Re: st: random sampling matching the characteristics of the sample

[David Hoaglin <dchoaglin@gmail.com>]

Andrea,

If we are to understand whether propensity scores are appropriate, you
must tell us more about the definition of the dummy variable, x.  If x
is the result of some sort of self-selection (e.g., the individuals
chose to participate in an activity), propensity scores might be
appropriate.  If x is an observed characteristic, but not a result of
such a choice (e.g., gender), propensity scores would not be
appropriate.

If the reviewer gave no explanation or reason for that request, I
would push back, saying that so far I have found no justification for
doing what the reviewer requested.  It is not unheard of for a
reviewer to be wrong.

In any event, you have not told us whether you have checked on the
assumption that the coefficients of age and size are the same for the
x=0 data and the x=1 data.  That is important, regardless of whether a
reviewer asked about it.

David Hoaglin

On Thu, May 3, 2012 at 6:01 AM, Andrea Rispoli <andrea.rspl@gmail.com> wrote:
> Thank you all very much for getting back to me! Unfortunately the rev
> does not provide much details.
> Based on your suggestions, I have now identified the command pscore
> and psmatch2 to use propensity scores.
>
> A brief follow up question: when I use pscore, I end up with a certain
> number of blocks, satosfying the balancing property.
> If I understood correctly the intuition behind this method, I shall
> select now some blocks of firms where the distribution of x is more
> balanced and that have high pscore and see whether my analysis hold in
> those groups. Is this correct? Or is there another method to chose the
> blocks to use?
> Thank you again!
*
*   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/