Using software like G*Power can calculate the statistical power and required sample size. But you need some similar studies to help your estimation.

The other thing for consideration is the size of population. As you have mentioned, you'll contact the experts for their opinions. How many potential respondents do you have? How representative it will be if you have 100 experts included?

The main thing that you will gain from a larger sample is increased precision, in the form of small standard errors (i.e., tighter confidence intervals). Suppose you found that 50% of a small sample agree with some statement, then the value in the population might range from 40% to 60% or even wider (i.e., low precision and a wide confidence interval). With a larger sample, the same 50% estimate might correspond to a range between 48% and 52% (i.e., higher precision and a tighter confidence interval).

If you are indeed trying to estimate descriptive statistics like percentages, there are a great many websites that will provide free, easy to easy sample size calculators.

The most important thing to consider during estimation is precision. If you want to be very precise you dont need to estimate sample sizes. As usually, for many reasons it is not possible so we should accept to loose precision adding some error to our estimations.

Finally, it depends on your research question and the main outcome you are looking for. Just to add some to the nice answer of Mr Morgan, you can look at this : http://www.openepi.com/

Best

You are asking opinions so some of the issues covered in most texts about power calculations may not apply. The issue here is actually simpler: it is a question of precision and you don't really need to refer to other studies to answer that - you need to ask yourself how precise you need to be. I think most statistical power programs will focus on calculating power for detecting differences (e.g. G*Power). What you need to do is to determine the sample size that gives you a 95% CI that is 'acceptable'. You are surveying opinion so you'll likely need to determine a % of the population that endorses a particular view with a CI that is (say) +/- 5%. https://www.surveysystem.com/sscalc.htm will do the calculations for you.

If there is some measured parameter to be estimated you will need to have some prior information to estimate the SD which would allow you to detemine n associated with a given width of CI (95% CI is =/- 1.96 SE and SE is SD/root n) [in vv simple terms]

Perhaps MORE important though is your sampling strategy. The precision of your estimate is of no importance if you are not providing an unbiased estimate of the population - this is entirely about sampling.

I'd be inclined to agree with the sentiments expressed by all four of the previous posters. What matters is the extent of uncertainty that you're prepared to tolerate.

It's helpful to see how previous investigators have approached the problem, the shortcomings that they have identified, so that you can (hopefully) avoid making the same errors/omissions.

At first sight, the larger the sample, the better. However, what is most important is the size of the pool of experts on which you can draw upon in the first place, but the larger the sample, the greater the expense, and more effort, but you have to allow for non-responders, etc. What is most important is that your sample is representative.

Bw

Chris