Choosing between OLS , RE, FE

Hello
I am analyzing the effect of LPI on export and import trade ( using panel data) when I ran OLS, the result was below

Code:

.  reg lex lgdp dis ll0 landlocked, cluster(country1)
note: landlocked omitted because of collinearity

Linear regression                               Number of obs     =        156
                                                F(3, 19)          =     125.91
                                                Prob > F          =     0.0000
                                                R-squared         =     0.8903
                                                Root MSE          =     .57966

                              (Std. Err. adjusted for 20 clusters in country1)
------------------------------------------------------------------------------
             |               Robust
         lex |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        lgdp |   .8434597    .074269    11.36   0.000      .688013    .9989065
         dis |  -.7162724   .0859011    -8.34   0.000    -.8960654   -.5364794
         ll0 |   1.771018    .588909     3.01   0.007     .5384175    3.003619
  landlocked |          0  (omitted)
       _cons |   3.902149   1.193095     3.27   0.004     1.404974    6.399325
------------------------------------------------------------------------------

The result shows that LPI has effect on export with statistically significant at 1% level
But when I ran RE and Fe the result was very different with OLS' result . All in RE and Fe, the result showed that LPI don't have any effect on export. More over , the result of F-test in fe and the p value Hausman test showed that Fe was best choice

Code:

. xtreg lex lgdp dis ll0 landlocked,fe
note: dis omitted because of collinearity
note: landlocked omitted because of collinearity

Fixed-effects (within) regression               Number of obs     =        156
Group variable: country1                        Number of groups  =         20

R-sq:                                           Obs per group:
     within  = 0.2828                                         min =          4
     between = 0.7417                                         avg =        7.8
     overall = 0.6698                                         max =          8

                                                F(2,134)          =      26.41
corr(u_i, Xb)  = -0.9032                        Prob > F          =     0.0000

------------------------------------------------------------------------------
         lex |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        lgdp |   2.068337   .2848885     7.26   0.000     1.504877    2.631797
         dis |          0  (omitted)
         ll0 |   -.861858   1.238365    -0.70   0.488    -3.311128    1.587412
  landlocked |          0  (omitted)
       _cons |  -9.631277   2.850162    -3.38   0.001     -15.2684   -3.994153
-------------+----------------------------------------------------------------
     sigma_u |  2.2327644
     sigma_e |   .4243563
         rho |  .96513702   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(19, 134) = 27.17                    Prob > F = 0.0000

Code:

. . xtreg lex lgdp dis ll0 landlocked,re
note: landlocked omitted because of collinearity

Random-effects GLS regression                   Number of obs     =        156
Group variable: country1                        Number of groups  =         20

R-sq:                                           Obs per group:
     within  = 0.2616                                         min =          4
     between = 0.9389                                         avg =        7.8
     overall = 0.8841                                         max =          8

                                                Wald chi2(3)      =     278.22
corr(u_i, X)   = 0 (assumed)                    Prob > chi2       =     0.0000

------------------------------------------------------------------------------
         lex |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        lgdp |   .9661854   .0729139    13.25   0.000     .8232767    1.109094
         dis |  -.7608128   .1180357    -6.45   0.000    -.9921585    -.529467
         ll0 |   .7772783   .7426183     1.05   0.295    -.6782268    2.232783
  landlocked |          0  (omitted)
       _cons |   4.390683   1.316084     3.34   0.001     1.811206     6.97016
-------------+----------------------------------------------------------------
     sigma_u |  .43224225
     sigma_e |   .4243563
         rho |  .50920534   (fraction of variance due to u_i)
------------------------------------------------------------------------------

Code:

. hausman fe re

                 ---- Coefficients ----
             |      (b)          (B)            (b-B)     sqrt(diag(V_b-V_B))
             |       fe           re         Difference          S.E.
-------------+----------------------------------------------------------------
        lgdp |    2.068337     .9661854        1.102151        .2753998
         ll0 |    -.861858     .7772783       -1.639136        .9909921
------------------------------------------------------------------------------
                           b = consistent under Ho and Ha; obtained from xtreg
            B = inconsistent under Ha, efficient under Ho; obtained from xtreg

    Test:  Ho:  difference in coefficients not systematic

                  chi2(2) = (b-B)'[(V_b-V_B)^(-1)](b-B)
                          =       19.30
                Prob>chi2 =      0.0001

What method should I use? Can you give me some advice?
Thanks so much