Re: st: Matched Pairs, Treatment-Control, Interrupted Panel Study -- What Model will Work?

[Jenny Day <jennieday@gmail.com>]

Hi Austin, Doug, Gene, and Statalist,

Thanks!!

My treatment variable is a loss of a daily metropolitan newspaper.  I have
a dataset of 40 cities that have seen a newspaper close since 2007.  I am
measuring newspaper content on certain regular key issues (e.g.,
redistricting) and also election outcomes (e.g., voter participation).
The cities range from small towns that went from one daily paper to none,
to large cities (e.g., Cincinnati) that went from two daily newspapers to
one.  Because the newspaper is lost all at once, the effects will be
immediate for some issues like coverage, and onset could be delayed for
other issues such as voter turnout.  For controls, I have Census data on
population, median income, land area of the metro region, and other
relevant variables.

As I said earlier, I have a balanced panel dataset from 1998 to 2011
(1989-2011 for the redistricting article searches) with few missing
values.  

Gene, yes, you are right about going to the literature.  I've never done
this type of study, and had not had a lot of luck identifying articles
that used datasets and methods similar to mine.  I will of course keep
looking, but Doug, your BACI suggestion has helped a lot.  Thanks for
that.  

In terms of the trends, most year-on-year graphs showed an inverted V
shape, where, say, the article counts on redistricting were low-ish in
1989, 1990, and 1991, in some cases doubled in 1999, 2000, and 2001, and
then dropped again in 2009, 2010, and 2011.  In almost all cases, the
control cities either did not drop, or did not drop as drastically as the
treatment cities, from the 1999-2001 time period to the 2009-2011 time
period.  

Does this help you to see where I am?

Thanks!!
Jennie



On 5/06/12 3:52 AM, "Austin Nichols" <austinnichols@gmail.com> wrote:

>Jenny Day <jennieday@gmail.com>:
>Some kind of regression is indicated--what is the outcome variable?
>What are the relevant control variables?
>Are there different pre- and post-treatment trends in both?
>
>On Fri, Jun 1, 2012 at 3:43 AM, Jenny Day <jennieday@gmail.com> wrote:
>> Hi Out There!
>>
>> I have a beautiful balanced panel dataset from 1998 to 2011, of election
>> results (incumbents re-elected, party turnover, voter turnout, etc.) for
>> municipalities in the USA, with no missing values.  I have matched 25
>> treatment cities to 25 control cities. All of the treatment cities
>> experienced a certain policy shock at different times between 2007 and
>> 2011; the control cities are similar in demographic composition and
>> region, but did not experience the shock.  I also have relevant control
>> variables from the Census data.
>>
>> In summary, I have a setup for a matched-pairs, treatment-control,
>> interrupted panel study.  I want to use a panel method to estimate the
>> effect of that shock.  I am writing to this list because I am not sure
>> what statistical method is suitable for this type of study.
>>
>> Does anyone have any suggestions?
>>
>> Thanks!!
>> Jennie
>
>*
>*   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/


*
*   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/