Neumayer,E
> with respect to detecting outliers, the STATA manual (R:
> regression diagnostics) says that following Belsley, Kuh
> and Welsch (1980), observations with DFITS values greater
> than 2*sqrt(k/n) deserve further attention, where k is the
> number of estimated coefficients and n is sample size.
> However, I have sometimes seen people stating that it is
> the absolute value of DFITS that matters, such that
> observations with abs(DFITS)>2*sqrt(k/n) deserve further
> attention. In many estimations, I get observations with a
> negative DFITS, so which formula is correct matters. Not
> having access to Belsley, Kuh and Welsch (1980) myself, I
> wonder whether anybody can shed light on this?
You don't need access to said text.
DFITS have the same signs as the corresponding residuals,
so that the rule of thumb quoted should be applied to absolute values.
This follows from the equations at [R] regress.
Nick
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
