In health related studies, can the level of precision be taken as 0.07 or 0.08 while calculating sample size?
Dear Deepika,
yes. From the point of view of the statistics theory there are no obstacles. Of course, if I correctly guessed that when you used the term "precision" then you mean "alpha" (type I error) [if in your question is alpha=0.08, then 1-alpha=0.92].
The sample size formulas depends on 5 main conditions:
1. the confidence coefficient (1-alpha),
2. the estimator (e.g. the variance of variable calculated from sample in preliminary research or taken from previous research),
3. the assumed maximum error of the estimate (ex ante value),
4. the scheme and type of random sampling which going to be used,
5. the number of general population (usually, when the sampling is without replacement).
Especially the former is arbitrary. In other words, the researcher chooses an alfa based on his/her knowledge about the investigated phenomenon and other elements of the context of conducted statistical inference. I recommend the estimation theory and the sampling theory. You can find explanations there.
Customs and expectations of your scientific community towards the desirable confidence coefficient may also be important, but these goes beyond statistics and methodology.
Best,
Slawek
Please check the following:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3895825/
https://www.yumpu.com/en/document/view/44454552/sample-size-determination-in-health-studies-tuberculosis
Article Sample Size Estimation in Diagnostic Test Studies of Biomedi...
Hello Depika,
I wasn't quite clear as to whether you are referring to:
1. Risk of type I error (as in hypothesis testing),
2. Risk of a C.I. completely missing the population target value, or
3. Maximum discrepancy between sample value and population value you'd be willing to accept in a given parameter estimate.
In any case, the answer is, "yes," you can set the value at whatever level makes sense for your research question, the variables involved, and the associated consequences of being wrong in your result.
Be aware, however, that especially in case (1) above, textbooks and many journal articles would have you believe that any values other than .10, .05, or .01 are simply not to be used. That's not true, of course. The values for Type I and Type II error risk should be chosen specifically to match the costs/consequences of a wrong decision regarding a hypothesis test.
The tongue in cheek quote from Rosnow and Rosenthal (1989 and elsewhere), "Surely God loves the .06 (blob) nearly as much as the .05" (https://micahallen.org/2012/02/06/surely-god-loves-the-06-blob-nearly-as-much-as-the-05/) challenges the lockstep convention.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size