Re: st: problem with `r(min)' and `r(mean)' macros

[Nick Cox <njcoxstata@gmail.com>]

I don't see any reason for r(min) and r(max), or their local macro
personas, to behave differently here. You were lucky in one case, but
not the other. That's my guess.

On Mon, Jan 28, 2013 at 3:14 PM, mario fiorini <mariofiorini73@gmail.com> wrote:

> thanks Nick,
> I could sense it was a precision issue, but I was very confused
> because it was only there for `r(min)' and not `r(max)'. [and sorry
> for the obviously stupid `r(mean)' check].
> now I understand both precision issues.
>
> Mario
>
>
> On 28 January 2013 16:01, Nick Cox <njcoxstata@gmail.com> wrote:
>> This is a precision issue, biting you twice over.
>>
>> First off, there is a loss of precision possible with using e.g.
>> `r(min)' rather than r(min) which is roughly parallel to the lower
>> precision of macros compared with scalars.
>>
>> !!! Don't use local macros if you seek maximum precision with non-integers. !!!
>>
>> Second, testing for exact equality can be tricky at the best of times
>> for reasons often rehearsed on this list and elsewhere.
>>
>> . search precision
>>
>> in an up-to-date Stata and start with Bill Gould's blog postings.
>>
>> Try this instead:
>>
>> set obs 1000
>> gen var1 = runiform()
>> su var1
>> scalar max = r(max)
>> scalar min = r(min)
>> count if var1 == max
>> count if var1 == min
>>
>> Whether the mean of random numbers on the unit interval is equal to
>> any particular random number is surely of less interest.
>>
>> Nick
>>
>>
>> On Mon, Jan 28, 2013 at 2:47 PM, mario fiorini <mariofiorini73@gmail.com> wrote:
>>> Dear statalist,
>>> I am summarising a variable and then trying to use the `r(.)' macros
>>> However, in some instances Stata does not seem to find any variables
>>> that take the `r(min)' and `r(mean)' values. For some reason, it does
>>> work with `r(max)' instead.
>>> The code below should clarify the problem. When I run it, I always get
>>> that no observation takes the min or the mean (why???), but there is
>>> an observation that takes the max.
>>> The problem does not depend on whether I use the original `r(min)' or
>>> a macro=`r(min)'.
>>> If it matters I am using a mac osx 10.8 with Stata 12.1 MP.
>>> Any hint would be appreciated because it is driving me crazy.
>>> Mario
>>>
>>>  // Start
>>> clear
>>> set obs 1000
>>> ge id = _n
>>> ge var1 = uniform()
>>> su var1
>>> local LCmin =`r(min)'
>>> local LCmean =`r(mean)'
>>> local LCmax =`r(max)'
>>>
>>> count if var1==`LCmin'
>>> count if var1==`LCmean'
>>> count if var1==`LCmax'
>>>  // Ends
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/