Outline of Lecture轮廓的演讲.pptVIP

January 25, 1999 PS581 lecture 3 Statistics Refresher: Topics Central tendency Expected value and means Dispersion Population variance, sample variance, standard deviations Measures of relations Covariation covariance matrices Correlations Sampling distributions Measures of Central Tendency In general: E[Y] = μY For discrete functions: For continuous functions: An unbiased estimator of the expected value: Rules for Expected Value E[a] = a -- the expected value of a constant is always a constant E[bX] = bE[X] E[X+W] = E[X] + E[W] E[a + bX] = E[a] + E[bX] = a + bE[X] Measures of Dispersion Var[X] = Cov[X,X] = E[X-E[X]]2 Sample variance: Standard deviation: Sample Std. Dev: Rules for Variance Manipulation Var[a] = 0 Var[bX] = b2 Var[X] From which we can deduce: Var[a+bX] = Var[a] + Var[bX] = b2 Var[X] Var[X + W] = Var[X] + Var[W] + 2Cov[X,W] Measures of Association Cov[X,Y] = E[(X - E[X])(Y - E[Y])] = E[XY] - E[X]E[Y] Sample Covariance: Correlation: Correlation restricts range to -1/+1 Rules of Covariance Manipulation Cov[a,Y] = 0 (why?) Cov[bX,Y] = bCov[X,Y] (why?) Cov[X + W,Y] = Cov[X,Y] + Cov[W,Y] Covariance Matrices In-Class Dataset: National Security Survey Review the Frequency Report Public perspectives on national security, domestic and international Telephone and Internet survey Dates: April 2005-June 2005 Knowledge, beliefs, policy preferences Class data: n=3006 Variable types Nominal Ordinal scales, Likert-type scales Ratio scales Stata format Characterizing Data Rolling in the data -- before modeling A Cautionary Tale Sample versus population statistics Properties of Standard Normal (Gaussian) Distributions Can be dramatically different than sample frequencies (especially small ones) Stata Tails go to plus/minus infinity The density of the distribution is key: +/- 1.96 std.s covers 95% of the distribution +/- 2.58 std.s covers 99% of the distribution Student’s t tables converge on Gaussian Standard Normal (Gaussian) Distributions S