Problem with hausman test

[Lenny Tran]

Im using stata 12
and I have a result like this

. hausman fe re

                 ---- Coefficients ----
             |      (b)          (B)            (b-B)     sqrt(diag(V_b-V_B))
             |       fe           re         Difference          S.E.
-------------+----------------------------------------------------------------
        a |   -.0129722    -.0151672         .002195               .
        b|   -.1919074    -.0865705       -.1053369        .0203032
         c |     .026702      19.1174        -19.0907        3.851953
        d |   -.3443036    -.4121315         .067828               .

why the variable a and d dont have the sqrt ?

[Carlo Lazzaro]

Lenny:
your (V_b-V_B) diagonal has negative values.

[Lenny Tran]

Lenny:
your (V_b-V_B) diagonal has negative values.

So , does the test still correct ?

[Carlo Lazzaro]

Lenny:
no, it's dubious at best.
You can give it another try with -sigmamore- or -sigmaless- options (see -hausman-menu).

[Lenny Tran]

Lenny:
no, it's dubious at best.
You can give it another try with -sigmamore- or -sigmaless- options (see -hausman-menu).

thanks, I've already tried it.

[Carlo Lazzaro]

Lenny:
https://www.stata.com/bookstore/micr...metrics-stata/, page 267-268, describes a robust -hausman- test (that is similar to the user-written command -xtoverid-, that you may want to -search-, install and try). The same textbook covers a bootstrap -hausman- test (see pages 443-444).