[Tutor] FYI: An Introduction to Interactive Programming in Python

[Steven D'Aprano]

On 10/04/13 06:04, Alan Gauld wrote:
> On 09/04/13 13:47, Steven D'Aprano wrote:
>
>>> Since when did 30 become a representative sample size?
>>
>> If they are randomly selected, 30 is likely plenty for a representative
>> sample size. In surveys, a sample size of 30 gives you a margin of error
>> of about 15%, which isn't too bad.
>
> Hmm, if I'd gone to my manager (and especially the finance director) with a project proposal based on assumptions with a 15% margin they'd have thrown me out.

The 15% figure comes from surveys where people are asked to rate something on some scale, say from 1 to 7. If your sample of 30 gives an average rating of 5, then a 15% margin of error corresponds to plus or minus 1: the true average is likely to be 4, 5 or 6. Given that the question being asking is fundamentally subjective, and you have no idea whether Fred's rating of 5 and Wilma's rating of 5 mean the same thing, I think that plus or minus 1 is pretty much as good as you can reasonably expect. Additional samples just adds spurious precision that looks good but doesn't *mean* anything.

But what do I know? :)



> My suspicion is they just asked around their local department!
> But maybe I'm just overly suspicious...

Nah, I reckon you've hit the nail right on the head.



-- 
Steven