n种统计方法的应用条件.docx

— parameter test numerical data 1.Z:independence,normality,homogeneity used in variance of population is known,or sample size is large(n＞50,n1+n2＞60) 2.t:independence,normality,homogeneity(INH) used in variance of population is unknown,or sample sizw is small one sample two independent sample paired sample (df=n-1) 3.ANOVA(F test):INH used in comparison of more than two means of groups. pletely randomized design(df of error =N-g) b.randomized block design(df of error=(g-1)(k-1)) c.latin square design (b and c can control confounding factors) further:want to know which pairwise groups have significant difference a.SNKq unplaned comparison b.Dunnett t:planed comparison c.LSDt:compare choosen two groups categorical data chi square test:usd in a.how closely an observed distribution matches expected distribution(goodness of fit) b.whether two variables are independent c.whether the difference of proportion is significant 1.fourfold table df=1 n≥40,E≥5:pearson chi square test 1≤E＜5:Yates continuity correction n＜40:fishers exact test paired fourfold table b＋c＜40:continuity correction used in:to same objects,different methods R*C table :df=(r-1)(c-1) cell numbers of E＜5 more than 1/5,use fishers exact test phi coefficient ,cramers V,contigency coefficient are the measures of association for two categorical variables 二.non parameter test wilcoxon signed rank test :used in paired sample and one sample ,when they do not satisfy the condition of parameter test.(statistics:T-,T+任选) 当 n 大于 50 无法查表时，T~N(n(n+1)/4,n(n+1)(2n+1)/24) 且当 n 不足够大时，需要 continuity correction 当秩次重复超过 1/5 时，Zc wilcoxon test for two independent sample:混合编秩，分别求和(T1,T2,选 n 较小者的 T) 3.rank sum test for ordinal data.对于单向有序，选用行列表检验时无法得出疗效差别时 4.K-W test for :the number of groups more than two(statistics:H) 三.regression simple linear regression :used in analysis of the influence of explanatory variable(independent variable) to the outcome variable(dependent variable) X,Y are numerical data X,Y have linear relationship