应用数理统计—主成分回归介绍

主成分回归介绍国际旅游外汇收入是国民收入是国民经济发展的重要组成部分，影响一个国家或地区旅游收入的因素包括自然、文化、社会、经济、交通等多方面的因素。《中国统计年鉴》把第三次产业划分为12个组成部分，分别为：一、提出问题x1：农林牧渔服务业x2：地质勘查水利管理业x3：交通运输仓储和邮电通讯业x4：批发零售贸易和餐食业x5：金融保险业x6：房地产业x7：社会服务业x8：卫生体育和社会福利业x9：教育文艺和广播x10：科学研究和综合艺术x11：党政机关x12：其他行业选自1998年我国31个省、市、自治区的数据。以旅游外汇收入（百万美圆）为因变量。自变量的单位为亿元人民币。数据略。InterceptCoefficients-205.236116.8459-1.756460.096008  标准 excel标准偏差excel标准偏差函数exl标准差函数国标检验抽样标准表免费下载红头文件格式标准下载 误差tStatP-valueXVariable1-1.4004522.8676-0.061240.951842XVariable22.67500118.575080.144010.887092XVariable33.3008772.4645561.3393390.197128XVariable4-0.944021.296117-0.728340.475774XVariable5-5.50164.508593-1.220250.238117XVariable64.0544343.9537451.0254670.318728XVariable74.1425.0699840.8169650.42463XVariable8-15.364910.82589-1.419270.172905XVariable917.367668.353372.0791210.052178XVariable109.07888310.147280.8947110.38275XVariable11-10.585.610696-1.885690.075582XVariable121.3507095.0015040.270060.790186这个模型是不理想的，一个最严重的问题是多重共线性的问题。线性回归模型的方差分析表方差来源自由度离差平方和方差F统计量显著性水平回归分析1211690140974178.310.513358.15025E-06残差18166789992661.04总计3113358039　利用主成分的互不相关性来建立应变量与主成分的回归，在理论上可以达到消除多重共线性。二、主成分回归方法原始数据观测矩阵主成分系数矩阵主成分得分矩阵根据最小二乘估计，则基于协方差矩阵的主成分回归基于相关系数矩阵的主成分回归主成分回归系数的协方差矩阵1、经济分析数据Y：进口总额X1：GDPX2：积累总额X3：消费总额求进口总额与GDP、积累总额和消费总额之间的回归方程。三、主成分回归的实例dataa;inputx1-x3y;cards;149.34.2108.115.9161.24.1114.816.4171.53.1123.219.0175.53.1126.919.1180.81.1132.118.8190.72.2137.720.4202.12.1146.022.7212.45.6154.126.5226.15.0162.328.1231.95.1164.327.6239.00.7167.626.3;procregoutest=b;modely=x1-x3/pcomit=1,2outvif;procprintdata=b;procstandarddata=aout=cmean=0std=1;varx1-x3y;procprincompdata=cout=dprefix=z;varx1-x3;procregdata=d;modely=z1z2/noint;run;AnalysisofVarianceSumofMeanSourceDFSquaresSquareFValuePr>FModel3204.7761468.25871285.61<.0001Error71.672950.23899CorrectedTotal10206.44909RootMSE0.48887R-Square0.9919DependentMean21.89091AdjR-Sq0.9884CoeffVar2.23321ParameterEstimatesParameterStandardVariableDFEstimateErrortValuePr>|t|Intercept1-10.127991.21216-8.36<.0001x11-0.051400.07028-0.730.4883x210.586950.094626.200.0004x310.286850.102212.810.0263EigenvaluesoftheCorrelationMatrixEigenvalueDifferenceProportionCumulative11.999154931.001000760.66640.666420.998154180.995463290.33270.999130.002690890.00091.0000EigenvectorsF1F2F3x10.706330-.0356890.706982x20.0435010.9990290.006971x30.706544-.025830-.707197Obsx1x2x3y*F1F2F31-1.509720.54571-1.53319-1.31852-2.125890.638660.0207222-1.113050.48507-1.20848-1.20848-1.618930.555540.0711133-0.76971-0.12127-0.80140-0.63625-1.11517-0.072980.0217304-0.63637-0.12127-0.62209-0.61424-0.89430-0.08237-0.0108135