What is appropriate to get a sample size for a feasibility study?
The researcher knows that they will able to recruit from 3 clinics and they want to test how many patients will be willing to fill out the survey. They are thinking that if 80% do that would be an acceptable amount.
Is this enough information to calculate the necessary sample size from each clinic?
Is this enough to show that a survey is "feasible"?
Size counts, and nobody would normally argue with 80% (see Yehuda Baruch in Human relations on response rates and samples). But the real point is not simply the sample size, but what it is that the researchers want to measure. if for example the 20% not responding were those showing the characteristics you wanted to measure you would have a problem. You would also want a non-response survey (see Baruch again) to indicate whether that was likely or not.
Good luck
Richard
Hi Richard,
Good point. So how do you suggest setting up the aim? Something like "80% of the entire population should respond to the questions, and of those 10% should fit the demographic of interest"
In my experience you will be very fortunate in getting an 80% response rate. In my experience response rates in such work vary from between 30-70% unless there is an incentive, which raises other issues.
Your questions do beg other questions. The critical issues are - what do you want to find out? Do you want to measure change? What are the concerns about response bias as % response rates are not the only issue here. Some questions can be answered with a 50% response rate from 30 respondents. Others require 1000s of respondents. Generally speaking we can measure Some change in the health care system with a 50% response rate on a sample of 600 per hospital but this is insensitive and is part of a national survey with a six figure sample size!
I agree, but in this case is is a clinic and everyone is asked these questions. So the only non-response would be people who refuse to answer.
It's a feasibility trial, so the main thing is to determine if adding these questions to the pre-existing infrastructure is do-able and will be people agree to answer.
Perhaps the best approach is to use a fixed N and determine the confidence interval around the estimate of a proportion?
OK, I see. Yes I agree with your solution. You can pretty much determine this from a table in a power analysis book.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size