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: fixed effects with multicollinearity
From
Nick Cox <[email protected]>
To
[email protected]
Subject
Re: st: fixed effects with multicollinearity
Date
Fri, 29 Jul 2011 11:11:08 +0100
What you have in mind is not especially clear. Quite what is
"accounting for" heteroscedasticity? Robust options don't do as much
as you appear to be hoping. Robust options just give a better
estimate of variance, or equivalently of standard errors; the
coefficients remain the same.
Sometimes, heteroscedasticity is tied up with nonlinearity, for
example, so that a different functional form, or transformation of
variables, is advisable.
In contrast, normality of error terms is the least important
assumption in regression, but looking at your residuals is not going
to do any harm. Plotting residual versus predicted is the best
graphical diagnostic, but often omitted by people in many disciplines.
Nick
On Fri, Jul 29, 2011 at 10:19 AM, Reddy, Colin <[email protected]> wrote:
> Yes, drop from the model.
>
> The FE estimation in Stata does mean centering in any case but apparently it removes the effects of other time invariant variables in the model if any. Should one then use the OLS plus dummies? Can one simply add the "robust" to the code for OLS plus dummies to account for any heteroskedacity? I guess a post-estimation normality check is also in order?
>
> Colin
> ________________________________________
> From: [email protected] [[email protected]] on behalf of Nick Cox [[email protected]]
> Sent: 29 July 2011 11:13 AM
> To: [email protected]
> Subject: Re: st: fixed effects with multicollinearity
>
> Yes, if by drop you mean omit from the model (not -drop- from the
> dataset). Best to do it on scientific, substantive or practical
> grounds if there is a choice.
>
> On Fri, Jul 29, 2011 at 10:07 AM, Reddy, Colin <[email protected]> wrote:
>> Thanks Daniel
>> So I guess the best is to drop one of the collinear variables.?
>>
>> Colin ________________________________________
>>
>> From: [email protected] [[email protected]] on behalf of daniel klein [[email protected]]
>> Sent: 29 July 2011 11:01 AM
>> To: [email protected]
>> Subject: Re: st: fixed effects with multicollinearity
>>
>> Colin,
>>
>> please note that mean centring does nothing to solve the underlying
>> problem of collinarity (if there is something like that)., see e.g.
>> Echambadi and Hess (2007) or Shieh, G. (2011).
>>
>> However, in another post
>> (http://www.stata.com/statalist/archive/2011-04/msg01204.html) Maarten
>> Buis pointed out that in the special case, where a variable is
>> interacted with itself, to model non-linearities, centering can help.
>>
>>
>> Echambadi and Hess (2007). Mean-Centering Does Not Alleviate
>> Collinearity Problems in Moderated Multiple Regression Models.
>> Marketing Science, 26: 438-445
>>
>> Shieh, G. (2011). Clarifying the role of mean centring in
>> multicollinearity of interaction effects. British Journal of
>> Mathematical and Statistical Psychology, 64: 1-12
*
* 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/