using jackknifeopts in the bootstrap command

[Andrew Xavier]

Hi,

Please see the code below

webuse pig, clear
bootstrap, reps(20) bca ties mse cluster(id) idcluster(xid) jackknifeopts(): mixed weight week || id: week

I am running bootstrap command with bca option with clustered data and there is a need to specify cluster(id) idcluster(xid)

Because I specify the option bca command it does seem that stata runs jacknife. When I looked at the jackknifeopts(), I am unclear whether I need to specify also the options cluster(id) idcluster(xid) within jackknifeopts(). When I do the following there is an error

webuse pig, clear
bootstrap, reps(20) bca ties mse cluster(id) idcluster(xid) jackknifeopts(, cluster(id) idcluster(xid)): mixed weight week || id: week

Please could anyone help me with this?

Additionally, I tried to run the jacknife command directly and it does not seem to allow the options and produces error

jacknife eclass, cluster(id) idcluster(xid): mixed weight week || id: week
jacknife rclass, cluster(id) idcluster(xid): mixed weight week || id: week

However, if I run the command without any option, the command runs fine

jacknife, cluster(id) idcluster(xid): mixed weight week || id: week

Please could anyone clarify this?

Thanks

Andrew

[Leonardo Guizzetti]

You don't need jacknifeopts() at all with -bootsrap, bca-. Yes, it does run a jackknife to calculate the "acceleration factor", but that is handled for you in the background. Additionally, if you plan to cluster, then you do need -cluster()- and -idcluster()- as options to -bootstrap-.

[Andrew Xavier]

Does the jacknife run in the background when I use "bootsrap, bca" take into account the options cluster and idclusterid, such as in the example below?

 
 bootstrap, reps(20) bca ties mse cluster(id) idcluster(xid): mixed weight week || id: week

Thanks
Andrew

[Leonardo Guizzetti]

As far as I'm aware, yes. Does the number of iterations equal the number of clusters minus 1?

Alternatively, use cluster robust variance estimates with REML and ditch bootstrapping.

[Andrew Xavier]

Thank you

I don't want to use the bootstrap facilities within the mixed command because I am trying to understand how the bootstrap command works


When I run the code

 
 webuse pig, clear bootstrap, reps(20) bca ties mse cluster(id) idcluster(xid) jackknifeopts(): mixed weight week || id: week

This is the output I get

Jackknife replications (48)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
................................................

Bootstrap replications (20)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
....................

Mixed-effects ML regression                     Number of obs     =        432
Group variable: id                              Number of groups  =         48
                                                Obs per group:
                                                              min =          9
                                                              avg =        9.0
                                                              max =          9
                                                Wald chi2(1)      =   16898.21
Log likelihood = -869.03825                     Prob > chi2       =     0.0000

                                     (Replications based on 48 clusters in id)
------------------------------------------------------------------------------
             |   Observed   Bstrap *
      weight | coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
        week |   6.209896    .047771   129.99   0.000     6.116266    6.303525
       _cons |   19.35561    .333134    58.10   0.000     18.70268    20.00854
------------------------------------------------------------------------------

------------------------------------------------------------------------------
                             |   Observed   Bstrap *          Normal-based
  Random-effects parameters  |   estimate   std. err.     [95% conf. interval]
-----------------------------+------------------------------------------------
id: Independent              |
                   var(week) |   .3680668    .058867      .2690229    .5035749
                  var(_cons) |   6.756364   1.445921      4.441692    10.27727
-----------------------------+------------------------------------------------
               var(Residual) |   1.598811   .3296255      1.067347    2.394907
------------------------------------------------------------------------------
LR test vs. linear model: chi2(2) = 764.42                Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference.

Jackknife replications seems to be 48 and there are 48 subjects and a total of 432 observations

Thanks

Andrew

[Andrew Xavier]

Thank you

I don't want to use the bootstrap facilities within the mixed command because I am trying to understand how the bootstrap command works


When I run the code

 
 webuse pig, clear bootstrap, reps(20) bca ties mse cluster(id) idcluster(xid) jackknifeopts(): mixed weight week || id: week

This is the output I get

Jackknife replications (48)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
................................................

Bootstrap replications (20)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
....................

Mixed-effects ML regression                     Number of obs     =        432
Group variable: id                              Number of groups  =         48
                                                Obs per group:
                                                              min =          9
                                                              avg =        9.0
                                                              max =          9
                                                Wald chi2(1)      =   16898.21
Log likelihood = -869.03825                     Prob > chi2       =     0.0000

                                     (Replications based on 48 clusters in id)
------------------------------------------------------------------------------
             |   Observed   Bstrap *
      weight | coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
        week |   6.209896    .047771   129.99   0.000     6.116266    6.303525
       _cons |   19.35561    .333134    58.10   0.000     18.70268    20.00854
------------------------------------------------------------------------------

------------------------------------------------------------------------------
                             |   Observed   Bstrap *          Normal-based
  Random-effects parameters  |   estimate   std. err.     [95% conf. interval]
-----------------------------+------------------------------------------------
id: Independent              |
                   var(week) |   .3680668    .058867      .2690229    .5035749
                  var(_cons) |   6.756364   1.445921      4.441692    10.27727
-----------------------------+------------------------------------------------
               var(Residual) |   1.598811   .3296255      1.067347    2.394907
------------------------------------------------------------------------------
LR test vs. linear model: chi2(2) = 764.42                Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference.

Jackknife replications seem to be 48 and there are 48 subjects and a total of 432 observations

Thanks

Andrew