Does anyone know of research on the effects of rounding on regression? e.g., when you ask people "How often have you _______?" you are more likely to get answers like 100, 200, etc. than 98, 203, etc. I'm interested in investigating this, but don't want to reinvent the wheel. thanks Peter Peter L. Flom, PhD Assistant Director, Statistics and Data Analysis Core Center for Drug Use and HIV Research National Development and Research Institutes 71 W. 23rd St www.peterflom.com New York, NY 10010 (212) 845-4485 (voice) (917) 438-0894 (fax)
Have you considered "?qr" and the references cited therein? hope this helps. spencer graves Peter Flom wrote:>Does anyone know of research on the effects of rounding on regression? > >e.g., when you ask people "How often have you _______?" you are more >likely to get answers like 100, 200, etc. than 98, 203, etc. > >I'm interested in investigating this, but don't want to reinvent the >wheel. > >thanks > >Peter > >Peter L. Flom, PhD >Assistant Director, Statistics and Data Analysis Core >Center for Drug Use and HIV Research >National Development and Research Institutes >71 W. 23rd St >www.peterflom.com >New York, NY 10010 >(212) 845-4485 (voice) >(917) 438-0894 (fax) > >______________________________________________ >R-help at stat.math.ethz.ch mailing list >https://www.stat.math.ethz.ch/mailman/listinfo/r-help > >
On 8 Nov 2003 at 10:25, Peter Flom wrote:
One reference for this is Seber (1977), "Linear Regression
Analysis",. page 158, "roundoff errors", and references therein.
(There is a new edition of this book, 02 (03?), which I haven't seen)
Synopsis: bias in \hat{\beta} does only depen on regressors which
have roundimg errors, so rounding errors in continuous variables will
not make bias in estimates for factors.
The usual variance estimate is biased upwards.
This should also be easy to investigate with simulation.
Kjetil Halvorsen
> Does anyone know of research on the effects of rounding on regression?
>
> e.g., when you ask people "How often have you _______?" you are
more
> likely to get answers like 100, 200, etc. than 98, 203, etc.
>
> I'm interested in investigating this, but don't want to reinvent
the
> wheel.
>
> thanks
>
> Peter
>
> Peter L. Flom, PhD
> Assistant Director, Statistics and Data Analysis Core
> Center for Drug Use and HIV Research
> National Development and Research Institutes
> 71 W. 23rd St
> www.peterflom.com
> New York, NY 10010
> (212) 845-4485 (voice)
> (917) 438-0894 (fax)
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help