1 数据
R可以方便的对面板数据进行分析。我们以EconomicData.sav的数据为例进行演示，基于中国30个省市2011年至2018年的面板数据估计外商直接投资（pfdi）、对外贸易（open）、劳动人口受教育年限（eduyear）对人均GDP的影响。
2 安装面板数据模型plm包
> install.packages("plm")
> library(plm)
3 导入数据
> library(haven)
> panelData=read_dta("D:/Desktop/EconomicData.dta")
> View(panelData)
4 设置面板数据格式
> PData = pdata.frame(panelData,index = c("prov","year"))
5 混合面板估计
> pooling = plm(pgdp ~ eduyear+pfdi+open, data = PData,model = "pooling")
> summary(pooling)
Pooling Model
Call:
plm(formula = pgdp ~ eduyear + pfdi + open, data = PData, model = "pooling")
Balanced Panel: n = 30, T = 8, N = 240
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-2.99231 -0.77270 -0.27814 0.58971 4.53531
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
(Intercept) -6.122498 1.126160 -5.4366 1.353e-07 ***
eduyear 1.114948 0.129990 8.5772 1.305e-15 ***
pfdi 4.131178 0.863377 4.7849 3.017e-06 ***
open 0.350009 0.042756 8.1861 1.689e-14 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Total Sum of Squares: 1482
Residual Sum of Squares: 340.19
R-Squared: 0.77045
Adj. R-Squared: 0.76754
F-statistic: 264.038 on 3 and 236 DF, p-value: < 2.22e-16
6 固定效应估计
> fixed = plm(pgdp ~ eduyear+pfdi+open, data = PData,model = "within")
> summary(fixed)
Oneway (individual) effect Within Model
Call:
plm(formula = pgdp ~ eduyear + pfdi + open, data = PData, model = "within")
Balanced Panel: n = 30, T = 8, N = 240
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-1.80547 -0.41447 -0.01872 0.34209 2.87503
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
eduyear 2.73086 0.22207 12.2974 < 2.2e-16 ***
pfdi 4.57606 0.78647 5.8185 2.233e-08 ***
open 1.17815 0.20618 5.7143 3.802e-08 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Total Sum of Squares: 250.78
Residual Sum of Squares: 107.76
R-Squared: 0.57032
Adj. R-Squared: 0.50389
F-statistic: 91.5838 on 3 and 207 DF, p-value: < 2.22e-16
7 估计随机效应
> random = plm(pgdp ~ eduyear+pfdi+open, data = PData,model = "random")
> summary(random)
Oneway (individual) effect Random Effect Model
(Swamy-Arora's transformation)
Call:
plm(formula = pgdp ~ eduyear + pfdi + open, data = PData, model = "random")
Balanced Panel: n = 30, T = 8, N = 240
Effects:
var std.dev share
idiosyncratic 0.5206 0.7215 0.392
individual 0.8059 0.8977 0.608
theta: 0.7267
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-2.014388 -0.461005 -0.097459 0.337473 3.199938
Coefficients:
Estimate Std. Error z-value Pr(>|z|)
(Intercept) -14.253480 1.684719 -8.4605 < 2.2e-16 ***
eduyear 2.020214 0.190513 10.6041 < 2.2e-16 ***
pfdi 3.424914 0.817656 4.1887 2.806e-05 ***
open 0.295670 0.082685 3.5759 0.0003491 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Total Sum of Squares: 342.76
Residual Sum of Squares: 151.91
R-Squared: 0.5568
Adj. R-Squared: 0.55116
Chisq: 296.487 on 3 DF, p-value: < 2.22e-16
8 豪斯曼检验
> phtest(random, fixed)
Hausman Test
data: pgdp ~ eduyear + pfdi + open
chisq = 69.079, df = 3, p-value = 6.722e-15
alternative hypothesis: one model is inconsistent
说明：p-value = 6.722e-15 < 0.05，因而拒绝原假设（随机效应），选择固定效应模型！