Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Hausman test (V_b-V_B is not positive definite)

    When checking to go for fixed or random effect model the hausman test returns: (V_b-V_B is not positive definite)

    Can someone explain what is going on?

    This is my input:

    . xtreg symAf1 scandal ov_cc ln_ta ln_rev max_severity scandalcheck gender_female ceo_left i.year, fe

    Fixed-effects (within) regression Number of obs = 3,466
    Group variable: isinN Number of groups = 348

    R-squared: Obs per group:
    Within = 0.0807 min = 9
    Between = 0.0105 avg = 10.0
    Overall = 0.0645 max = 10

    F(17,3101) = 16.02
    corr(u_i, Xb) = -0.0939 Prob > F = 0.0000

    -------------------------------------------------------------------------------
    symAf1 | Coefficient Std. err. t P>|t| [95% conf. interval]
    --------------+----------------------------------------------------------------
    scandal | 1.549997 .5297553 2.93 0.003 .5112905 2.588704
    ov_cc | -2.00458 .7226264 -2.77 0.006 -3.421455 -.5877055
    ln_ta | -.5018562 .705017 -0.71 0.477 -1.884204 .8804912
    ln_rev | -.0598661 .7202941 -0.08 0.934 -1.472168 1.352436
    max_severity | -.6217514 .3120612 -1.99 0.046 -1.233619 -.0098839
    scandalcheck | -.7569737 .5122314 -1.48 0.140 -1.761321 .2473733
    gender_female | .4642782 .9205489 0.50 0.614 -1.340669 2.269225
    ceo_left | -.6074617 .5859492 -1.04 0.300 -1.756349 .5414262
    |
    year |
    2011 | -.8468325 .5626149 -1.51 0.132 -1.949968 .2563029
    2012 | -3.476982 .5655172 -6.15 0.000 -4.585808 -2.368156
    2013 | -5.135926 .5710935 -8.99 0.000 -6.255685 -4.016166
    2014 | -4.233517 .5810325 -7.29 0.000 -5.372764 -3.094269
    2015 | -2.752737 .5803346 -4.74 0.000 -3.890616 -1.614857
    2016 | -1.941065 .5854607 -3.32 0.001 -3.088995 -.7931351
    2017 | -1.374717 .5922833 -2.32 0.020 -2.536025 -.2134099
    2018 | .7371546 .5988942 1.23 0.218 -.4371147 1.911424
    2019 | .9679316 .6284136 1.54 0.124 -.2642174 2.20008
    |
    _cons | 18.10061 12.98603 1.39 0.163 -7.36148 43.56269
    --------------+----------------------------------------------------------------
    sigma_u | 3.07719
    sigma_e | 7.3779773
    rho | .14817774 (fraction of variance due to u_i)
    -------------------------------------------------------------------------------
    F test that all u_i=0: F(347, 3101) = 1.62 Prob > F = 0.0000

    . estimates store fe

    . xtreg symAf1 scandal ov_cc ln_ta ln_rev max_severity scandalcheck gender_female ceo_left i.year, re

    Random-effects GLS regression Number of obs = 3,466
    Group variable: isinN Number of groups = 348

    R-squared: Obs per group:
    Within = 0.0790 min = 9
    Between = 0.0378 avg = 10.0
    Overall = 0.0729 max = 10

    Wald chi2(17) = 279.74
    corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000

