Statistics question: What % must I survey to ensure validity?

Factual Questions
1

I have an urgent need for information, and I hope the Teeming Millions can help.

I need to determine the opinions of an organization of about 18,000 people. I want my results to be valid, i.e., be a relatively reliable representation of the attitudes of the group. How many people (out of 18,000) must I send my survey to, to ensure a valid response? Is there a flat percentage that is generally accepted?

2

Sample size calculator. It isn’t a flat percentage.

And be careful mailing surveys. A lot of people won’t return them, and the people that do won’t necessarily be an accurate sample.

3

I think if you get a 5-10% response rate you’ll be doing ok for a population that size.

However, if it’s a self administered questionaire you’ll have to send to a much larger sample of the group to get that kind of response.

4

If your purpose is to get a reasonably accurate percentage of the number of people in your population believing statement “X” or holds view “y”, mailing the survey out is a waste of time.

All you find out is the percentage of people who RETURNED the survey held such a position or view. No statements can be accurately made of the population as a whole, unless you KNOW the propensity of each type of viewholder in returning the survey.

5

Just for the record, things like margin of error and all that depend only on the sample size, not the sample size as a percentage of the total population.

6

A big THANK YOU to all of you. :slight_smile:

7

Not true. If my population of interest is my immediate family, and I survey four people, I’ve got zero margin of error.

8

OK, fine, but we don’t usually deal with small populations like that.

9

I might add, Marrs could send out a few hundred surveys, making sure each of the 18,000 had an equal chance of being included in the sample and keep track of the responses and who didn’t respond at all. Then contact virtually all those who didn’t.

He couldn’t use mathematical or statistical inference, as people might answer differently by mail or by phone call, but he possibly could get a sufficient level of knowledge for his purpose.