Survey Collection - Sample Selection Problem

[kruupy]

Hi All,

Thank you all for your help with this issue, I greatly appreciate it.

I have an industry population (number of businesses) of 6000. This industry population can be broken down in 3 different states. State X owns 3000 of the population, State Y owns 2000 and Z owns 1000.

Now, if I want to conduct a survey and draw conclusions on this population, statistical calculators seem to indicate that I need to survey 362 businesses (95% confidence level, 5% margin of error). These have to be randomly drawn from the population right? - or is there a required distribution of the 362 sample from each state?

Further more, is there a way that I can gather data to make conclusions about each state aswell as at a national level (without blowing out the required sample size for each state)?

Also, does the size of the business matter to the data? i.e if I sample someone who owns half the industry, should this person have a higher weight?

How is the sample size effected if I used industry value weights as opposed to one business, one vote methodology?

Thanks Again.

[trubador]

First of all, I would do a power analysis to determine the sample size. You can find plenty of resources if you are interested.

Let's say you have decided to sample 360 observations from the population. Then it would be fine to breakdown the sample into three strata (i.e. states) in proportion to their size and do the random sampling for each. Which would give you a sample of 180, 120 and 60 observations for each states. Now you can make inferences either at the national or at the state level.

It really depends on your study and your research question, but size of the business could matter. If there is such a dominant firm in the industry, then you may want to make sure that your sample includes that observation. So yes, you may need to apply a postsampling weighting scheme in order to adjust the responses for better representativeness.

Or, you can also define an additional categorical variable that groups businesses according to their size (e.g. small, medium, large). And try to balance your draws based on this indicator as well (i.e. not to over/under sample any category).

After all, the whole point of sampling is to represent the population as close as possible.