    -------------------------------------------------------------------------------
    symAf1 | Coefficient Std. err. z P>|z| [95% conf. interval]
    --------------+----------------------------------------------------------------
    scandal | 1.426404 .5052439 2.82 0.005 .4361439 2.416663
    ov_cc | -.3704852 .4421185 -0.84 0.402 -1.237022 .4960512
    ln_ta | -.0782882 .1594136 -0.49 0.623 -.3907331 .2341567
    ln_rev | -.3988242 .1825832 -2.18 0.029 -.7566807 -.0409678
    max_severity | -.5711134 .2877424 -1.98 0.047 -1.135078 -.0071488
    scandalcheck | -.6836392 .4743194 -1.44 0.149 -1.613288 .2460097
    gender_female | .0034313 .654536 0.01 0.996 -1.279436 1.286298
    ceo_left | -.7298028 .5687446 -1.28 0.199 -1.844522 .3849162
    |
    year |
    2011 | -.8456373 .5604301 -1.51 0.131 -1.94406 .2527855
    2012 | -3.479558 .5602224 -6.21 0.000 -4.577574 -2.381542
    2013 | -5.143497 .5613902 -9.16 0.000 -6.243802 -4.043193
    2014 | -4.23394 .5648867 -7.50 0.000 -5.341097 -3.126782
    2015 | -2.763149 .5633926 -4.90 0.000 -3.867379 -1.65892
    2016 | -1.968737 .5632368 -3.50 0.000 -3.072661 -.8648132
    2017 | -1.378294 .5633773 -2.45 0.014 -2.482493 -.2740948
    2018 | .7463297 .5632567 1.33 0.185 -.3576331 1.850293
    2019 | .9926731 .5823127 1.70 0.088 -.1486389 2.133985
    |
    _cons | 15.77485 3.370074 4.68 0.000 9.169627 22.38007
    --------------+----------------------------------------------------------------
    sigma_u | 1.8488825
    sigma_e | 7.3779773
    rho | .0590872 (fraction of variance due to u_i)
    -------------------------------------------------------------------------------

    . estimates store re

    . hausman fe re

    ---- Coefficients ----
    | (b) (B) (b-B) sqrt(diag(V_b-V_B))
    | fe re Difference Std. err.
    -------------+----------------------------------------------------------------
    scandal | 1.549997 1.426404 .1235934 .1592774
    ov_cc | -2.00458 -.3704852 -1.634095 .5715944
    ln_ta | -.5018562 -.0782882 -.423568 .6867578
    ln_rev | -.0598661 -.3988242 .3389581 .6967689
    max_severity | -.6217514 -.5711134 -.050638 .1207748
    scandalcheck | -.7569737 -.6836392 -.0733346 .1933962
    gender_fem~e | .4642782 .0034313 .4608469 .6472967
    ceo_left | -.6074617 -.7298028 .1223411 .140947
    year |
    2011 | -.8468325 -.8456373 -.0011952 .0495338
    2012 | -3.476982 -3.479558 .0025756 .0772047
    2013 | -5.135926 -5.143497 .0075716 .1048275
    2014 | -4.233517 -4.23394 .000423 .1360212
    2015 | -2.752737 -2.763149 .0104129 .1392015
    2016 | -1.941065 -1.968737 .0276719 .1597766
    2017 | -1.374717 -1.378294 .0035768 .1827719
    2018 | .7371546 .7463297 -.0091751 .2035095
    2019 | .9679316 .9926731 -.0247415 .2362531
    ------------------------------------------------------------------------------
    b = Consistent under H0 and Ha; obtained from xtreg.
    B = Inconsistent under Ha, efficient under H0; obtained from xtreg.

    Test of H0: Difference in coefficients not systematic

    chi2(17) = (b-B)'[(V_b-V_B)^(-1)](b-B)
    = 9.69
    Prob > chi2 = 0.9160
    (V_b-V_B is not positive definite)

    . hausman fe re, sigmamore

    Note: the rank of the differenced variance matrix (10) does not equal the number of coefficients being tested
    (17); be sure this is what you expect, or there may be problems computing the test. Examine the output
    of your estimators for anything unexpected and possibly consider scaling your variables so that the
    coefficients are on a similar scale.

