Here is the case: comparing systolic blood pressure (in mmHg) in a trial of new drug and placebo.
Now I know that the formula of minimum sample size for this case is:
2.((Za + Zb).SD/d)2
Red: 2 times the square of ((Za+Zb).SD/d).
Za and Zb is constant for type one error and type two error.
SD is a standard deviation of the mean (found in literature)
d is the clinically important difference of the mean
What I'm trying to ask is: what if we cannot find the standard deviation from any literature? There is no previous research conducted about this topic. What number should we put there in the SD?
It is not uncommon that the SD is not known for a variable with respect to a particular problem. In any case, the SD value you should use is an estimate of the parametric standard deviation (i.e., chances are very good that the parameter is unknown). Your response variable has been investigated in many populations so you could calculate an "average" SD from such studies or estimate via resampling.
In as much as I agreed with the procedure by David about on the use of parametric standard deviation base on re-sampling, care should be taken about the population in question since systolic is much dependent on genetic make up of the population in question (African-American potentially might have high systolic) a re-sampling from a pool of database of population can give sporious standard deviation
Dewi -
Might you try a small pilot study?
Cheers - Jim
PS - Also, from other sources you might determine a range of possibilities and use that to see how sensitive your results are to that range.
To answer your question; there are 3 basic ways of obtaining the standard deviation
1. estimate from the previous studies
2. conduct a pilot study and calculate
3. Use best guess = data range (High-low) divided by 4
You can choose the one which is possible
You might as well refer to stats 101 sample size requirements on youtube
https://www.youtube.com/watch?v=x_H-dr7s1zU&t=62s
Happy learning
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size