Re: margin of error of mean from 3-valued (0,100,200) mutiple-choice survey question

On Jan 4, 10:45 pm, jgsurv...@xxxxxxxxx wrote:
Since you only wanted a rough estimate, I'd go with

MoE = 2*69/sqrt(1000) and SigDiff = 2*69*sqrt(1/1000 + 1/1000)

This is great-- exactly what I was looking for.

For SigDiff, is it 2*69*sqrt(1/1000 + 1/1000) or 2*69*sqrt(1/1000 +
1/500), given that one of my samples is only 500 responses?

The latter. I actually oversimplified a bit. If you have samples of
size n1 and n2 with sample variances var1 and var2, the sigDiff is
approximately 2*sqrt(var1/n1 + var2/n2). (Note that this formula uses
the variances, not the standard deviations.)

Finally, is there a good place online where you'd recommend a beginner
like me could learn a little more about the math involved in computing
the formulas above?  aka "margin of error for dummies"?  This is both
to satisfy my own curiosity as well as being able to intelligently
answer questions from my users if I'm asked.

This I am not sure of. I would start with wikipedia. You should look
for "comparing two means" or something like that. Maybe some of the
regulars here follow this exchange and can jump in.

The more important question is: Is the mean a
representative measure to describe the phenomenon
you are interested in? If the answer to that question
is "Yes", then you can use the standard deviation as
well.

Yes, I think so.  Or at least I think think the mean is
"representative enough" given the somewhat arbitrary definition of
this index in the first place.

Thanks so much for your help Gus, you've helped me alot here.

.

Relevant Pages
Quantcast