Re: lognormal parameters for Kol-Smir test

[heckert@nist.gov]

Quoting Ryan Fitzgerald <ryan.fitzgerald@nist.gov>:

Ryan,

Enter 

   LET SIGMA = <value>

A suggestion on the lognormal plot.

  1) Typically for the lognormal, the

       LOGNORMAL KS PLOT Y

     will give a better estimate than

       LOGNORMAL PPCC PLOT

  2) Entering the following command before the
     LOGNORMAL KS PLOT command can sometimes
     improve the estimate even more:

        SET PPCC PLOT LOCA SCALE BIWEIGHT

     This uses the biweight fitted line on the probability
     plot to estimate location and scale rather than
     the regular least squares line.  The basic effect
     is to minimize the effect of extreme points.  So the
     larger the value of SIGMA (i.e., the more heavy-tailed
     the data are), the more this is likely to help.

     Since the PPCC plot is invariant to scale, this command
     has no effect for the PPCC plot.  To reset this command,
     enter

        SET PPCC PLOT LOCA SCALE DEFAULT


Alan
     
     

> 
> Hi.
> 
> I am using the kolmogorov-smirnov goodness of fit test to see if my 
> data are lognormal. I first get parameter estimates from maximum 
> likelihood (2 parameter lognormal) or PPCC (3 parameter 
> lognormal).  The question is, how do I specify the shape parameter 
> for kol-smir?  According to the Dataplot help page, the parameter is 
> called "SD".  But I find that the kol-smir test statistic is the same 
> regardless of what I put for SD. It is only sensitive to scale and 
> location (ksloc and ksscale respectively).
> 
> When I use efits, I get a smaller test value, so presumably it is 
> doing something I'm not.  I've tried using pieces of the efits code, 
> but haven't fixed the problem. I appreciate any help you can offer.
> 
> 
> 
>