Hausman test: different results for different components estimator

[Kalle75]

I am testing a number of random effects equations for possible correlation between the random effects and the independant varables. I am using the Hausman test in Eviews. My problem is that I get widely different results depending upon which estimator for components variance I have used. Swamy-Arora signals the existence of correlation for almost all the equations. Wansbeek-Kapteyn does not confirm correlation for any of the equations. And Wallace-Hussein reports that the test result is invalid for quite a few of the equations and gives a clear signal for the other equations: no statistically confirmed correlation.
So my question is: is there any arguments for putting more trust in any of these results than in the other two? Or should I choose the FE model just to avoid any risk?

Don't bother! I have found the ANSWER.
As no one has had the interest to try an answer to my question I have to answer it myself ! With some help from Google I managed to localize an article that presented an answer to precisely my question: “Relative Performance of Components Variance Estimators in Random Effects Models” by Teimour Mohammadi, Iranian Journal of Economic Research Vol. 17, No. 50, Spring 2012, pp. 83-98 (www.sid.ir/en/VEWSSID/J_pdf/93820125004.pdf ). In large-sample context the Swamy-Arora estimator is often the default method. In this study though the small-sample performance of the three estimators, Swamy-Arora, Wallace-Hussein and Wansbeek-Kapteyn is compared in different respects, among them how reliably they answer to Hausmans test (small sample = few groups). And the results are quite different from the large-sample results According to the study, in small-sample contexts the Swamy-Arora estimator is clearly inferior to the other two estimators. So obviously, for small samples Wallace-Husseins or Wansbeek-Kapteyn method is to prefer.