Re: Linux Market share?

From: Larry Qualig (removethispartlqualig_at_uku.co.uk)
Date: 07/13/05 

Date: Tue, 12 Jul 2005 20:54:51 -0400

"TokaMundo" <TokaMundo@weedizgood.org> wrote in message
news:9nn8d19jiajm0ksl237gkedaktjj3lvob2@4ax.com...
> On 12 Jul 2005 07:22:43 -0700, lqualig@uku.co.uk Gave us:
>
>>For the nth time nobody is going to calculate marketshare by looking at
>>sales and/or download numbers. All you need to do is to survey 10,000
>>random computer users and you have a valid statistical sampling. No
>>need to count or guess how many live CD's somebody copied.
>
>
> Not true. A survey of 10k Boeing workers will yield a different
> result than 10k average consumers ...

Perhaps this is why I said to survey *random* computer users. For the
obvious reason that you'll get skewed results if you surveyed employees of
Sun Micro, Redhat, Msft or Boeing.

> So a statistical sampling among
> average consumers is not a valid one. A statistical sampling among
> college CS or engineering students, however, would be.

If you're looking for marketshare then it needs to be random. Once you start
hand selecting "college CS students" the results become invalid. A large
enough random sampling would likely include a fair number of college
students, lawyers, government workers and garbage men. If you're out to find
what OS people in the world are using you need to survey all cross-sections
of the general population.

Assuming there are 500 million computer users on this planet and you want
the survey to be accurate to within 1%.

You would need to survey 6806 users to get a statistical result that has a
90% confidence level of being accurate to within 1%. To be 95% confident
that the results are accurate you'd need to survey 9604 users. To have a 99%
confidence level in the survey results you would need to survey 16,589
users. The number I picked (10,000) will give you a 99% confidence level
that the survey is accurate to within 1.2%.

Note: Do *not* confuse a 1.2% accuracy rate with Linux having an installed
user base of x% +/- 1.2%. The way to read this is that whatever "x%" comes
out to... that number will be accurate to within 1.2%.

Relevant Pages

  • Re: Linux Market share?
    ... >> Assuming there are 500 million computer users on this planet and you want ... >> the survey to be accurate to within 1%. ... >> confidence level in the survey results you would need to survey 16,589 ... > indicates some training in statistics, so surely you also know that the ...
    (alt.os.linux.suse)
  • Re: Linux Market share?
    ... > Assuming there are 500 million computer users on this planet and you want ... > the survey to be accurate to within 1%. ... > confidence level in the survey results you would need to survey 16,589 ... I'm curious why you state an assumption of 500 million computer users? ...
    (alt.os.linux.suse)
  • Re: Sample size - How big should it be?
    ... A sample based on a random number generator I suspect would produce a better sample. ... You can try and predict what it the standard deviation going to be but you really have to wait till the end of the survey to work it out for sure. ... Tell us more about what you're trying to measure, and what significance level or confidence level you're shooting for. ...
    (sci.stat.edu)
  • Re: Anyone good at statistics...??
    ... Just wondering about all those surveys that ask hundreds or thousands of people what they think of Darwinism, Creationism, etc: I'm told that when the figures are extrapolated to produce an estimate for the entire population, there's some kind of 'confidence level' that can be calculated from the size of the sample. ... Hoe does the 'three-ness' of your survey show up in your final results? ... Margin of Error and Confidence Levels Made Simple ... For sample sizes way below that, the usual assumption of normality no longer holds. ...
    (talk.origins)
  • Re: Canon Reliability
    ... the statistical sampling math isn't all that hard yet Navas always gets his panties in a bunch when the survey results for a particular product or service don't align with what he's purchased. ... Yet the survey isn't asking subscribers "which is the best carrier" it's simply asking them a series of questions of their experiences with their own carrier. ...
    (rec.photo.digital)