相关系数是没有控制其它因素下简单相关系数，而回归系数是控制了其它因素下的偏相关系数，简单相关系数具有欺骗性，偏相关系数同回归系数的符号一般一致。

你用的面板数据模型是什么？

Fixed Effect？

如果是，

符号不同是正常的，

因为在FE模型中，你剔除了固定效果；

而OLS估计就是有偏的。

恭喜你，你也许是得到了有趣的发现！

谢谢您。

关键是我们是如何取舍结果呢？如果符号不同，其经济含义就不同的。

OLS：

reg lngdp lnziben lnrenli jiaoyi

Source | SS df MS Number of obs = 206
-------------+------------------------------ F( 3, 202) = 230.92
Model | 22.405809 3 7.46860301 Prob > F = 0.0000
Residual | 6.53313413 202 .032342248 R-squared = 0.7742
-------------+------------------------------ Adj R-squared = 0.7709
Total | 28.9389431 205 .141165576 Root MSE = .17984

------------------------------------------------------------------------------
lngdp | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lnziben | .6305782 .0310074 20.34 0.000 .5694386 .6917178
lnrenli | .4865001 .1325911 3.67 0.000 .2250599 .7479403
jiaoyi | .2054489 .0636243 3.23 0.001 .0799959 .330902
_cons | 1.56149 .323377 4.83 0.000 .9238624 2.199117
------------------------------------------------------------------------------

PANEL DATA：

xtreg lngdp lnziben lnrenli jiaoyi,fe

Fixed-effects (within) regression Number of obs = 206
Group variable (i): diqu Number of groups = 19

R-sq: within = 0.8292 Obs per group: min = 8
between = 0.4789 avg = 10.8
overall = 0.6339 max = 11

F(3,184) = 297.76
corr(u_i, Xb) = -0.3768 Prob > F = 0.0000

------------------------------------------------------------------------------
lngdp | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lnziben | .8394926 .0358939 23.39 0.000 .7686761 .9103092
lnrenli | .2477823 .1191919 2.08 0.039 .0126238 .4829409
jiaoyi | -.0774143 .0786784 -0.98 0.326 -.232642 .0778135
_cons | -.099759 .2914722 -0.34 0.733 -.6748164 .4752983
-------------+----------------------------------------------------------------
sigma_u | .21868982
sigma_e | .12023762
rho | .7678782 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(18, 184) = 14.88 Prob > F = 0.0000

解释变量jiaoyi符号变成了负的。而这不符合我的经济涵义啊？xiexie