I should have added that an answer to your second question: adding
indicator variables for "year" would be a good start to modeling the
time effect, but only a start.
-Steve
On Mon, Jul 27, 2009 at 11:58 AM, <[email protected]> wrote:
> Rose, I've not tried any of these programs and I am not expert in this
> area. However, with only five years of data, you do not have enough
> time clusters to rely on the two-way clustering methods. In fact, with
> so few year clusters, Petersen's simulations seem to show that two-way
> clustering is not necessary (see Figure 7 of his paper). So, I
> suggest that you cluster on company and model time in another way,
> perhaps with random coefficient models. As Petersen stresses, you must
> still test the assumption that your model is correct.
>
> In any case, I suggest that you read at least the summaries and
> conclusions of Petersen and the Cameron, Gelbach, and Miller papers.
>
> Good luck
>
>
>
>
>
> On Mon, Jul 27, 2009 at 11:03 AM, <[email protected]> wrote:
>> Steve, thanks a lot.I got probit2.ado in the website you posted.
>>
>> However, I had two doubts.
>>
>> Firstly, My sample is pooled cross-section data.
>> I estimate standard error clustered by company and year. Should I
>> still take year dummies varialbes as independent variables? In fact,
>> when I include year dummies as independent variables, the standard error
>> for them are missing. why?
>>
>> Probit with 2D clustered SEs Number of obs = 4396
>> F( 9, .) = .
>> Prob > F = .
>> Number of clusters (code) = 1285 R-squared = .
>> Number of clusters (year) = 5 Root MSE = .
>> ------------------------------------------------------------------------------
>> | Coef. Std. Err. z P>|z| [95% Conf. Interval]
>> -------------+----------------------------------------------------------------
>> x1 | .1120743 .024432 4.59 0.000 .0641884 .1599602
>> x2 | -.4514519 .2984397 -1.51 0.130 -1.036383 .1334791
>> x3 | -.8063046 .2314595 -3.48 0.000 -1.259957 -.3526524
>> x4 | -.3627517 .1717722 -2.11 0.035 -.699419 -.0260843
>> x5 | -.7701916 .2277355 -3.38 0.001 -1.216545 -.3238382
>> year1 | -.0380545 . . . . .
>> year2 | -.0613474 . . . . .
>> year3 | -.1183585 . . . . .
>> year4 | -.1321975 . . . . .
>> _cons | -1.997594 .2141259 -9.33 0.000 -2.417273 -1.577915
>> ------------------------------------------------------------------------------
>>
>> SE clustered by code and year
>>
>>
>> Secondly, why the equation statistics is F, not chi2? why is missing?
>>
>> Thank you for any help!
>>
>> Sincerely,
>> From Rose.
>>
>> ----- Original Message -----
>> From: [email protected]
>> To: [email protected]
>> Subject: Re: st: how to adjust standard error with -cluster- option by two dimensions?
>> Date: 2009-7-27 20:19:39
>>
>> --
>> You can download a Stata program at:
>> http://www.econ.ucdavis.edu/faculty/dlmiller/statafiles/#_jmp0_
>>
>> This code was used in: “Robust Inference with multi-way clustering” by
>> Cameron, Gelbach, and Miller (2006, NBER TWP 327) available at:
>> http://ideas.repec.org/p/nbr/nberte/0327.html#_jmp0_
>>
>> Other Stata programs that have adapted or extended this approach can
>> be found on the web page:
>> http://www.kellogg.northwestern.edu/faculty/petersen/htm/papers/se/se_programming.htm
>>
>> to accompany a paper:
>> http://www.kellogg.northwestern.edu/faculty/petersen/htm/papers/standarderror.htm
>>
>>
*
* 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/
