[讨论][求助]还是关于pool data与panel data的差异问题

7395

收藏 2007-05-04

最近做毕业论文，用到了面板数据的分析。平时读书太少，临时抱佛脚；在坛子里面看到了各位学友以前的讨论，但是还是不甚明晰，所以再弱弱的问一次，恳请高人作答。
1.本文的论文是探讨3年30个省市自治区，多种因素（5种以上）对经济发展的影响。在高铁梅的书里看到了根据数据的“长宽”结构决定用Pool还是Panel。在这里我是否是该用panel？
2.看到前面的讨论中，说两者的计量分析结果是一致的，我做了对比。
 在分析选项为none——none——no weight时，结果差异不大。(前为panel，后为pool)
 Dependent Variable: INCOME Method: Panel Least Squares Date: 05/04/07 Time: 15:54 Sample: 2003 2005 Cross-sections included: 30 Total panel (balanced) observations: 90 Variable Coefficient Std. Error t-Statistic Prob. E1 124.7870 6.325950 19.72621 0.0000 E2 9.266186 6.812881 1.360098 0.1774 E3 -23.34554 10.84067 -2.153515 0.0341 E4 18.44371 4.128810 4.467077 0.0000 R-squared 0.813462 Mean dependent var 2949.047 Adjusted R-squared 0.806955 S.D. dependent var 1424.912 S.E. of regression 626.0624 Akaike info criterion 15.76020 Sum squared resid 33708052 Schwarz criterion 15.87131 Log likelihood -705.2092 F-statistic 125.0106 Durbin-Watson stat 0.486574 Prob(F-statistic) 0.000000 
Dependent Variable: INCOME Method: Pooled Least Squares Date: 05/04/07 Time: 15:54 Sample: 2003 2005 Included observations: 90 Cross-sections included: 30 Total pool (balanced) observations: 2700 Variable Coefficient Std. Error t-Statistic Prob. E1 124.7870 1.129835 110.4471 0.0000 E2 9.266186 1.216802 7.615193 0.0000 E3 -23.34554 1.936178 -12.05754 0.0000 E4 18.44371 0.737419 25.01118 0.0000 R-squared 0.813462 Mean dependent var 2949.047 Adjusted R-squared 0.813254 S.D. dependent var 1417.236 S.E. of regression 612.4456 Akaike info criterion 15.67428 Sum squared resid 1.01E+09 Schwarz criterion 15.68302 Log likelihood -21156.27 F-statistic 3918.937 Durbin-Watson stat 1.398221 Prob(F-statistic) 0.000000 分析选项改动后，差异就相当大。fixed——none——no weight
Dependent Variable: INCOME Method: Panel Least Squares Date: 05/04/07 Time: 16:01 Sample: 2003 2005 Cross-sections included: 30 Total panel (balanced) observations: 90 Variable Coefficient Std. Error t-Statistic Prob. C 2797.634 271.3935 10.30841 0.0000 E1 14.97280 11.20799 1.335904 0.1870 E2 0.351364 2.790877 0.125897 0.9003 E3 -30.70877 4.438541 -6.918662 0.0000 E4 6.485118 2.345146 2.765337 0.0077 Effects Specification Cross-section fixed (dummy variables) R-squared 0.987260 Mean dependent var 2949.047 Adjusted R-squared 0.979753 S.D. dependent var 1424.912 S.E. of regression 202.7546 Akaike info criterion 13.74297 Sum squared resid 2302129. Schwarz criterion 14.68734 Log likelihood -584.4336 F-statistic 131.5050 Durbin-Watson stat 2.308158 Prob(F-statistic) 0.000000 
