它参数间的偏差不显著;所有参数的均方误都是未转化的资料大;(3)对试验资料进行转化,有利于提高提高遗传力的大小,降低参数的误差.由于林木造林试验多采用4-8株小区,所以可以得出结论;转化分析法都要优于原模型分析法.建议在林木遗传育种实践中采用转化分析法来处理非平衡试验资料.

Monte Carlo simulation was used to evaluate statistical effects of transformation analysis. In order to reduce the workload and make the results of the study more universal, 5 tests balanced data about single factor randomized block design （RCB, 100sets of data per test） were simulated. According to the breeding experience of Chinese fir （afforestation survival rate is 83%）, random deeth of individual plants were completed by use of delete command in MATLAB language, unbalanced experimental data had been got. Analytical contents of unbalanced data between before and after transformation used to assess the statistical effectiveness and accuracy of transformation analysis have： （1） there exist or not negative variance components; （2） bias of parameters, significance testing of bias, and size of mean square error. （3） Test data transforma- tion influence on the size and error of heritability. The results of the study were as follows： （1） transformation analysis can eliminate negative variance components; （2） In bias of the parameters, sizes of bias and mean square error, the results of the test I to III are consistent： beside test data not to transform, bias of Vb reached significant levels, bias of other factors are not significant; the mean square error of all factors is always bigger in the original test. （3） All tests show that transformation of test type benefits to increase the size of heritability, decreasing error of heri- tability. Because of field test of forest, 4-8individuals in plot is widely used, so it can be concluded： transformation analysis is better than the original model analysis method. It was suggested that transformation analysis would be used in the domain of genetic breeding of forest trees in order to deal with unbalanced test data.