Re: Get rid of Tables

From: scdsc (scdsc_at_wintec.ac.nz)
Date: 02/08/05 

Date: Tue, 8 Feb 2005 20:43:27 +0000 (UTC)

I've often wondered how many real life data sets are, in fact,
sufficiently normal (or whatever the model assumes they are) to
justify more than two or perhaps three decimal places at the most when
calculating probabilities. Some folk with a good computer but who
feel a little uncomfortable with statistics get an unwarranted feeling
of security by having lots of decimal places.

Someone's law says "Statistics impress as to the square of the number
of significant digits"

Large_Nassau_Grouper@Yahoo.com wrote:
>
>"Luis A. Afonso" wrote:
>> Get rid of Tables (a quite presumptuous title)
>>
>Not really.
>
>> Do use a program that evaluate de Cumulative Normal Distribution
>
>There are many of those around.
>
>
>> Get exact 7 decimal places values:
>
>Don't you mean the approximate value correct to 7 decimal places?
>
>>
>> Provides de Integrals: Example: z=1.2345678
>
>But when would you ever have the need to have that many decimals
>in the value of z?
>
>>
>> from 0 to z 0.3915043
>> from -z to z 0.7830086
>> < = z 0.8915043
>
>or, more accurately, .8915043004722... and I have a program on
>my PC, named ET (Electronic Tables) that can calculate the tail
>areas of most of the commonly used probability distributions
>(normal, Student t, F chi-square, exponential, Weibull, lognormal,
>noncenral chi-sq, gamma, beta, ...) up to 100 places of decimal
>accuracy. But who needs it???
>
>-- Bob.