Re: st: Autocorrelation tests for FE models

[Kyle Hood <kyle.hood@yale.edu>]

There is no conflict, here.  Stata has not reported any standard errors 
for rho estimated using the xtregar command; this command is not testing 
for serial autocorrelation, per se, although if you can find a standard 
error for rho in the e() structure, maybe you can do the test yourself.  
The other command you report the results for is testing for serial 
autocorrelation, and we only know the p-value associated with this 
test.  Be careful not to confuse statistical significance and effect 
size -- both of these commands are telling you different things.  The 
latter test strengthens the argument that xtregar should be used 
(instead of xtreg, for example).

P K wrote:
> Hi,
>
> I test a fixed effects panel model with 61 groups for 10 years and tested for autocorrelation with the xtregar and the xtserial command, which lead to different findings. 
> Xtregar doesn't indicate too high autocorrelation (rho_ar =.12), while xtserial indicates autocorrelation (Woolridge test: Prob > F =      0.0041)
>
> My questions therefore are:
> Which test is more "reliable", and what can I do about the autocorrelation indicated by the Woolridge test (how to correct for it)?
>
> Thanks,
> Pat
>
>
> PS: Here is the output for each test:
>
> 1) xtregar command in STATA:  rho_ar   .12069893   => doesn't seem to indicate too high autocorrelation
>
> OUTPUT:
>
> FE (within) regression with AR(1) disturbances  Number of obs      =       426
> Group variable: number                          Number of groups   =        61
>
> R-sq:  within  = 0.0206                         Obs per group: min =         5
> between = 0.0017                                        avg =       7.0
> overall = 0.0011                                        max =         8
>
> F(6,359)           =      1.26
> corr(u_i, Xb)  = -0.8001                        Prob > F           =    0.2762
>
>
> AverCh~Eadj2       Coef.   Std. Err.      t    P>t     [95% Conf. Interval]
>
> duration_c~e   -.4899444    .687246    -0.71   0.476    -1.841478    .8615894
> sizelnempl   -1.867376   2.287751    -0.82   0.415    -6.366453      2.6317
> age   -.0368308   .2794663    -0.13   0.895    -.5864275    .5127659
> prior_slac~2    .4346291   .3607199     1.20   0.229    -.2747605    1.144019
> ROEadj_1    .0288556   .0507004     0.57   0.570    -.0708516    .1285627
> external_c~y    2.555537   1.347617     1.90   0.059    -.0946773    5.205752
> _cons    16.30932   21.37442     0.76   0.446    -25.72549    58.34412
>
> rho_ar   .12069893
> sigma_u   6.6460257
> sigma_e   12.876146
> rho_fov   .21036696   (fraction of variance because of u_i)
>
> F test that all u_i=0:     F(60,359) =     0.49              Prob > F = 0.9994
>
>
> 3) xtserial command in STATA: Woolridge test indicates that there is autocorrelation
>
> OUTPUT:
> Linear regression                                      Number of obs =     414
> F(  6,    60) =    2.31
> Prob > F      =  0.0450
> R-squared     =  0.0252
> Root MSE      =  16.212
>
> (Std. Err. adjusted for 61 clusters in number)
>
> Robust
> D.          
> AverCh~Eadj2       Coef.   Std. Err.      t    P>t     [95% Conf. Interval]
>
> duration_c~e 
> D1.    .3378522   .8894139     0.38   0.705    -1.441241    2.116945
> ukdum 
> D1.   (dropped)
> frendum 
> D1.   (dropped)
> sizelnempl 
> D1.   -4.680739   2.989539    -1.57   0.123    -10.66071    1.299229
> age 
> D1.   -.3837846    .391899    -0.98   0.331    -1.167699    .4001301
> prior_slac~2 
> D1.    .2684736    .315091     0.85   0.398    -.3618022    .8987494
> ROEadj_1 
> D1.    .1397712   .0695373     2.01   0.049      .000676    .2788664
> external_c~y 
> D1.    1.648788   1.834245     0.90   0.372    -2.020249    5.317824
>
>
> Woolridge test for autocorrelation in panel data
> H0: no first-order autocorrelation
> F(  1,      60) =      8.903
> Prob > F =      0.0041
>
>
>       
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>   


--
Kyle Hood
Department of Economics
Yale University
New Haven, CT
website: http://www.econ.yale.edu/~kkh25/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/