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: collin
From
Nick Cox <[email protected]>
To
[email protected]
Subject
Re: st: collin
Date
Sat, 12 Mar 2011 16:16:48 +0000
Your variables
y y98 y99 y00 y01 y02
should all be included in y*. Please show those too.
Nick
On Sat, Mar 12, 2011 at 4:10 PM, Aggie Chidlow
<[email protected]> wrote:
> Here are the results for sum y*
>
> Variable | Obs Mean Std. Dev. Min Max
> -------------+-------------------------------------------------------------------
> y_hat | 2251 .3609488 .1824771 4.26e-06 1
On Sat, Mar 12, 2011 at 3:52 PM, DE SOUZA Eric
<[email protected]> wrote:
>> This is exactly what I thought you had, not just collinearity but perfect collinearity.
>> The question is: why are you getting perfectly collinearity?
>> Your y's appear to be constants.
>> Could you produce the results of -summarize y*- ?
Aggie Chidlow
>> Thank you for your advice... will definetly look this reference up.
>>
>> When I run my model with all dummies as the reviewer wants me to:
>>
>> probit y x1 x2 x3 lnx4 x5 y98 y99 y00 y01 y02
>>
>> where:
>> y98=463
>> y99=494
>> y00=425
>> y01=406
>> y02=376
>> y03=88 -not included in the model due to dummies trap
>>
>> I get the regression results that say the follwing:
>> note: y00 omitted because of collinearity
>> note: y01 omitted because of collinearity
>> note: y02 omitted because of collinearity
>>
>> The coefficients for y00 y01 and y02 are not reported in the model and there is a note which says y00 (omitted); y01 (omitted) and y02 (omitted).
>>
>> By the way the collin for year dummies is as follow:
>> Collinearity Diagnostics
>>
>> SQRT R-
>> Variable VIF VIF Tolerance Squared
>> ----------------------------------------------------
>> y98 -3.37e+13 . -0.0000 1.0000
>> y99 -3.53e+13 . -0.0000 1.0000
>> y00 -3.16e+13 . -0.0000 1.0000
>> y01 -3.05e+13 . -0.0000 1.0000
>> y02 -2.87e+13 . -0.0000 1.0000
>> y03 -7.74e+12 . -0.0000 1.0000
>> ----------------------------------------------------
>> Mean VIF -2.79e+13
>>
>> Cond
>> Eigenval Index
>> ---------------------------------
>> 1 2.0000 1.0000
>> 2 1.0000 1.4142
>> 3 1.0000 1.4142
>> 4 1.0000 1.4142
>> 5 1.0000 1.4142
>> 6 1.0000 1.4142
>> 7 0.0000 .
>> ---------------------------------
>> Condition Number .
>> Eigenvalues & Cond Index computed from scaled raw sscp (w/ intercept)
>> Det(correlation matrix) -0.0000
>>
>>
>> On Sat, Mar 12, 2011 at 11:16 AM, DE SOUZA Eric <[email protected]> wrote:
>>> I haven't been following this thread till now.
>>> Jeffrey Wooldridge in his introductory textbook (page 99, international edition) does not encourage use of the VIF . The variance of a coefficient depends on three factors: the standard error of the regression, the total sample variation in the variable attached to the coefficient and the partial R2 . Concentrating on the partial R2 has no justification, even less so the rule of 10.
>>>
>>> However, in this case, the referee will probably have to be satisfied in some way or the other.
>>>
>>> Aggie, when you say that the dummies were dropped on account of collinearity, what exactly do you mean?
>>>> From: Aggie Chidlow <[email protected]>
>>>> I was appreciate some help regarding "collin"
>>>>
>>>> I just got a paper back from a reviewer and he/she wants me to
>>>> include all my year dummies (i.e. y98 y99 y00 y01 y02 y03) in the
>>>> following
>>>> model: probit y x1 x2 x3 lnx4 x5 y98 y99 y00 y01 y02
>>>>
>>>> Previusly in the model I only included two year dummies (i.e y99 and
>>>> y01) as the others we omitted automatically due to collinearity.
>>>> I mentioned that in the paper, however, he/she says it is
>>>> unsatisfactory and I should include them all and than comment on VIF.
>>>>
>>>> Please, can somebody tell me how I can go about this?
*
* 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/