Hi,
While reading detail about various procedure for calculating sample size I got various problem while dealing with real situation.
First thing on what basis we should calculate sample size either we look after the study design (cross sectional, case control or cohort) or sampling techniques involved in the study ?
Also there are formula for calculating sample size for every test parametric and non-parametric , also for regression methods like logistic or other .
So on what basis we should calculate the adequate sample size ?
Guide me. Provide the materials also if possible.
I have read WHO book on sample size also. But my doubts is still unexplained.
The sample size is first dependent on the degree of closeness you require to the 'truth'. Google 'standard error' or get a statistics textbook that explains.
Hi Sir,
I am from statistics background only , my graduation subject is statistics and Masters specialization is in Biostatistics.
I am very well aware of standard error.
.
Before putting question in RG , I have gone through various books.
Theoretical calculations of sample size is easy, but in practical situation what to consider is still my doubts.
Thanks
This is based on my experience in clinical trials.
To us, sample size calculations are a guide. We never know what is the right or wrong number to recruit in reality because we depend on a lot of pilot studies to get an estimate of the effect sizes which could potentially be wrong due to selection bias. The number also depends on the amount of funding the investigator has got. Practicality is important.
We usually based our sample size calculation on the primary outcome measure. N will be calculated differently depending on the nature of the outcome (e.g continuous, binary, repeated measures etc). The UCLA web (link attached below) is a good place to get started
https://stats.idre.ucla.edu/other/dae/
Hope this helps.
You could start by quantitatively defining adequate. I want to be certain that the probability of a genetic mutation in my corn plants does not move into "natural" populations. I accept that anything less than 5% probability equates to certainty.
Hi, Ravi
Very good question . The question can be answered in two ways. One for those who are are students of Statistics and the in other way for those who are non-Statistician. As you are a Statistician, remember the two papers which you have studied in MSc namely, Sampling Theory and Design of Experiment. Now first identify your sample belongs to which one. Whether, it is a random sample or a randomized sample. Is this not be kept in mind first.
As you mention, a formal power analysis will require different calculations of sample size for every different kind of parameter you plan to estimate (means, proportions, slopes, etc.), and for each of these, you will need to have some idea of the effect size you want to be able to detect (and generally, also some idea of the relevant variance).
As a practical matter, it might be sufficient for your purposes to simply remember that the margin of error (for 95% confidence) in estimating a proportion from a sample, when the true population proportion is close to 0.5, is approximately equal to the reciprocal of the square root of the sample size [MOE=1.96xSE ~= 2x(.5)/sqrt(n)]. For example, a sample size of 400 will yield a 95% c.i. of about plus or minus 5 percentage points around the sample percent on a yes/no variable that is split 50/50. A sample of 100 yields an MOE of about 10 percentage points; a sample of 1600 yields and MOE of about 2.5 percentage points. The true but unknown split does not have to be all that close to 50/50; 60/40 or 40/60 is close enough.
With this in mind, all you need to do is decide how small you need the MOE to be to draw meaningful conclusions about proportions for purposes of your particular study. 10 percentage points may be small enough for exploratory research, but if the conclusions will lead to life-or-death decisions, 2.5 percentage points may be too large. If the MOE for a proportion is small enough for purposes of your study, then the MOE for estimates of other parameters (though different), will likely be "small enough" as well.
My comments assume simple random sampling in a survey-type research context; stratified, clustered, or other complex sample designs require more complex MOE calculations. And as others have suggested, randomized experimental groups raise further issues.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size