Hausman test for panel data

[dabobson]

Hi,
I am using panel data, and trying to decide whether to use random or fixed effects. There is one variable in particular that I suspect is correlated with the intercepts (violating the assumptions for random effects estimation) and this seems to be confirmed by the fact that the coefficient for this variable changes considerably when moving from the fixed effects to random effects estimation. All of the other coefficients remain around the same. When I run a Hausman test the null hypothesis is not rejected, i.e. no difference is found between the two coefficient matrices, implying that i can safely use random effects. Should I ignore my preconceptions and visual inspection of the data on the basis that the hausman test has found the difference to be insignificant or is there a way that I can legitimately focus the test on the "problem" variable? I am concerned that the relatively large number of degrees of freedom (27), with most of the coefficients having almost no difference, is meaning that the null hypothesis is not rejected despite a relatively large difference in one of the variables.
Any comments would be most welcome.
Thank you.