Announcement

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

  • Interpreting Hausman test results

    I am deciding between a panel regression with fixed effects or random effects. The outcome variable here is labor force participation rate "lfp_rate", and one independent variable is poverty rates "pov_". I have data between 2011-2019, but data completeness varies by both country and year.
    dataex countrycode lfp_rate pov_ year ----------------------- copy starting from the next line ----------------------- [CODE] * Example generated by -dataex-. To install: ssc install dataex clear input str3 countrycode double lfp_rate float pov_ int year "ARG" 59.96 1.4 2013 "ARG" 59.6 1.1 2011 "ARG" 58.81 1.1 2019 "ARG" 58.81 .9 2017 "ARG" 58.81 1.5 2018 "ARG" 60.29 1 2012 "ARG" 60.36 1.2 2015 "ARG" 60.65 .9 2016 "ARG" 59.44 . 2014 "ARM" 59.39 1.1 2013 "ARM" 63.02 1.4 2017 "ARM" 60.82 .9 2019 "ARM" 59.42 .9 2014 "ARM" 62.29 1.5 2015 "ARM" 59.67 1.3 2018 "ARM" 54.11 1.8 2012 "ARM" 60.72 1.2 2016 "ARM" 58.65 1.2 2011
    To decide between the two type panel regression methods, I followed instructions here

    where I ran both first a fixed effects regression:
    ```
    xtreg lfp_rate pov_ , fe
    estimates store fixed

    Fixed-effects (within) regression Number of obs = 571
    Group variable: numeric_co~e Number of groups = 124

    R-sq: Obs per group:
    within = 0.0136 min = 1
    between = 0.0766 avg = 4.6
    overall = 0.0138 max = 9

    F(1,446) = 6.14
    corr(u_i, Xb) = -0.4866 Prob > F = 0.0136


    lfp_rate Coef. Std. Err. t P>t [95% Conf. Interval]

    pov_ -.3172859 .1280625 -2.48 0.014 -.5689667 -.0656051
    _cons 62.67351 .6149928 101.91 0.000 61.46486 63.88215

    sigma_u 13.733666
    sigma_e 3.1857261
    rho .94893979 (fraction of variance due to u_i)

    F test that all u_i=0: F(123, 446) = 31.33 Prob > F = 0.0000
    ```




    And then ran a random effects model:
    ```
    // Random effecs
    xtreg lfp_rate pov_, re
    estimates store random
    Random-effects GLS regression Number of obs = 571
    Group variable: numeric_co~e Number of groups = 124

    R-sq: Obs per group:
    within = 0.0136 min = 1
    between = 0.0766 avg = 4.6
    overall = 0.0138 max = 9

    Wald chi2(1) = 3.70
    corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0544


    lfp_rate Coef. Std. Err. z P>z [95% Conf. Interval]

    pov_ .0917747 .0477002 1.92 0.054 -.0017161 .1852654
    _cons 59.40576 1.089927 54.50 0.000 57.26954 61.54198

    sigma_u 10.228442
    sigma_e 3.1857261
    rho .91157217 (fraction of variance due to u_i)
    ```



    Then ran a Hausman test to decide between the two models:
    ```
    hausman fixed random
    Coefficients ----
    (b) (B) (b-B) sqrt(diag(V_b-V_B))
    fixed random Difference S.E.

    pov_ -.3172859 .0917747 -.4090606 .1188473

    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(1) = (b-B)'[(V_b-V_B)^(-1)](b-B)
    = 11.85
    Prob>chi2 = 0.0006
    ```


    Based on my understanding, I should use a fixed effects regression model if Prob>chi2 is statistically significant at .05, and given that the result is Prob>chi2 = 0.0006, then I should use an FE model?

    Are there other checks I can run to assess whether to proceed with an FE or RE model?

  • #2
    Meshal:
    as .hausma- reject the null, you should actually go -fe-.
    But my guess is that you have a more substantive issue: the right-hand side of your regression equation includes one predictor only; it's hard to believe that this specifcation can give a true and fair view of the data generating process you're investigating.
    Kind regards,
    Carlo
    (Stata 19.0)

    Comment

    Working...
    X