Dependent Variable: INCOME Method: Pooled Least Squares Date: 05/04/07 Time: 16:01 Sample: 2003 2005 Included observations: 90 Cross-sections included: 30 Total pool (balanced) observations: 2700 Variable Coefficient Std. Error t-Statistic Prob. C 50.59930 29.89669 1.692471 0.0907 E1 123.2763 1.444399 85.34780 0.0000 E2 8.923536 1.239617 7.198623 0.0000 E3 -23.19858 1.947932 -11.90934 0.0000 E4 18.06670 0.773909 23.34472 0.0000 Fixed Effects (Cross) (随带询问：为何我做出的以下结果都是一样的，而书上出来的都不一样) _1--C 3.81E-12 _2--C 3.81E-12 _3--C 3.81E-12 _4--C 3.81E-12 _5--C 3.81E-12 _6--C 3.81E-12 _7--C 3.81E-12 _8--C 3.81E-12 _9--C 3.81E-12 _10--C 3.81E-12 _11--C 3.81E-12 _12--C 3.81E-12 _13--C 3.81E-12 _14--C 3.81E-12 _15--C 3.81E-12 _16--C 3.81E-12 _17--C 3.81E-12 _18--C 3.81E-12 _19--C 3.81E-12 _20--C 3.81E-12 _21--C 3.81E-12 _22--C 3.81E-12 _23--C 3.81E-12 _24--C 3.81E-12 _25--C 3.81E-12 _26--C 3.81E-12 _27--C 3.81E-12 _28--C 3.81E-12 _29--C 3.81E-12 _30--C 3.81E-12 
 Effects Specification Cross-section fixed (dummy variables) R-squared 0.813662 Mean dependent var 2949.047 Adjusted R-squared 0.811356 S.D. dependent var 1417.236 S.E. of regression 615.5512 Akaike info criterion 15.69543 Sum squared resid 1.01E+09 Schwarz criterion 15.76973 Log likelihood -21154.83 F-statistic 352.7679 Durbin-Watson stat 1.397094 Prob(F-statistic) 0.000000 
 分析条件：fixed——none——cross section weights,差异很大
Dependent Variable: INCOME Method: Panel EGLS (Cross-section weights) Date: 05/04/07 Time: 16:06 Sample: 2003 2005 Cross-sections included: 30 Total panel (balanced) observations: 90 Linear estimation after one-step weighting matrix Variable Coefficient Std. Error t-Statistic Prob. C 3402.656 131.6626 25.84375 0.0000 E1 -10.84125 5.874811 -1.845378 0.0703 E2 -2.793492 1.923607 -1.452216 0.1520 E3 -18.58849 2.702947 -6.877121 0.0000 E4 -0.418080 1.412075 -0.296075 0.7683 Effects Specification Cross-section fixed (dummy variables) Weighted Statistics R-squared 0.998406 Mean dependent var 5028.896 Adjusted R-squared 0.997467 S.D. dependent var 3294.579 S.E. of regression 165.8197 Sum squared resid 1539786. F-statistic 1062.945 Durbin-Watson stat 2.262978 Prob(F-statistic) 0.000000 Unweighted Statistics R-squared 0.984702 Mean dependent var 2949.047 Sum squared resid 2764482. Durbin-Watson stat 1.984506 
Dependent Variable: INCOME Method: Pooled EGLS (Cross-section weights) Date: 05/04/07 Time: 16:06 Sample: 2003 2005 Included observations: 90 Cross-sections included: 30 Total pool (balanced) observations: 2700 Linear estimation after one-step weighting matrix Variable Coefficient Std. Error t-Statistic Prob. C 50.59930 29.89669 1.692471 0.0907 E1 123.2763 1.444399 85.34780 0.0000 E2 8.923536 1.239617 7.198623 0.0000 E3 -23.19858 1.947932 -11.90934 0.0000 E4 18.06670 0.773909 23.34472 0.0000 Fixed Effects (Cross) _1--C 1.96E-12 _2--C 1.96E-12 _3--C 1.96E-12 _4--C 1.96E-12 _5--C 1.96E-12 _6--C 1.96E-12 _7--C 1.96E-12 _8--C 1.96E-12 _9--C 1.96E-12 _10--C 1.96E-12 _11--C 1.96E-12 _12--C 1.96E-12 _13--C 1.96E-12 _14--C 1.96E-12 15--C 1.96E-12 _16--C 1.96E-12 _17--C 1.96E-12 _18--C 1.96E-12 _19--C 