    ---- Coefficients ----
    | (b) (B) (b-B) sqrt(diag(V_b-V_B))
    | fe re Difference Std. err.
    -------------+----------------------------------------------------------------
    scandal | 1.549997 1.426404 .1235934 .1596002
    ov_cc | -2.00458 -.3704852 -1.634095 .5717619
    ln_ta | -.5018562 -.0782882 -.423568 .6868905
    ln_rev | -.0598661 -.3988242 .3389581 .6969055
    max_severity | -.6217514 -.5711134 -.050638 .1209226
    scandalcheck | -.7569737 -.6836392 -.0733346 .1936449
    gender_fem~e | .4642782 .0034313 .4608469 .6475368
    ceo_left | -.6074617 -.7298028 .1223411 .141393
    year |
    2011 | -.8468325 -.8456373 -.0011952 .0506922
    2012 | -3.476982 -3.479558 .0025756 .0779607
    2013 | -5.135926 -5.143497 .0075716 .1053966
    2014 | -4.233517 -4.23394 .000423 .1364756
    2015 | -2.752737 -2.763149 .0104129 .1396445
    2016 | -1.941065 -1.968737 .0276719 .1601696
    2017 | -1.374717 -1.378294 .0035768 .1831236
    2018 | .7371546 .7463297 -.0091751 .2038325
    2019 | .9679316 .9926731 -.0247415 .2365594
    ------------------------------------------------------------------------------
    b = Consistent under H0 and Ha; obtained from xtreg.
    B = Inconsistent under Ha, efficient under H0; obtained from xtreg.

    Test of H0: Difference in coefficients not systematic

    chi2(10) = (b-B)'[(V_b-V_B)^(-1)](b-B)
    = 11.18
    Prob > chi2 = 0.3434
    (V_b-V_B is not positive definite)

    . hausman fe re, sigmaless

    Note: the rank of the differenced variance matrix (10) does not equal the number of coefficients being tested
    (17); be sure this is what you expect, or there may be problems computing the test. Examine the output
    of your estimators for anything unexpected and possibly consider scaling your variables so that the
    coefficients are on a similar scale.

    ---- Coefficients ----
    | (b) (B) (b-B) sqrt(diag(V_b-V_B))
    | fe re Difference Std. err.
    -------------+----------------------------------------------------------------
    scandal | 1.549997 1.426404 .1235934 .1595709
    ov_cc | -2.00458 -.3704852 -1.634095 .5716571
    ln_ta | -.5018562 -.0782882 -.423568 .6867646
    ln_rev | -.0598661 -.3988242 .3389581 .6967777
    max_severity | -.6217514 -.5711134 -.050638 .1209004
    scandalcheck | -.7569737 -.6836392 -.0733346 .1936094
    gender_fem~e | .4642782 .0034313 .4608469 .647418
    ceo_left | -.6074617 -.7298028 .1223411 .1413671
    year |
    2011 | -.8468325 -.8456373 -.0011952 .0506829
    2012 | -3.476982 -3.479558 .0025756 .0779464
    2013 | -5.135926 -5.143497 .0075716 .1053773
    2014 | -4.233517 -4.23394 .000423 .1364506
    2015 | -2.752737 -2.763149 .0104129 .1396189
    2016 | -1.941065 -1.968737 .0276719 .1601402
    2017 | -1.374717 -1.378294 .0035768 .18309
    2018 | .7371546 .7463297 -.0091751 .2037951
    2019 | .9679316 .9926731 -.0247415 .236516
    ------------------------------------------------------------------------------
    b = Consistent under H0 and Ha; obtained from xtreg.
    B = Inconsistent under Ha, efficient under H0; obtained from xtreg.

    Test of H0: Difference in coefficients not systematic

    chi2(10) = (b-B)'[(V_b-V_B)^(-1)](b-B)
    = 11.19
    Prob > chi2 = 0.3431
    (V_b-V_B is not positive definite)

  • #2
    Frank:
    welcome to this forum.
    Unfortunately, what you experienced happens quite frequently.
    You may want to consider the community-contributed modules -xtoverid- and -mundlak- to tes which specification is the way to go.
    Kind regards,
    Carlo
    (Stata 19.0)

    Comment


    • #3
      Thanks for your reaction Carlo, I will look in to it.

      Comment

      Working...
      X