Click here to download free Microsoft Visual FoxPro tools!
| Navigational map -- for text only please go to the bottom of the page ||Forums|
Readers' Choice Awards (siteadm) Fri, 13 Jun
Wow!!! 300 hundred people!! (dhmjr) Sat, 14 Jun
Statistical Validity (sdugan) Sun, 15 Jun

Nope

Posted by: CRConrad
Date posted: Sun, 15 Jun 1997

Sean Dugan, Research Editor, writes:

> [ . . . ] yields an error of 5.7% in a worst case scenario (a score of 50%).


On a _yes-no_ question, where you ask all 300 respondents the _same_ question, that is correct, yes.

But in this case, with an unlimited (?) number of alternative answers, I think you'd have to regard _each_product_ as such a yes-no question.
("Is Iomega Jaz the best product in this or that category? Is Microsoft Windows NT the best product in this or that category?" etc, etc . . . )

And then the sample size is basically, since your questions weren't put that way, only those who have selected one particular product.
But I'm not quite sure about this; I'll have to check it up in my old textbooks and get back to you. Oh, by the way: One thing I _am_
half-sure of, is that population size doesn't really matter, once your sample is in the hundreds, and population in the tens of thousands.

The main issue is of course that you have made what is basically a _usage_ _survey_; emphasis on both words,
since it's not a survey of what people would _choose_, but what they _use_ -- how many alternatives had the respondents even _tried_?
Nor is it an expression of their _choice_, in the way that a self-selecting poll is. A survey is the best for finding tangible facts, of course,
but to see what _choice_ people make in any reasonable meaning of the word "choice", the biases introduced by a self-selecting sample
_are_ in fact a large part of precisely what you are (or _should_ be, if you honestly wanted a *Readers'* *Choice*) after!



Sincerely,


Christian R. Conrad


Respond | Search | Help


For questions or comments, email the Forums Editor

kristin_kueter@infoworld.com