st: Randomly selecting variables in panel, collapsing, and simulating

[John Antonakis <john.antonakis@unil.ch>]

Hi:

I want to do something that is quite straightforward, but don't yet have 
the programming skills to do this, so I would appreciate some help.  
Basically, I want to select randomly within clusters, then collapse, 
then estimate a regression model that I would like to simulate.  I would 
like to define it as an e-class program so that I can easily incorporate 
a simulation in there.

I have the following data structure (I have added a line in-between 
clusters to facilitate viewing):

i      y      x     z     n
1     5      8    3    1
1     9    10    5    2
1     8    12    6    3
1     6      9    4    4

2     5      7    2    1
2     9    13    4    2
2     8      8    7    3
2     6      8    6    4
2     8      8    7    5
2     8      8    7    6

3     5      7    2    1
3     9    11    7    2
3     8    11    5    3


Where y is the dependent variable, and x and z are covariates (note, I 
have more covariates, so I would like to use something general that 
works even with more covariates), "i" is the panel identifier and "n" is 
number of observations in each cluster. Note that each cluster does not 
have the same number of observations

I would like to do the following:

1)  by cluster (i.e., within in each cluster), randomly split the 
observations into two groups, with an equal amount of observations (if 
number of observations are even); if the number of observations are odd, 
then one of the group will have 1 more observation than the other. Thus, 
I would now have this:

i      y      x     z     n   g
1     8    12    6    3   1
1     5      8    3    1   1
1     6      9    4    4   2
1     9    10    5    2   2

2     9    13    4    2   1
2     5      7    2    1   1
2     8      8    7    3   1
2     8      8    7    5   2
2     6      8    6    4   2
2     8      8    7    6   2

3     5      7    2    1   1
3     8    11    5    3   2
3     9    11    7    2   2

2) collapse observations at the group level for each cluster to get:

i      g     y        x      z    
1    1    6.5    10    4.5
1    2    7.5    9.5    4.5

2    1    7.3    9.3    4.3
2    2    7.3    8      6.7

3    1    5       7         2
3    2    8.5    11       6

3) then I want to the data as below, so that I can regress the y of the 
first observation in each cluster to the x and z of the second 
observation in each cluster

i       y        x       z 
1    7.5    9.5     4.5
2    7.3    8        6.7
3    8.5    11         2

4) once I have this I want to simulate it, using something like this:

capture program drop sim
version 10.1
program define sim, eclass

[here is the code I would like to insert to make points 1-4 above run]

simulate _b _se, reps(20) seed (123) : sim,

foreach v in x z {
 gen t_`v' = _b_`v'/_se_`v'
 gen p_`v' = 2*(1-normal(abs(t_`v')))

Thank you.
John.
--

____________________________________________________

Prof. John Antonakis
Associate Dean Faculty of Business and Economics
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland

Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305

Faculty page:
http://www.hec.unil.ch/people/jantonakis&cl=en

Personal page:
http://www.hec.unil.ch/jantonakis
____________________________________________________

*
*   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/