-0.45970-1.33395-0.37008-0.68027-0.64421-1.30669-0.0725826-0.12970-0.66697-0.09869-0.32813-0.19035-0.65915-0.02655370.25031-0.727610.303550.178070.35962-0.74367-0.04278180.593651.394580.696101.014400.971801.35406-0.06286391.050321.030781.093501.366541.559320.96405-0.023574101.243661.091411.190421.256491.767001.015220.044988111.48033-1.576481.350350.970381.93110-1.662660.080613AnalysisofVarianceSumofMeanSourceDFSquaresSquareFValuePr>FModel29.882784.94139379.38<.0001Error90.117220.01302UncorrectedTotal1110.0000ParameterEstimatesParameterStandardVariableDFEstimateErrortValuePr>|t|F110.689980.0255227.03<.0001F210.191300.036125.300.0005可见，系数的符号没有与经济概念相悖。一般建议，当VIF>10时，多重共线性是严重的。2、朗莱用美国联邦政府雇员人数Y和国民总产出隐含平减指数X1，国民总产出X2，失业人数X3，武装力量人数X4，14岁及以上非慈善机构人口数X5，时间变量X6。朗莱所用数据是美国47—62年数据，该例是主成分回归用得较早的例子。EigenvaluesoftheCorrelationMatrix（相关系数矩阵的特征根）EigenvalueDifferenceProportionCumulative（特征根）（差值）(贡献率）（累计贡献率）14.603377453.428037110.76720.767221.175340350.971915180.19590.963130.203425170.188496890.03390.997040.014928280.012376240.00250.999550.002552040.002175330.00040.999960.000376710.00011.0000Eigenvectors（特征向量）Prin1Prin2Prin3Prin4Prin5Prin6x10.4618350.057843-.149120-.7928740.337934-.135193x20.4615040.053211-.2776810.121625-.1495500.818485x30.321317-.5955130.728306-.0076450.0092350.107451x40.2015100.7981930.5616070.0772550.0242530.017970x50.462279-.045544-.1959850.5897430.548569-.311589x60.4649400.000619-.1281160.052285-.749556-.450388Prin1Prin2Prin3Prin4Prin5Prin63.47885-0.75147-0.307950.164240.008797-0.0025793.01051-0.84904-0.64223-0.125920.061546-0.0119802.34330-1.540000.493430.008820.005746-0.0050622.09390-1.276320.111290.06126-0.0618450.0136771.438241.235790.02909-0.097460.0522570.042682……0.099510.693490.097570.10111-0.0988080.0189260.449430.54784-0.29295-0.01756-0.083762-0.0141390.955060.42945-0.44524-0.11933-0.023694-0.0271541.81710-0.863170.67742-0.187060.021671-0.0081081.93999-0.38657-0.26596-0.14392-0.0366860.0235302.36112-0.49910-0.36567-0.06160-0.016235-0.0043603.07803-0.989950.201960.068110.0564270.0013393.34476-0.17667-0.423850.259680.0580920.008939SumofMeanSourceDFSquaresSquareFValuePr>FModel64985048308447.22<.0001Error9158361759.57184CorrectedTotal15514340ParameterEstimatesParameterStandardVariableDFEstimateErrortValuePr>|t|Intercept13865051225163.150.0116x1113.7116211.684241.170.2707x210.008460.004611.840.0995x310.094050.067201.400.1952x410.205620.029486.97<.0001x51-0.004350.03111-0.140.8918x61-199.2021362.67100-3.180.0112SAS的回归分析（REG）过程中，带有主成分回归的功能，在这个功能中，SAS不仅用因变量的标准化值建立了与主成分之间的回归方程，并且将回归方程还原为以原始变量为自变量，以因变量Y为被解释变量的模型。