 |
 |
The design of a sampling scheme requires knowledge of the particulars of the situation. Simple answers can be misleading.. It is common in survey sample design to assume the worst case scenario for a dichotomous variable, 50%. When there are many variables it is a useful starting point unless you KNOW that the worst case is some other percentage point. Do you anticipate extremely large or small percentages? Seldom is the most effective sample design a simple random sample. What measurement processes do you anticipate using? mail? phone? in person? Existing records? Why would people respond to you? Legal obligation like a census? Employees or program beneficiaries? Out of the goodness of their hearts? Do you wish to compare and contrast different subgroups? genders? regions? departments? What kind of margin of error (95% confidence interval) do you wish to achieve? Do you want to make conclusions about single variables, or about the relations of variables? Do you have a list of the pop members or will you have to do field listings? Are respondents clustered: families, classrooms, military units, etc.? Are there auxiliary variables that you know pop values for? Is the marginal cost of additional cases very large? I.e., pretesting, instrument development, printing setup, data prep, analysis, report writing, etc. don't vary much with the size of the sample. Phone charges, interviewer time, data entry do vary with the number of cases. Art [log in to unmask] Social Research Consultants University Park, MD USA (301) 864-5570 Are there subgroups Louise Swainston wrote: >Hi Allstat, >This may sound like a very simple question or it may not. > >If I have a population of 2.3 million people and I want to draw some >conclusions about them, what size sample do I need to take? I know nothing >about the population I am sampling from, so I don't know what's >representative, is there some rule of thumb about sample sizes? I am likely >to be looking at a number of different things in the population of which I >know nothing, so I can't just assume that 50% do something and calculate >sample sizes that way can I? > >Regards >Louise > > > >