多重共线性案例分析
多重共线性的侦查
(1) 2 值高且不显著的t 值四个中占了两个,有 R 理由怀疑该模型中存在着较为严重的多重共线 性。
(2)考察回归元之间的相关系数
回归元的两两相关系数
LOG(X1) LOG(X1) LOG(X2) LOG(X3) LOG(X4) 1 0.907175 0.972459 0.979005 LOG(X2) 0.907175 1 0.946751 0.933064 LOG(X3) 0.972459 0.946751 1 0.954277 LOG(X4) 0.979005 0.933064 0.954277 1
Mean dependent var S.D. dependent var Akaike info criterion Schwarz criterion Hannan-Quinn criter. Durbin-Watson stat
3.663887 0.187659 -4.152987 -3.906140 -4.090906 1.826069
R 2 = 0.9661
R 2 = 0.9815
∧
ቤተ መጻሕፍቲ ባይዱ
∧
∧
∧
R 2 = 0.9786
F 值 = 336.1808
p值 = 0.0000
多重共线性的再检验: (1)回归元的相关系数
LOG(X1) LOG(X1) LOG(X2) LOG(X3) 1 0.907175 0.972459 LOG(X2) 0.907175 1 0.946751 LOG(X3) 0.972459 0.946751 1
第十二讲
多重共线性案例分析
多重共线性的侦查与处理
Y = 每人的子鸡消费量,磅
研究美国每人的子鸡消费量 令:
X1 = 每人实际可支配收入,美元
X 2 = 每磅子鸡实际零售价格,美元
X 3 = 每磅猪肉实际零售价格,美元
X 4 = 每磅牛肉实际零售价格,美元
获取数据
1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 27.80000 29.90000 29.80000 30.80000 31.20000 33.30000 35.60000 36.40000 36.70000 38.40000 40.40000 40.30000 41.80000 40.40000 40.70000 40.10000 42.70000 44.10000 46.70000 50.60000 50.10000 51.70000 52.90000 397.5000 413.3000 439.2000 459.7000 492.9000 528.6000 560.3000 624.6000 666.4000 717.8000 768.2000 843.3000 911.6000 931.1000 1021.500 1165.900 1349.600 1449.400 1575.500 1759.100 1994.200 2258.100 2478.700 42.20000 38.10000 40.30000 39.50000 37.30000 38.10000 39.30000 37.80000 38.40000 40.10000 38.60000 39.80000 39.70000 52.10000 48.90000 58.30000 57.90000 56.50000 63.70000 61.60000 58.90000 66.40000 70.40000 50.70000 52.00000 54.00000 55.30000 54.70000 63.70000 69.80000 65.90000 64.50000 70.00000 73.20000 67.80000 79.10000 95.40000 94.20000 123.5000 129.9000 117.6000 130.9000 129.8000 128.0000 141.0000 168.2000 78.30000 79.20000 79.20000 79.20000 77.40000 80.20000 80.40000 83.90000 85.50000 93.70000 106.1000 104.8000 114.0000 124.1000 127.6000 142.9000 143.6000 139.2000 165.5000 203.3000 219.6000 221.6000 232.6000 65.80000 66.90000 67.80000 69.60000 68.70000 73.60000 76.30000 77.20000 78.10000 84.70000 93.30000 89.70000 100.7000 113.5000 115.3000 136.7000 139.2000 132.0000 132.1000 154.4000 174.9000 180.8000 189.4000
② ln( X 2 ) = 1.2332 − 0.4693ln( X 1 ) + 0.6694 ln( X 3 ) + 0.5955ln( X 4 )
R 2 = 0.9338
p值 = 0.0000
③ ln( X 3 ) = −1.0127 + 0.6618ln( X 1 ) + 0.8287 ln( X 2 ) − 0.4695ln( X 4 )
ln Y = 2.1898 + 0.3426ln X1 − 0.5046ln X 2 + 0.1485ln X3 + 0.0911ln X 4 se = ( 0.1557) ( 0.0833) ( 0.1109 ) ( 0.0997 ) ( 0.1007 ) (1.49 ) ( 0.90 ) ( −4.55) t = (14.06 ) ( 4.11)
R 2 = 0.9759 R 2 = 0.9721
F = 256.08
p值 = 0.0000
④ ln( X 4 ) = −0.7057 + 0.6956 ln( X1 ) + 0.7219 ln( X 2 ) − 0.4598ln( X 3 )
R 2 = 0.9763
R 2 = 0.9726
F = 261.61 p值 = 0.0000
回归元之间的相关系数均大于0.8, 回归元之间的相关系数均大于 ,表明多重共线性是 严重的
(3)辅助回归
① ln( X 1 ) = 0.9461 − 0.8324 ln( X 2 ) + 0.9483ln( X 3 ) + 1.0176 ln( X 4 )
R 2 = 0.9846
R 2 = 0.9822 F = 406.0592 p值 = 0.0000 R 2 = 0.9428 F = 104.41
0.0000 0.0000 0.0000 0.2395
R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood F-statistic Prob(F-statistic)
0.981509 0.978590 0.027459 0.014326 52.24812 336.1808 0.000000
∧
∧
∧
∧
∧
p值 = ( 0.0000 ) ( 0.0007 )
( 0.0002 )
( 0.1535)
( 0.3776 )
R 2 = 0.9823
R 2 = 0.9783
F = 249.9282
p值 = 0.0000
通过总体显著性检验 ln ln ln ln t 检验: X1、 X 2 通过显著性检验; X 3、 X 4 未 通过显著性检验 R 2 很大
② ln( X 2 ) = 1.4259 − 0.0967 ln( X 1 ) + 0.6939 ln( X 3 )
R 2 = 0.8997 F = 89.69
R 2 = 0.9692
F = 314.89
R 2 = 0.8897
p值 = 0.0000
③ ln( X 3 ) = −0.8689 + 0.4276 ln( X 1 ) + 0.6246 ln( X 2 )
应用克莱因法则: 辅助模型①的 R 2 大于回归方程的 R 2
0.9822 > 0.9783
多重共线性问题比较严重
多重共线性的处理 (1)不做处理。 作为分析者,如果你主要关注的因素 是可支配收入与子鸡价格对子鸡消费量 的影响的话,两个统计量都通过了显著 性的 t 检验,对多重共线性可以不做处理。
回归元之间的相关系数均大于0.8, 回归元之间的相关系数均大于 ,表明多重共线性是 严重的
(2)辅助回归
① ln( X 1 ) = 0.7800 − 0.3347 ln( X 2 ) + 1.6443ln( X 3 )
R 2 = 0.9474
F = 180.24
R 2 = 0.9422
p值 = 0.0000
Variable
Coefficient
Std. Error
t-Statistic
Prob.
C LOG(X1) LOG(X2) LOG(X3)
2.125498 0.405924 -0.438825 0.106656
0.137882 0.044791 0.083332 0.087838
15.41533 9.062535 -5.265956 1.214228
14.06283 0.0007 0.0002 0.1535 0.3776
0.0000
R-squared 0.982313 Adjusted R-squared 0.978383 S.E. of regression 0.027591 Sum squared resid 0.013703 Log likelihood 52.75935 F-statistic 249.9282 Prob(F-statistic) 0.000000