Jackknife for xtpcse

[Laurence Vedelsdal]

Hi,

I am trying to conduct a Jackknife estimation.

The following command is able to be executed, even though the output looks dubious:

jackknife coef=property_rights: xtpcse tea_opp property_rights lag1_log_gdp_per_capita_ppp_cons unemployment,
pairwise correlation(ar1)

However, with the above command it excludes a single observation at a time.
My interest is not if the findinds hold when excluding individual observations (577) but rather excluding countries (100). So I attempted the following, adding cluster(country):

jackknife coef=property_rights: xtpcse tea_opp property_rights lag1_log_gdp_per_capita_ppp_cons unemployment,
pairwise correlation(ar1) cluster(country_id)

But i retrieve the following error:

varlist not allowed

Does anyone know how to exclude specific clusters rather than observations for the specified regression?

//Laurence

[William Lisowski]

From the output of help jackknife, cluster() is an option to the jackknife command, rather than to the xtpcse command being run by jackknife, so at a minimum your command needs to be something like

jackknife coef=property_rights, cluster(country_id): xtpcse tea_opp property_rights lag1_log_gdp_per_capita_ppp_cons unemployment, pairwise correlation(ar1)

The error message apparently generated by xtpcse seems ... misleading, to be generous about it.

[Laurence Vedelsdal]

Thank you for your help William.

Unfortunately, the output still looks odd.

. jackknife coef=property_rights, cluster(country_id): xtpcse tea_opp property_rights lag1_log_gdp_per_capita_ppp_cons unemployment, pairwise correlation(ar1)
(running xtpcse on estimation sample)

Jackknife replications (100)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
..................................................    50
..................................................   100

Jackknife results                               Number of obs     =        577
                                                Replications      =        100

      command:  xtpcse tea_opp property_rights lag1_log_gdp_per_capita_ppp_cons unemployment, pairwise correlation(ar1)
         coef:  property_rights
          n():  e(N)

                            (Replications based on 100 clusters in country_id)
------------------------------------------------------------------------------
             |              Jackknife
             |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        coef |          3          .        .       .            .           .
------------------------------------------------------------------------------

Do you have any suggestions for what I need to adjust/correct?

[William Lisowski]

I have no ideas; I use neither jackknife nor xtpcse. But it would perhaps help someone else who is more familiar than I am with xtpcse to see the output from xtpcse run without the jackknife.

[Laurence Vedelsdal]

Of couse, thank you for the suggestion. The command and output are as follows:

jackknife coef=property_rights, cluster(country_id) eclass: xtpcse tea_opp property_rights lag1_log_gdp_per_capita_ppp_cons unemployment, pairwise correlation(ar1)
(running xtpcse on estimation sample)

Jackknife replications (34)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
..................................

Jackknife results                               Number of obs     =        249
                                                Replications      =         34

      command:  xtpcse tea_opp property_rights lag1_log_gdp_per_capita_ppp_cons unemployment, pairwise correlation(ar1)
         coef:  property_rights
          n():  e(N)

                             (Replications based on 34 clusters in country_id)
------------------------------------------------------------------------------
             |              Jackknife
             |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        coef |          9          .        .       .            .           .
------------------------------------------------------------------------------

[Laurence Vedelsdal]

For anyone interested, I shifted focus from the coefficients to the standard errors and retrived useful output from the command shown below.

jackknife, cluster(country_id): xtpcse tea_opp property_rights lag1_log_gdp_per_capita_ppp_cons unemployment, pairwise correlation(ar1)
(running xtpcse on estimation sample)

Jackknife replications (34)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
..................................

Prais-Winsten regression, correlated panels corrected standard errors (PCSEs)

Group variable:   country_id                    Number of obs     =        249
Time variable:    year                          Number of groups  =         34
Panels:           correlated (unbalanced)       Obs per group:
Autocorrelation:  common AR(1)                                min =          1
Sigma computed by pairwise selection                          avg =  7.3235294
                                                              max =         11
Estimated covariances      =       595          R-squared         =     0.3157
Estimated autocorrelations =         1          Wald chi2(3)      =          .
Estimated coefficients     =         4          Prob > chi2       =     0.0774

                                                 (Replications based on 34 clusters in country_id)
--------------------------------------------------------------------------------------------------
                                 |              Jackknife
                         tea_opp |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------------------+----------------------------------------------------------------
                 property_rights |   .0240794   .2181675     0.11   0.913    -.4197858    .4679446
lag1_log_gdp_per_capita_ppp_cons |   2.168841   2.910386     0.75   0.461    -3.752385    8.090067
                    unemployment |  -.1190462   .0647625    -1.84   0.075    -.2508065    .0127141
                           _cons |  -3.614457   13.07118    -0.28   0.784    -30.20798    22.97906
---------------------------------+----------------------------------------------------------------
                             rho |   .5591998
--------------------------------------------------------------------------------------------------

Thanks again to you William :-)