0
$\begingroup$

My interest is to perform a statistically significant survey on a population of 1700 people, that can be described in different categories, so each person belongs to only one category.
I have two categories of interest: Cat1 is comprised of lets say 700 people, and Cat2 is comprised of 200 people.
My survey should be done by enough people so each one of these two groups of people are significantly represented, with a confidence level of 95% and a margin of error of 5%, although these values can be changed.

I know that the more margin of error I have, the smaller the sample. Up to which value could it be justified?

Also, what difference does it make to the sample size to consider the population finite or infinite?

$\endgroup$

1 Answer 1

1
$\begingroup$

Welcome to CV. There are a number of confused statements in you question.

First: Your aim is not to "perform a statistically significant survey" as that isn't really a meaningful phrase. Your aim should be to test a hypothesis, or answer a research question, or estimate a parameter, or something like that.

Second, your whole second paragraph is problematic. It is unclear what it would mean for a group to be "significantly represented". You may mean that you want the mean (or median, or something) to be estimated to a certain precision.

Third, your third paragraph isn't wrong, but it's sort of backward. The usual way of looking at this would be to say "the smaller the sample, the higher the margin of error". As for how much can be justified, that's really up to you -- it is a substantive question, not a statistical one, and varies by field. For some cases (e.g. the size of a ball bearing) we may want a very, very precise estimate. For others, e.g. the amount of milk in a liter package, much less precision is needed.

The size of the population makes relatively little difference unless the sample size is a large proportion of the population. That may be the case here, as the population is pretty small. You can look up the "finite population correction".

Finally, to do a power analysis you need all but one of the following:

  • Effect size you want to be able to detect and the test you will use to detect it
  • Acceptable type 1 error
  • Desired power
  • Sample size

Then, for many tests, you can use a canned program such as GPower or PASS or the power functions in R, and so on, but for others, you many need to simulated data.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.