[Theforum] quick survey feedback

Luther, Ron Ron.Luther at compaq.com
Mon Dec 3 16:07:30 CST 2001


Hi Aardvark,


To be an 'anal techie stat guy' for a minute, I'm not sure
'self-selected' surveys like this one count as "random samples".  So the
question of statistical significance [strictly speaking] isn't 100%
applicable.

The true answer is "it depends" ..... on what you are trying to do and
at what level of significance.  [I've had studies where I needed "30" to
be significant ... and I had one (incredibly stupid) study where I
needed a sample of "250,000" to hit the level of significance the
client, (AT&T), required. {Yes - we went ahead and actually did that
study - not one of my finer technical moments - but the client was
happy.}]




Now that that's out of my system .....  I think we're using this for
'rough directional purposes' rather than generating 'hard' numbers .....
sooooooooo

.... most stat folks are willing to work with any dataset > 30  ... 'cuz
that's 'pretty much' a good ballpark for when the law of large numbers
and normal approximations kick in ...


HTH,

RonL.

-----Original Message-----
From: aardvark [mailto:roselli at earthlink.net]

anyone know how many we need to be statistically significant?





More information about the theforum mailing list