On Thu, 07 Dec 2006 07:21:34 -0800, SMS <scharf.steven@geemail.com>
wrote in <45783179$0$82525$742ec2ed@news.sonic.net>:

>Thomas T. Veldhouse wrote:
>> In alt.cellular.sprintpcs SMS <scharf.steven@geemail.com> wrote:
>> <snip>
>>> 220,000,000) is used. If the 43,000 responses were divided by 20 cities
>>> equally, this would be 2000 responses per city, the margin of error
>>> would be less than 2%. No one can argue with these results.
>>>
>>
>> I am sure John Navas can argue with them ... in fact, I am sure he already
>> has.
>
>Many people don't understand, or do understand but lie about, the
>margins of error in surveys like this.

Either you don't, or you're being deliberately misleading.

>They'll claim that 43,000 surveys
>out of 200,000,000 qualified respondents is not valid, because "gee,
>what if those 43,000 respondents happen to be all biased one way or
>another.

The problem is self-selected sample from a non-representative universe,
a problem that size cannot overcome.

>It doesn't work that way of course. You can get a very low
>margin of error with a relatively small sample.

Actually not.

>In reality the CR sample
>was extremely large, and even when you break it down by city, as they
>did, the sample was still large.

But still suffering from the above defect.

--
Best regards, FAQ FOR CINGULAR WIRELESS:
John Navas <http://en.wikibooks.org/wiki/Cingular_Wireless_FAQ>