生物统计13相关与回归分析.DOC

醫資系 Lecture#13 p. PAGE 14 生物統計學 96-1 200 生物統計學13：相關係數與回歸分析 0 0. Definition Before we conduct any type of analysis, we should always create a two-way scatter plot of the data. Correlation is the quantification strength of linear association among continuous random variables. 1. Pearson’s Correlation Coefficient - The linear relationship between two r.v.s X and Y is denote byρ(rho), which is the average of the product of the standard normal deviates of X and Y. - The estimator of the population correlation is known as Pearson’s coefficient of correlation, which is denoted by r. ρ= E {[(X –μX )/σX][(Y –μY )/σY]}, population correlation r = =, sample correlation = - Note that, -1≤ r ≤ 1, and r = 1 or -1 when there is an exact linear relationship between x and y. As the relationship deviates from perfect linearity, r moves away from ±1 and closer to 0. - If y increases as x increases, r 0, positively correlated; if y decreases as x increases, r 0, negatively correlated. (1) Linear relationship, invalid for a nonlinear relationship. (2) Highly sensitive to outliers, extreme values. (3) Not to be extrapolated beyond the observed ranges of the variables. (4) The correlation coefficient merely tells us that a linear relationship exists; it does not specify whether the relationship is cause-and-effect. 1.1 Hypothesis Testing - Evaluate whether any correlation exists between the r.v.s X and Y; i.e., testing H0 :ρ= 0 (there is no association) vs H1 :ρ≠ 0. Test statistic T = = , where the estimated standard error of r is . - Under H0, the quantity T ~ t(n – 2), if X and Y are normally distributed. (Valid only for testing H0 :ρ= 0, not for H0 :ρ=ρ0 → sampling dist of r is skewed) 2. Spearman’s Rank Correlation Coefficient - Pearson’s correlation is sensitive to outliers, like other parametric techniques; - Spearman’s rank correlation calculates the correlation based on the ranked values of x and y a nonparametric method. - A measure of concordance of the ranks for the o