I suspect that your data is non-normal.  You might try the diagnostics in
the nortest package and refer to the text Thode (2002), Testing for
Normality, Marcel Decker, quoted in the references to that package.  A QQ
diagram might help to reveal the problems with your data.

John Frain

On 22/02/07, Simon P. Kempf <[EMAIL PROTECTED]> wrote:
>
> Hello R-Users,
>
>
>
> The following questions are not R-technical, but more of general
> statistical
> nature.
>
>
>
> 1. NORMALITY
>
> I built a normal linear regression model and now I want to check for the
> residual normality assumption. If I check the distribution graphically and
> look at the descriptive characteristics (skewness and kurtosis are below
> 1),
> I would confirm that the residuals are normally distributed.
>
>
>
> > basicStats(IQR.in.mi02.nw.tdv.mix$residuals)
>
>             round.ans..digits...6.
>
> nobs                  19316.000000
>
> NAs                       0.000000
>
> Minimum                  -0.639527
>
> Maximum                   0.693383
>
> 1. Quartile              -0.083753
>
> 3. Quartile               0.088886
>
> Mean                      0.000000
>
> Median                    0.004641
>
> Sum                       0.000000
>
> SE Mean                   0.001047
>
> LCL Mean                 -0.002053
>
> UCL Mean                  0.002053
>
> Variance                  0.021186
>
> Stdev                     0.145554
>
> Skewness                 -0.164821
>
> Kurtosis                  0.937282
>
>
>
> However, when I use the jarque.bera.test(), the assumption of normality is
> rejected.
>
>
>
> > jarque.bera.test(IQR.in.mi02.nw.tdv.mix$residuals)
>
>
>
>         Jarque Bera Test
>
>
>
> data:  IQR.in.mi02.nw.tdv.mix$residuals
>
> X-squared = 795.1296, df = 2, p-value < 2.2e-16
>
>
>
> Therefore, I am wondering how good are the diagnostic test for the
> normality
> assumption? Do you they work for large sample as well? If not, what other
> diagnostic measures for normality exist for large samples?
>
>
>
> In many statistic textbooks it is mentioned that for large samples the
> significance tests are independent from the distributions of residuals.
> Why
> is that?
>
>
>
> 2. RAMSEY TEST
>
> I made several scatterplots (residuals ~ predictors). Here everything
> looks
> fine but when I apply the RAMSEY resettest(), the assumption of linearity
> is
> rejected. My model does not show any signs of multicollinearity and all
> independent variables have strong linear relationship with the dependent
> variable. Therefore, I am wondering how well the RAMSEY resettest works in
> general. What other tests exist?
>
>
>
> Again, I would like to apologize for the asking no specific R-technical
> questions. But I would really appreciate any help.
>
>
>
> Thanks,
>
>
>
> Simon
>
>
>
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> [email protected] mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>



-- 
John C Frain
Trinity College Dublin
Dublin 2
Ireland
www.tcd.ie/Economics/staff/frainj/home.html
mailto:[EMAIL PROTECTED]
mailto:[EMAIL PROTECTED]

        [[alternative HTML version deleted]]

______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

Reply via email to