Bert Jung <[email protected]>:
No, clustering by state allows for serial correlation within state,
but clustering by state-year combination does not. You cannot do
two-way clustering with two years (equivalent to having two clusters
in the time dimension), either.
On Sun, Sep 27, 2009 at 12:52 PM, Bert Jung <[email protected]> wrote:
> Thanks Austin. In principle, assuming the number of states is
> sufficiently "large", wouldn't it be sensible to cluster on both
> variables? If so is creating a new variable with the each state-year
> combination the right way forward?
>
> Thanks,
> Bert
>
>
> On Sun, Sep 27, 2009 at 8:26 AM, Austin Nichols <[email protected]> wrote:
>> Bert Jung <[email protected]>:
>> You should use vce(cluster state) rather than
>> egen g=group(state year)
>> reg y x, vce(cluster g)
>> assuming you have 51 or so clusters with the first method.
>>
>>
>> On Sun, Sep 27, 2009 at 12:15 AM, Bert Jung <[email protected]> wrote:
>>> Dear Statalisters,
>>>
>>> I have a repeated cross-section of individuals in states (2 time
>>> periods) and want to run a difference in difference with state-year
>>> clusters. Since these cluster variables are not nested, how can I
>>> include this in -vce(cluster XX)-?
>>>
>>> Cheers,
>>> Bert
>>
>> *
>> * 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/
>
*
* 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/