1.96E-12 20--C 1.96E-12 _21--C 1.96E-12 _22--C 1.96E-12 _23--C 1.96E-12 _24--C 1.96E-12 _25--C 1.96E-12 _26--C 1.96E-12 _27--C 1.96E-12 _28--C 1.96E-12 _29--C 1.96E-12 _30--C 1.96E-12 Effects Specification Cross-section fixed (dummy variables) Weighted Statistics R-squared 0.813662 Mean dependent var 2949.047 Adjusted R-squared 0.811356 S.D. dependent var 1417.236 S.E. of regression 615.5512 Sum squared resid 1.01E+09 F-statistic 352.7679 Durbin-Watson stat 1.397094 Prob(F-statistic) 0.000000 Unweighted Statistics R-squared 0.813662 Mean dependent var 2949.047 Sum squared resid 1.01E+09 Durbin-Watson stat 1.397094 

3. 在部分分析中，我希望进行某一年的横截面数据分析。不知道我的做法正确与否。我分别在panel和pool中分别通过改变时间区域的办法进行了横截面分析。结果也差异很大。
分析条件：none——none——no weights,除特征值外差异不大。（前为panel，后为pool）
Dependent Variable: INCOME Method: Panel Least Squares Date: 05/04/07 Time: 16:13 Sample: 2003 2003 Cross-sections included: 30 Total panel (balanced) observations: 30 Variable Coefficient Std. Error t-Statistic Prob. E1 48.15397 18.23247 2.641110 0.0138 E2 -2.293155 21.92957 -0.104569 0.9175 E3 102.3355 28.70613 3.564937 0.0014 E4 77.26180 31.29374 2.468922 0.0204 R-squared 0.847892 Mean dependent var 2742.348 Adjusted R-squared 0.830341 S.D. dependent var 1272.425 S.E. of regression 524.1083 Akaike info criterion 15.48484 Sum squared resid 7141927. Schwarz criterion 15.67167 Log likelihood -228.2726 F-statistic 48.31023 Durbin-Watson stat 0.000000 Prob(F-statistic) 0.000000 （随带问一声，横截面数据是否不用考虑DW检验） Dependent Variable: INCOME Method: Pooled Least Squares Date: 05/04/07 Time: 16:13 Sample: 2003 2003 Included observations: 30 Cross-sections included: 30 Total pool (balanced) observations: 900 Variable Coefficient Std. Error t-Statistic Prob. E1 48.15397 3.105834 15.50436 0.0000 E2 -2.293155 3.735622 -0.613862 0.5395 E3 102.3355 4.889982 20.92759 0.0000 E4 77.26180 5.330773 14.49355 0.0000 R-squared 0.847892 Mean dependent var 2742.348 Adjusted R-squared 0.847382 S.D. dependent var 1251.733 S.E. of regression 489.0061 Akaike info criterion 15.22706 Sum squared resid 2.14E+08 Schwarz criterion 15.24841 Log likelihood -6848.178 F-statistic 1664.845 Durbin-Watson stat 0.000000 Prob(F-statistic) 0.000000 
分析条件：fixed——none——no weights时，panel分析提示“near singular matrix”，pool则可顺利做出
以下是pool结果：
Dependent Variable: INCOME Method: Pooled Least Squares Date: 05/04/07 Time: 16:17 Sample: 2003 2003 Included observations: 30 Cross-sections included: 30 Total pool (balanced) observations: 900 Variable Coefficient Std. Error t-Statistic Prob. C -1094.718 49.69840 -22.02724 0.0000 E1 45.93406 2.531145 18.14754 0.0000 E2 38.15520 3.553257 10.73809 0.0000 E3 127.5180 4.142868 30.78013 0.0000 E4 93.39735 4.402320 21.21549 0.0000 Fixed Effects (Cross) _1--C -5.70E-12 _2--C -5.70E-12 _3--C -5.70E-12 _4--C -5.70E-12 _5--C -5.70E-12 _6--C -5.70E-12 _7--C -5.70E-12 _8--C -5.70E-12 _9--C -5.70E-12 _10--C -5.70E-12 _11--C -5.70E-12 _12--C -5.70E-12 _13--C -5.70E-12 _14--C -5.70E-12 _15--C -5.70E-12 _16--C -5.70E-12 _17--C -5.70E-12 _18--C -5.70E-12 _19--C -5.70E-12 _20--C -5.70E-12 _21--C -5.70E-12 _22--C -5.70E-12 _23--C -5.70E-12 _24--C -5.70E-12 _25--C -5.70E-12 _26--C -5.70E-12 _27--C -5.70E-12 _28--C -5.70E-12 _29--C -5.70E-12 _30--C -5.70E-12 Effects Specification Cross-section fixed (dummy variables) R-squared 0.902512 Mean dependent var 2742.348 Adjusted R-squared 0.898797 S.D. dependent var 1251.733 S.E. of regression 398.2068 Akaike info criterion 14.84887 Sum squared resid 1.37E+08 Schwarz criterion 15.03029 Log likelihood -6647.990 F-statistic 242.9433 Durbin-Watson stat 0.000000 Prob(F-statistic) 0.000000 
分析条件：none——none——cross section weights, 差异很大
Dependent Variable: INCOME Method: Panel EGLS (Cross-section weights) Date: 05/04/07 Time: 16:20 Sample: 2003 2003 Cross-sections included: 30 Total panel (balanced) observations: 30 Linear estimation after one-step weighting matrix Variable Coefficient Std. Error t-Statistic Prob. E1 43.11071 5.190409 8.305841 0.0000 E2 -7.385865 3.803325 -1.941950 0.0631 E3 110.2336 6.742339 16.34945 0.0000 E4 98.89031 14.35453 6.889137 0.0000 Weighted Statistics R-squared 0.997489 Mean dependent var 7643.457 Adjusted R-squared 0.997199 S.D. dependent var 9186.062 S.E. of regression 486.1690 Sum squared resid 6145367. F-statistic 3442.461 Durbin-Watson stat 0.000000 Prob(F-statistic) 0.000000 Unweighted Statistics R-squared 0.845013 Mean dependent var 2742.348 Sum squared resid 7277102. Durbin-Watson stat 0.000000 
Dependent Variable: INCOME Method: Pooled EGLS (Cross-section weights) Date: 05/04/07 Time: 16:20 Sample: 2003 2003 Included observations: 30 Cross-sections included: 30 Total pool (balanced) observations: 900 Linear estimation after one-step weighting matrix Variable Coefficient Std. Error t-Statistic Prob. E1 48.15397 3.105834 15.50436 0.0000 E2 -2.293155 3.735622 -0.613862 0.5395 E3 102.3355 4.889982 20.92759 0.0000 E4 77.26180 5.330773 14.49355 0.0000 Weighted Statistics R-squared 0.847892 Mean dependent var 2742.348 Adjusted R-squared 0.847382 S.D. dependent var 1251.733 S.E. of regression 489.0061 Sum squared resid 2.14E+08 F-statistic 1664.845 Durbin-Watson stat 0.000000 Prob(F-statistic) 0.000000 Unweighted Statistics R-squared 0.847892 Mean dependent var 2742.348 Sum squared resid 2.14E+08 Durbin-Watson stat 0.000000 

各位达人：
以上就是我论文写作的部分问题，恳请各位达人给于帮助，老板的刀已经架上脖子了……
1. panel和Pool的差异何在，本文的数据结构应该用哪一种？（3年30个省，多种因素（5种以上）对经济发展的影响）
2. 为何在添加分析条件和两者结果差距很大，不添加条件显然通不过检验。
3. 做横截面分析，该用哪一种，或者说两种方法都不对，该用什么方法
4. 随带提出的两个问题（在上文中已注明）
 Fixed Effects (Cross) (随带询问：为何我做出的以下结果都是一样的，而书上出来的都不一样) Durbin-Watson stat 0.000000 （随带问一声，横截面数据是否不用考虑DW检验） 
本人平时荒废了学业了，以上问题可能对各位达人来说都不是问题，求助求助……