http://qs321.pair.com?node_id=612779

Hi folks,

Following on from my question the other day (Stats: Testing whether data is normally (Gaussian) distributed), here's a simple normality test.

This is the Jarque-Bera test, "a goodness-of-fit measure of departure from normality, based on the sample kurtosis and skewness".

Kurtosis and skewness are calculated like this.

Comments very welcome, if I've gone wrong here please do point it out!

Input $source should be a 1-D PDL, type float (it'll probably work with other types, but be less useful!).

Output $JB is a number indicating nearness to the Normal distribution. High number = not Normal.

Warning: this can produce NaN for data where standard deviation =0, so if you plan to sort it afterwards, you'll need to weed out the NaN ones (as I discovered for myself {'sort subroutine edge cases'}).

I hope this is useful to someone.

Best wishes, andye

        use PDL;

        my ($mean,$std_dev,$median,$min,$max,$adev,$rms) = stats($sour
+ce);
        
        my $skewness = sclr(sum(   
        
                         ($source - $mean)**3
        
                            ) / (
    
                         (nelem($source)-1) * $std_dev**3 
                        
                       ));
        
        my $exs_kurtosis = sclr(sum(   
        
                         ($source - $mean)**4
        
                            ) / (
    
                         (nelem($source)-1) * $std_dev**4 
                        
                       ) -3);
        
        my $JB    = 
                    ( nelem($source) / 6 )
                    *
                    (
                        $skewness**2
                        + ($exs_kurtosis**2 / 4)
                    )
                  ;

[download]

Back to Cool Uses for Perl