Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: -xtpcse and period dummies- no full table reported due to asymmetric and singular variance matrix
From
Nick Cox <[email protected]>
To
[email protected]
Subject
Re: st: -xtpcse and period dummies- no full table reported due to asymmetric and singular variance matrix
Date
Mon, 14 Mar 2011 13:13:33 +0000
So, you are getting a clear signal that the model is a very poor idea
for your data. Why not back off and consider a different treatment of
time, e.g. not using dummies but a linear, quadratic or spline fit?
Nick
On Mon, Mar 14, 2011 at 1:01 PM, Felix Wädlich <[email protected]> wrote:
> Hey Justina,
>
> I include gdp as a measure of market size and gdppc as an indicator
> for development or wealth.
>
> I tried to build up the covariates, but the problem remains: as soon
> as I include the year dummies (yr*), I get the breakdown message. It
> probably is a very basic problem, but I just dont see what I am doing
> wrong, and also what else I can do. Any more thoughts?
>
> Best,
> Felix
>
>
> 2011/3/14 Justina Fischer <[email protected]>:
>> Hi
>>
>> try to build up the covariates step-by-step and observe when breakdown occurs.
>>
>> Why do you include gdppc and gdp simulatenously ?
>>
>> Justina
>> -------- Original-Nachricht --------
>>> Datum: Mon, 14 Mar 2011 13:39:28 +0100
>>> Von: "Felix Wädlich" <[email protected]>
>>> An: [email protected]
>>> Betreff: Re: st: -xtpcse and period dummies- no full table reported due to asymmetric and singular variance matrix
>>
>>> Dear Clive,
>>>
>>> thanks for ur suggestion, but unfortunately that did not do the job.
>>> Still getting the same warning message.Any other tip?
>>>
>>> Best,
>>> Felix
>>>
>>> . xtpcse c_capital c_loggdppc c_loggdp c_lgovcon c_lgrowth
>>> c_lregionothers yr*, pairwise c(ar1)
>>>
>>> Number of gaps in sample: 8
>>> (note: computations for rho restarted at each gap)
>>> note: yr1 omitted because of collinearity
>>> note: yr2 omitted because of collinearity
>>> (note: estimates of rho outside [-1,1] bounded to be in the range [-1,1])
>>> (note: at least one disturbance covariance assumed 0, no common time
>>> periods
>>> between panels)
>>> Warning: variance matrix is nonsymmetric or highly singular
>>>
>>> Prais-Winsten regression, correlated panels corrected standard errors
>>> (PCSEs)
>>>
>>> Group variable: Country Number of obs
>>> = 2814
>>> Time variable: year Number of
>>> groups = 134
>>> Panels: correlated (unbalanced) Obs per group: min
>>> = 2
>>> Autocorrelation: common AR(1)
>>> avg = 21
>>> Sigma computed by pairwise selection
>>> max = 25
>>> Estimated covariances = 9045 R-squared
>>> = 0.0493
>>> Estimated autocorrelations = 1 Wald chi2(0)
>>> = .
>>> Estimated coefficients = 30 Prob > chi2
>>> = .
>>>
>>> ------------------------------------------------------------------------------
>>> | Panel-corrected
>>> c_capital | Coef. Std. Err. z P>|z| [95%
>>> Conf. Interval]
>>> -------------+----------------------------------------------------------------
>>> c_loggdppc | -.5241978 . . .
>>> . .
>>> c_loggdp | .110086 . . .
>>> . .
>>> c_lgovcon | .0016855 . . .
>>> . .
>>> c_lgrowth | .0008101 . . .
>>> . .
>>> c_lregiono~s | .1941566 . . .
>>> . .
>>> yr1 | (omitted)
>>> yr2 | (omitted)
>>> yr3 | .022493 . . .
>>> . .
>>> yr4 | .0273882 . . .
>>> . .
>>> yr5 | .0115461 . . .
>>> . .
>>> yr6 | .0337352 . . .
>>> . .
>>> yr7 | .0598745 . . .
>>> . .
>>> yr8 | .0291169 . . .
>>> . .
>>> yr9 | .0211556 . . .
>>> . .
>>> yr10 | .0255307 . . .
>>> . .
>>> yr11 | .0242099 . . .
>>> . .
>>> yr12 | .0043918 . . .
>>> . .
>>> yr13 | -.0247601 . . .
>>> . .
>>> yr14 | .0009665 . . .
>>> . .
>>> yr15 | .1060365 . . .
>>> . .
>>> yr16 | .0805885 . . .
>>> . .
>>> yr17 | .0332214 . . .
>>> . .
>>> yr18 | .0832608 . . .
>>> . .
>>> yr19 | .0403092 . . .
>>> . .
>>> yr20 | .0195583 . . .
>>> . .
>>> yr21 | .0004272 . . .
>>> . .
>>> yr22 | .013192 . . .
>>> . .
>>> yr23 | .0135001 . . .
>>> . .
>>> yr24 | .0059468 . . .
>>> . .
>>> yr25 | .0014196 . . .
>>> . .
>>> yr26 | -.00606 . . .
>>> . .
>>> _cons | -.0334178 . . .
>>> . .
>>> -------------+----------------------------------------------------------------
>>> rho | .8822325
>>> ------------------------------------------------------------------------------
>>>
>>>
>>> 2011/3/13 Clive Nicholas <[email protected]>:
>>> > Felix Waedlich:
>>> >
>>> >> I have to run a Prais-Winsten-regression with Period-Dummies and
>>> panel-corrected standard errors [by xtpcse, corr(ar1)], since my (unbalanced)
>>> panel needs correction for serial correlation, groupwise heteroskedasticity
>>> and contemporaneous correlation. I need the period dummies to control for
>>> common shocks and trends, since I have a Diffusion variable respectively
>>> spatial lag, which requires a conservative test against alternative external
>>> influences.
>>> >> Without period dummies, there appears to be no problem: Stata reports
>>> me all the coefficients, the z-values, etc. But when I include the period
>>> dummies, Stata says "warning: variance matrix is nonsymmetric or highly
>>> singular", and only lists the coefficients, but no standard errors, no z-values,
>>> etc.. Since this is a standard specification in my field of research, I am
>>> wondering how I can solve this problem. Any suggestions? Help much
>>> appreciated
>>> >> I read the only article on Statalist about this error message, but it
>>> really does not help me, and i dont really see how my data suffers from
>>> these problems.
>>> >
>>> > Try centering your continously measured variables by period and then
>>> > run -xtpcse, c(ar1)- again. Doing this comes at the cost of robbing
>>> > your model of explaining the variance in your response variable that
>>> > it would otherwise in a 'normal' model.
*
* 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/